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);
23 dir_glob_foreach file_get_contents file_set_contents lock_file
24 lock_file_full AT_FDCWD O_PATH $IPV4RE $IPV6RE
28 use PVE
::AccessControl
;
30 use PVE
::Syscall
qw(:fsmount);
32 use PVE
::GuestHelpers
qw(safe_string_ne safe_num_ne safe_boolean_ne);
35 use Time
::HiRes qw
(gettimeofday
);
38 require PVE
::Network
::SDN
::Zones
;
42 my $LXC_CONFIG_PATH = '/usr/share/lxc/config';
44 my $nodename = PVE
::INotify
::nodename
();
46 my $cpuinfo= PVE
::ProcFSTools
::read_cpuinfo
();
49 my $vmlist = PVE
::Cluster
::get_vmlist
();
51 return $res if !$vmlist || !$vmlist->{ids
};
52 my $ids = $vmlist->{ids
};
54 foreach my $vmid (keys %$ids) {
55 next if !$vmid; # skip CT0
56 my $d = $ids->{$vmid};
57 next if !$d->{node
} || $d->{node
} ne $nodename;
58 next if !$d->{type
} || $d->{type
} ne 'lxc';
59 $res->{$vmid} = { type
=> 'lxc', vmid
=> $vmid };
64 # container status helpers
66 sub list_active_containers
{
68 my $filename = "/proc/net/unix";
70 # similar test is used by lcxcontainers.c: list_active_containers
73 my $fh = IO
::File-
>new ($filename, "r");
76 while (defined(my $line = <$fh>)) {
77 if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) {
79 if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) {
90 # warning: this is slow
94 my $active_hash = list_active_containers
();
96 return 1 if defined($active_hash->{$vmid});
101 sub get_container_disk_usage
{
102 my ($vmid, $pid) = @_;
104 return PVE
::Tools
::df
("/proc/$pid/root/", 1);
107 my $last_proc_vmid_stat;
109 my $parse_cpuacct_stat = sub {
110 my ($vmid, $unprivileged) = @_;
112 my $raw = read_cgroup_value
('cpuacct', $vmid, $unprivileged, 'cpuacct.stat', 1);
116 if ($raw =~ m/^user (\d+)\nsystem (\d+)\n/) {
126 our $vmstatus_return_properties = {
127 vmid
=> get_standard_option
('pve-vmid'),
129 description
=> "LXC Container status.",
131 enum
=> ['stopped', 'running'],
134 description
=> "Maximum memory in bytes.",
140 description
=> "Maximum SWAP memory in bytes.",
146 description
=> "Root disk size in bytes.",
152 description
=> "Container name.",
157 description
=> "Uptime.",
160 renderer
=> 'duration',
163 description
=> "Maximum usable CPUs.",
168 description
=> "The current config lock, if any.",
173 description
=> "The current configured tags, if any.",
182 my $list = $opt_vmid ?
{ $opt_vmid => { type
=> 'lxc', vmid
=> $opt_vmid }} : config_list
();
184 my $active_hash = list_active_containers
();
186 my $cpucount = $cpuinfo->{cpus
} || 1;
188 my $cdtime = gettimeofday
;
190 my $uptime = (PVE
::ProcFSTools
::read_proc_uptime
(1))[0];
191 my $clock_ticks = POSIX
::sysconf
(&POSIX
::_SC_CLK_TCK
);
193 my $unprivileged = {};
195 foreach my $vmid (keys %$list) {
196 my $d = $list->{$vmid};
198 eval { $d->{pid
} = find_lxc_pid
($vmid) if defined($active_hash->{$vmid}); };
199 warn $@ if $@; # ignore errors (consider them stopped)
201 $d->{status
} = $active_hash->{$vmid} ?
'running' : 'stopped';
203 my $cfspath = PVE
::LXC
::Config-
>cfs_config_path($vmid);
204 my $conf = PVE
::Cluster
::cfs_read_file
($cfspath) || {};
206 $unprivileged->{$vmid} = $conf->{unprivileged
};
208 $d->{name
} = $conf->{'hostname'} || "CT$vmid";
209 $d->{name
} =~ s/[\s]//g;
211 $d->{cpus
} = $conf->{cores
} || $conf->{cpulimit
};
212 $d->{cpus
} = $cpucount if !$d->{cpus
};
214 $d->{lock} = $conf->{lock} || '';
215 $d->{tags
} = $conf->{tags
} if defined($conf->{tags
});
218 my $res = get_container_disk_usage
($vmid, $d->{pid
});
219 $d->{disk
} = $res->{used
};
220 $d->{maxdisk
} = $res->{total
};
223 # use 4GB by default ??
224 if (my $rootfs = $conf->{rootfs
}) {
225 my $rootinfo = PVE
::LXC
::Config-
>parse_ct_rootfs($rootfs);
226 $d->{maxdisk
} = $rootinfo->{size
} || (4*1024*1024*1024);
228 $d->{maxdisk
} = 4*1024*1024*1024;
234 $d->{maxmem
} = ($conf->{memory
}||512)*1024*1024;
235 $d->{maxswap
} = ($conf->{swap
}//0)*1024*1024;
246 $d->{template
} = PVE
::LXC
::Config-
>is_template($conf);
247 $d->{lock} = $conf->{lock} if $conf->{lock};
250 foreach my $vmid (keys %$list) {
251 my $d = $list->{$vmid};
254 next if !$pid; # skip stopped CTs
256 my $proc_pid_stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
257 $d->{uptime
} = int(($uptime - $proc_pid_stat->{starttime
}) / $clock_ticks); # the method lxcfs uses
259 my $unpriv = $unprivileged->{$vmid};
261 if (-d
'/sys/fs/cgroup/memory') {
262 my $memory_stat = read_cgroup_list
('memory', $vmid, $unpriv, 'memory.stat');
263 my $mem_usage_in_bytes = read_cgroup_value
('memory', $vmid, $unpriv, 'memory.usage_in_bytes');
265 $d->{mem
} = $mem_usage_in_bytes - $memory_stat->{total_cache
};
266 $d->{swap
} = read_cgroup_value
('memory', $vmid, $unpriv, 'memory.memsw.usage_in_bytes') - $mem_usage_in_bytes;
272 if (-d
'/sys/fs/cgroup/blkio') {
273 my $blkio_bytes = read_cgroup_value
('blkio', $vmid, 0, 'blkio.throttle.io_service_bytes', 1); # don't check if unpriv
274 my @bytes = split(/\n/, $blkio_bytes);
275 foreach my $byte (@bytes) {
276 if (my ($key, $value) = $byte =~ /(Read|Write)\s+(\d+)/) {
277 $d->{diskread
} += $2 if $key eq 'Read';
278 $d->{diskwrite
} += $2 if $key eq 'Write';
286 if (-d
'/sys/fs/cgroup/cpuacct') {
287 my $pstat = $parse_cpuacct_stat->($vmid, $unpriv);
289 my $used = $pstat->{utime} + $pstat->{stime
};
291 my $old = $last_proc_vmid_stat->{$vmid};
293 $last_proc_vmid_stat->{$vmid} = {
301 my $dtime = ($cdtime - $old->{time}) * $cpucount * $cpuinfo->{user_hz
};
304 my $dutime = $used - $old->{used
};
306 $d->{cpu
} = (($dutime/$dtime)* $cpucount) / $d->{cpus
};
307 $last_proc_vmid_stat->{$vmid} = {
313 $d->{cpu
} = $old->{cpu
};
320 my $netdev = PVE
::ProcFSTools
::read_proc_net_dev
();
322 foreach my $dev (keys %$netdev) {
323 next if $dev !~ m/^veth([1-9]\d*)i/;
325 my $d = $list->{$vmid};
329 $d->{netout
} += $netdev->{$dev}->{receive
};
330 $d->{netin
} += $netdev->{$dev}->{transmit
};
337 sub read_cgroup_list
($$$$) {
338 my ($group, $vmid, $unprivileged, $name) = @_;
340 my $content = read_cgroup_value
($group, $vmid, $unprivileged, $name, 1);
342 return { split(/\s+/, $content) };
345 sub read_cgroup_value
($$$$$) {
346 my ($group, $vmid, $unprivileged, $name, $full) = @_;
348 my $nsdir = $unprivileged ?
'' : 'ns/';
349 my $path = "/sys/fs/cgroup/$group/lxc/$vmid/${nsdir}$name";
351 return PVE
::Tools
::file_get_contents
($path) if $full;
353 return PVE
::Tools
::file_read_firstline
($path);
356 sub write_cgroup_value
{
357 my ($group, $vmid, $name, $value) = @_;
359 my $path = "/sys/fs/cgroup/$group/lxc/$vmid/$name";
360 PVE
::ProcFSTools
::write_proc_entry
($path, $value) if -e
$path;
364 sub find_lxc_console_pids
{
368 PVE
::Tools
::dir_glob_foreach
('/proc', '\d+', sub {
371 my $cmdline = PVE
::Tools
::file_read_firstline
("/proc/$pid/cmdline");
374 my @args = split(/\0/, $cmdline);
376 # search for lxc-console -n <vmid>
377 return if scalar(@args) != 3;
378 return if $args[1] ne '-n';
379 return if $args[2] !~ m/^\d+$/;
380 return if $args[0] !~ m
|^(/usr/bin
/)?lxc-console
$|;
384 push @{$res->{$vmid}}, $pid;
396 $pid = $1 if $line =~ m/^PID:\s+(\d+)$/;
398 PVE
::Tools
::run_command
(['lxc-info', '-n', $vmid, '-p'], outfunc
=> $parser);
400 die "unable to get PID for CT $vmid (not running?)\n" if !$pid;
407 sysopen(my $fd, "/proc/$pid", O_RDONLY
| O_DIRECTORY
)
408 or die "failed to open /proc/$pid pid fd\n";
415 # Find the pid and open:
416 my $pid = find_lxc_pid
($vmid);
417 my $fd = open_pid_fd
($pid);
420 my $pid2 = find_lxc_pid
($vmid);
422 return () if $pid != $pid2;
429 # Find the parent pid via proc and open it:
430 my $stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
431 my $ppid = $stat->{ppid
} // die "failed to get parent pid\n";
433 my $fd = open_pid_fd
($ppid);
436 $stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
437 my $ppid2 = $stat->{ppid
} // die "failed to get parent pid for verification\n";
439 return () if $ppid != $ppid2;
443 # Note: we cannot use Net:IP, because that only allows strict
445 sub parse_ipv4_cidr
{
446 my ($cidr, $noerr) = @_;
448 if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) && ($2 <= 32)) {
449 return { address
=> $1, netmask
=> $PVE::Network
::ipv4_reverse_mask-
>[$2] };
452 return undef if $noerr;
454 die "unable to parse ipv4 address/mask\n";
457 sub get_cgroup_subsystems
{
460 my $data = PVE
::Tools
::file_get_contents
('/proc/self/cgroup');
461 while ($data =~ /^\d+:([^:\n]*):.*$/gm) {
464 $v1->{$_} = 1 foreach split(/,/, $type);
469 return wantarray ?
($v1, $v2) : $v1;
472 # With seccomp trap to userspace we now have the ability to optionally forward
473 # certain syscalls to the "host" to handle (via our pve-lxc-syscalld daemon).
475 # This means that there are cases where we need to create an extra seccomp
476 # profile for the container to load.
478 # This returns a configuration snippet added to the raw lxc config.
479 sub make_seccomp_config
{
480 my ($conf, $vmid, $conf_dir, $unprivileged, $features) = @_;
481 # User-configured profile has precedence, note that the user's entry would
482 # be written 'after' this line anyway...
483 if (PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.seccomp.profile')) {
484 # Warn the user if this conflicts with a feature:
485 my $warn = join(', ', grep { $features->{$_} } qw(keyctl mknod));
486 warn "explicitly configured lxc.seccomp.profile overrides the following settings: $warn\n"
487 if length($warn) > 0;
491 # Privileged containers keep using the default (which is already part of
492 # the files included via lxc.include, so we don't need to write it out,
493 # that way it stays admin-configurable via /usr/share/lxc/config/... as
495 return '' if !$unprivileged;
498 keyctl
=> ['errno 38'],
503 # Unprivileged containers will get keyctl() disabled by default as a
504 # workaround for systemd-networkd behavior. But we have an option to
505 # explicitly enable it:
506 if ($features->{keyctl
}) {
507 delete $rules->{keyctl
};
510 # By default, unprivileged containers cannot use `mknod` at all.
511 # Since lxc 3.2, we can use seccomp's trap to userspace feature for this,
512 # but for now this is experimental, so it has to be enabled via a feature
514 # Note that we only handle block and char devices (like lxd), the rest we
515 # leave up to the kernel. We may in the future remove this if seccomp gets
516 # a way to tell the kernel to "continue" a syscall.
517 if ($features->{mknod
}) {
518 my ($ok, $kernel) = PVE
::ProcFSTools
::check_kernel_release
(5, 3);
520 die "'mknod' feature requested, but kernel too old (found $kernel, required >= 5.3)\n";
523 $raw_conf .= "lxc.seccomp.notify.proxy = unix:/run/pve/lxc-syscalld.sock\n";
524 $raw_conf .= "lxc.seccomp.notify.cookie = $vmid\n";
527 # condition: (mode & S_IFMT) == S_IFCHR
528 'notify [1,8192,SCMP_CMP_MASKED_EQ,61440]',
529 # condition: (mode & S_IFMT) == S_IFBLK
530 'notify [1,24576,SCMP_CMP_MASKED_EQ,61440]',
532 $rules->{mknodat
} = [
533 # condition: (mode & S_IFMT) == S_IFCHR
534 'notify [2,8192,SCMP_CMP_MASKED_EQ,61440]',
535 # condition: (mode & S_IFMT) == S_IFBLK
536 'notify [2,24576,SCMP_CMP_MASKED_EQ,61440]',
540 # Now build the custom seccomp rule text...
541 my $extra_rules = join("\n", map {
543 map { "$syscall $_" } $rules->{$syscall}->@*
544 } sort keys %$rules) . "\n";
546 return $raw_conf if $extra_rules eq "\n";
548 # We still have the "most common" config readily available, so don't write
550 if ($raw_conf eq '' && $extra_rules eq "keyctl errno 38\n") {
551 # we have no extra $raw_conf and use the same we had in pve 6.1:
552 return "lxc.seccomp.profile = $LXC_CONFIG_PATH/pve-userns.seccomp\n";
555 # Write the rule file to the container's config path:
556 my $rule_file = "$conf_dir/rules.seccomp";
557 my $rule_data = file_get_contents
("$LXC_CONFIG_PATH/common.seccomp")
559 file_set_contents
($rule_file, $rule_data);
560 $raw_conf .= "lxc.seccomp.profile = $rule_file\n";
565 # Since lxc-3.0.2 we can have lxc generate a profile for the container
566 # automatically. The default should be equivalent to the old
567 # `lxc-container-default-cgns` profile.
569 # Additionally this also added `lxc.apparmor.raw` which can be used to inject
570 # additional lines into the profile. We can use that to allow mounting specific
572 sub make_apparmor_config
{
573 my ($conf, $unprivileged, $features) = @_;
575 # user-configured profile has precedence, but first we go through our own
576 # code to figure out whether we should warn the user:
578 my $raw = "lxc.apparmor.profile = generated\n";
581 if ($features->{fuse
}) {
582 # For the informational warning:
583 push @profile_uses, 'features:fuse';
586 # There's lxc.apparmor.allow_nesting now, which will add the necessary
587 # apparmor lines, create an apparmor namespace for the container, but also
588 # adds proc and sysfs mounts to /dev/.lxc/{proc,sys}. These do not have
589 # lxcfs mounted over them, because that would prevent the container from
590 # mounting new instances of them for nested containers.
591 if ($features->{nesting
}) {
592 push @profile_uses, 'features:nesting';
593 $raw .= "lxc.apparmor.allow_nesting = 1\n"
595 # In the default profile in /etc/apparmor.d we patch this in because
596 # otherwise a container can for example run `chown` on /sys, breaking
597 # access to it for non-CAP_DAC_OVERRIDE tools on the host:
598 $raw .= "lxc.apparmor.raw = deny mount -> /proc/,\n";
599 $raw .= "lxc.apparmor.raw = deny mount -> /sys/,\n";
600 # Preferably we could use the 'remount' flag but this does not sit well
601 # with apparmor_parser currently:
602 # mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,
605 if (my $mount = $features->{mount
}) {
606 push @profile_uses, 'features:mount';
607 foreach my $fs (PVE
::Tools
::split_list
($mount)) {
608 $raw .= "lxc.apparmor.raw = mount fstype=$fs,\n";
614 if (PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.apparmor.profile')) {
615 if (length(my $used = join(', ', @profile_uses))) {
616 warn "explicitly configured lxc.apparmor.profile overrides the following settings: $used\n";
624 sub update_lxc_config
{
625 my ($vmid, $conf) = @_;
627 my $dir = "/var/lib/lxc/$vmid";
629 if ($conf->{template
}) {
631 unlink "$dir/config";
638 die "missing 'arch' - internal error" if !$conf->{arch
};
639 $raw .= "lxc.arch = $conf->{arch}\n";
641 my $custom_idmap = PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.idmap');
642 my $unprivileged = $conf->{unprivileged
} || $custom_idmap;
644 my $ostype = $conf->{ostype
} || die "missing 'ostype' - internal error";
646 my $cfgpath = '/usr/share/lxc/config';
647 my $inc = "$cfgpath/$ostype.common.conf";
648 $inc ="$cfgpath/common.conf" if !-f
$inc;
649 $raw .= "lxc.include = $inc\n";
651 $inc = "$cfgpath/$ostype.userns.conf";
652 $inc = "$cfgpath/userns.conf" if !-f
$inc;
653 $raw .= "lxc.include = $inc\n";
656 my $features = PVE
::LXC
::Config-
>parse_features($conf->{features
});
658 $raw .= make_seccomp_config
($conf, $vmid, $dir, $unprivileged, $features);
659 $raw .= make_apparmor_config
($conf, $unprivileged, $features);
660 if ($features->{fuse
}) {
661 $raw .= "lxc.apparmor.raw = mount fstype=fuse,\n";
662 $raw .= "lxc.mount.entry = /dev/fuse dev/fuse none bind,create=file 0 0\n";
665 # WARNING: DO NOT REMOVE this without making sure that loop device nodes
666 # cannot be exposed to the container with r/w access (cgroup perms).
667 # When this is enabled mounts will still remain in the monitor's namespace
668 # after the container unmounted them and thus will not detach from their
669 # files while the container is running!
670 $raw .= "lxc.monitor.unshare = 1\n";
672 my $cgv1 = get_cgroup_subsystems
();
674 # Should we read them from /etc/subuid?
675 if ($unprivileged && !$custom_idmap) {
676 $raw .= "lxc.idmap = u 0 100000 65536\n";
677 $raw .= "lxc.idmap = g 0 100000 65536\n";
680 if (!PVE
::LXC
::Config-
>has_dev_console($conf)) {
681 $raw .= "lxc.console.path = none\n";
682 $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n" if $cgv1->{devices
};
685 my $ttycount = PVE
::LXC
::Config-
>get_tty_count($conf);
686 $raw .= "lxc.tty.max = $ttycount\n";
688 # some init scripts expect a linux terminal (turnkey).
689 $raw .= "lxc.environment = TERM=linux\n";
691 my $utsname = $conf->{hostname
} || "CT$vmid";
692 $raw .= "lxc.uts.name = $utsname\n";
694 if ($cgv1->{memory
}) {
695 my $memory = $conf->{memory
} || 512;
696 my $swap = $conf->{swap
} // 0;
698 my $lxcmem = int($memory*1024*1024);
699 $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n";
701 my $lxcswap = int(($memory + $swap)*1024*1024);
702 $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n";
706 if (my $cpulimit = $conf->{cpulimit
}) {
707 $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n";
708 my $value = int(100000*$cpulimit);
709 $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n";
712 my $shares = $conf->{cpuunits
} || 1024;
713 $raw .= "lxc.cgroup.cpu.shares = $shares\n";
716 die "missing 'rootfs' configuration\n"
717 if !defined($conf->{rootfs
});
719 my $mountpoint = PVE
::LXC
::Config-
>parse_ct_rootfs($conf->{rootfs
});
721 $raw .= "lxc.rootfs.path = $dir/rootfs\n";
723 foreach my $k (sort keys %$conf) {
724 next if $k !~ m/^net(\d+)$/;
726 my $d = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$k});
727 $raw .= "lxc.net.$ind.type = veth\n";
728 $raw .= "lxc.net.$ind.veth.pair = veth${vmid}i${ind}\n";
729 $raw .= "lxc.net.$ind.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr
});
730 $raw .= "lxc.net.$ind.name = $d->{name}\n" if defined($d->{name
});
731 $raw .= "lxc.net.$ind.mtu = $d->{mtu}\n" if defined($d->{mtu
});
734 if ($cgv1->{cpuset
}) {
736 if (my $lxcconf = $conf->{lxc
}) {
737 foreach my $entry (@$lxcconf) {
738 my ($k, $v) = @$entry;
739 $had_cpuset = 1 if $k eq 'lxc.cgroup.cpuset.cpus';
744 my $cores = $conf->{cores
};
745 if (!$had_cpuset && $cores) {
746 my $cpuset = eval { PVE
::CpuSet-
>new_from_cgroup('lxc', 'effective_cpus') };
747 $cpuset = PVE
::CpuSet-
>new_from_cgroup('', 'effective_cpus') if !$cpuset;
748 my @members = $cpuset->members();
749 while (scalar(@members) > $cores) {
750 my $randidx = int(rand(scalar(@members)));
751 $cpuset->delete($members[$randidx]);
752 splice(@members, $randidx, 1); # keep track of the changes
754 $raw .= "lxc.cgroup.cpuset.cpus = ".$cpuset->short_string()."\n";
758 File
::Path
::mkpath
("$dir/rootfs");
760 PVE
::Tools
::file_set_contents
("$dir/config", $raw);
763 # verify and cleanup nameserver list (replace \0 with ' ')
764 sub verify_nameserver_list
{
765 my ($nameserver_list) = @_;
768 foreach my $server (PVE
::Tools
::split_list
($nameserver_list)) {
769 PVE
::JSONSchema
::pve_verify_ip
($server);
773 return join(' ', @list);
776 sub verify_searchdomain_list
{
777 my ($searchdomain_list) = @_;
780 foreach my $server (PVE
::Tools
::split_list
($searchdomain_list)) {
781 # todo: should we add checks for valid dns domains?
785 return join(' ', @list);
788 sub get_console_command
{
789 my ($vmid, $conf, $escapechar) = @_;
791 # '-1' as $escapechar disables keyboard escape sequence
792 # any other passed char (a-z) will result in <Ctrl+$escapechar q>
794 my $cmode = PVE
::LXC
::Config-
>get_cmode($conf);
797 if ($cmode eq 'console') {
798 push @$cmd, 'lxc-console', '-n', $vmid, '-t', 0;
799 push @$cmd, '-e', $escapechar if $escapechar;
800 } elsif ($cmode eq 'tty') {
801 push @$cmd, 'lxc-console', '-n', $vmid;
802 push @$cmd, '-e', $escapechar if $escapechar;
803 } elsif ($cmode eq 'shell') {
804 push @$cmd, 'lxc-attach', '--clear-env', '-n', $vmid;
806 die "internal error";
812 sub get_primary_ips
{
815 # return data from net0
817 return undef if !defined($conf->{net0
});
818 my $net = PVE
::LXC
::Config-
>parse_lxc_network($conf->{net0
});
820 my $ipv4 = $net->{ip
};
822 if ($ipv4 =~ /^(dhcp|manual)$/) {
828 my $ipv6 = $net->{ip6
};
830 if ($ipv6 =~ /^(auto|dhcp|manual)$/) {
837 return ($ipv4, $ipv6);
840 sub delete_mountpoint_volume
{
841 my ($storage_cfg, $vmid, $volume) = @_;
843 return if PVE
::LXC
::Config-
>classify_mountpoint($volume) ne 'volume';
845 my ($vtype, $name, $owner) = PVE
::Storage
::parse_volname
($storage_cfg, $volume);
847 if ($vmid == $owner) {
848 PVE
::Storage
::vdisk_free
($storage_cfg, $volume);
850 warn "ignore deletion of '$volume', CT $vmid isn't the owner!\n";
854 sub destroy_lxc_container
{
855 my ($storage_cfg, $vmid, $conf, $replacement_conf) = @_;
857 PVE
::LXC
::Config-
>foreach_mountpoint($conf, sub {
858 my ($ms, $mountpoint) = @_;
859 delete_mountpoint_volume
($storage_cfg, $vmid, $mountpoint->{volume
});
862 rmdir "/var/lib/lxc/$vmid/rootfs";
863 unlink "/var/lib/lxc/$vmid/config";
864 rmdir "/var/lib/lxc/$vmid";
865 if (defined $replacement_conf) {
866 PVE
::LXC
::Config-
>write_config($vmid, $replacement_conf);
868 PVE
::LXC
::Config-
>destroy_config($vmid);
872 sub vm_stop_cleanup
{
873 my ($storage_cfg, $vmid, $conf, $keepActive) = @_;
875 return if $keepActive;
878 my $vollist = PVE
::LXC
::Config-
>get_vm_volumes($conf);
879 PVE
::Storage
::deactivate_volumes
($storage_cfg, $vollist);
881 warn $@ if $@; # avoid errors - just warn
885 my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_;
887 if ($newnet->{type
} ne 'veth') {
888 # for when there are physical interfaces
889 die "cannot update interface of type $newnet->{type}";
892 my $veth = "veth${vmid}i${netid}";
893 my $eth = $newnet->{name
};
895 if (my $oldnetcfg = $conf->{$opt}) {
896 my $oldnet = PVE
::LXC
::Config-
>parse_lxc_network($oldnetcfg);
898 if (safe_string_ne
($oldnet->{hwaddr
}, $newnet->{hwaddr
}) ||
899 safe_string_ne
($oldnet->{name
}, $newnet->{name
})) {
901 PVE
::Network
::veth_delete
($veth);
902 delete $conf->{$opt};
903 PVE
::LXC
::Config-
>write_config($vmid, $conf);
905 hotplug_net
($vmid, $conf, $opt, $newnet, $netid);
908 if (safe_string_ne
($oldnet->{bridge
}, $newnet->{bridge
}) ||
909 safe_num_ne
($oldnet->{tag
}, $newnet->{tag
}) ||
910 safe_num_ne
($oldnet->{firewall
}, $newnet->{firewall
})) {
912 if ($oldnet->{bridge
}) {
913 PVE
::Network
::tap_unplug
($veth);
914 foreach (qw(bridge tag firewall)) {
915 delete $oldnet->{$_};
917 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($oldnet);
918 PVE
::LXC
::Config-
>write_config($vmid, $conf);
922 PVE
::Network
::SDN
::Zones
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
924 PVE
::Network
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
927 # This includes the rate:
928 foreach (qw(bridge tag firewall rate)) {
929 $oldnet->{$_} = $newnet->{$_} if $newnet->{$_};
931 } elsif (safe_string_ne
($oldnet->{rate
}, $newnet->{rate
})) {
932 # Rate can be applied on its own but any change above needs to
933 # include the rate in tap_plug since OVS resets everything.
934 PVE
::Network
::tap_rate_limit
($veth, $newnet->{rate
});
935 $oldnet->{rate
} = $newnet->{rate
}
937 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($oldnet);
938 PVE
::LXC
::Config-
>write_config($vmid, $conf);
941 hotplug_net
($vmid, $conf, $opt, $newnet, $netid);
944 update_ipconfig
($vmid, $conf, $opt, $eth, $newnet, $rootdir);
948 my ($vmid, $conf, $opt, $newnet, $netid) = @_;
950 my $veth = "veth${vmid}i${netid}";
951 my $vethpeer = $veth . "p";
952 my $eth = $newnet->{name
};
955 PVE
::Network
::SDN
::Zones
::veth_create
($veth, $vethpeer, $newnet->{bridge
}, $newnet->{hwaddr
});
956 PVE
::Network
::SDN
::Zones
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
958 PVE
::Network
::veth_create
($veth, $vethpeer, $newnet->{bridge
}, $newnet->{hwaddr
});
959 PVE
::Network
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
962 # attach peer in container
963 my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ];
964 PVE
::Tools
::run_command
($cmd);
966 # link up peer in container
967 $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up' ];
968 PVE
::Tools
::run_command
($cmd);
970 my $done = { type
=> 'veth' };
971 foreach (qw(bridge tag firewall hwaddr name)) {
972 $done->{$_} = $newnet->{$_} if $newnet->{$_};
974 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($done);
976 PVE
::LXC
::Config-
>write_config($vmid, $conf);
979 sub update_ipconfig
{
980 my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_;
982 my $lxc_setup = PVE
::LXC
::Setup-
>new($conf, $rootdir);
984 my $optdata = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$opt});
989 PVE
::Tools
::run_command
(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs);
991 my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) };
993 my $change_ip_config = sub {
994 my ($ipversion) = @_;
996 my $family_opt = "-$ipversion";
997 my $suffix = $ipversion == 4 ?
'' : $ipversion;
1001 my $newip = $newnet->{$ip};
1002 my $newgw = $newnet->{$gw};
1003 my $oldip = $optdata->{$ip};
1004 my $oldgw = $optdata->{$gw};
1006 my $change_ip = safe_string_ne
($oldip, $newip);
1007 my $change_gw = safe_string_ne
($oldgw, $newgw);
1009 return if !$change_ip && !$change_gw;
1011 # step 1: add new IP, if this fails we cancel
1012 my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/);
1013 if ($change_ip && $is_real_ip) {
1014 eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); };
1021 # step 2: replace gateway
1022 # If this fails we delete the added IP and cancel.
1023 # If it succeeds we save the config and delete the old IP, ignoring
1024 # errors. The config is then saved.
1025 # Note: 'ip route replace' can add
1029 if ($is_real_ip && !PVE
::Network
::is_ip_in_cidr
($newgw, $newip, $ipversion)) {
1030 &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth);
1032 &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw);
1036 # the route was not replaced, the old IP is still available
1037 # rollback (delete new IP) and cancel
1039 eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); };
1040 warn $@ if $@; # no need to die here
1045 eval { &$ipcmd($family_opt, 'route', 'del', 'default'); };
1046 # if the route was not deleted, the guest might have deleted it manually
1050 if ($oldgw && $oldip && !PVE
::Network
::is_ip_in_cidr
($oldgw, $oldip)) {
1051 eval { &$ipcmd($family_opt, 'route', 'del', $oldgw, 'dev', $eth); };
1052 # warn if the route was deleted manually
1057 # from this point on we save the configuration
1058 # step 3: delete old IP ignoring errors
1059 if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) {
1060 # We need to enable promote_secondaries, otherwise our newly added
1061 # address will be removed along with the old one.
1064 if ($ipversion == 4) {
1065 &$nscmd({ outfunc
=> sub { $promote = int(shift) } },
1066 'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries");
1067 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1");
1069 &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth);
1071 warn $@ if $@; # no need to die here
1073 if ($ipversion == 4) {
1074 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote");
1078 foreach my $property ($ip, $gw) {
1079 if ($newnet->{$property}) {
1080 $optdata->{$property} = $newnet->{$property};
1082 delete $optdata->{$property};
1085 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($optdata);
1086 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1087 $lxc_setup->setup_network($conf);
1090 &$change_ip_config(4);
1091 &$change_ip_config(6);
1095 my $open_namespace = sub {
1096 my ($vmid, $pid, $kind) = @_;
1097 sysopen my $fd, "/proc/$pid/ns/$kind", O_RDONLY
1098 or die "failed to open $kind namespace of container $vmid: $!\n";
1102 my $enter_namespace = sub {
1103 my ($vmid, $pid, $kind, $type) = @_;
1104 my $fd = $open_namespace->($vmid, $pid, $kind);
1105 PVE
::Tools
::setns
(fileno($fd), $type)
1106 or die "failed to enter $kind namespace of container $vmid: $!\n";
1110 my $get_container_namespace = sub {
1111 my ($vmid, $pid, $kind) = @_;
1114 if (!defined($pid)) {
1115 # Pin the pid while we're grabbing its stuff from /proc
1116 ($pid, $pidfd) = open_lxc_pid
($vmid)
1117 or die "failed to open pidfd of container $vmid\'s init process\n";
1120 return $open_namespace->($vmid, $pid, $kind);
1123 my $do_syncfs = sub {
1124 my ($vmid, $pid, $socket) = @_;
1126 &$enter_namespace($vmid, $pid, 'mnt', PVE
::Tools
::CLONE_NEWNS
);
1128 # Tell the parent process to start reading our /proc/mounts
1129 print {$socket} "go\n";
1132 # Receive /proc/self/mounts
1133 my $mountdata = do { local $/ = undef; <$socket> };
1136 # Now sync all mountpoints...
1137 my $mounts = PVE
::ProcFSTools
::parse_mounts
($mountdata);
1138 foreach my $mp (@$mounts) {
1139 my ($what, $dir, $fs) = @$mp;
1140 next if $fs eq 'fuse.lxcfs';
1141 eval { PVE
::Tools
::sync_mountpoint
($dir); };
1146 sub sync_container_namespace
{
1148 my $pid = find_lxc_pid
($vmid);
1150 # SOCK_DGRAM is nicer for barriers but cannot be slurped
1151 socketpair my $pfd, my $cfd, AF_UNIX
, SOCK_STREAM
, PF_UNSPEC
1152 or die "failed to create socketpair: $!\n";
1155 die "fork failed: $!\n" if !defined($child);
1160 &$do_syncfs($vmid, $pid, $cfd);
1170 die "failed to enter container namespace\n" if $go ne "go\n";
1172 open my $mounts, '<', "/proc/$child/mounts"
1173 or die "failed to open container's /proc/mounts: $!\n";
1174 my $mountdata = do { local $/ = undef; <$mounts> };
1176 print {$pfd} $mountdata;
1179 while (waitpid($child, 0) != $child) {}
1180 die "failed to sync container namespace\n" if $? != 0;
1183 sub template_create
{
1184 my ($vmid, $conf) = @_;
1186 my $storecfg = PVE
::Storage
::config
();
1188 PVE
::LXC
::Config-
>foreach_mountpoint($conf, sub {
1189 my ($ms, $mountpoint) = @_;
1191 my $volid = $mountpoint->{volume
};
1193 die "Template feature is not available for '$volid'\n"
1194 if !PVE
::Storage
::volume_has_feature
($storecfg, 'template', $volid);
1197 PVE
::LXC
::Config-
>foreach_mountpoint($conf, sub {
1198 my ($ms, $mountpoint) = @_;
1200 my $volid = $mountpoint->{volume
};
1202 PVE
::Storage
::activate_volumes
($storecfg, [$volid]);
1204 my $template_volid = PVE
::Storage
::vdisk_create_base
($storecfg, $volid);
1205 $mountpoint->{volume
} = $template_volid;
1206 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq "rootfs");
1209 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1212 sub check_ct_modify_config_perm
{
1213 my ($rpcenv, $authuser, $vmid, $pool, $newconf, $delete) = @_;
1215 return 1 if $authuser eq 'root@pam';
1218 my ($opt, $delete) = @_;
1219 if ($opt eq 'cores' || $opt eq 'cpuunits' || $opt eq 'cpulimit') {
1220 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']);
1221 } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) {
1222 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']);
1224 my $data = $opt eq 'rootfs' ? PVE
::LXC
::Config-
>parse_ct_rootfs($newconf->{$opt})
1225 : PVE
::LXC
::Config-
>parse_ct_mountpoint($newconf->{$opt});
1226 raise_perm_exc
("mount point type $data->{type} is only allowed for root\@pam")
1227 if $data->{type
} ne 'volume';
1228 } elsif ($opt eq 'memory' || $opt eq 'swap') {
1229 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']);
1230 } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' ||
1231 $opt eq 'searchdomain' || $opt eq 'hostname') {
1232 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']);
1233 } elsif ($opt eq 'features') {
1234 # For now this is restricted to root@pam
1235 raise_perm_exc
("changing feature flags is only allowed for root\@pam");
1236 } elsif ($opt eq 'hookscript') {
1237 # For now this is restricted to root@pam
1238 raise_perm_exc
("changing the hookscript is only allowed for root\@pam");
1240 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']);
1244 foreach my $opt (keys %$newconf) {
1247 foreach my $opt (@$delete) {
1255 my ($vmid, $storage_cfg, $conf, $noerr) = @_;
1257 my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1258 my $volid_list = PVE
::LXC
::Config-
>get_vm_volumes($conf);
1262 PVE
::LXC
::Config-
>foreach_mountpoint_reverse($conf, sub {
1263 my ($ms, $mountpoint) = @_;
1265 my $volid = $mountpoint->{volume
};
1266 my $mount = $mountpoint->{mp
};
1268 return if !$volid || !$mount;
1270 my $mount_path = "$rootdir/$mount";
1271 $mount_path =~ s!/+!/!g;
1273 return if !PVE
::ProcFSTools
::is_mounted
($mount_path);
1276 PVE
::Tools
::run_command
(['umount', '-d', $mount_path]);
1288 return $res; # tell caller if (some) umounts failed for the noerr case
1292 my ($vmid, $storage_cfg, $conf, $ignore_ro) = @_;
1294 my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1295 File
::Path
::make_path
($rootdir);
1297 my $volid_list = PVE
::LXC
::Config-
>get_vm_volumes($conf);
1298 PVE
::Storage
::activate_volumes
($storage_cfg, $volid_list);
1300 my (undef, $rootuid, $rootgid) = parse_id_maps
($conf);
1303 PVE
::LXC
::Config-
>foreach_mountpoint($conf, sub {
1304 my ($ms, $mountpoint) = @_;
1306 $mountpoint->{ro
} = 0 if $ignore_ro;
1308 mountpoint_mount
($mountpoint, $rootdir, $storage_cfg, undef, $rootuid, $rootgid);
1312 warn "mounting container failed\n";
1313 umount_all
($vmid, $storage_cfg, $conf, 1);
1321 sub mountpoint_mount_path
{
1322 my ($mountpoint, $storage_cfg, $snapname) = @_;
1324 return mountpoint_mount
($mountpoint, undef, $storage_cfg, $snapname);
1332 if ($line =~ m
@^(/dev/loop\d
+):@) {
1336 my $cmd = ['losetup', '--associated', $path];
1337 PVE
::Tools
::run_command
($cmd, outfunc
=> $parser);
1341 # Run a function with a file attached to a loop device.
1342 # The loop device is always detached afterwards (or set to autoclear).
1343 # Returns the loop device.
1344 sub run_with_loopdev
{
1345 my ($func, $file, $readonly) = @_;
1346 my $device = query_loopdev
($file);
1347 # Try to reuse an existing device
1349 # We assume that whoever setup the loop device is responsible for
1357 if ($line =~ m
@^(/dev/loop\d
+)$@) {
1367 push @$losetup_cmd, '-r' if $readonly;
1368 PVE
::Tools
::run_command
($losetup_cmd, outfunc
=> $parser);
1369 die "failed to setup loop device for $file\n" if !$device;
1370 eval { &$func($device); };
1372 PVE
::Tools
::run_command
(['losetup', '-d', $device]);
1377 # In scalar mode: returns a file handle to the deepest directory node.
1378 # In list context: returns a list of:
1379 # * the deepest directory node
1380 # * the 2nd deepest directory (parent of the above)
1381 # * directory name of the last directory
1382 # So that the path $2/$3 should lead to $1 afterwards.
1383 sub walk_tree_nofollow
($$$;$$) {
1384 my ($start, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1386 sysopen(my $fd, $start, O_PATH
| O_DIRECTORY
)
1387 or die "failed to open start directory $start: $!\n";
1389 return walk_tree_nofollow_fd
($start, $fd, $subdir, $mkdir, $rootuid, $rootgid);
1393 sub walk_tree_nofollow_fd
($$$$;$$) {
1394 my ($start_dirname, $start_fd, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1396 # splitdir() returns '' for empty components including the leading /
1397 my @comps = grep { length($_)>0 } File
::Spec-
>splitdir($subdir);
1400 my $dir = $start_dirname;
1401 my $last_component = undef;
1403 foreach my $component (@comps) {
1404 $dir .= "/$component";
1405 my $next = PVE
::Tools
::openat
(fileno($fd), $component, O_NOFOLLOW
| O_DIRECTORY
);
1408 # failed, check for symlinks and try to create the path
1409 die "symlink encountered at: $dir\n" if $! == ELOOP
|| $! == ENOTDIR
;
1410 die "cannot open directory $dir: $!\n" if !$mkdir;
1412 # We don't check for errors on mkdirat() here and just try to
1413 # openat() again, since at least one error (EEXIST) is an
1414 # expected possibility if multiple containers start
1415 # simultaneously. If someone else injects a symlink now then
1416 # the subsequent openat() will fail due to O_NOFOLLOW anyway.
1417 PVE
::Tools
::mkdirat
(fileno($fd), $component, 0755);
1419 $next = PVE
::Tools
::openat
(fileno($fd), $component, O_NOFOLLOW
| O_DIRECTORY
);
1420 die "failed to create path: $dir: $!\n" if !$next;
1422 PVE
::Tools
::fchownat
(fileno($next), '', $rootuid, $rootgid, PVE
::Tools
::AT_EMPTY_PATH
)
1423 if defined($rootuid) && defined($rootgid);
1426 close $second if defined($last_component) && $second != $start_fd;
1427 $last_component = $component;
1432 return ($fd, defined($last_component) && $second, $last_component) if wantarray;
1433 close $second if defined($last_component) && $second != $start_fd;
1437 # To guard against symlink attack races against other currently running
1438 # containers with shared recursive bind mount hierarchies we prepare a
1439 # directory handle for the directory we're mounting over to verify the
1440 # mountpoint afterwards.
1441 sub __bindmount_prepare
{
1442 my ($hostroot, $dir) = @_;
1443 my $srcdh = walk_tree_nofollow
($hostroot, $dir, 0);
1447 # Assuming we mount to rootfs/a/b/c, verify with the directory handle to 'b'
1448 # ($parentfd) that 'b/c' (openat($parentfd, 'c')) really leads to the directory
1449 # we intended to bind mount.
1450 sub __bindmount_verify
{
1451 my ($srcdh, $parentfd, $last_dir, $ro) = @_;
1454 # Open the mount point path coming from the parent directory since the
1455 # filehandle we would have gotten as first result of walk_tree_nofollow
1456 # earlier is still a handle to the underlying directory instead of the
1458 $destdh = PVE
::Tools
::openat
(fileno($parentfd), $last_dir, PVE
::Tools
::O_PATH
| O_NOFOLLOW
| O_DIRECTORY
);
1459 die "failed to open mount point: $!\n" if !$destdh;
1462 # no separate function because 99% of the time it's the wrong thing to use.
1463 if (syscall(PVE
::Syscall
::faccessat
, fileno($destdh), $dot, &POSIX
::W_OK
, 0) != -1) {
1464 die "failed to mark bind mount read only\n";
1466 die "read-only check failed: $!\n" if $! != EROFS
;
1469 # For the rootfs we don't have a parentfd so we open the path directly.
1470 # Note that this means bindmounting any prefix of the host's
1471 # /var/lib/lxc/$vmid path into another container is considered a grave
1473 sysopen $destdh, $last_dir, O_PATH
| O_DIRECTORY
;
1474 die "failed to open mount point: $!\n" if !$destdh;
1477 my ($srcdev, $srcinode) = stat($srcdh);
1478 my ($dstdev, $dstinode) = stat($destdh);
1482 return ($srcdev == $dstdev && $srcinode == $dstinode);
1485 # Perform the actual bind mounting:
1486 sub __bindmount_do
{
1487 my ($dir, $dest, $ro, @extra_opts) = @_;
1488 PVE
::Tools
::run_command
(['mount', '-o', 'bind', @extra_opts, $dir, $dest]);
1490 eval { PVE
::Tools
::run_command
(['mount', '-o', 'bind,remount,ro', $dest]); };
1492 warn "bindmount error\n";
1493 # don't leave writable bind-mounts behind...
1494 PVE
::Tools
::run_command
(['umount', $dest]);
1501 my ($dir, $parentfd, $last_dir, $dest, $ro, @extra_opts) = @_;
1503 my $srcdh = __bindmount_prepare
('/', $dir);
1505 __bindmount_do
($dir, $dest, $ro, @extra_opts);
1507 if (!__bindmount_verify
($srcdh, $parentfd, $last_dir, $ro)) {
1508 PVE
::Tools
::run_command
(['umount', $dest]);
1509 die "detected mount path change at: $dir\n";
1513 # Cleanup $rootdir a bit (double and trailing slashes), build the mount path
1514 # from $rootdir and $mount and walk the path from $rootdir to the final
1515 # directory to check for symlinks.
1516 sub __mount_prepare_rootdir
{
1517 my ($rootdir, $mount, $rootuid, $rootgid) = @_;
1518 $rootdir =~ s!/+!/!g;
1519 $rootdir =~ s!/+$!!;
1520 my $mount_path = "$rootdir/$mount";
1521 my ($mpfd, $parentfd, $last_dir) = walk_tree_nofollow
($rootdir, $mount, 1, $rootuid, $rootgid);
1522 return ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir);
1525 # use $rootdir = undef to just return the corresponding mount path
1526 sub mountpoint_mount
{
1527 my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1528 return __mountpoint_mount
($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, undef);
1531 sub mountpoint_stage
{
1532 my ($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1533 my ($path, $loop, $dev) =
1534 __mountpoint_mount
($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid, 1);
1536 if (!defined($path)) {
1537 return undef if $! == ENOSYS
;
1538 die "failed to mount subvolume: $!\n";
1541 # We clone the mount point and leave it there in order to keep them connected to eg. loop
1542 # devices in case we're hotplugging (which would allow contaienrs to unmount the new mount
1545 my $fd = PVE
::Tools
::open_tree
(&AT_FDCWD
, $stage_dir, &OPEN_TREE_CLOEXEC
| &OPEN_TREE_CLONE
)
1546 or die "open_tree() on mount point failed: $!\n";
1548 return wantarray ?
($path, $loop, $dev, $fd) : $fd;
1551 sub mountpoint_insert_staged
{
1552 my ($mount_fd, $rootdir_fd, $mp_dir, $opt, $rootuid, $rootgid) = @_;
1554 if (!defined($rootdir_fd)) {
1555 sysopen($rootdir_fd, '.', O_PATH
| O_DIRECTORY
)
1556 or die "failed to open '.': $!\n";
1559 my $dest_fd = walk_tree_nofollow_fd
('/', $rootdir_fd, $mp_dir, 1, $rootuid, $rootgid);
1561 PVE
::Tools
::move_mount
(
1566 &MOVE_MOUNT_F_EMPTY_PATH
| &MOVE_MOUNT_T_EMPTY_PATH
,
1567 ) or die "failed to move '$opt' into container hierarchy: $!\n";
1570 # Use $stage_mount, $rootdir is treated as a temporary path to "stage" the file system. The user
1571 # can then open a file descriptor to it which can be used with the `move_mount` syscall.
1572 # Note that if the kernel does not support the new mount API, this will not perform any action
1573 # and return `undef` with $! = ENOSYS.
1574 sub __mountpoint_mount
{
1575 my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, $stage_mount) = @_;
1577 if (defined($stage_mount) && !PVE
::LXC
::Tools
::can_use_new_mount_api
()) {
1582 # When staging mount points we always mount to $rootdir directly (iow. as if `mp=/`).
1583 # This is required since __mount_prepare_rootdir() will return handles to the parent directory
1584 # which we use in __bindmount_verify()!
1585 my $mount = $stage_mount ?
'/': $mountpoint->{mp
};
1587 my $volid = $mountpoint->{volume
};
1588 my $type = $mountpoint->{type
};
1589 my $quota = !$snapname && !$mountpoint->{ro
} && $mountpoint->{quota
};
1592 return if !$volid || !$mount;
1597 my ($mpfd, $parentfd, $last_dir);
1599 if (defined($rootdir)) {
1600 ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir) =
1601 __mount_prepare_rootdir
($rootdir, $mount, $rootuid, $rootgid);
1604 if (defined($stage_mount)) {
1605 $mount_path = $rootdir;
1608 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1610 die "unknown snapshot path for '$volid'" if !$storage && defined($snapname);
1614 if (my $mountopts = $mountpoint->{mountoptions
}) {
1615 my @opts = split(/;/, $mountpoint->{mountoptions
});
1616 push @$optlist, grep { PVE
::LXC
::Config
::is_valid_mount_option
($_) } @opts;
1619 my $acl = $mountpoint->{acl
};
1620 if (defined($acl)) {
1621 push @$optlist, ($acl ?
'acl' : 'noacl');
1624 my $optstring = join(',', @$optlist);
1625 my $readonly = $mountpoint->{ro
};
1628 @extra_opts = ('-o', $optstring) if $optstring;
1632 my $scfg = PVE
::Storage
::storage_config
($storage_cfg, $storage);
1634 my $path = PVE
::Storage
::map_volume
($storage_cfg, $volid, $snapname);
1636 $path = PVE
::Storage
::path
($storage_cfg, $volid, $snapname) if !defined($path);
1638 my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1639 PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1641 $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files
1643 if ($format eq 'subvol') {
1645 my (undef, $name) = PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1646 if (defined($snapname)) {
1647 $name .= "\@$snapname";
1648 if ($scfg->{type
} eq 'zfspool') {
1649 PVE
::Tools
::run_command
(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', "$scfg->{pool}/$name", $mount_path]);
1651 die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n";
1654 if (defined($acl) && $scfg->{type
} eq 'zfspool') {
1655 my $acltype = ($acl ?
'acltype=posixacl' : 'acltype=noacl');
1656 PVE
::Tools
::run_command
(['zfs', 'set', $acltype, "$scfg->{pool}/$name"]);
1658 bindmount
($path, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts);
1659 warn "cannot enable quota control for bind mounted subvolumes\n" if $quota;
1662 return wantarray ?
($path, 0, undef) : $path;
1663 } elsif ($format eq 'raw' || $format eq 'iso') {
1664 # NOTE: 'mount' performs canonicalization without the '-c' switch, which for
1665 # device-mapper devices is special-cased to use the /dev/mapper symlinks.
1666 # Our autodev hook expects the /dev/dm-* device currently
1667 # and will create the /dev/mapper symlink accordingly
1668 $path = Cwd
::realpath
($path);
1669 die "failed to get device path\n" if !$path;
1670 ($path) = ($path =~ /^(.*)$/s); #untaint
1674 if ($format eq 'iso') {
1675 PVE
::Tools
::run_command
(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]);
1676 } elsif ($isBase || defined($snapname)) {
1677 PVE
::Tools
::run_command
(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]);
1680 push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0';
1682 push @extra_opts, '-o', 'ro' if $readonly;
1683 PVE
::Tools
::run_command
(['mount', @extra_opts, $path, $mount_path]);
1687 my $use_loopdev = 0;
1688 if ($scfg->{path
}) {
1689 $mounted_dev = run_with_loopdev
($domount, $path, $readonly);
1691 } elsif ($scfg->{type
} eq 'drbd' || $scfg->{type
} eq 'lvm' ||
1692 $scfg->{type
} eq 'rbd' || $scfg->{type
} eq 'lvmthin') {
1693 $mounted_dev = $path;
1696 die "unsupported storage type '$scfg->{type}'\n";
1698 return wantarray ?
($path, $use_loopdev, $mounted_dev) : $path;
1700 die "unsupported image format '$format'\n";
1702 } elsif ($type eq 'device') {
1703 push @extra_opts, '-o', 'ro' if $readonly;
1704 push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0' if $quota;
1705 # See the NOTE above about devicemapper canonicalization
1706 my ($devpath) = (Cwd
::realpath
($volid) =~ /^(.*)$/s); # realpath() taints
1707 PVE
::Tools
::run_command
(['mount', @extra_opts, $volid, $mount_path]) if $mount_path;
1708 return wantarray ?
($volid, 0, $devpath) : $volid;
1709 } elsif ($type eq 'bind') {
1710 die "directory '$volid' does not exist\n" if ! -d
$volid;
1711 bindmount
($volid, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts) if $mount_path;
1712 warn "cannot enable quota control for bind mounts\n" if $quota;
1713 return wantarray ?
($volid, 0, undef) : $volid;
1716 die "unsupported storage";
1719 sub mountpoint_hotplug
($$$) {
1720 my ($vmid, $conf, $opt, $mp, $storage_cfg) = @_;
1722 my (undef, $rootuid, $rootgid) = PVE
::LXC
::parse_id_maps
($conf);
1724 # We do the rest in a fork with an unshared mount namespace, because:
1725 # -) change our papparmor profile to that of /usr/bin/lxc-start
1726 # -) we're now going to 'stage' # the mountpoint, then grab it, then move into the
1727 # container's namespace, then mount it.
1729 PVE
::Tools
::run_fork
(sub {
1730 # Pin the container pid longer, we also need to get its monitor/parent:
1731 my ($ct_pid, $ct_pidfd) = open_lxc_pid
($vmid)
1732 or die "failed to open pidfd of container $vmid\'s init process\n";
1734 my ($monitor_pid, $monitor_pidfd) = open_ppid
($ct_pid)
1735 or die "failed to open pidfd of container $vmid\'s monitor process\n";
1737 my $ct_mnt_ns = $get_container_namespace->($vmid, $ct_pid, 'mnt');
1738 my $monitor_mnt_ns = $get_container_namespace->($vmid, $monitor_pid, 'mnt');
1740 # Grab a file descriptor to our apparmor label file so we can change into the 'lxc-start'
1741 # profile to lower our privileges to the same level we have in the start hook:
1742 sysopen(my $aa_fd, "/proc/self/attr/current", O_WRONLY
)
1743 or die "failed to open '/proc/self/attr/current' for writing: $!\n";
1744 # But switch namespaces first, to make sure the namespace switches aren't blocked by
1747 # Change into the monitor's mount namespace. We "pin" the mount into the monitor's
1748 # namespace for it to remain active there since the container will be able to unmount
1749 # hotplugged mount points and thereby potentially free up loop devices, which is a security
1751 PVE
::Tools
::setns
(fileno($monitor_mnt_ns), PVE
::Tools
::CLONE_NEWNS
);
1753 or die "failed to change root directory within the monitor's mount namespace: $!\n";
1755 my $dir = get_staging_mount_path
($opt);
1757 # Now switch our apparmor profile before mounting:
1758 my $data = 'changeprofile /usr/bin/lxc-start';
1759 if (syswrite($aa_fd, $data, length($data)) != length($data)) {
1760 die "failed to change apparmor profile: $!\n";
1762 # Check errors on close as well:
1764 or die "failed to change apparmor profile (close() failed): $!\n";
1766 my $mount_fd = mountpoint_stage
($mp, $dir, $storage_cfg, undef, $rootuid, $rootgid);
1768 PVE
::Tools
::setns
(fileno($ct_mnt_ns), PVE
::Tools
::CLONE_NEWNS
);
1770 or die "failed to change root directory within the container's mount namespace: $!\n";
1772 mountpoint_insert_staged
($mount_fd, undef, $mp->{mp
}, $opt, $rootuid, $rootgid);
1776 # Create a directory in the mountpoint staging tempfs.
1777 sub get_staging_mount_path
($) {
1780 my $target = get_staging_tempfs
() . "/$opt";
1781 if (!mkdir($target) && $! != EEXIST
) {
1782 die "failed to create directory $target: $!\n";
1788 # Mount /run/pve/mountpoints as tmpfs
1789 sub get_staging_tempfs
() {
1790 # We choose a path in /var/lib/lxc/ here because the lxc-start apparmor profile restricts most
1792 my $target = '/var/lib/lxc/.pve-staged-mounts';
1793 if (!mkdir($target)) {
1794 return $target if $! == EEXIST
;
1795 die "failed to create directory $target: $!\n";
1798 PVE
::Tools
::mount
("none", $target, 'tmpfs', 0, "size=8k,mode=755")
1799 or die "failed to mount $target as tmpfs: $!\n";
1805 my ($dev, $rootuid, $rootgid) = @_;
1807 PVE
::Tools
::run_command
(['mkfs.ext4', '-O', 'mmp',
1808 '-E', "root_owner=$rootuid:$rootgid",
1813 my ($storage_cfg, $volid, $rootuid, $rootgid) = @_;
1815 if ($volid =~ m!^/dev/.+!) {
1820 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1822 die "cannot format volume '$volid' with no storage\n" if !$storage;
1824 PVE
::Storage
::activate_volumes
($storage_cfg, [$volid]);
1826 my $path = PVE
::Storage
::map_volume
($storage_cfg, $volid);
1828 $path = PVE
::Storage
::path
($storage_cfg, $volid) if !defined($path);
1830 my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1831 PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1833 die "cannot format volume '$volid' (format == $format)\n"
1834 if $format ne 'raw';
1836 mkfs
($path, $rootuid, $rootgid);
1840 my ($storecfg, $vollist) = @_;
1842 foreach my $volid (@$vollist) {
1843 eval { PVE
::Storage
::vdisk_free
($storecfg, $volid); };
1849 my ($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid) = @_;
1851 my $needs_chown = 0;
1854 my $scfg = PVE
::Storage
::storage_config
($storecfg, $storage);
1855 # fixme: use better naming ct-$vmid-disk-X.raw?
1859 if ($scfg->{type
} eq 'dir' || $scfg->{type
} eq 'nfs' || $scfg->{type
} eq 'cifs' ) {
1861 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw',
1865 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'subvol',
1869 } elsif ($scfg->{type
} eq 'zfspool') {
1871 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'subvol',
1874 } elsif ($scfg->{type
} eq 'drbd' || $scfg->{type
} eq 'lvm' || $scfg->{type
} eq 'lvmthin') {
1876 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1879 } elsif ($scfg->{type
} eq 'rbd') {
1881 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1884 die "unable to create containers on storage type '$scfg->{type}'\n";
1886 format_disk
($storecfg, $volid, $rootuid, $rootgid) if $do_format;
1889 # in case formatting got interrupted:
1890 if (defined($volid)) {
1891 eval { PVE
::Storage
::vdisk_free
($storecfg, $volid); };
1897 return ($volid, $needs_chown);
1900 our $NEW_DISK_RE = qr/^([^:\s]+):(\d+(\.\d+)?)$/;
1902 my ($storecfg, $vmid, $settings, $conf, $pending) = @_;
1907 my (undef, $rootuid, $rootgid) = PVE
::LXC
::parse_id_maps
($conf);
1908 my $chown_vollist = [];
1910 PVE
::LXC
::Config-
>foreach_mountpoint($settings, sub {
1911 my ($ms, $mountpoint) = @_;
1913 my $volid = $mountpoint->{volume
};
1914 my $mp = $mountpoint->{mp
};
1916 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1918 if ($storage && ($volid =~ $NEW_DISK_RE)) {
1919 my ($storeid, $size_gb) = ($1, $2);
1921 my $size_kb = int(${size_gb
}*1024) * 1024;
1923 my $needs_chown = 0;
1924 ($volid, $needs_chown) = alloc_disk
($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid);
1925 push @$chown_vollist, $volid if $needs_chown;
1926 push @$vollist, $volid;
1927 $mountpoint->{volume
} = $volid;
1928 $mountpoint->{size
} = $size_kb * 1024;
1930 $conf->{pending
}->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1932 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1935 # use specified/existing volid/dir/device
1936 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1940 PVE
::Storage
::activate_volumes
($storecfg, $chown_vollist, undef);
1941 foreach my $volid (@$chown_vollist) {
1942 my $path = PVE
::Storage
::path
($storecfg, $volid, undef);
1943 chown($rootuid, $rootgid, $path);
1945 PVE
::Storage
::deactivate_volumes
($storecfg, $chown_vollist, undef);
1947 # free allocated images on error
1949 destroy_disks
($storecfg, $vollist);
1955 sub update_disksize
{
1956 my ($vmid, $conf, $all_volumes) = @_;
1959 my $prefix = "CT $vmid:";
1961 my $update_mp = sub {
1962 my ($key, $mp, @param) = @_;
1963 my $size = $all_volumes->{$mp->{volume
}}->{size
} // 0;
1965 if (!defined($mp->{size
}) || $size != $mp->{size
}) {
1967 print "$prefix updated volume size of '$mp->{volume}' in config.\n";
1968 $mp->{size
} = $size;
1969 my $nomp = 1 if ($key eq 'rootfs');
1970 $conf->{$key} = PVE
::LXC
::Config-
>print_ct_mountpoint($mp, $nomp);
1974 PVE
::LXC
::Config-
>foreach_mountpoint($conf, $update_mp);
1980 my ($vmid, $conf, $all_volumes) = @_;
1983 my $prefix = "CT $vmid:";
1985 # Note: it is allowed to define multiple storage entries with the same path
1986 # (alias), so we need to check both 'volid' and real 'path' (two different
1987 # volid can point to the same path).
1989 # used and unused disks
1993 foreach my $opt (keys %$conf) {
1994 next if ($opt !~ m/^unused\d+$/);
1995 my $vol = $all_volumes->{$conf->{$opt}};
1996 $refpath->{$vol->{path
}} = $vol->{volid
};
1999 foreach my $key (keys %$all_volumes) {
2000 my $vol = $all_volumes->{$key};
2001 my $in_use = PVE
::LXC
::Config-
>is_volume_in_use($conf, $vol->{volid
});
2002 my $path = $vol->{path
};
2005 $refpath->{$path} = $key;
2006 delete $orphans->{$path};
2008 if ((!$orphans->{$path}) && (!$refpath->{$path})) {
2009 $orphans->{$path} = $key;
2014 for my $key (keys %$orphans) {
2015 my $disk = $orphans->{$key};
2016 my $unused = PVE
::LXC
::Config-
>add_unused_volume($conf, $disk);
2020 print "$prefix add unreferenced volume '$disk' as '$unused' to config.\n";
2028 my ($cfg, $vmid) = @_;
2030 my $info = PVE
::Storage
::vdisk_list
($cfg, undef, $vmid);
2032 my $all_volumes = {};
2033 foreach my $storeid (keys %$info) {
2034 foreach my $item (@{$info->{$storeid}}) {
2035 my $volid = $item->{volid
};
2036 next if !($volid && $item->{size
});
2037 $item->{path
} = PVE
::Storage
::path
($cfg, $volid);
2038 $all_volumes->{$volid} = $item;
2042 return $all_volumes;
2046 my ($vmid, $nolock, $dryrun) = @_;
2048 my $cfg = PVE
::Storage
::config
();
2050 # FIXME: Remove once our RBD plugin can handle CT and VM on a single storage
2051 # see: https://pve.proxmox.com/pipermail/pve-devel/2018-July/032900.html
2052 foreach my $stor (keys %{$cfg->{ids
}}) {
2053 delete($cfg->{ids
}->{$stor}) if !$cfg->{ids
}->{$stor}->{content
}->{rootdir
};
2056 print "rescan volumes...\n";
2057 my $all_volumes = scan_volids
($cfg, $vmid);
2059 my $updatefn = sub {
2063 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2065 PVE
::LXC
::Config-
>check_lock($conf);
2068 foreach my $volid (keys %$all_volumes) {
2069 my $info = $all_volumes->{$volid};
2070 $vm_volids->{$volid} = $info if $info->{vmid
} == $vmid;
2073 my $upu = update_unused
($vmid, $conf, $vm_volids);
2074 my $upd = update_disksize
($vmid, $conf, $vm_volids);
2075 $changes = $upu || $upd;
2077 PVE
::LXC
::Config-
>write_config($vmid, $conf) if $changes && !$dryrun;
2080 if (defined($vmid)) {
2084 PVE
::LXC
::Config-
>lock_config($vmid, $updatefn, $vmid);
2087 my $vmlist = config_list
();
2088 foreach my $vmid (keys %$vmlist) {
2092 PVE
::LXC
::Config-
>lock_config($vmid, $updatefn, $vmid);
2099 # bash completion helper
2101 sub complete_os_templates
{
2102 my ($cmdname, $pname, $cvalue) = @_;
2104 my $cfg = PVE
::Storage
::config
();
2108 if ($cvalue =~ m/^([^:]+):/) {
2112 my $vtype = $cmdname eq 'restore' ?
'backup' : 'vztmpl';
2113 my $data = PVE
::Storage
::template_list
($cfg, $storeid, $vtype);
2116 foreach my $id (keys %$data) {
2117 foreach my $item (@{$data->{$id}}) {
2118 push @$res, $item->{volid
} if defined($item->{volid
});
2125 my $complete_ctid_full = sub {
2128 my $idlist = vmstatus
();
2130 my $active_hash = list_active_containers
();
2134 foreach my $id (keys %$idlist) {
2135 my $d = $idlist->{$id};
2136 if (defined($running)) {
2137 next if $d->{template
};
2138 next if $running && !$active_hash->{$id};
2139 next if !$running && $active_hash->{$id};
2148 return &$complete_ctid_full();
2151 sub complete_ctid_stopped
{
2152 return &$complete_ctid_full(0);
2155 sub complete_ctid_running
{
2156 return &$complete_ctid_full(1);
2166 my $lxc = $conf->{lxc
};
2167 foreach my $entry (@$lxc) {
2168 my ($key, $value) = @$entry;
2169 # FIXME: remove the 'id_map' variant when lxc-3.0 arrives
2170 next if $key ne 'lxc.idmap' && $key ne 'lxc.id_map';
2171 if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) {
2172 my ($type, $ct, $host, $length) = ($1, $2, $3, $4);
2173 push @$id_map, [$type, $ct, $host, $length];
2175 $rootuid = $host if $type eq 'u';
2176 $rootgid = $host if $type eq 'g';
2179 die "failed to parse idmap: $value\n";
2183 if (!@$id_map && $conf->{unprivileged
}) {
2184 # Should we read them from /etc/subuid?
2185 $id_map = [ ['u', '0', '100000', '65536'],
2186 ['g', '0', '100000', '65536'] ];
2187 $rootuid = $rootgid = 100000;
2190 return ($id_map, $rootuid, $rootgid);
2193 sub userns_command
{
2196 return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--'];
2202 my ($vmid, $conf, $skiplock) = @_;
2204 # apply pending changes while starting
2205 if (scalar(keys %{$conf->{pending
}})) {
2206 my $storecfg = PVE
::Storage
::config
();
2207 PVE
::LXC
::Config-
>vmconfig_apply_pending($vmid, $conf, $storecfg);
2208 $conf = PVE
::LXC
::Config-
>load_config($vmid); # update/reload
2211 update_lxc_config
($vmid, $conf);
2213 my $skiplock_flag_fn = "/run/lxc/skiplock-$vmid";
2216 open(my $fh, '>', $skiplock_flag_fn) || die "failed to open $skiplock_flag_fn for writing: $!\n";
2220 my $cmd = ['systemctl', 'start', "pve-container\@$vmid"];
2222 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'pre-start', 1);
2223 eval { PVE
::Tools
::run_command
($cmd); };
2225 unlink $skiplock_flag_fn;
2228 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'post-start');
2233 # Helper to stop a container completely and make sure it has stopped completely.
2234 # This is necessary because we want the post-stop hook to have completed its
2235 # unmount-all step, but post-stop happens after lxc puts the container into the
2237 # $kill - if true it will always do an immediate hard-stop
2238 # $shutdown_timeout - the timeout to wait for a gracefull shutdown
2239 # $kill_after_timeout - if true, send a hardstop if shutdown timed out
2241 my ($vmid, $kill, $shutdown_timeout, $kill_after_timeout) = @_;
2243 # Open the container's command socket.
2244 my $path = "\0/var/lib/lxc/$vmid/command";
2245 my $sock = IO
::Socket
::UNIX-
>new(
2246 Type
=> SOCK_STREAM
(),
2250 return if $! == ECONNREFUSED
; # The container is not running
2251 die "failed to open container ${vmid}'s command socket: $!\n";
2254 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2255 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'pre-stop');
2257 # Stop the container:
2259 my $cmd = ['lxc-stop', '-n', $vmid];
2262 push @$cmd, '--kill'; # doesn't allow timeouts
2264 # lxc-stop uses a default timeout
2265 push @$cmd, '--nokill' if !$kill_after_timeout;
2267 if (defined($shutdown_timeout)) {
2268 push @$cmd, '--timeout', $shutdown_timeout;
2269 # Give run_command 5 extra seconds
2270 $shutdown_timeout += 5;
2274 eval { PVE
::Tools
::run_command
($cmd, timeout
=> $shutdown_timeout) };
2279 my $result = <$sock>;
2281 return if !defined $result; # monitor is gone and the ct has stopped.
2282 die "container did not stop\n";
2286 my ($vmid, $timeout, $skiplock) = @_;
2288 PVE
::LXC
::Config-
>lock_config($vmid, sub {
2289 return if !check_running
($vmid);
2291 vm_stop
($vmid, 0, $timeout, 1); # kill if timeout exceeds
2293 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2294 vm_start
($vmid, $conf);
2301 return PVE
::Tools
::run_fork
(sub {
2302 # Unshare the mount namespace
2303 die "failed to unshare mount namespace: $!\n"
2304 if !PVE
::Tools
::unshare
(PVE
::Tools
::CLONE_NEWNS
);
2305 PVE
::Tools
::run_command
(['mount', '--make-rslave', '/']);
2310 my $copy_volume = sub {
2311 my ($src_volid, $src, $dst_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid) = @_;
2313 my $src_mp = { volume
=> $src_volid, mp
=> '/', ro
=> 1 };
2314 $src_mp->{type
} = PVE
::LXC
::Config-
>classify_mountpoint($src_volid);
2316 my $dst_mp = { volume
=> $dst_volid, mp
=> '/', ro
=> 0 };
2317 $dst_mp->{type
} = PVE
::LXC
::Config-
>classify_mountpoint($dst_volid);
2323 mountpoint_mount
($src_mp, $src, $storage_cfg, $snapname, $rootuid, $rootgid);
2324 push @mounted, $src;
2326 mountpoint_mount
($dst_mp, $dest, $storage_cfg, undef, $rootuid, $rootgid);
2327 push @mounted, $dest;
2331 PVE
::Tools
::run_command
(['/usr/bin/rsync', '--stats', '-X', '-A', '--numeric-ids',
2332 '-aH', '--whole-file', '--sparse', '--one-file-system',
2333 "--bwlimit=$bwlimit", "$src/", $dest]);
2337 # Wait for rsync's children to release dest so that
2338 # consequent file operations (umount, remove) are possible
2339 while ((system {"fuser"} "fuser", "-s", $dest) == 0) {sleep 1};
2341 foreach my $mount (reverse @mounted) {
2342 eval { PVE
::Tools
::run_command
(['/bin/umount', $mount], errfunc
=> sub{})};
2343 warn "Can't umount $mount\n" if $@;
2346 # If this fails they're used as mount points in a concurrent operation
2347 # (which should not happen but there's also no real need to get rid of them).
2354 # Should not be called after unsharing the mount namespace!
2356 my ($mp, $vmid, $storage, $storage_cfg, $conf, $snapname, $bwlimit) = @_;
2358 die "cannot copy volumes of type $mp->{type}\n" if $mp->{type
} ne 'volume';
2359 File
::Path
::make_path
("/var/lib/lxc/$vmid");
2360 my $dest = "/var/lib/lxc/$vmid/.copy-volume-1";
2361 my $src = "/var/lib/lxc/$vmid/.copy-volume-2";
2363 # get id's for unprivileged container
2364 my (undef, $rootuid, $rootgid) = parse_id_maps
($conf);
2366 # Allocate the disk before unsharing in order to make sure zfs subvolumes
2367 # are visible in this namespace, otherwise the host only sees the empty
2368 # (not-mounted) directory.
2371 # Make sure $mp contains a correct size.
2372 $mp->{size
} = PVE
::Storage
::volume_size_info
($storage_cfg, $mp->{volume
});
2374 ($new_volid, $needs_chown) = alloc_disk
($storage_cfg, $vmid, $storage, $mp->{size
}/1024, $rootuid, $rootgid);
2376 PVE
::Storage
::activate_volumes
($storage_cfg, [$new_volid], undef);
2377 my $path = PVE
::Storage
::path
($storage_cfg, $new_volid, undef);
2378 chown($rootuid, $rootgid, $path);
2382 $copy_volume->($mp->{volume
}, $src, $new_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid);
2386 PVE
::Storage
::vdisk_free
($storage_cfg, $new_volid)
2387 if defined($new_volid);