7 use Errno
qw(ELOOP ENOTDIR EROFS ECONNREFUSED ENOSYS EEXIST);
8 use Fcntl
qw(O_RDONLY O_WRONLY O_NOFOLLOW O_DIRECTORY);
11 use IO
::Poll
qw(POLLIN POLLHUP);
15 use Time
::HiRes qw
(gettimeofday
);
17 use PVE
::AccessControl
;
20 use PVE
::Exception
qw(raise_perm_exc);
21 use PVE
::GuestHelpers
qw(safe_string_ne safe_num_ne safe_boolean_ne);
23 use PVE
::JSONSchema
qw(get_standard_option);
26 use PVE
::RESTEnvironment
;
39 use PVE
::Syscall
qw(:fsmount);
43 use PVE
::LXC
::Monitor
;
48 require PVE
::Network
::SDN
::Zones
;
52 my $LXC_CONFIG_PATH = '/usr/share/lxc/config';
54 my $nodename = PVE
::INotify
::nodename
();
56 my $cpuinfo= PVE
::ProcFSTools
::read_cpuinfo
();
58 our $NEW_DISK_RE = qr/^([^:\s]+):(\d+(\.\d+)?)$/;
61 my $vmlist = PVE
::Cluster
::get_vmlist
();
63 return $res if !$vmlist || !$vmlist->{ids
};
64 my $ids = $vmlist->{ids
};
66 foreach my $vmid (keys %$ids) {
67 next if !$vmid; # skip CT0
68 my $d = $ids->{$vmid};
69 next if !$d->{node
} || $d->{node
} ne $nodename;
70 next if !$d->{type
} || $d->{type
} ne 'lxc';
71 $res->{$vmid} = { type
=> 'lxc', vmid
=> $vmid };
76 # container status helpers
78 sub list_active_containers
{
80 my $filename = "/proc/net/unix";
82 # similar test is used by lcxcontainers.c: list_active_containers
85 my $fh = IO
::File-
>new ($filename, "r");
88 while (defined(my $line = <$fh>)) {
89 if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) {
91 if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) {
102 # warning: this is slow
106 my $active_hash = list_active_containers
();
108 return 1 if defined($active_hash->{$vmid});
113 sub get_container_disk_usage
{
114 my ($vmid, $pid) = @_;
116 return PVE
::Tools
::df
("/proc/$pid/root/", 1);
119 my $last_proc_vmid_stat;
121 our $vmstatus_return_properties = {
122 vmid
=> get_standard_option
('pve-vmid'),
124 description
=> "LXC Container status.",
126 enum
=> ['stopped', 'running'],
129 description
=> "Maximum memory in bytes.",
135 description
=> "Maximum SWAP memory in bytes.",
141 description
=> "Root disk size in bytes.",
147 description
=> "Container name.",
152 description
=> "Uptime.",
155 renderer
=> 'duration',
158 description
=> "Maximum usable CPUs.",
163 description
=> "The current config lock, if any.",
168 description
=> "The current configured tags, if any.",
177 my $list = $opt_vmid ?
{ $opt_vmid => { type
=> 'lxc', vmid
=> int($opt_vmid) }} : config_list
();
179 my $active_hash = list_active_containers
();
181 my $cpucount = $cpuinfo->{cpus
} || 1;
183 my $cdtime = gettimeofday
;
185 my $uptime = (PVE
::ProcFSTools
::read_proc_uptime
(1))[0];
186 my $clock_ticks = POSIX
::sysconf
(&POSIX
::_SC_CLK_TCK
);
188 my $unprivileged = {};
190 foreach my $vmid (keys %$list) {
191 my $d = $list->{$vmid};
193 eval { $d->{pid
} = int(find_lxc_pid
($vmid)) if defined($active_hash->{$vmid}); };
194 warn $@ if $@; # ignore errors (consider them stopped)
196 $d->{status
} = $active_hash->{$vmid} ?
'running' : 'stopped';
198 my $cfspath = PVE
::LXC
::Config-
>cfs_config_path($vmid);
199 my $conf = PVE
::Cluster
::cfs_read_file
($cfspath) || {};
201 $unprivileged->{$vmid} = $conf->{unprivileged
};
203 $d->{name
} = $conf->{'hostname'} || "CT$vmid";
204 $d->{name
} =~ s/[\s]//g;
206 $d->{cpus
} = $conf->{cores
} || $conf->{cpulimit
};
207 $d->{cpus
} = $cpucount if !$d->{cpus
};
209 $d->{tags
} = $conf->{tags
} if defined($conf->{tags
});
212 my $res = get_container_disk_usage
($vmid, $d->{pid
});
213 $d->{disk
} = int($res->{used
});
214 $d->{maxdisk
} = int($res->{total
});
217 # use 4GB by default ??
218 if (my $rootfs = $conf->{rootfs
}) {
219 my $rootinfo = PVE
::LXC
::Config-
>parse_volume('rootfs', $rootfs);
220 $d->{maxdisk
} = $rootinfo->{size
} || (4*1024*1024*1024);
222 $d->{maxdisk
} = 4*1024*1024*1024;
228 $d->{maxmem
} = ($conf->{memory
}||512)*1024*1024;
229 $d->{maxswap
} = ($conf->{swap
}//0)*1024*1024;
240 $d->{template
} = 1 if PVE
::LXC
::Config-
>is_template($conf);
241 $d->{lock} = $conf->{lock} if $conf->{lock};
244 foreach my $vmid (keys %$list) {
245 my $d = $list->{$vmid};
248 next if !$pid; # skip stopped CTs
250 my $proc_pid_stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
251 $d->{uptime
} = int(($uptime - $proc_pid_stat->{starttime
}) / $clock_ticks); # the method lxcfs uses
253 my $unpriv = $unprivileged->{$vmid};
255 my $cgroups = PVE
::LXC
::CGroup-
>new($vmid);
257 if (defined(my $mem = $cgroups->get_memory_stat())) {
258 $d->{mem
} = int($mem->{mem
});
259 $d->{swap
} = int($mem->{swap
});
265 if (defined(my $blkio = $cgroups->get_io_stats())) {
266 $d->{diskread
} = int($blkio->{diskread
});
267 $d->{diskwrite
} = int($blkio->{diskwrite
});
273 if (defined(my $cpu = $cgroups->get_cpu_stat())) {
274 # Total time (in milliseconds) used up by the cpu.
275 my $used_ms = $cpu->{utime} + $cpu->{stime
};
277 my $old = $last_proc_vmid_stat->{$vmid};
279 $last_proc_vmid_stat->{$vmid} = {
287 my $delta_ms = ($cdtime - $old->{time}) * $cpucount * 1000.0;
288 if ($delta_ms > 1000.0) {
289 my $delta_used_ms = $used_ms - $old->{used
};
290 $d->{cpu
} = (($delta_used_ms / $delta_ms) * $cpucount) / $d->{cpus
};
291 $last_proc_vmid_stat->{$vmid} = {
297 $d->{cpu
} = $old->{cpu
};
304 my $netdev = PVE
::ProcFSTools
::read_proc_net_dev
();
306 foreach my $dev (keys %$netdev) {
307 next if $dev !~ m/^veth([1-9]\d*)i/;
309 my $d = $list->{$vmid};
313 $d->{netout
} += $netdev->{$dev}->{receive
};
314 $d->{netin
} += $netdev->{$dev}->{transmit
};
321 sub find_lxc_console_pids
{
325 PVE
::Tools
::dir_glob_foreach
('/proc', '\d+', sub {
328 my $cmdline = PVE
::Tools
::file_read_firstline
("/proc/$pid/cmdline");
331 my @args = split(/\0/, $cmdline);
333 # search for lxc-console -n <vmid>
334 return if scalar(@args) != 3;
335 return if $args[1] ne '-n';
336 return if $args[2] !~ m/^\d+$/;
337 return if $args[0] !~ m
|^(/usr/bin
/)?lxc-console
$|;
341 push @{$res->{$vmid}}, $pid;
353 $pid = $1 if $line =~ m/^PID:\s+(\d+)$/;
355 PVE
::Tools
::run_command
(['lxc-info', '-n', $vmid, '-p'], outfunc
=> $parser);
357 die "unable to get PID for CT $vmid (not running?)\n" if !$pid;
364 sysopen(my $fd, "/proc/$pid", O_RDONLY
| O_DIRECTORY
)
365 or die "failed to open /proc/$pid pid fd\n";
372 # Find the pid and open:
373 my $pid = find_lxc_pid
($vmid);
374 my $fd = open_pid_fd
($pid);
377 my $pid2 = find_lxc_pid
($vmid);
379 return () if $pid != $pid2;
386 # Find the parent pid via proc and open it:
387 my $stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
388 my $ppid = $stat->{ppid
} // die "failed to get parent pid\n";
390 my $fd = open_pid_fd
($ppid);
393 $stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
394 my $ppid2 = $stat->{ppid
} // die "failed to get parent pid for verification\n";
396 return () if $ppid != $ppid2;
400 # Note: we cannot use Net:IP, because that only allows strict
402 sub parse_ipv4_cidr
{
403 my ($cidr, $noerr) = @_;
405 if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) && ($2 <= 32)) {
406 return { address
=> $1, netmask
=> $PVE::Network
::ipv4_reverse_mask-
>[$2] };
409 return undef if $noerr;
411 die "unable to parse ipv4 address/mask\n";
414 # With seccomp trap to userspace we now have the ability to optionally forward
415 # certain syscalls to the "host" to handle (via our pve-lxc-syscalld daemon).
417 # This means that there are cases where we need to create an extra seccomp
418 # profile for the container to load.
420 # This returns a configuration snippet added to the raw lxc config.
421 sub make_seccomp_config
{
422 my ($conf, $vmid, $conf_dir, $unprivileged, $features) = @_;
423 # User-configured profile has precedence, note that the user's entry would
424 # be written 'after' this line anyway...
425 if (PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.seccomp.profile')) {
426 # Warn the user if this conflicts with a feature:
427 my $warn = join(', ', grep { $features->{$_} } qw(keyctl mknod));
428 warn "explicitly configured lxc.seccomp.profile overrides the following settings: $warn\n"
429 if length($warn) > 0;
433 # Privileged containers keep using the default (which is already part of
434 # the files included via lxc.include, so we don't need to write it out,
435 # that way it stays admin-configurable via /usr/share/lxc/config/... as
437 return '' if !$unprivileged;
440 keyctl
=> ['errno 38'],
442 # Disable btrfs ioctrls since they don't work particularly well in user namespaces.
443 # Particularly, without the mount option to enable rmdir removing snapshots, user
444 # namespaces can create snapshots but neither `show` or `delete` them, which is quite
445 # horrible, so for now, just disable this entirely:
447 # BTRFS_IOCTL_MAGIC 0x94, _IOC type shift is 8,
448 # so `(req & 0xFF00) == 0x9400` is a btrfs ioctl and gets an EPERM
449 ioctl => ['errno 1 [1,0x9400,SCMP_CMP_MASKED_EQ,0xff00]'],
454 # Unprivileged containers will get keyctl() disabled by default as a
455 # workaround for systemd-networkd behavior. But we have an option to
456 # explicitly enable it:
457 if ($features->{keyctl
}) {
458 delete $rules->{keyctl
};
461 # By default, unprivileged containers cannot use `mknod` at all.
462 # Since lxc 3.2, we can use seccomp's trap to userspace feature for this,
463 # but for now this is experimental, so it has to be enabled via a feature
465 # Note that we only handle block and char devices (like lxd), the rest we
466 # leave up to the kernel. We may in the future remove this if seccomp gets
467 # a way to tell the kernel to "continue" a syscall.
468 if ($features->{mknod
}) {
469 my ($ok, $kernel) = PVE
::ProcFSTools
::check_kernel_release
(5, 3);
471 die "'mknod' feature requested, but kernel too old (found $kernel, required >= 5.3)\n";
474 $raw_conf .= "lxc.seccomp.notify.proxy = unix:/run/pve/lxc-syscalld.sock\n";
475 $raw_conf .= "lxc.seccomp.notify.cookie = $vmid\n";
478 # condition: (mode & S_IFMT) == S_IFCHR
479 'notify [1,8192,SCMP_CMP_MASKED_EQ,61440]',
480 # condition: (mode & S_IFMT) == S_IFBLK
481 'notify [1,24576,SCMP_CMP_MASKED_EQ,61440]',
483 $rules->{mknodat
} = [
484 # condition: (mode & S_IFMT) == S_IFCHR
485 'notify [2,8192,SCMP_CMP_MASKED_EQ,61440]',
486 # condition: (mode & S_IFMT) == S_IFBLK
487 'notify [2,24576,SCMP_CMP_MASKED_EQ,61440]',
491 # Now build the custom seccomp rule text...
492 my $extra_rules = join("\n", map {
494 map { "$syscall $_" } $rules->{$syscall}->@*
495 } sort keys %$rules) . "\n";
497 return $raw_conf if $extra_rules eq "\n";
499 # We still have the "most common" config readily available, so don't write
501 if ($raw_conf eq '' && $extra_rules eq "keyctl errno 38\n") {
502 # we have no extra $raw_conf and use the same we had in pve 6.1:
503 return "lxc.seccomp.profile = $LXC_CONFIG_PATH/pve-userns.seccomp\n";
506 # Write the rule file to the container's config path:
507 my $rule_file = "$conf_dir/rules.seccomp";
508 my $rule_data = file_get_contents
("$LXC_CONFIG_PATH/common.seccomp")
510 file_set_contents
($rule_file, $rule_data);
511 $raw_conf .= "lxc.seccomp.profile = $rule_file\n";
516 # Since lxc-3.0.2 we can have lxc generate a profile for the container
517 # automatically. The default should be equivalent to the old
518 # `lxc-container-default-cgns` profile.
520 # Additionally this also added `lxc.apparmor.raw` which can be used to inject
521 # additional lines into the profile. We can use that to allow mounting specific
523 sub make_apparmor_config
{
524 my ($conf, $unprivileged, $features) = @_;
526 # user-configured profile has precedence, but first we go through our own
527 # code to figure out whether we should warn the user:
529 my $raw = "lxc.apparmor.profile = generated\n";
532 if ($features->{fuse
}) {
533 # For the informational warning:
534 push @profile_uses, 'features:fuse';
537 # There's lxc.apparmor.allow_nesting now, which will add the necessary
538 # apparmor lines, create an apparmor namespace for the container, but also
539 # adds proc and sysfs mounts to /dev/.lxc/{proc,sys}. These do not have
540 # lxcfs mounted over them, because that would prevent the container from
541 # mounting new instances of them for nested containers.
542 if ($features->{nesting
}) {
543 push @profile_uses, 'features:nesting';
544 $raw .= "lxc.apparmor.allow_nesting = 1\n"
546 # In the default profile in /etc/apparmor.d we patch this in because
547 # otherwise a container can for example run `chown` on /sys, breaking
548 # access to it for non-CAP_DAC_OVERRIDE tools on the host:
549 $raw .= "lxc.apparmor.raw = deny mount -> /proc/,\n";
550 $raw .= "lxc.apparmor.raw = deny mount -> /sys/,\n";
551 # Preferably we could use the 'remount' flag but this does not sit well
552 # with apparmor_parser currently:
553 # mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,
556 if (my $mount = $features->{mount
}) {
557 push @profile_uses, 'features:mount';
558 foreach my $fs (PVE
::Tools
::split_list
($mount)) {
559 $raw .= "lxc.apparmor.raw = mount fstype=$fs,\n";
565 if (PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.apparmor.profile')) {
566 if (length(my $used = join(', ', @profile_uses))) {
567 warn "explicitly configured lxc.apparmor.profile overrides the following settings: $used\n";
575 sub update_lxc_config
{
576 my ($vmid, $conf) = @_;
578 my $dir = "/var/lib/lxc/$vmid";
580 if ($conf->{template
}) {
582 unlink "$dir/config";
587 my ($lxc_major, $lxc_minor) = get_lxc_version
();
591 if ($lxc_major >= 4) {
592 # Explicitly don't use relative directories, which is the default, but
593 # note that we do this mostly because they are only applied for *some*
594 # cgroups. Our pve-container@.service now starts lxc-start with `-F`,
595 # so we also don't need to worry about the new monitor cgroup to
597 $raw .= "lxc.cgroup.relative = 0\n";
599 # To make things easier, let's keep our previous cgroup layout and
600 # simply move the monitor outside:
601 $raw .= "lxc.cgroup.dir.monitor = lxc.monitor/$vmid\n";
602 # cgroup namespace separation for stronger limits:
603 $raw .= "lxc.cgroup.dir.container = lxc/$vmid\n";
604 $raw .= "lxc.cgroup.dir.container.inner = ns\n";
607 die "missing 'arch' - internal error" if !$conf->{arch
};
608 $raw .= "lxc.arch = $conf->{arch}\n";
610 my $custom_idmap = PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.idmap');
611 my $unprivileged = $conf->{unprivileged
} || $custom_idmap;
613 my $ostype = $conf->{ostype
} || die "missing 'ostype' - internal error";
615 File
::Path
::mkpath
($dir);
617 my $cfgpath = '/usr/share/lxc/config';
618 my $inc = "$cfgpath/$ostype.common.conf";
619 $inc ="$cfgpath/common.conf" if !-f
$inc;
620 $raw .= "lxc.include = $inc\n";
622 $inc = "$cfgpath/$ostype.userns.conf";
623 $inc = "$cfgpath/userns.conf" if !-f
$inc;
624 $raw .= "lxc.include = $inc\n";
627 my $features = PVE
::LXC
::Config-
>parse_features($conf->{features
});
629 $raw .= make_seccomp_config
($conf, $vmid, $dir, $unprivileged, $features);
630 $raw .= make_apparmor_config
($conf, $unprivileged, $features);
631 if ($features->{fuse
}) {
632 $raw .= "lxc.apparmor.raw = mount fstype=fuse,\n";
633 $raw .= "lxc.mount.entry = /dev/fuse dev/fuse none bind,create=file 0 0\n";
636 if ($unprivileged && !$features->{force_rw_sys
}) {
637 # unpriv. CT default to sys:rw, but that doesn't always plays well with
638 # systemd, e.g., systemd-networkd https://systemd.io/CONTAINER_INTERFACE/
639 $raw .= "lxc.mount.auto = sys:mixed\n";
642 # WARNING: DO NOT REMOVE this without making sure that loop device nodes
643 # cannot be exposed to the container with r/w access (cgroup perms).
644 # When this is enabled mounts will still remain in the monitor's namespace
645 # after the container unmounted them and thus will not detach from their
646 # files while the container is running!
647 $raw .= "lxc.monitor.unshare = 1\n";
649 my ($cgv1, $cgv2) = PVE
::CGroup
::get_cgroup_controllers
();
651 # Should we read them from /etc/subuid?
652 if ($unprivileged && !$custom_idmap) {
653 $raw .= "lxc.idmap = u 0 100000 65536\n";
654 $raw .= "lxc.idmap = g 0 100000 65536\n";
657 if (!PVE
::LXC
::Config-
>has_dev_console($conf)) {
658 $raw .= "lxc.console.path = none\n";
659 if ($cgv1->{devices
}) {
660 $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n";
661 } elsif (defined($cgv2)) {
662 $raw .= "lxc.cgroup2.devices.deny = c 5:1 rwm\n";
666 my $ttycount = PVE
::LXC
::Config-
>get_tty_count($conf);
667 $raw .= "lxc.tty.max = $ttycount\n";
669 # some init scripts expect a linux terminal (turnkey).
670 $raw .= "lxc.environment = TERM=linux\n";
672 my $utsname = $conf->{hostname
} || "CT$vmid";
673 $raw .= "lxc.uts.name = $utsname\n";
675 if ($cgv1->{memory
}) {
676 my $memory = $conf->{memory
} || 512;
677 my $swap = $conf->{swap
} // 0;
679 my $lxcmem = int($memory*1024*1024);
680 $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n";
682 my $lxcswap = int(($memory + $swap)*1024*1024);
683 $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n";
684 } elsif ($cgv2->{memory
}) {
685 my $memory = $conf->{memory
} || 512;
686 my $swap = $conf->{swap
} // 0;
688 # cgroup memory usage is limited by the hard 'max' limit (OOM-killer enforced) and the soft
689 # 'high' limit (cgroup processes get throttled and put under heavy reclaim pressure).
690 my ($lxc_mem_max, $lxc_mem_high) = PVE
::LXC
::Config
::calculate_memory_constraints
($memory);
691 $raw .= "lxc.cgroup2.memory.max = $lxc_mem_max\n";
692 $raw .= "lxc.cgroup2.memory.high = $lxc_mem_high\n";
694 my $lxcswap = int($swap*1024*1024);
695 $raw .= "lxc.cgroup2.memory.swap.max = $lxcswap\n";
699 if (my $cpulimit = $conf->{cpulimit
}) {
700 $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n";
701 my $value = int(100000*$cpulimit);
702 $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n";
705 my $shares = PVE
::CGroup
::clamp_cpu_shares
($conf->{cpuunits
});
706 $raw .= "lxc.cgroup.cpu.shares = $shares\n";
707 } elsif ($cgv2->{cpu
}) {
709 if (my $cpulimit = $conf->{cpulimit
}) {
710 my $value = int(100000*$cpulimit);
711 $raw .= "lxc.cgroup2.cpu.max = $value 100000\n";
714 if (defined(my $shares = $conf->{cpuunits
})) {
715 $shares = PVE
::CGroup
::clamp_cpu_shares
($shares);
716 $raw .= "lxc.cgroup2.cpu.weight = $shares\n";
720 die "missing 'rootfs' configuration\n"
721 if !defined($conf->{rootfs
});
723 my $mountpoint = PVE
::LXC
::Config-
>parse_volume('rootfs', $conf->{rootfs
});
725 $raw .= "lxc.rootfs.path = $dir/rootfs\n";
727 foreach my $k (sort keys %$conf) {
728 next if $k !~ m/^net(\d+)$/;
730 my $d = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$k});
731 $raw .= "lxc.net.$ind.type = veth\n";
732 $raw .= "lxc.net.$ind.veth.pair = veth${vmid}i${ind}\n";
733 $raw .= "lxc.net.$ind.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr
});
734 $raw .= "lxc.net.$ind.name = $d->{name}\n" if defined($d->{name
});
736 # Keep container from starting with invalid mtu configuration
737 if (my $mtu = $d->{mtu
}) {
738 my $bridge_mtu = PVE
::Network
::read_bridge_mtu
($d->{bridge
});
739 die "$k: MTU size '$mtu' is bigger than bridge MTU '$bridge_mtu'\n"
740 if ($mtu > $bridge_mtu);
742 $raw .= "lxc.net.$ind.mtu = $mtu\n";
745 # Starting with lxc 4.0, we do not patch lxc to execute our up-scripts.
746 if ($lxc_major >= 4) {
747 $raw .= "lxc.net.$ind.script.up = /usr/share/lxc/lxcnetaddbr\n";
752 if (my $lxcconf = $conf->{lxc
}) {
753 foreach my $entry (@$lxcconf) {
754 my ($k, $v) = @$entry;
755 $had_cpuset = 1 if $k eq 'lxc.cgroup.cpuset.cpus' || $k eq 'lxc.cgroup2.cpuset.cpus';
761 my ($cpuset_cgroup, $cpuset_version) = eval { PVE
::CGroup
::cpuset_controller_path
() };
762 if (defined($cpuset_cgroup)) {
763 $cpuset = eval { PVE
::CpuSet-
>new_from_path("$cpuset_cgroup/lxc", 1) }
764 || PVE
::CpuSet-
>new_from_path($cpuset_cgroup, 1);
766 my $cores = $conf->{cores
};
767 if (!$had_cpuset && $cores && $cpuset) {
768 my @members = $cpuset->members();
769 while (scalar(@members) > $cores) {
770 my $randidx = int(rand(scalar(@members)));
771 $cpuset->delete($members[$randidx]);
772 splice(@members, $randidx, 1); # keep track of the changes
774 my $ver = $cpuset_version == 1 ?
'' : '2';
775 $raw .= "lxc.cgroup$ver.cpuset.cpus = ".$cpuset->short_string()."\n";
778 File
::Path
::mkpath
("$dir/rootfs");
780 PVE
::Tools
::file_set_contents
("$dir/config", $raw);
783 # verify and cleanup nameserver list (replace \0 with ' ')
784 sub verify_nameserver_list
{
785 my ($nameserver_list) = @_;
788 foreach my $server (PVE
::Tools
::split_list
($nameserver_list)) {
789 PVE
::LXC
::Config
::verify_ip_with_ll_iface
($server);
793 return join(' ', @list);
796 sub verify_searchdomain_list
{
797 my ($searchdomain_list) = @_;
800 foreach my $server (PVE
::Tools
::split_list
($searchdomain_list)) {
801 # todo: should we add checks for valid dns domains?
805 return join(' ', @list);
808 sub get_console_command
{
809 my ($vmid, $conf, $escapechar) = @_;
811 # '-1' as $escapechar disables keyboard escape sequence
812 # any other passed char (a-z) will result in <Ctrl+$escapechar q>
814 my $cmode = PVE
::LXC
::Config-
>get_cmode($conf);
817 if ($cmode eq 'console') {
818 push @$cmd, 'lxc-console', '-n', $vmid, '-t', 0;
819 push @$cmd, '-e', $escapechar if $escapechar;
820 } elsif ($cmode eq 'tty') {
821 push @$cmd, 'lxc-console', '-n', $vmid;
822 push @$cmd, '-e', $escapechar if $escapechar;
823 } elsif ($cmode eq 'shell') {
824 push @$cmd, 'lxc-attach', '--clear-env', '-n', $vmid;
826 die "internal error";
832 sub get_primary_ips
{
835 # return data from net0
837 return undef if !defined($conf->{net0
});
838 my $net = PVE
::LXC
::Config-
>parse_lxc_network($conf->{net0
});
840 my $ipv4 = $net->{ip
};
842 if ($ipv4 =~ /^(dhcp|manual)$/) {
848 my $ipv6 = $net->{ip6
};
850 if ($ipv6 =~ /^(auto|dhcp|manual)$/) {
857 return ($ipv4, $ipv6);
860 sub delete_mountpoint_volume
{
861 my ($storage_cfg, $vmid, $volume) = @_;
863 return if PVE
::LXC
::Config-
>classify_mountpoint($volume) ne 'volume';
865 my ($vtype, $name, $owner) = PVE
::Storage
::parse_volname
($storage_cfg, $volume);
867 if ($vmid == $owner) {
868 PVE
::Storage
::vdisk_free
($storage_cfg, $volume);
870 warn "ignore deletion of '$volume', CT $vmid isn't the owner!\n";
874 sub destroy_lxc_container
{
875 my ($storage_cfg, $vmid, $conf, $replacement_conf, $purge_unreferenced) = @_;
878 my $remove_volume = sub {
879 my ($ms, $mountpoint) = @_;
881 my $volume = $mountpoint->{volume
};
883 return if $volids->{$volume};
884 $volids->{$volume} = 1;
886 delete_mountpoint_volume
($storage_cfg, $vmid, $volume);
888 PVE
::LXC
::Config-
>foreach_volume_full($conf, {include_unused
=> 1}, $remove_volume);
890 PVE
::LXC
::Config-
>foreach_volume_full($conf->{pending
}, {include_unused
=> 1}, $remove_volume);
892 if ($purge_unreferenced) { # also remove unreferenced disk
893 my $vmdisks = PVE
::Storage
::vdisk_list
($storage_cfg, undef, $vmid, undef, 'rootdir');
894 PVE
::Storage
::foreach_volid
($vmdisks, sub {
895 my ($volid, $sid, $volname, $d) = @_;
896 eval { PVE
::Storage
::vdisk_free
($storage_cfg, $volid) };
901 rmdir "/var/lib/lxc/$vmid/rootfs";
902 unlink "/var/lib/lxc/$vmid/config";
903 rmdir "/var/lib/lxc/$vmid";
904 if (defined $replacement_conf) {
905 PVE
::LXC
::Config-
>write_config($vmid, $replacement_conf);
907 PVE
::LXC
::Config-
>destroy_config($vmid);
911 sub vm_stop_cleanup
{
912 my ($storage_cfg, $vmid, $conf, $keepActive) = @_;
914 return if $keepActive;
917 my $vollist = PVE
::LXC
::Config-
>get_vm_volumes($conf);
918 PVE
::Storage
::deactivate_volumes
($storage_cfg, $vollist);
920 warn $@ if $@; # avoid errors - just warn
923 sub net_tap_plug
: prototype($$) {
924 my ($iface, $net) = @_;
926 if (defined($net->{link_down
})) {
927 PVE
::Tools
::run_command
(['/sbin/ip', 'link', 'set', 'dev', $iface, 'down']);
928 # Don't add disconnected interfaces to the bridge, otherwise e.g. applying any network
929 # change (e.g. `ifreload -a`) could (re-)activate it unintentionally.
933 my ($bridge, $tag, $firewall, $trunks, $rate, $hwaddr) =
934 $net->@{'bridge', 'tag', 'firewall', 'trunks', 'rate', 'hwaddr'};
937 PVE
::Network
::SDN
::Zones
::tap_plug
($iface, $bridge, $tag, $firewall, $trunks, $rate);
938 PVE
::Network
::SDN
::Zones
::add_bridge_fdb
($iface, $hwaddr, $bridge, $firewall);
940 PVE
::Network
::tap_plug
($iface, $bridge, $tag, $firewall, $trunks, $rate, { mac
=> $hwaddr });
943 PVE
::Tools
::run_command
(['/sbin/ip', 'link', 'set', 'dev', $iface, 'up']);
947 my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_;
949 if ($newnet->{type
} ne 'veth') {
950 # for when there are physical interfaces
951 die "cannot update interface of type $newnet->{type}";
954 my $veth = "veth${vmid}i${netid}";
955 my $eth = $newnet->{name
};
957 if (my $oldnetcfg = $conf->{$opt}) {
958 my $oldnet = PVE
::LXC
::Config-
>parse_lxc_network($oldnetcfg);
960 if (safe_string_ne
($oldnet->{hwaddr
}, $newnet->{hwaddr
}) ||
961 safe_string_ne
($oldnet->{name
}, $newnet->{name
})) {
963 PVE
::Network
::veth_delete
($veth);
964 delete $conf->{$opt};
965 PVE
::LXC
::Config-
>write_config($vmid, $conf);
967 hotplug_net
($vmid, $conf, $opt, $newnet, $netid);
970 if (safe_string_ne
($oldnet->{bridge
}, $newnet->{bridge
}) ||
971 safe_num_ne
($oldnet->{tag
}, $newnet->{tag
}) ||
972 safe_num_ne
($oldnet->{firewall
}, $newnet->{firewall
}) ||
973 safe_boolean_ne
($oldnet->{link_down
}, $newnet->{link_down
})
976 if ($oldnet->{bridge
}) {
977 PVE
::Network
::tap_unplug
($veth);
978 foreach (qw(bridge tag firewall)) {
979 delete $oldnet->{$_};
981 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($oldnet);
982 PVE
::LXC
::Config-
>write_config($vmid, $conf);
985 PVE
::LXC
::net_tap_plug
($veth, $newnet);
987 # This includes the rate:
988 foreach (qw(bridge tag firewall rate link_down)) {
989 $oldnet->{$_} = $newnet->{$_} if $newnet->{$_};
991 } elsif (safe_string_ne
($oldnet->{rate
}, $newnet->{rate
})) {
992 # Rate can be applied on its own but any change above needs to
993 # include the rate in tap_plug since OVS resets everything.
994 PVE
::Network
::tap_rate_limit
($veth, $newnet->{rate
});
995 $oldnet->{rate
} = $newnet->{rate
}
997 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($oldnet);
998 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1001 hotplug_net
($vmid, $conf, $opt, $newnet, $netid);
1004 update_ipconfig
($vmid, $conf, $opt, $eth, $newnet, $rootdir);
1008 my ($vmid, $conf, $opt, $newnet, $netid) = @_;
1010 my $veth = "veth${vmid}i${netid}";
1011 my $vethpeer = $veth . "p";
1012 my $eth = $newnet->{name
};
1015 PVE
::Network
::SDN
::Zones
::veth_create
($veth, $vethpeer, $newnet->{bridge
}, $newnet->{hwaddr
});
1017 PVE
::Network
::veth_create
($veth, $vethpeer, $newnet->{bridge
}, $newnet->{hwaddr
});
1020 PVE
::LXC
::net_tap_plug
($veth, $newnet);
1022 # attach peer in container
1023 my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ];
1024 PVE
::Tools
::run_command
($cmd);
1026 # link up peer in container
1027 $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up' ];
1028 PVE
::Tools
::run_command
($cmd);
1030 my $done = { type
=> 'veth' };
1031 foreach (qw(bridge tag firewall hwaddr name link_down)) {
1032 $done->{$_} = $newnet->{$_} if $newnet->{$_};
1034 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($done);
1036 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1039 sub update_ipconfig
{
1040 my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_;
1042 my $lxc_setup = PVE
::LXC
::Setup-
>new($conf, $rootdir);
1044 my $optdata = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$opt});
1048 my $cmdargs = shift;
1049 PVE
::Tools
::run_command
(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs);
1051 my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) };
1053 my $change_ip_config = sub {
1054 my ($ipversion) = @_;
1056 my $family_opt = "-$ipversion";
1057 my $suffix = $ipversion == 4 ?
'' : $ipversion;
1058 my $gw= "gw$suffix";
1059 my $ip= "ip$suffix";
1061 my $newip = $newnet->{$ip};
1062 my $newgw = $newnet->{$gw};
1063 my $oldip = $optdata->{$ip};
1064 my $oldgw = $optdata->{$gw};
1066 my $change_ip = safe_string_ne
($oldip, $newip);
1067 my $change_gw = safe_string_ne
($oldgw, $newgw);
1069 return if !$change_ip && !$change_gw;
1071 # step 1: add new IP, if this fails we cancel
1072 my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/);
1073 if ($change_ip && $is_real_ip) {
1074 eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); };
1081 # step 2: replace gateway
1082 # If this fails we delete the added IP and cancel.
1083 # If it succeeds we save the config and delete the old IP, ignoring
1084 # errors. The config is then saved.
1085 # Note: 'ip route replace' can add
1089 if ($is_real_ip && !PVE
::Network
::is_ip_in_cidr
($newgw, $newip, $ipversion)) {
1090 &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth);
1092 &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw);
1096 # the route was not replaced, the old IP is still available
1097 # rollback (delete new IP) and cancel
1099 eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); };
1100 warn $@ if $@; # no need to die here
1105 eval { &$ipcmd($family_opt, 'route', 'del', 'default'); };
1106 # if the route was not deleted, the guest might have deleted it manually
1110 if ($oldgw && $oldip && !PVE
::Network
::is_ip_in_cidr
($oldgw, $oldip)) {
1111 eval { &$ipcmd($family_opt, 'route', 'del', $oldgw, 'dev', $eth); };
1112 # warn if the route was deleted manually
1117 # from this point on we save the configuration
1118 # step 3: delete old IP ignoring errors
1119 if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) {
1120 # We need to enable promote_secondaries, otherwise our newly added
1121 # address will be removed along with the old one.
1124 if ($ipversion == 4) {
1125 &$nscmd({ outfunc
=> sub { $promote = int(shift) } },
1126 'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries");
1127 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1");
1129 &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth);
1131 warn $@ if $@; # no need to die here
1133 if ($ipversion == 4) {
1134 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote");
1138 foreach my $property ($ip, $gw) {
1139 if ($newnet->{$property}) {
1140 $optdata->{$property} = $newnet->{$property};
1142 delete $optdata->{$property};
1145 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($optdata);
1146 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1147 $lxc_setup->setup_network($conf);
1150 &$change_ip_config(4);
1151 &$change_ip_config(6);
1155 my $open_namespace = sub {
1156 my ($vmid, $pid, $kind) = @_;
1157 sysopen my $fd, "/proc/$pid/ns/$kind", O_RDONLY
1158 or die "failed to open $kind namespace of container $vmid: $!\n";
1162 my $enter_namespace = sub {
1163 my ($vmid, $pid, $kind, $type) = @_;
1164 my $fd = $open_namespace->($vmid, $pid, $kind);
1165 PVE
::Tools
::setns
(fileno($fd), $type)
1166 or die "failed to enter $kind namespace of container $vmid: $!\n";
1170 my $get_container_namespace = sub {
1171 my ($vmid, $pid, $kind) = @_;
1174 if (!defined($pid)) {
1175 # Pin the pid while we're grabbing its stuff from /proc
1176 ($pid, $pidfd) = open_lxc_pid
($vmid)
1177 or die "failed to open pidfd of container $vmid\'s init process\n";
1180 return $open_namespace->($vmid, $pid, $kind);
1183 my $do_syncfs = sub {
1184 my ($vmid, $pid, $socket) = @_;
1186 &$enter_namespace($vmid, $pid, 'mnt', PVE
::Tools
::CLONE_NEWNS
);
1188 # Tell the parent process to start reading our /proc/mounts
1189 print {$socket} "go\n";
1192 # Receive /proc/self/mounts
1193 my $mountdata = do { local $/ = undef; <$socket> };
1209 # Now sync all mountpoints...
1210 my $mounts = PVE
::ProcFSTools
::parse_mounts
($mountdata);
1211 foreach my $mp (@$mounts) {
1212 my ($what, $dir, $fs) = @$mp;
1213 next if $nosyncfs{$fs};
1214 eval { PVE
::Tools
::sync_mountpoint
($dir); };
1219 sub sync_container_namespace
{
1221 my $pid = find_lxc_pid
($vmid);
1223 # SOCK_DGRAM is nicer for barriers but cannot be slurped
1224 socketpair my $pfd, my $cfd, AF_UNIX
, SOCK_STREAM
, PF_UNSPEC
1225 or die "failed to create socketpair: $!\n";
1228 die "fork failed: $!\n" if !defined($child);
1233 &$do_syncfs($vmid, $pid, $cfd);
1243 die "failed to enter container namespace\n" if $go ne "go\n";
1245 open my $mounts, '<', "/proc/$child/mounts"
1246 or die "failed to open container's /proc/mounts: $!\n";
1247 my $mountdata = do { local $/ = undef; <$mounts> };
1249 print {$pfd} $mountdata;
1252 while (waitpid($child, 0) != $child) {}
1253 die "failed to sync container namespace\n" if $? != 0;
1256 sub template_create
{
1257 my ($vmid, $conf) = @_;
1259 my $storecfg = PVE
::Storage
::config
();
1261 PVE
::LXC
::Config-
>foreach_volume($conf, sub {
1262 my ($ms, $mountpoint) = @_;
1264 my $volid = $mountpoint->{volume
};
1266 die "Template feature is not available for '$volid'\n"
1267 if !PVE
::Storage
::volume_has_feature
($storecfg, 'template', $volid);
1270 PVE
::LXC
::Config-
>foreach_volume($conf, sub {
1271 my ($ms, $mountpoint) = @_;
1273 my $volid = $mountpoint->{volume
};
1275 PVE
::Storage
::activate_volumes
($storecfg, [$volid]);
1277 my $template_volid = PVE
::Storage
::vdisk_create_base
($storecfg, $volid);
1278 $mountpoint->{volume
} = $template_volid;
1279 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq "rootfs");
1282 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1285 sub check_ct_modify_config_perm
{
1286 my ($rpcenv, $authuser, $vmid, $pool, $oldconf, $newconf, $delete, $unprivileged) = @_;
1288 return 1 if $authuser eq 'root@pam';
1289 my $storage_cfg = PVE
::Storage
::config
();
1292 my ($opt, $delete) = @_;
1293 if ($opt eq 'cores' || $opt eq 'cpuunits' || $opt eq 'cpulimit') {
1294 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']);
1295 } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) {
1296 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']);
1298 my $data = PVE
::LXC
::Config-
>parse_volume($opt, $newconf->{$opt});
1299 raise_perm_exc
("mount point type $data->{type} is only allowed for root\@pam")
1300 if $data->{type
} ne 'volume';
1301 my $volid = $data->{volume
};
1302 if ($volid =~ $NEW_DISK_RE) {
1304 $rpcenv->check($authuser, "/storage/$sid", ['Datastore.AllocateSpace']);
1306 PVE
::Storage
::check_volume_access
(
1315 } elsif ($opt eq 'memory' || $opt eq 'swap') {
1316 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']);
1317 } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' ||
1318 $opt eq 'searchdomain' || $opt eq 'hostname') {
1319 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']);
1320 } elsif ($opt eq 'features') {
1321 raise_perm_exc
("changing feature flags for privileged container is only allowed for root\@pam")
1324 my $nesting_changed = 0;
1325 my $other_changed = 0;
1327 my $features = PVE
::LXC
::Config-
>parse_features($newconf->{$opt});
1328 if (defined($oldconf) && $oldconf->{$opt}) {
1329 # existing container with features
1330 my $old_features = PVE
::LXC
::Config-
>parse_features($oldconf->{$opt});
1331 for my $feature ((keys %$old_features, keys %$features)) {
1332 my $old = $old_features->{$feature} // '';
1333 my $new = $features->{$feature} // '';
1335 if ($feature eq 'nesting') {
1336 $nesting_changed = 1;
1345 # new container or no features defined
1346 if (scalar(keys %$features) == 1 && $features->{nesting
}) {
1347 $nesting_changed = 1;
1348 } elsif (scalar(keys %$features) > 0) {
1353 my $features = PVE
::LXC
::Config-
>parse_features($oldconf->{$opt});
1354 if (scalar(keys %$features) == 1 && $features->{nesting
}) {
1355 $nesting_changed = 1;
1356 } elsif (scalar(keys %$features) > 0) {
1360 raise_perm_exc
("changing feature flags (except nesting) is only allowed for root\@pam")
1362 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Allocate'])
1363 if $nesting_changed;
1364 } elsif ($opt eq 'hookscript') {
1365 # For now this is restricted to root@pam
1366 raise_perm_exc
("changing the hookscript is only allowed for root\@pam");
1367 } elsif ($opt eq 'tags') {
1368 my $old = $oldconf->{$opt};
1369 my $new = $delete ?
'' : $newconf->{$opt};
1370 PVE
::GuestHelpers
::assert_tag_permissions
($vmid, $old, $new, $rpcenv, $authuser);
1372 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']);
1376 foreach my $opt (keys %$newconf) {
1379 foreach my $opt (@$delete) {
1387 my ($vmid, $storage_cfg, $conf, $noerr) = @_;
1389 my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1390 my $volid_list = PVE
::LXC
::Config-
>get_vm_volumes($conf);
1394 PVE
::LXC
::Config-
>foreach_volume_full($conf, {'reverse' => 1}, sub {
1395 my ($ms, $mountpoint) = @_;
1397 my $volid = $mountpoint->{volume
};
1398 my $mount = $mountpoint->{mp
};
1400 return if !$volid || !$mount;
1402 my $mount_path = "$rootdir/$mount";
1403 $mount_path =~ s!/+!/!g;
1405 return if !PVE
::ProcFSTools
::is_mounted
($mount_path);
1408 PVE
::Tools
::run_command
(['umount', '-d', $mount_path]);
1420 return $res; # tell caller if (some) umounts failed for the noerr case
1424 my ($vmid, $storage_cfg, $conf, $ignore_ro) = @_;
1426 my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1427 File
::Path
::make_path
($rootdir);
1429 my $volid_list = PVE
::LXC
::Config-
>get_vm_volumes($conf);
1430 PVE
::Storage
::activate_volumes
($storage_cfg, $volid_list);
1432 my (undef, $rootuid, $rootgid) = parse_id_maps
($conf);
1435 PVE
::LXC
::Config-
>foreach_volume($conf, sub {
1436 my ($ms, $mountpoint) = @_;
1438 $mountpoint->{ro
} = 0 if $ignore_ro;
1440 mountpoint_mount
($mountpoint, $rootdir, $storage_cfg, undef, $rootuid, $rootgid);
1444 warn "mounting container failed\n";
1445 umount_all
($vmid, $storage_cfg, $conf, 1);
1453 sub mountpoint_mount_path
{
1454 my ($mountpoint, $storage_cfg, $snapname) = @_;
1456 return mountpoint_mount
($mountpoint, undef, $storage_cfg, $snapname);
1464 if ($line =~ m
@^(/dev/loop\d
+):@) {
1468 my $cmd = ['losetup', '--associated', $path];
1469 PVE
::Tools
::run_command
($cmd, outfunc
=> $parser);
1473 # Run a function with a file attached to a loop device.
1474 # The loop device is always detached afterwards (or set to autoclear).
1475 # Returns the loop device.
1476 sub run_with_loopdev
{
1477 my ($func, $file, $readonly) = @_;
1478 my $device = query_loopdev
($file);
1479 # Try to reuse an existing device
1481 # We assume that whoever setup the loop device is responsible for
1489 if ($line =~ m
@^(/dev/loop\d
+)$@) {
1499 push @$losetup_cmd, '-r' if $readonly;
1500 PVE
::Tools
::run_command
($losetup_cmd, outfunc
=> $parser);
1501 die "failed to setup loop device for $file\n" if !$device;
1502 eval { &$func($device); };
1504 PVE
::Tools
::run_command
(['losetup', '-d', $device]);
1509 # In scalar mode: returns a file handle to the deepest directory node.
1510 # In list context: returns a list of:
1511 # * the deepest directory node
1512 # * the 2nd deepest directory (parent of the above)
1513 # * directory name of the last directory
1514 # So that the path $2/$3 should lead to $1 afterwards.
1515 sub walk_tree_nofollow
($$$;$$) {
1516 my ($start, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1518 sysopen(my $fd, $start, O_PATH
| O_DIRECTORY
)
1519 or die "failed to open start directory $start: $!\n";
1521 return walk_tree_nofollow_fd
($start, $fd, $subdir, $mkdir, $rootuid, $rootgid);
1525 sub walk_tree_nofollow_fd
($$$$;$$) {
1526 my ($start_dirname, $start_fd, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1528 # splitdir() returns '' for empty components including the leading /
1529 my @comps = grep { length($_)>0 } File
::Spec-
>splitdir($subdir);
1532 my $dir = $start_dirname;
1533 my $last_component = undef;
1535 foreach my $component (@comps) {
1536 $dir .= "/$component";
1537 my $next = PVE
::Tools
::openat
(fileno($fd), $component, O_NOFOLLOW
| O_DIRECTORY
);
1540 # failed, check for symlinks and try to create the path
1541 die "symlink encountered at: $dir\n" if $! == ELOOP
|| $! == ENOTDIR
;
1542 die "cannot open directory $dir: $!\n" if !$mkdir;
1544 # We don't check for errors on mkdirat() here and just try to
1545 # openat() again, since at least one error (EEXIST) is an
1546 # expected possibility if multiple containers start
1547 # simultaneously. If someone else injects a symlink now then
1548 # the subsequent openat() will fail due to O_NOFOLLOW anyway.
1549 PVE
::Tools
::mkdirat
(fileno($fd), $component, 0755);
1551 $next = PVE
::Tools
::openat
(fileno($fd), $component, O_NOFOLLOW
| O_DIRECTORY
);
1552 die "failed to create path: $dir: $!\n" if !$next;
1554 PVE
::Tools
::fchownat
(fileno($next), '', $rootuid, $rootgid, PVE
::Tools
::AT_EMPTY_PATH
)
1555 if defined($rootuid) && defined($rootgid);
1558 close $second if defined($last_component) && $second != $start_fd;
1559 $last_component = $component;
1564 return ($fd, defined($last_component) && $second, $last_component) if wantarray;
1565 close $second if defined($last_component) && $second != $start_fd;
1569 # To guard against symlink attack races against other currently running
1570 # containers with shared recursive bind mount hierarchies we prepare a
1571 # directory handle for the directory we're mounting over to verify the
1572 # mountpoint afterwards.
1573 sub __bindmount_prepare
{
1574 my ($hostroot, $dir) = @_;
1575 my $srcdh = walk_tree_nofollow
($hostroot, $dir, 0);
1579 # Assuming we mount to rootfs/a/b/c, verify with the directory handle to 'b'
1580 # ($parentfd) that 'b/c' (openat($parentfd, 'c')) really leads to the directory
1581 # we intended to bind mount.
1582 sub __bindmount_verify
{
1583 my ($srcdh, $parentfd, $last_dir, $ro) = @_;
1586 # Open the mount point path coming from the parent directory since the
1587 # filehandle we would have gotten as first result of walk_tree_nofollow
1588 # earlier is still a handle to the underlying directory instead of the
1590 $destdh = PVE
::Tools
::openat
(fileno($parentfd), $last_dir, PVE
::Tools
::O_PATH
| O_NOFOLLOW
| O_DIRECTORY
);
1591 die "failed to open mount point: $!\n" if !$destdh;
1594 # no separate function because 99% of the time it's the wrong thing to use.
1595 if (syscall(PVE
::Syscall
::faccessat
, fileno($destdh), $dot, &POSIX
::W_OK
, 0) != -1) {
1596 die "failed to mark bind mount read only\n";
1598 die "read-only check failed: $!\n" if $! != EROFS
;
1601 # For the rootfs we don't have a parentfd so we open the path directly.
1602 # Note that this means bindmounting any prefix of the host's
1603 # /var/lib/lxc/$vmid path into another container is considered a grave
1605 sysopen $destdh, $last_dir, O_PATH
| O_DIRECTORY
;
1606 die "failed to open mount point: $!\n" if !$destdh;
1609 my ($srcdev, $srcinode) = stat($srcdh);
1610 my ($dstdev, $dstinode) = stat($destdh);
1614 return ($srcdev == $dstdev && $srcinode == $dstinode);
1617 # Perform the actual bind mounting:
1618 sub __bindmount_do
{
1619 my ($dir, $dest, $ro, @extra_opts) = @_;
1620 PVE
::Tools
::run_command
(['mount', '-o', 'bind', @extra_opts, $dir, $dest]);
1622 eval { PVE
::Tools
::run_command
(['mount', '-o', 'bind,remount,ro', $dest]); };
1624 warn "bindmount error\n";
1625 # don't leave writable bind-mounts behind...
1626 PVE
::Tools
::run_command
(['umount', $dest]);
1633 my ($dir, $parentfd, $last_dir, $dest, $ro, @extra_opts) = @_;
1635 my $srcdh = __bindmount_prepare
('/', $dir);
1637 __bindmount_do
($dir, $dest, $ro, @extra_opts);
1639 if (!__bindmount_verify
($srcdh, $parentfd, $last_dir, $ro)) {
1640 PVE
::Tools
::run_command
(['umount', $dest]);
1641 die "detected mount path change at: $dir\n";
1645 # Cleanup $rootdir a bit (double and trailing slashes), build the mount path
1646 # from $rootdir and $mount and walk the path from $rootdir to the final
1647 # directory to check for symlinks.
1648 sub __mount_prepare_rootdir
{
1649 my ($rootdir, $mount, $rootuid, $rootgid) = @_;
1650 $rootdir =~ s!/+!/!g;
1651 $rootdir =~ s!/+$!!;
1652 my $mount_path = "$rootdir/$mount";
1653 my ($mpfd, $parentfd, $last_dir) = walk_tree_nofollow
($rootdir, $mount, 1, $rootuid, $rootgid);
1654 return ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir);
1657 # use $rootdir = undef to just return the corresponding mount path
1658 sub mountpoint_mount
{
1659 my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1660 return __mountpoint_mount
($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, undef);
1663 sub mountpoint_stage
{
1664 my ($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1665 my ($path, $loop, $dev) =
1666 __mountpoint_mount
($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid, 1);
1668 if (!defined($path)) {
1669 return undef if $! == ENOSYS
;
1670 die "failed to mount subvolume: $!\n";
1673 # We clone the mount point and leave it there in order to keep them connected to eg. loop
1674 # devices in case we're hotplugging (which would allow contaienrs to unmount the new mount
1677 my $fd = PVE
::Tools
::open_tree
(&AT_FDCWD
, $stage_dir, &OPEN_TREE_CLOEXEC
| &OPEN_TREE_CLONE
)
1678 or die "open_tree() on mount point failed: $!\n";
1680 return wantarray ?
($path, $loop, $dev, $fd) : $fd;
1683 sub mountpoint_insert_staged
{
1684 my ($mount_fd, $rootdir_fd, $mp_dir, $opt, $rootuid, $rootgid) = @_;
1686 if (!defined($rootdir_fd)) {
1687 sysopen($rootdir_fd, '.', O_PATH
| O_DIRECTORY
)
1688 or die "failed to open '.': $!\n";
1691 my $dest_fd = walk_tree_nofollow_fd
('/', $rootdir_fd, $mp_dir, 1, $rootuid, $rootgid);
1693 PVE
::Tools
::move_mount
(
1698 &MOVE_MOUNT_F_EMPTY_PATH
| &MOVE_MOUNT_T_EMPTY_PATH
,
1699 ) or die "failed to move '$opt' into container hierarchy: $!\n";
1702 # Use $stage_mount, $rootdir is treated as a temporary path to "stage" the file system. The user
1703 # can then open a file descriptor to it which can be used with the `move_mount` syscall.
1704 # Note that if the kernel does not support the new mount API, this will not perform any action
1705 # and return `undef` with $! = ENOSYS.
1706 sub __mountpoint_mount
{
1707 my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, $stage_mount) = @_;
1709 if (defined($stage_mount) && !PVE
::LXC
::Tools
::can_use_new_mount_api
()) {
1714 # When staging mount points we always mount to $rootdir directly (iow. as if `mp=/`).
1715 # This is required since __mount_prepare_rootdir() will return handles to the parent directory
1716 # which we use in __bindmount_verify()!
1717 my $mount = $stage_mount ?
'/': $mountpoint->{mp
};
1719 my $volid = $mountpoint->{volume
};
1720 my $type = $mountpoint->{type
};
1721 my $quota = !$snapname && !$mountpoint->{ro
} && $mountpoint->{quota
};
1724 return if !$volid || !$mount;
1729 my ($mpfd, $parentfd, $last_dir);
1731 if (defined($rootdir)) {
1732 ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir) =
1733 __mount_prepare_rootdir
($rootdir, $mount, $rootuid, $rootgid);
1736 if (defined($stage_mount)) {
1737 $mount_path = $rootdir;
1740 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1742 die "unknown snapshot path for '$volid'" if !$storage && defined($snapname);
1746 if (my $mountopts = $mountpoint->{mountoptions
}) {
1747 my @opts = split(/;/, $mountpoint->{mountoptions
});
1748 push @$optlist, grep { PVE
::LXC
::Config
::is_valid_mount_option
($_) } @opts;
1751 my $acl = $mountpoint->{acl
};
1752 if (defined($acl)) {
1753 push @$optlist, ($acl ?
'acl' : 'noacl');
1756 my $optstring = join(',', @$optlist);
1757 my $readonly = $mountpoint->{ro
};
1760 @extra_opts = ('-o', $optstring) if $optstring;
1764 my $scfg = PVE
::Storage
::storage_config
($storage_cfg, $storage);
1766 my $path = PVE
::Storage
::map_volume
($storage_cfg, $volid, $snapname);
1768 $path = PVE
::Storage
::path
($storage_cfg, $volid, $snapname) if !defined($path);
1770 my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1771 PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1773 $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files
1775 if ($format eq 'subvol') {
1777 my (undef, $name) = PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1778 if (defined($snapname)) {
1779 $name .= "\@$snapname";
1780 if ($scfg->{type
} eq 'zfspool') {
1781 PVE
::Tools
::run_command
(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', "$scfg->{pool}/$name", $mount_path]);
1783 die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n";
1786 if (defined($acl) && $scfg->{type
} eq 'zfspool') {
1787 my $acltype = ($acl ?
'acltype=posixacl' : 'acltype=noacl');
1788 PVE
::Tools
::run_command
(['zfs', 'set', $acltype, "$scfg->{pool}/$name"]);
1790 bindmount
($path, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts);
1791 warn "cannot enable quota control for bind mounted subvolumes\n" if $quota;
1794 return wantarray ?
($path, 0, undef) : $path;
1795 } elsif ($format eq 'raw' || $format eq 'iso') {
1796 # NOTE: 'mount' performs canonicalization without the '-c' switch, which for
1797 # device-mapper devices is special-cased to use the /dev/mapper symlinks.
1798 # Our autodev hook expects the /dev/dm-* device currently
1799 # and will create the /dev/mapper symlink accordingly
1800 $path = Cwd
::realpath
($path);
1801 die "failed to get device path\n" if !$path;
1802 ($path) = ($path =~ /^(.*)$/s); #untaint
1806 if ($format eq 'iso') {
1807 PVE
::Tools
::run_command
(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]);
1808 } elsif ($isBase || defined($snapname)) {
1809 PVE
::Tools
::run_command
(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]);
1812 push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0';
1814 push @extra_opts, '-o', 'ro' if $readonly;
1815 PVE
::Tools
::run_command
(['mount', @extra_opts, $path, $mount_path]);
1819 my $use_loopdev = 0;
1820 if ($scfg->{content
}->{rootdir
}) {
1821 if ($scfg->{path
}) {
1822 $mounted_dev = run_with_loopdev
($domount, $path, $readonly);
1825 $mounted_dev = $path;
1829 die "storage '$storage' does not support containers\n";
1831 return wantarray ?
($path, $use_loopdev, $mounted_dev) : $path;
1833 die "unsupported image format '$format'\n";
1835 } elsif ($type eq 'device') {
1836 push @extra_opts, '-o', 'ro' if $readonly;
1837 push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0' if $quota;
1838 # See the NOTE above about devicemapper canonicalization
1839 my ($devpath) = (Cwd
::realpath
($volid) =~ /^(.*)$/s); # realpath() taints
1840 PVE
::Tools
::run_command
(['mount', @extra_opts, $volid, $mount_path]) if $mount_path;
1841 return wantarray ?
($volid, 0, $devpath) : $volid;
1842 } elsif ($type eq 'bind') {
1843 die "directory '$volid' does not exist\n" if ! -d
$volid;
1844 bindmount
($volid, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts) if $mount_path;
1845 warn "cannot enable quota control for bind mounts\n" if $quota;
1846 return wantarray ?
($volid, 0, undef) : $volid;
1849 die "unsupported storage";
1852 sub mountpoint_hotplug
:prototype($$$$$) {
1853 my ($vmid, $conf, $opt, $mp, $storage_cfg) = @_;
1855 my (undef, $rootuid, $rootgid) = PVE
::LXC
::parse_id_maps
($conf);
1857 # We do the rest in a fork with an unshared mount namespace, because:
1858 # -) change our papparmor profile to that of /usr/bin/lxc-start
1859 # -) we're now going to 'stage' # the mountpoint, then grab it, then move into the
1860 # container's namespace, then mount it.
1862 PVE
::Tools
::run_fork
(sub {
1863 # Pin the container pid longer, we also need to get its monitor/parent:
1864 my ($ct_pid, $ct_pidfd) = open_lxc_pid
($vmid)
1865 or die "failed to open pidfd of container $vmid\'s init process\n";
1867 my ($monitor_pid, $monitor_pidfd) = open_ppid
($ct_pid)
1868 or die "failed to open pidfd of container $vmid\'s monitor process\n";
1870 my $ct_mnt_ns = $get_container_namespace->($vmid, $ct_pid, 'mnt');
1871 my $monitor_mnt_ns = $get_container_namespace->($vmid, $monitor_pid, 'mnt');
1873 # Grab a file descriptor to our apparmor label file so we can change into the 'lxc-start'
1874 # profile to lower our privileges to the same level we have in the start hook:
1875 sysopen(my $aa_fd, "/proc/self/attr/current", O_WRONLY
)
1876 or die "failed to open '/proc/self/attr/current' for writing: $!\n";
1877 # But switch namespaces first, to make sure the namespace switches aren't blocked by
1880 # Change into the monitor's mount namespace. We "pin" the mount into the monitor's
1881 # namespace for it to remain active there since the container will be able to unmount
1882 # hotplugged mount points and thereby potentially free up loop devices, which is a security
1884 PVE
::Tools
::setns
(fileno($monitor_mnt_ns), PVE
::Tools
::CLONE_NEWNS
);
1886 or die "failed to change root directory within the monitor's mount namespace: $!\n";
1888 my $dir = get_staging_mount_path
($opt);
1890 # Now switch our apparmor profile before mounting:
1891 my $data = 'changeprofile /usr/bin/lxc-start';
1892 if (syswrite($aa_fd, $data, length($data)) != length($data)) {
1893 die "failed to change apparmor profile: $!\n";
1895 # Check errors on close as well:
1897 or die "failed to change apparmor profile (close() failed): $!\n";
1899 my $mount_fd = mountpoint_stage
($mp, $dir, $storage_cfg, undef, $rootuid, $rootgid);
1901 PVE
::Tools
::setns
(fileno($ct_mnt_ns), PVE
::Tools
::CLONE_NEWNS
);
1903 or die "failed to change root directory within the container's mount namespace: $!\n";
1905 mountpoint_insert_staged
($mount_fd, undef, $mp->{mp
}, $opt, $rootuid, $rootgid);
1909 # Create a directory in the mountpoint staging tempfs.
1910 sub get_staging_mount_path
($) {
1913 my $target = get_staging_tempfs
() . "/$opt";
1914 if (!mkdir($target) && $! != EEXIST
) {
1915 die "failed to create directory $target: $!\n";
1921 # Mount /run/pve/mountpoints as tmpfs
1922 sub get_staging_tempfs
() {
1923 # We choose a path in /var/lib/lxc/ here because the lxc-start apparmor profile restricts most
1925 my $target = '/var/lib/lxc/.pve-staged-mounts';
1926 if (!mkdir($target)) {
1927 return $target if $! == EEXIST
;
1928 die "failed to create directory $target: $!\n";
1931 PVE
::Tools
::mount
("none", $target, 'tmpfs', 0, "size=8k,mode=755")
1932 or die "failed to mount $target as tmpfs: $!\n";
1938 my ($dev, $rootuid, $rootgid) = @_;
1946 "root_owner=$rootuid:$rootgid",
1951 # a hack to print only the relevant stuff, i.e., the one which could help on repair
1952 if ($line =~ /^(Creating filesystem|Filesystem UUID|Superblock backups|\s+\d+, \d)/) {
1958 print STDERR
"$line\n" if $line && $line !~ /^mke2fs \d\.\d/;
1964 my ($storage_cfg, $volid, $rootuid, $rootgid) = @_;
1966 if ($volid =~ m!^/dev/.+!) {
1971 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1973 die "cannot format volume '$volid' with no storage\n" if !$storage;
1975 PVE
::Storage
::activate_volumes
($storage_cfg, [$volid]);
1977 my $path = PVE
::Storage
::map_volume
($storage_cfg, $volid);
1979 $path = PVE
::Storage
::path
($storage_cfg, $volid) if !defined($path);
1981 my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1982 PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1984 die "cannot format volume '$volid' (format == $format)\n"
1985 if $format ne 'raw';
1987 mkfs
($path, $rootuid, $rootgid);
1991 my ($storecfg, $vollist) = @_;
1993 foreach my $volid (@$vollist) {
1994 eval { PVE
::Storage
::vdisk_free
($storecfg, $volid); };
2000 my ($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid) = @_;
2002 my $needs_chown = 0;
2005 my $scfg = PVE
::Storage
::storage_config
($storecfg, $storage);
2006 # fixme: use better naming ct-$vmid-disk-X.raw?
2010 if ($scfg->{content
}->{rootdir
} && $scfg->{path
}) {
2011 if ($size_kb > 0 && !($scfg->{type
} eq 'btrfs' && $scfg->{quotas
})) {
2012 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
2015 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'subvol', undef, $size_kb);
2018 } elsif ($scfg->{type
} eq 'zfspool') {
2019 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'subvol', undef, $size_kb);
2021 } elsif ($scfg->{content
}->{rootdir
}) {
2022 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
2025 die "content type 'rootdir' is not available or configured on storage '$storage'\n";
2027 format_disk
($storecfg, $volid, $rootuid, $rootgid) if $do_format;
2030 # in case formatting got interrupted:
2031 if (defined($volid)) {
2032 eval { PVE
::Storage
::vdisk_free
($storecfg, $volid); };
2038 return ($volid, $needs_chown);
2042 my ($storecfg, $vmid, $settings, $conf, $pending) = @_;
2047 my (undef, $rootuid, $rootgid) = PVE
::LXC
::parse_id_maps
($conf);
2048 my $chown_vollist = [];
2050 PVE
::LXC
::Config-
>foreach_volume($settings, sub {
2051 my ($ms, $mountpoint) = @_;
2053 my $volid = $mountpoint->{volume
};
2054 my $mp = $mountpoint->{mp
};
2056 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
2058 if ($storage && ($volid =~ $NEW_DISK_RE)) {
2059 my ($storeid, $size_gb) = ($1, $2);
2061 my $size_kb = int(${size_gb
}*1024) * 1024;
2063 my $needs_chown = 0;
2064 ($volid, $needs_chown) = alloc_disk
($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid);
2065 push @$chown_vollist, $volid if $needs_chown;
2066 push @$vollist, $volid;
2067 $mountpoint->{volume
} = $volid;
2068 $mountpoint->{size
} = $size_kb * 1024;
2070 $conf->{pending
}->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
2072 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
2075 # use specified/existing volid/dir/device
2076 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
2080 PVE
::Storage
::activate_volumes
($storecfg, $chown_vollist, undef);
2081 foreach my $volid (@$chown_vollist) {
2082 my $path = PVE
::Storage
::path
($storecfg, $volid, undef);
2083 chown($rootuid, $rootgid, $path);
2085 PVE
::Storage
::deactivate_volumes
($storecfg, $chown_vollist, undef);
2087 # free allocated images on error
2089 destroy_disks
($storecfg, $vollist);
2095 sub update_disksize
{
2096 my ($vmid, $conf, $all_volumes) = @_;
2099 my $prefix = "CT $vmid:";
2101 my $update_mp = sub {
2102 my ($key, $mp, @param) = @_;
2103 my $size = $all_volumes->{$mp->{volume
}}->{size
} // 0;
2105 if (!defined($mp->{size
}) || $size != $mp->{size
}) {
2107 print "$prefix updated volume size of '$mp->{volume}' in config.\n";
2108 $mp->{size
} = $size;
2109 my $no_mp = $key eq 'rootfs'; # rootfs is handled different from other mount points
2110 $conf->{$key} = PVE
::LXC
::Config-
>print_ct_mountpoint($mp, $no_mp);
2114 PVE
::LXC
::Config-
>foreach_volume($conf, $update_mp);
2120 my ($vmid, $conf, $all_volumes) = @_;
2123 my $prefix = "CT $vmid:";
2125 # Note: it is allowed to define multiple storage entries with the same path
2126 # (alias), so we need to check both 'volid' and real 'path' (two different
2127 # volid can point to the same path).
2129 # used and unused disks
2133 foreach my $opt (keys %$conf) {
2134 next if ($opt !~ m/^unused\d+$/);
2135 my $vol = $all_volumes->{$conf->{$opt}};
2136 $refpath->{$vol->{path
}} = $vol->{volid
};
2139 foreach my $key (keys %$all_volumes) {
2140 my $vol = $all_volumes->{$key};
2141 my $in_use = PVE
::LXC
::Config-
>is_volume_in_use($conf, $vol->{volid
});
2142 my $path = $vol->{path
};
2145 $refpath->{$path} = $key;
2146 delete $orphans->{$path};
2148 if ((!$orphans->{$path}) && (!$refpath->{$path})) {
2149 $orphans->{$path} = $key;
2154 for my $key (keys %$orphans) {
2155 my $disk = $orphans->{$key};
2156 my $unused = PVE
::LXC
::Config-
>add_unused_volume($conf, $disk);
2160 print "$prefix add unreferenced volume '$disk' as '$unused' to config.\n";
2168 my ($cfg, $vmid) = @_;
2170 my $info = PVE
::Storage
::vdisk_list
($cfg, undef, $vmid, undef, 'rootdir');
2172 my $all_volumes = {};
2173 foreach my $storeid (keys %$info) {
2174 foreach my $item (@{$info->{$storeid}}) {
2175 my $volid = $item->{volid
};
2176 next if !($volid && $item->{size
});
2177 $item->{path
} = PVE
::Storage
::path
($cfg, $volid);
2178 $all_volumes->{$volid} = $item;
2182 return $all_volumes;
2186 my ($vmid, $nolock, $dryrun) = @_;
2188 my $cfg = PVE
::Storage
::config
();
2190 print "rescan volumes...\n";
2191 my $all_volumes = scan_volids
($cfg, $vmid);
2193 my $updatefn = sub {
2197 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2199 PVE
::LXC
::Config-
>check_lock($conf);
2202 foreach my $volid (keys %$all_volumes) {
2203 my $info = $all_volumes->{$volid};
2204 $vm_volids->{$volid} = $info if $info->{vmid
} == $vmid;
2207 my $upu = update_unused
($vmid, $conf, $vm_volids);
2208 my $upd = update_disksize
($vmid, $conf, $vm_volids);
2209 $changes = $upu || $upd;
2211 PVE
::LXC
::Config-
>write_config($vmid, $conf) if $changes && !$dryrun;
2214 if (defined($vmid)) {
2218 PVE
::LXC
::Config-
>lock_config($vmid, $updatefn, $vmid);
2221 my $vmlist = config_list
();
2222 foreach my $vmid (keys %$vmlist) {
2226 PVE
::LXC
::Config-
>lock_config($vmid, $updatefn, $vmid);
2233 # bash completion helper
2235 sub complete_os_templates
{
2236 my ($cmdname, $pname, $cvalue) = @_;
2238 my $cfg = PVE
::Storage
::config
();
2242 if ($cvalue =~ m/^([^:]+):/) {
2246 my $vtype = $cmdname eq 'restore' ?
'backup' : 'vztmpl';
2247 my $data = PVE
::Storage
::template_list
($cfg, $storeid, $vtype);
2250 foreach my $id (keys %$data) {
2251 foreach my $item (@{$data->{$id}}) {
2252 push @$res, $item->{volid
} if defined($item->{volid
});
2259 my $complete_ctid_full = sub {
2262 my $idlist = vmstatus
();
2264 my $active_hash = list_active_containers
();
2268 foreach my $id (keys %$idlist) {
2269 my $d = $idlist->{$id};
2270 if (defined($running)) {
2271 next if $d->{template
};
2272 next if $running && !$active_hash->{$id};
2273 next if !$running && $active_hash->{$id};
2282 return &$complete_ctid_full();
2285 sub complete_ctid_stopped
{
2286 return &$complete_ctid_full(0);
2289 sub complete_ctid_running
{
2290 return &$complete_ctid_full(1);
2300 my $lxc = $conf->{lxc
};
2301 foreach my $entry (@$lxc) {
2302 my ($key, $value) = @$entry;
2304 next if $key ne 'lxc.idmap';
2306 if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) {
2307 my ($type, $ct, $host, $length) = ($1, $2, $3, $4);
2308 push @$id_map, [$type, $ct, $host, $length];
2310 $rootuid = $host if $type eq 'u';
2311 $rootgid = $host if $type eq 'g';
2314 die "failed to parse idmap: $value\n";
2318 if (!@$id_map && $conf->{unprivileged
}) {
2319 # Should we read them from /etc/subuid?
2320 $id_map = [ ['u', '0', '100000', '65536'],
2321 ['g', '0', '100000', '65536'] ];
2322 $rootuid = $rootgid = 100000;
2325 return ($id_map, $rootuid, $rootgid);
2328 sub validate_id_maps
{
2331 # $mappings->{$type}->{$side} = [ { line => $line, start => $start, count => $count }, ... ]
2332 # $type: either "u" or "g"
2333 # $side: either "container" or "host"
2334 # $line: index of this mapping in @$id_map
2335 # $start, $count: interval of this mapping
2336 my $mappings = { u
=> {}, g
=> {} };
2337 for (my $i = 0; $i < scalar(@$id_map); $i++) {
2338 my ($type, $ct_start, $host_start, $count) = $id_map->[$i]->@*;
2339 my $sides = $mappings->{$type};
2340 push $sides->{host
}->@*, { line
=> $i, start
=> $host_start, count
=> $count };
2341 push $sides->{container
}->@*, { line
=> $i, start
=> $ct_start, count
=> $count };
2344 # find the first conflict between two consecutive mappings when sorted by their start id
2345 for my $type (qw(u g)) {
2346 for my $side (qw(container host)) {
2347 my @entries = sort { $a->{start
} <=> $b->{start
} } $mappings->{$type}->{$side}->@*;
2348 for my $idx (1..scalar(@entries) - 1) {
2349 my $previous = $entries[$idx - 1];
2350 my $current = $entries[$idx];
2351 if ($previous->{start
} + $previous->{count
} > $current->{start
}) {
2352 my $conflict = $current->{start
};
2353 my @previous_line = $id_map->[$previous->{line
}]->@*;
2354 my @current_line = $id_map->[$current->{line
}]->@*;
2355 die "invalid map entry '@current_line': $side ${type}id $conflict "
2356 ."is also mapped by entry '@previous_line'\n";
2363 sub userns_command
{
2366 return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--'];
2371 my sub print_ct_stderr_log
{
2373 my $log = eval { file_get_contents
("/run/pve/ct-$vmid.stderr") };
2376 while ($log =~ /^\h*(lxc-start:?\s+$vmid:?\s*\S+\s*)?(.*?)\h*$/gm) {
2378 print STDERR
"$line\n";
2381 my sub print_ct_warn_log
{
2383 my $log_fn = "/run/pve/ct-$vmid.warnings";
2384 my $log = eval { file_get_contents
($log_fn) };
2387 while ($log =~ /^\h*\s*(.*?)\h*$/gm) {
2388 PVE
::RESTEnvironment
::log_warn
($1);
2390 unlink $log_fn or warn "could not unlink '$log_fn' - $!\n";
2393 my sub monitor_state_change
($$) {
2394 my ($monitor_socket, $vmid) = @_;
2395 die "no monitor socket\n" if !defined($monitor_socket);
2398 my ($type, $name, $value) = PVE
::LXC
::Monitor
::read_lxc_message
($monitor_socket);
2400 die "monitor socket: got EOF\n" if !defined($type);
2402 next if $name ne "$vmid" || $type ne 'STATE';
2404 if ($value eq PVE
::LXC
::Monitor
::STATE_STARTING
) {
2405 alarm(0); # don't timeout after seeing the starting state
2406 } elsif ($value eq PVE
::LXC
::Monitor
::STATE_ABORTING
||
2407 $value eq PVE
::LXC
::Monitor
::STATE_STOPPING
||
2408 $value eq PVE
::LXC
::Monitor
::STATE_STOPPED
) {
2410 } elsif ($value eq PVE
::LXC
::Monitor
::STATE_RUNNING
) {
2413 warn "unexpected message from monitor socket - " .
2414 "type: '$type' - value: '$value'\n";
2418 my sub monitor_start
($$) {
2419 my ($monitor_socket, $vmid) = @_;
2421 my $success = eval {
2422 PVE
::Tools
::run_with_timeout
(10, \
&monitor_state_change
, $monitor_socket, $vmid)
2425 warn "problem with monitor socket, but continuing anyway: $err\n";
2426 } elsif (!$success) {
2427 print_ct_stderr_log
($vmid);
2428 die "startup for container '$vmid' failed\n";
2433 my ($vmid, $conf, $skiplock, $debug) = @_;
2435 # apply pending changes while starting
2436 if (scalar(keys %{$conf->{pending
}})) {
2437 my $storecfg = PVE
::Storage
::config
();
2438 PVE
::LXC
::Config-
>vmconfig_apply_pending($vmid, $conf, $storecfg);
2439 PVE
::LXC
::Config-
>write_config($vmid, $conf);
2440 $conf = PVE
::LXC
::Config-
>load_config($vmid); # update/reload
2443 update_lxc_config
($vmid, $conf);
2446 my ($id_map, undef, undef) = PVE
::LXC
::parse_id_maps
($conf);
2447 PVE
::LXC
::validate_id_maps
($id_map);
2449 warn "lxc.idmap: $@" if $@;
2451 my $skiplock_flag_fn = "/run/lxc/skiplock-$vmid";
2454 open(my $fh, '>', $skiplock_flag_fn) || die "failed to open $skiplock_flag_fn for writing: $!\n";
2458 my $storage_cfg = PVE
::Storage
::config
();
2459 my $vollist = PVE
::LXC
::Config-
>get_vm_volumes($conf);
2461 PVE
::Storage
::activate_volumes
($storage_cfg, $vollist);
2463 my $monitor_socket = eval { PVE
::LXC
::Monitor
::get_monitor_socket
() };
2466 unlink "/run/pve/ct-$vmid.stderr"; # systemd does not truncate log files
2468 my $is_debug = $debug || (!defined($debug) && $conf->{debug
});
2469 my $base_unit = $is_debug ?
'pve-container-debug' : 'pve-container';
2471 my $cmd = ['systemctl', 'start', "$base_unit\@$vmid"];
2473 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'pre-start', 1);
2477 monitor_start
($monitor_socket, $vmid) if defined($monitor_socket);
2479 # if debug is requested, print the log it also when the start succeeded
2480 print_ct_stderr_log
($vmid) if $is_debug;
2482 print_ct_warn_log
($vmid); # always print warn log, if any
2485 unlink $skiplock_flag_fn;
2488 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'post-start');
2493 # Helper to stop a container completely and make sure it has stopped completely.
2494 # This is necessary because we want the post-stop hook to have completed its
2495 # unmount-all step, but post-stop happens after lxc puts the container into the
2497 # $kill - if true it will always do an immediate hard-stop
2498 # $shutdown_timeout - the timeout to wait for a gracefull shutdown
2499 # $kill_after_timeout - if true, send a hardstop if shutdown timed out
2501 my ($vmid, $kill, $shutdown_timeout, $kill_after_timeout) = @_;
2503 # Open the container's command socket.
2504 my $path = "\0/var/lib/lxc/$vmid/command";
2505 my $sock = IO
::Socket
::UNIX-
>new(
2506 Type
=> SOCK_STREAM
(),
2510 return if $! == ECONNREFUSED
; # The container is not running
2511 die "failed to open container ${vmid}'s command socket: $!\n";
2514 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2515 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'pre-stop');
2517 # Stop the container:
2519 my $cmd = ['lxc-stop', '-n', $vmid];
2522 push @$cmd, '--kill'; # doesn't allow timeouts
2524 # lxc-stop uses a default timeout
2525 push @$cmd, '--nokill' if !$kill_after_timeout;
2527 if (defined($shutdown_timeout)) {
2528 push @$cmd, '--timeout', $shutdown_timeout;
2529 # Give run_command 5 extra seconds
2530 $shutdown_timeout += 5;
2534 eval { run_command
($cmd, timeout
=> $shutdown_timeout) };
2536 # Wait until the command socket is closed.
2537 # In case the lxc-stop call failed, reading from the command socket may block forever,
2538 # so poll with another timeout to avoid freezing the shutdown task.
2542 my $poll = IO
::Poll-
>new();
2543 $poll->mask($sock => POLLIN
| POLLHUP
); # watch for input and EOF events
2544 $poll->poll($shutdown_timeout); # IO::Poll timeout is in seconds
2545 return if ($poll->events($sock) & POLLHUP
);
2547 my $result = <$sock>;
2548 return if !defined $result; # monitor is gone and the ct has stopped.
2551 die "container did not stop\n";
2555 my ($vmid, $timeout, $skiplock) = @_;
2557 PVE
::LXC
::Config-
>lock_config($vmid, sub {
2558 return if !check_running
($vmid);
2560 vm_stop
($vmid, 0, $timeout, 1); # kill if timeout exceeds
2562 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2563 vm_start
($vmid, $conf);
2570 return PVE
::Tools
::run_fork
(sub {
2571 # Unshare the mount namespace
2572 die "failed to unshare mount namespace: $!\n"
2573 if !PVE
::Tools
::unshare
(PVE
::Tools
::CLONE_NEWNS
);
2574 run_command
(['mount', '--make-rslave', '/']);
2579 my $copy_volume = sub {
2580 my ($src_volid, $src, $dst_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid) = @_;
2582 my $src_mp = { volume
=> $src_volid, mp
=> '/', ro
=> 1 };
2583 $src_mp->{type
} = PVE
::LXC
::Config-
>classify_mountpoint($src_volid);
2585 my $dst_mp = { volume
=> $dst_volid, mp
=> '/', ro
=> 0 };
2586 $dst_mp->{type
} = PVE
::LXC
::Config-
>classify_mountpoint($dst_volid);
2592 mountpoint_mount
($src_mp, $src, $storage_cfg, $snapname, $rootuid, $rootgid);
2593 push @mounted, $src;
2595 mountpoint_mount
($dst_mp, $dest, $storage_cfg, undef, $rootuid, $rootgid);
2596 push @mounted, $dest;
2609 '--one-file-system',
2610 "--bwlimit=$bwlimit",
2617 # Wait for rsync's children to release dest so that
2618 # consequent file operations (umount, remove) are possible
2619 while ((system {"fuser"} "fuser", "-s", $dest) == 0) {sleep 1};
2621 foreach my $mount (reverse @mounted) {
2622 eval { run_command
(['/bin/umount', $mount], errfunc
=> sub{})};
2623 warn "Can't umount $mount\n" if $@;
2626 # If this fails they're used as mount points in a concurrent operation
2627 # (which should not happen but there's also no real need to get rid of them).
2634 # Should not be called after unsharing the mount namespace!
2636 my ($mp, $vmid, $storage, $storage_cfg, $conf, $snapname, $bwlimit) = @_;
2638 die "cannot copy volumes of type $mp->{type}\n" if $mp->{type
} ne 'volume';
2639 File
::Path
::make_path
("/var/lib/lxc/$vmid");
2640 my $dest = "/var/lib/lxc/$vmid/.copy-volume-1";
2641 my $src = "/var/lib/lxc/$vmid/.copy-volume-2";
2643 # get id's for unprivileged container
2644 my (undef, $rootuid, $rootgid) = parse_id_maps
($conf);
2646 # Allocate the disk before unsharing in order to make sure zfs subvolumes
2647 # are visible in this namespace, otherwise the host only sees the empty
2648 # (not-mounted) directory.
2651 # Make sure $mp contains a correct size.
2652 $mp->{size
} = PVE
::Storage
::volume_size_info
($storage_cfg, $mp->{volume
});
2654 ($new_volid, $needs_chown) = alloc_disk
($storage_cfg, $vmid, $storage, $mp->{size
}/1024, $rootuid, $rootgid);
2656 PVE
::Storage
::activate_volumes
($storage_cfg, [$new_volid], undef);
2657 my $path = PVE
::Storage
::path
($storage_cfg, $new_volid, undef);
2658 chown($rootuid, $rootgid, $path);
2662 $copy_volume->($mp->{volume
}, $src, $new_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid);
2666 PVE
::Storage
::vdisk_free
($storage_cfg, $new_volid)
2667 if defined($new_volid);
2674 sub get_lxc_version
() {
2676 run_command
([qw(lxc-start --version)], outfunc
=> sub {
2678 # We only parse out major & minor version numbers.
2679 if ($line =~ /^(\d+)\.(\d+)(?:\D.*)?$/) {
2680 $version = [$1, $2];
2684 die "failed to get lxc version\n" if !defined($version);
2687 return $version->@*;
2692 if (PVE
::CGroup
::cgroup_mode
() == 2) {
2693 PVE
::LXC
::Command
::freeze
($vmid, 30);
2695 PVE
::LXC
::CGroup-
>new($vmid)->freeze_thaw(1);
2701 if (PVE
::CGroup
::cgroup_mode
() == 2) {
2702 PVE
::LXC
::Command
::unfreeze
($vmid, 30);
2704 PVE
::LXC
::CGroup-
>new($vmid)->freeze_thaw(0);