2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <linux/loop.h>
30 #include <selinux/selinux.h>
37 #include <sys/mount.h>
38 #include <sys/personality.h>
39 #include <sys/prctl.h>
40 #include <sys/types.h>
45 #include "sd-daemon.h"
48 #include "alloc-util.h"
50 #include "base-filesystem.h"
51 #include "blkid-util.h"
52 #include "btrfs-util.h"
55 #include "capability-util.h"
56 #include "cgroup-util.h"
58 #include "dev-setup.h"
59 #include "dissect-image.h"
64 #include "format-util.h"
67 #include "hexdecoct.h"
68 #include "hostname-util.h"
69 #include "id128-util.h"
71 #include "loop-util.h"
72 #include "loopback-setup.h"
73 #include "machine-image.h"
77 #include "mount-util.h"
78 #include "netlink-util.h"
79 #include "nspawn-cgroup.h"
80 #include "nspawn-expose-ports.h"
81 #include "nspawn-mount.h"
82 #include "nspawn-network.h"
83 #include "nspawn-patch-uid.h"
84 #include "nspawn-register.h"
85 #include "nspawn-seccomp.h"
86 #include "nspawn-settings.h"
87 #include "nspawn-setuid.h"
88 #include "nspawn-stub-pid1.h"
89 #include "parse-util.h"
90 #include "path-util.h"
91 #include "process-util.h"
93 #include "random-util.h"
94 #include "raw-clone.h"
96 #include "selinux-util.h"
97 #include "signal-util.h"
98 #include "socket-util.h"
99 #include "stat-util.h"
100 #include "stdio-util.h"
101 #include "string-util.h"
103 #include "terminal-util.h"
104 #include "udev-util.h"
105 #include "umask-util.h"
106 #include "user-util.h"
109 /* Note that devpts's gid= parameter parses GIDs as signed values, hence we stay away from the upper half of the 32bit
110 * UID range here. We leave a bit of room at the lower end and a lot of room at the upper end, so that other subsystems
111 * may have their own allocation ranges too. */
112 #define UID_SHIFT_PICK_MIN ((uid_t) UINT32_C(0x00080000))
113 #define UID_SHIFT_PICK_MAX ((uid_t) UINT32_C(0x6FFF0000))
115 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
116 * nspawn_notify_socket_path is relative to the container
117 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
120 #define EXIT_FORCE_RESTART 133
122 typedef enum ContainerStatus
{
123 CONTAINER_TERMINATED
,
127 typedef enum LinkJournal
{
134 static char *arg_directory
= NULL
;
135 static char *arg_template
= NULL
;
136 static char *arg_chdir
= NULL
;
137 static char *arg_pivot_root_new
= NULL
;
138 static char *arg_pivot_root_old
= NULL
;
139 static char *arg_user
= NULL
;
140 static sd_id128_t arg_uuid
= {};
141 static char *arg_machine
= NULL
;
142 static const char *arg_selinux_context
= NULL
;
143 static const char *arg_selinux_apifs_context
= NULL
;
144 static const char *arg_slice
= NULL
;
145 static bool arg_private_network
= false;
146 static bool arg_read_only
= false;
147 static StartMode arg_start_mode
= START_PID1
;
148 static bool arg_ephemeral
= false;
149 static LinkJournal arg_link_journal
= LINK_AUTO
;
150 static bool arg_link_journal_try
= false;
151 static uint64_t arg_caps_retain
=
152 (1ULL << CAP_AUDIT_CONTROL
) |
153 (1ULL << CAP_AUDIT_WRITE
) |
154 (1ULL << CAP_CHOWN
) |
155 (1ULL << CAP_DAC_OVERRIDE
) |
156 (1ULL << CAP_DAC_READ_SEARCH
) |
157 (1ULL << CAP_FOWNER
) |
158 (1ULL << CAP_FSETID
) |
159 (1ULL << CAP_IPC_OWNER
) |
161 (1ULL << CAP_LEASE
) |
162 (1ULL << CAP_LINUX_IMMUTABLE
) |
163 (1ULL << CAP_MKNOD
) |
164 (1ULL << CAP_NET_BIND_SERVICE
) |
165 (1ULL << CAP_NET_BROADCAST
) |
166 (1ULL << CAP_NET_RAW
) |
167 (1ULL << CAP_SETFCAP
) |
168 (1ULL << CAP_SETGID
) |
169 (1ULL << CAP_SETPCAP
) |
170 (1ULL << CAP_SETUID
) |
171 (1ULL << CAP_SYS_ADMIN
) |
172 (1ULL << CAP_SYS_BOOT
) |
173 (1ULL << CAP_SYS_CHROOT
) |
174 (1ULL << CAP_SYS_NICE
) |
175 (1ULL << CAP_SYS_PTRACE
) |
176 (1ULL << CAP_SYS_RESOURCE
) |
177 (1ULL << CAP_SYS_TTY_CONFIG
);
178 static CustomMount
*arg_custom_mounts
= NULL
;
179 static unsigned arg_n_custom_mounts
= 0;
180 static char **arg_setenv
= NULL
;
181 static bool arg_quiet
= false;
182 static bool arg_register
= true;
183 static bool arg_keep_unit
= false;
184 static char **arg_network_interfaces
= NULL
;
185 static char **arg_network_macvlan
= NULL
;
186 static char **arg_network_ipvlan
= NULL
;
187 static bool arg_network_veth
= false;
188 static char **arg_network_veth_extra
= NULL
;
189 static char *arg_network_bridge
= NULL
;
190 static char *arg_network_zone
= NULL
;
191 static unsigned long arg_personality
= PERSONALITY_INVALID
;
192 static char *arg_image
= NULL
;
193 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
194 static ExposePort
*arg_expose_ports
= NULL
;
195 static char **arg_property
= NULL
;
196 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
197 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
198 static bool arg_userns_chown
= false;
199 static int arg_kill_signal
= 0;
200 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
201 static SettingsMask arg_settings_mask
= 0;
202 static int arg_settings_trusted
= -1;
203 static char **arg_parameters
= NULL
;
204 static const char *arg_container_service_name
= "systemd-nspawn";
205 static bool arg_notify_ready
= false;
206 static bool arg_use_cgns
= true;
207 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
208 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
;
209 static void *arg_root_hash
= NULL
;
210 static size_t arg_root_hash_size
= 0;
212 static void help(void) {
213 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
214 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
215 " -h --help Show this help\n"
216 " --version Print version string\n"
217 " -q --quiet Do not show status information\n"
218 " -D --directory=PATH Root directory for the container\n"
219 " --template=PATH Initialize root directory from template directory,\n"
221 " -x --ephemeral Run container with snapshot of root directory, and\n"
222 " remove it after exit\n"
223 " -i --image=PATH File system device or disk image for the container\n"
224 " --root-hash=HASH Specify verity root hash\n"
225 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
226 " -b --boot Boot up full system (i.e. invoke init)\n"
227 " --chdir=PATH Set working directory in the container\n"
228 " --pivot-root=PATH[:PATH]\n"
229 " Pivot root to given directory in the container\n"
230 " -u --user=USER Run the command under specified user or uid\n"
231 " -M --machine=NAME Set the machine name for the container\n"
232 " --uuid=UUID Set a specific machine UUID for the container\n"
233 " -S --slice=SLICE Place the container in the specified slice\n"
234 " --property=NAME=VALUE Set scope unit property\n"
235 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
236 " --private-users[=UIDBASE[:NUIDS]]\n"
237 " Similar, but with user configured UID/GID range\n"
238 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
239 " --private-network Disable network in container\n"
240 " --network-interface=INTERFACE\n"
241 " Assign an existing network interface to the\n"
243 " --network-macvlan=INTERFACE\n"
244 " Create a macvlan network interface based on an\n"
245 " existing network interface to the container\n"
246 " --network-ipvlan=INTERFACE\n"
247 " Create a ipvlan network interface based on an\n"
248 " existing network interface to the container\n"
249 " -n --network-veth Add a virtual Ethernet connection between host\n"
251 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
252 " Add an additional virtual Ethernet link between\n"
253 " host and container\n"
254 " --network-bridge=INTERFACE\n"
255 " Add a virtual Ethernet connection to the container\n"
256 " and attach it to an existing bridge on the host\n"
257 " --network-zone=NAME Similar, but attach the new interface to an\n"
258 " an automatically managed bridge interface\n"
259 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
260 " Expose a container IP port on the host\n"
261 " -Z --selinux-context=SECLABEL\n"
262 " Set the SELinux security context to be used by\n"
263 " processes in the container\n"
264 " -L --selinux-apifs-context=SECLABEL\n"
265 " Set the SELinux security context to be used by\n"
266 " API/tmpfs file systems in the container\n"
267 " --capability=CAP In addition to the default, retain specified\n"
269 " --drop-capability=CAP Drop the specified capability from the default set\n"
270 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
271 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
272 " host, try-guest, try-host\n"
273 " -j Equivalent to --link-journal=try-guest\n"
274 " --read-only Mount the root directory read-only\n"
275 " --bind=PATH[:PATH[:OPTIONS]]\n"
276 " Bind mount a file or directory from the host into\n"
278 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
279 " Similar, but creates a read-only bind mount\n"
280 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
281 " --overlay=PATH[:PATH...]:PATH\n"
282 " Create an overlay mount from the host to \n"
284 " --overlay-ro=PATH[:PATH...]:PATH\n"
285 " Similar, but creates a read-only overlay mount\n"
286 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
287 " --register=BOOLEAN Register container as machine\n"
288 " --keep-unit Do not register a scope for the machine, reuse\n"
289 " the service unit nspawn is running in\n"
290 " --volatile[=MODE] Run the system in volatile mode\n"
291 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
292 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
293 , program_invocation_short_name
);
296 static int custom_mount_check_all(void) {
299 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
300 CustomMount
*m
= &arg_custom_mounts
[i
];
302 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
304 if (arg_userns_chown
) {
305 log_error("--private-users-chown may not be combined with custom root mounts.");
307 } else if (arg_uid_shift
== UID_INVALID
) {
308 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
317 static int detect_unified_cgroup_hierarchy(const char *directory
) {
321 /* Allow the user to control whether the unified hierarchy is used */
322 e
= getenv("UNIFIED_CGROUP_HIERARCHY");
324 r
= parse_boolean(e
);
326 return log_error_errno(r
, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
328 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
330 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
335 /* Otherwise inherit the default from the host system */
336 r
= cg_all_unified();
338 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
340 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
341 * routine only detects 231, so we'll have a false negative here for 230. */
342 r
= systemd_installation_has_version(directory
, 230);
344 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
346 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
348 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
349 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
350 /* Mixed cgroup hierarchy support was added in 233 */
351 r
= systemd_installation_has_version(directory
, 233);
353 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
355 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
357 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
359 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
364 static void parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
367 r
= getenv_bool(name
);
371 log_warning_errno(r
, "Failed to parse %s from environment, defaulting to false.", name
);
372 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
375 static void parse_mount_settings_env(void) {
379 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
383 if (streq(e
, "network")) {
384 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
388 r
= parse_boolean(e
);
390 log_warning_errno(r
, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
393 arg_mount_settings
&= ~MOUNT_APPLY_APIVFS_RO
;
395 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
;
397 arg_mount_settings
&= ~MOUNT_APPLY_APIVFS_NETNS
;
400 static int parse_argv(int argc
, char *argv
[]) {
418 ARG_NETWORK_INTERFACE
,
423 ARG_NETWORK_VETH_EXTRA
,
433 ARG_PRIVATE_USERS_CHOWN
,
438 static const struct option options
[] = {
439 { "help", no_argument
, NULL
, 'h' },
440 { "version", no_argument
, NULL
, ARG_VERSION
},
441 { "directory", required_argument
, NULL
, 'D' },
442 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
443 { "ephemeral", no_argument
, NULL
, 'x' },
444 { "user", required_argument
, NULL
, 'u' },
445 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
446 { "as-pid2", no_argument
, NULL
, 'a' },
447 { "boot", no_argument
, NULL
, 'b' },
448 { "uuid", required_argument
, NULL
, ARG_UUID
},
449 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
450 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
451 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
452 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
453 { "bind", required_argument
, NULL
, ARG_BIND
},
454 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
455 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
456 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
457 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
458 { "machine", required_argument
, NULL
, 'M' },
459 { "slice", required_argument
, NULL
, 'S' },
460 { "setenv", required_argument
, NULL
, 'E' },
461 { "selinux-context", required_argument
, NULL
, 'Z' },
462 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
463 { "quiet", no_argument
, NULL
, 'q' },
464 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
465 { "register", required_argument
, NULL
, ARG_REGISTER
},
466 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
467 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
468 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
469 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
470 { "network-veth", no_argument
, NULL
, 'n' },
471 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
472 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
473 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
474 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
475 { "image", required_argument
, NULL
, 'i' },
476 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
477 { "port", required_argument
, NULL
, 'p' },
478 { "property", required_argument
, NULL
, ARG_PROPERTY
},
479 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
480 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
481 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
482 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
483 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
484 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
485 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
486 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
492 uint64_t plus
= 0, minus
= 0;
493 bool mask_all_settings
= false, mask_no_settings
= false;
498 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options
, NULL
)) >= 0)
510 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
516 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
522 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
528 arg_ephemeral
= true;
532 r
= free_and_strdup(&arg_user
, optarg
);
536 arg_settings_mask
|= SETTING_USER
;
539 case ARG_NETWORK_ZONE
: {
542 j
= strappend("vz-", optarg
);
546 if (!ifname_valid(j
)) {
547 log_error("Network zone name not valid: %s", j
);
552 free(arg_network_zone
);
553 arg_network_zone
= j
;
555 arg_network_veth
= true;
556 arg_private_network
= true;
557 arg_settings_mask
|= SETTING_NETWORK
;
561 case ARG_NETWORK_BRIDGE
:
563 if (!ifname_valid(optarg
)) {
564 log_error("Bridge interface name not valid: %s", optarg
);
568 r
= free_and_strdup(&arg_network_bridge
, optarg
);
575 arg_network_veth
= true;
576 arg_private_network
= true;
577 arg_settings_mask
|= SETTING_NETWORK
;
580 case ARG_NETWORK_VETH_EXTRA
:
581 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
583 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
585 arg_private_network
= true;
586 arg_settings_mask
|= SETTING_NETWORK
;
589 case ARG_NETWORK_INTERFACE
:
591 if (!ifname_valid(optarg
)) {
592 log_error("Network interface name not valid: %s", optarg
);
596 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
599 arg_private_network
= true;
600 arg_settings_mask
|= SETTING_NETWORK
;
603 case ARG_NETWORK_MACVLAN
:
605 if (!ifname_valid(optarg
)) {
606 log_error("MACVLAN network interface name not valid: %s", optarg
);
610 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
613 arg_private_network
= true;
614 arg_settings_mask
|= SETTING_NETWORK
;
617 case ARG_NETWORK_IPVLAN
:
619 if (!ifname_valid(optarg
)) {
620 log_error("IPVLAN network interface name not valid: %s", optarg
);
624 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
629 case ARG_PRIVATE_NETWORK
:
630 arg_private_network
= true;
631 arg_settings_mask
|= SETTING_NETWORK
;
635 if (arg_start_mode
== START_PID2
) {
636 log_error("--boot and --as-pid2 may not be combined.");
640 arg_start_mode
= START_BOOT
;
641 arg_settings_mask
|= SETTING_START_MODE
;
645 if (arg_start_mode
== START_BOOT
) {
646 log_error("--boot and --as-pid2 may not be combined.");
650 arg_start_mode
= START_PID2
;
651 arg_settings_mask
|= SETTING_START_MODE
;
655 r
= sd_id128_from_string(optarg
, &arg_uuid
);
657 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
659 if (sd_id128_is_null(arg_uuid
)) {
660 log_error("Machine UUID may not be all zeroes.");
664 arg_settings_mask
|= SETTING_MACHINE_ID
;
673 arg_machine
= mfree(arg_machine
);
675 if (!machine_name_is_valid(optarg
)) {
676 log_error("Invalid machine name: %s", optarg
);
680 r
= free_and_strdup(&arg_machine
, optarg
);
687 arg_selinux_context
= optarg
;
691 arg_selinux_apifs_context
= optarg
;
695 arg_read_only
= true;
696 arg_settings_mask
|= SETTING_READ_ONLY
;
700 case ARG_DROP_CAPABILITY
: {
703 _cleanup_free_
char *t
= NULL
;
705 r
= extract_first_word(&p
, &t
, ",", 0);
707 return log_error_errno(r
, "Failed to parse capability %s.", t
);
712 if (streq(t
, "all")) {
713 if (c
== ARG_CAPABILITY
)
714 plus
= (uint64_t) -1;
716 minus
= (uint64_t) -1;
720 cap
= capability_from_name(t
);
722 log_error("Failed to parse capability %s.", t
);
726 if (c
== ARG_CAPABILITY
)
727 plus
|= 1ULL << (uint64_t) cap
;
729 minus
|= 1ULL << (uint64_t) cap
;
733 arg_settings_mask
|= SETTING_CAPABILITY
;
738 arg_link_journal
= LINK_GUEST
;
739 arg_link_journal_try
= true;
742 case ARG_LINK_JOURNAL
:
743 if (streq(optarg
, "auto")) {
744 arg_link_journal
= LINK_AUTO
;
745 arg_link_journal_try
= false;
746 } else if (streq(optarg
, "no")) {
747 arg_link_journal
= LINK_NO
;
748 arg_link_journal_try
= false;
749 } else if (streq(optarg
, "guest")) {
750 arg_link_journal
= LINK_GUEST
;
751 arg_link_journal_try
= false;
752 } else if (streq(optarg
, "host")) {
753 arg_link_journal
= LINK_HOST
;
754 arg_link_journal_try
= false;
755 } else if (streq(optarg
, "try-guest")) {
756 arg_link_journal
= LINK_GUEST
;
757 arg_link_journal_try
= true;
758 } else if (streq(optarg
, "try-host")) {
759 arg_link_journal
= LINK_HOST
;
760 arg_link_journal_try
= true;
762 log_error("Failed to parse link journal mode %s", optarg
);
770 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
772 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
774 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
778 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
780 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
782 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
787 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
788 if (r
== -EADDRNOTAVAIL
)
789 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
791 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
793 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
799 if (!env_assignment_is_valid(optarg
)) {
800 log_error("Environment variable assignment '%s' is not valid.", optarg
);
804 n
= strv_env_set(arg_setenv
, optarg
);
808 strv_free(arg_setenv
);
811 arg_settings_mask
|= SETTING_ENVIRONMENT
;
819 case ARG_SHARE_SYSTEM
:
820 /* We don't officially support this anymore, except for compat reasons. People should use the
821 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
822 arg_clone_ns_flags
= 0;
826 r
= parse_boolean(optarg
);
828 log_error("Failed to parse --register= argument: %s", optarg
);
836 arg_keep_unit
= true;
839 case ARG_PERSONALITY
:
841 arg_personality
= personality_from_string(optarg
);
842 if (arg_personality
== PERSONALITY_INVALID
) {
843 log_error("Unknown or unsupported personality '%s'.", optarg
);
847 arg_settings_mask
|= SETTING_PERSONALITY
;
853 arg_volatile_mode
= VOLATILE_YES
;
857 m
= volatile_mode_from_string(optarg
);
859 log_error("Failed to parse --volatile= argument: %s", optarg
);
862 arg_volatile_mode
= m
;
865 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
869 r
= expose_port_parse(&arg_expose_ports
, optarg
);
871 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
873 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
875 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
879 if (strv_extend(&arg_property
, optarg
) < 0)
884 case ARG_PRIVATE_USERS
: {
889 else if (!in_charset(optarg
, DIGITS
))
890 /* do *not* parse numbers as booleans */
891 boolean
= parse_boolean(optarg
);
893 if (boolean
== false) {
894 /* no: User namespacing off */
895 arg_userns_mode
= USER_NAMESPACE_NO
;
896 arg_uid_shift
= UID_INVALID
;
897 arg_uid_range
= UINT32_C(0x10000);
898 } else if (boolean
== true) {
899 /* yes: User namespacing on, UID range is read from root dir */
900 arg_userns_mode
= USER_NAMESPACE_FIXED
;
901 arg_uid_shift
= UID_INVALID
;
902 arg_uid_range
= UINT32_C(0x10000);
903 } else if (streq(optarg
, "pick")) {
904 /* pick: User namespacing on, UID range is picked randomly */
905 arg_userns_mode
= USER_NAMESPACE_PICK
;
906 arg_uid_shift
= UID_INVALID
;
907 arg_uid_range
= UINT32_C(0x10000);
909 _cleanup_free_
char *buffer
= NULL
;
910 const char *range
, *shift
;
912 /* anything else: User namespacing on, UID range is explicitly configured */
914 range
= strchr(optarg
, ':');
916 buffer
= strndup(optarg
, range
- optarg
);
922 r
= safe_atou32(range
, &arg_uid_range
);
924 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
928 r
= parse_uid(shift
, &arg_uid_shift
);
930 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
932 arg_userns_mode
= USER_NAMESPACE_FIXED
;
935 if (arg_uid_range
<= 0) {
936 log_error("UID range cannot be 0.");
940 arg_settings_mask
|= SETTING_USERNS
;
945 if (userns_supported()) {
946 arg_userns_mode
= USER_NAMESPACE_PICK
;
947 arg_uid_shift
= UID_INVALID
;
948 arg_uid_range
= UINT32_C(0x10000);
950 arg_settings_mask
|= SETTING_USERNS
;
955 case ARG_PRIVATE_USERS_CHOWN
:
956 arg_userns_chown
= true;
958 arg_settings_mask
|= SETTING_USERNS
;
961 case ARG_KILL_SIGNAL
:
962 arg_kill_signal
= signal_from_string_try_harder(optarg
);
963 if (arg_kill_signal
< 0) {
964 log_error("Cannot parse signal: %s", optarg
);
968 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
973 /* no → do not read files
974 * yes → read files, do not override cmdline, trust only subset
975 * override → read files, override cmdline, trust only subset
976 * trusted → read files, do not override cmdline, trust all
979 r
= parse_boolean(optarg
);
981 if (streq(optarg
, "trusted")) {
982 mask_all_settings
= false;
983 mask_no_settings
= false;
984 arg_settings_trusted
= true;
986 } else if (streq(optarg
, "override")) {
987 mask_all_settings
= false;
988 mask_no_settings
= true;
989 arg_settings_trusted
= -1;
991 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
994 mask_all_settings
= false;
995 mask_no_settings
= false;
996 arg_settings_trusted
= -1;
999 mask_all_settings
= true;
1000 mask_no_settings
= false;
1001 arg_settings_trusted
= false;
1007 if (!path_is_absolute(optarg
)) {
1008 log_error("Working directory %s is not an absolute path.", optarg
);
1012 r
= free_and_strdup(&arg_chdir
, optarg
);
1016 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1019 case ARG_PIVOT_ROOT
:
1020 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1022 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1024 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1027 case ARG_NOTIFY_READY
:
1028 r
= parse_boolean(optarg
);
1030 log_error("%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1033 arg_notify_ready
= r
;
1034 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1037 case ARG_ROOT_HASH
: {
1041 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1043 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1044 if (l
< sizeof(sd_id128_t
)) {
1045 log_error("Root hash must be at least 128bit long: %s", optarg
);
1050 free(arg_root_hash
);
1052 arg_root_hash_size
= l
;
1060 assert_not_reached("Unhandled option");
1063 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
1064 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
1065 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
1066 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
1068 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1069 arg_mount_settings
|= MOUNT_USE_USERNS
;
1071 if (arg_private_network
)
1072 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1074 parse_mount_settings_env();
1076 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1077 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1078 arg_register
= false;
1079 if (arg_start_mode
!= START_PID1
) {
1080 log_error("--boot cannot be used without namespacing.");
1085 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1086 arg_userns_chown
= true;
1088 if (arg_keep_unit
&& cg_pid_get_owner_uid(0, NULL
) >= 0) {
1089 log_error("--keep-unit may not be used when invoked from a user session.");
1093 if (arg_directory
&& arg_image
) {
1094 log_error("--directory= and --image= may not be combined.");
1098 if (arg_template
&& arg_image
) {
1099 log_error("--template= and --image= may not be combined.");
1103 if (arg_ephemeral
&& arg_template
&& !arg_directory
) {
1104 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1105 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1106 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1109 arg_directory
= arg_template
;
1110 arg_template
= NULL
;
1113 if (arg_template
&& !(arg_directory
|| arg_machine
)) {
1114 log_error("--template= needs --directory= or --machine=.");
1118 if (arg_ephemeral
&& arg_template
) {
1119 log_error("--ephemeral and --template= may not be combined.");
1123 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
)) {
1124 log_error("--ephemeral and --link-journal= may not be combined.");
1128 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported()) {
1129 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1133 if (arg_userns_chown
&& arg_read_only
) {
1134 log_error("--read-only and --private-users-chown may not be combined.");
1138 if (arg_network_bridge
&& arg_network_zone
) {
1139 log_error("--network-bridge= and --network-zone= may not be combined.");
1143 if (argc
> optind
) {
1144 arg_parameters
= strv_copy(argv
+ optind
);
1145 if (!arg_parameters
)
1148 arg_settings_mask
|= SETTING_START_MODE
;
1151 /* Load all settings from .nspawn files */
1152 if (mask_no_settings
)
1153 arg_settings_mask
= 0;
1155 /* Don't load any settings from .nspawn files */
1156 if (mask_all_settings
)
1157 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1159 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? 1ULL << CAP_NET_ADMIN
: 0)) & ~minus
;
1161 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1163 arg_container_service_name
= e
;
1165 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1167 arg_use_cgns
= cg_ns_supported();
1171 r
= custom_mount_check_all();
1178 static int verify_arguments(void) {
1179 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
) {
1180 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1184 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
)) {
1185 log_error("Cannot combine --private-users with read-write mounts.");
1189 if (arg_volatile_mode
!= VOLATILE_NO
&& arg_read_only
) {
1190 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1194 if (arg_expose_ports
&& !arg_private_network
) {
1195 log_error("Cannot use --port= without private networking.");
1199 #ifndef HAVE_LIBIPTC
1200 if (arg_expose_ports
) {
1201 log_error("--port= is not supported, compiled without libiptc support.");
1206 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1207 arg_kill_signal
= SIGRTMIN
+3;
1212 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1215 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1218 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1221 if (uid
!= UID_INVALID
) {
1222 uid
+= arg_uid_shift
;
1224 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1228 if (gid
!= GID_INVALID
) {
1229 gid
+= (gid_t
) arg_uid_shift
;
1231 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1235 if (lchown(p
, uid
, gid
) < 0)
1241 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1244 q
= prefix_roota(root
, path
);
1245 if (mkdir(q
, mode
) < 0) {
1246 if (errno
== EEXIST
)
1251 return userns_lchown(q
, uid
, gid
);
1254 static int setup_timezone(const char *dest
) {
1255 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
1256 const char *where
, *check
, *what
;
1262 /* Fix the timezone, if possible */
1263 r
= readlink_malloc("/etc/localtime", &p
);
1265 log_warning("host's /etc/localtime is not a symlink, not updating container timezone.");
1266 /* to handle warning, delete /etc/localtime and replace it
1267 * with a symbolic link to a time zone data file.
1270 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1275 z
= path_startswith(p
, "../usr/share/zoneinfo/");
1277 z
= path_startswith(p
, "/usr/share/zoneinfo/");
1279 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1283 where
= prefix_roota(dest
, "/etc/localtime");
1284 r
= readlink_malloc(where
, &q
);
1286 y
= path_startswith(q
, "../usr/share/zoneinfo/");
1288 y
= path_startswith(q
, "/usr/share/zoneinfo/");
1290 /* Already pointing to the right place? Then do nothing .. */
1291 if (y
&& streq(y
, z
))
1295 check
= strjoina("/usr/share/zoneinfo/", z
);
1296 check
= prefix_roota(dest
, check
);
1297 if (laccess(check
, F_OK
) < 0) {
1298 log_warning("Timezone %s does not exist in container, not updating container timezone.", z
);
1302 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1303 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1305 "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1309 what
= strjoina("../usr/share/zoneinfo/", z
);
1310 if (symlink(what
, where
) < 0) {
1311 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1313 "Failed to correct timezone of container, ignoring: %m");
1317 r
= userns_lchown(where
, 0, 0);
1319 return log_warning_errno(r
, "Failed to chown /etc/localtime: %m");
1324 static int resolved_running(void) {
1325 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1328 /* Check if resolved is running */
1330 r
= sd_bus_open_system(&bus
);
1334 return bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1337 static int setup_resolv_conf(const char *dest
) {
1338 _cleanup_free_
char *resolved
= NULL
, *etc
= NULL
;
1344 if (arg_private_network
)
1347 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1349 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1353 where
= strjoina(etc
, "/resolv.conf");
1354 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1356 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1360 if (access("/usr/lib/systemd/resolv.conf", F_OK
) >= 0 &&
1361 resolved_running() > 0) {
1363 /* resolved is enabled on the host. In this, case bind mount its static resolv.conf file into the
1364 * container, so that the container can use the host's resolver. Given that network namespacing is
1365 * disabled it's only natural of the container also uses the host's resolver. It also has the big
1366 * advantage that the container will be able to follow the host's DNS server configuration changes
1369 if (found
== 0) /* missing? */
1370 (void) touch(resolved
);
1372 r
= mount_verbose(LOG_DEBUG
, "/usr/lib/systemd/resolv.conf", resolved
, NULL
, MS_BIND
, NULL
);
1374 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1377 /* If that didn't work, let's copy the file */
1378 r
= copy_file("/etc/resolv.conf", where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, COPY_REFLINK
);
1380 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1381 * resolved or something similar runs inside and the symlink points there.
1383 * If the disk image is read-only, there's also no point in complaining.
1385 log_full_errno(IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1386 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1390 r
= userns_lchown(where
, 0, 0);
1392 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1397 static int setup_boot_id(const char *dest
) {
1398 sd_id128_t rnd
= SD_ID128_NULL
;
1399 const char *from
, *to
;
1402 /* Generate a new randomized boot ID, so that each boot-up of
1403 * the container gets a new one */
1405 from
= prefix_roota(dest
, "/run/proc-sys-kernel-random-boot-id");
1406 to
= prefix_roota(dest
, "/proc/sys/kernel/random/boot_id");
1408 r
= sd_id128_randomize(&rnd
);
1410 return log_error_errno(r
, "Failed to generate random boot id: %m");
1412 r
= id128_write(from
, ID128_UUID
, rnd
, false);
1414 return log_error_errno(r
, "Failed to write boot id: %m");
1416 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1418 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
1419 MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1421 (void) unlink(from
);
1425 static int copy_devnodes(const char *dest
) {
1427 static const char devnodes
[] =
1438 _cleanup_umask_ mode_t u
;
1444 /* Create /dev/net, so that we can create /dev/net/tun in it */
1445 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1446 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1448 NULSTR_FOREACH(d
, devnodes
) {
1449 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1452 from
= strappend("/dev/", d
);
1453 to
= prefix_root(dest
, from
);
1455 if (stat(from
, &st
) < 0) {
1457 if (errno
!= ENOENT
)
1458 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1460 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
1462 log_error("%s is not a char or block device, cannot copy.", from
);
1466 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1467 /* Explicitly warn the user when /dev is already populated. */
1468 if (errno
== EEXIST
)
1469 log_notice("%s/dev is pre-mounted and pre-populated. If a pre-mounted /dev is provided it needs to be an unpopulated file system.", dest
);
1471 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
1473 /* Some systems abusively restrict mknod but
1474 * allow bind mounts. */
1477 return log_error_errno(r
, "touch (%s) failed: %m", to
);
1478 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
1480 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
1483 r
= userns_lchown(to
, 0, 0);
1485 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
1492 static int setup_pts(const char *dest
) {
1493 _cleanup_free_
char *options
= NULL
;
1498 if (arg_selinux_apifs_context
)
1499 (void) asprintf(&options
,
1500 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
1501 arg_uid_shift
+ TTY_GID
,
1502 arg_selinux_apifs_context
);
1505 (void) asprintf(&options
,
1506 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
1507 arg_uid_shift
+ TTY_GID
);
1512 /* Mount /dev/pts itself */
1513 p
= prefix_roota(dest
, "/dev/pts");
1514 if (mkdir(p
, 0755) < 0)
1515 return log_error_errno(errno
, "Failed to create /dev/pts: %m");
1516 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
1519 r
= userns_lchown(p
, 0, 0);
1521 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
1523 /* Create /dev/ptmx symlink */
1524 p
= prefix_roota(dest
, "/dev/ptmx");
1525 if (symlink("pts/ptmx", p
) < 0)
1526 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
1527 r
= userns_lchown(p
, 0, 0);
1529 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
1531 /* And fix /dev/pts/ptmx ownership */
1532 p
= prefix_roota(dest
, "/dev/pts/ptmx");
1533 r
= userns_lchown(p
, 0, 0);
1535 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
1540 static int setup_dev_console(const char *dest
, const char *console
) {
1541 _cleanup_umask_ mode_t u
;
1550 r
= chmod_and_chown(console
, 0600, arg_uid_shift
, arg_uid_shift
);
1552 return log_error_errno(r
, "Failed to correct access mode for TTY: %m");
1554 /* We need to bind mount the right tty to /dev/console since
1555 * ptys can only exist on pts file systems. To have something
1556 * to bind mount things on we create a empty regular file. */
1558 to
= prefix_roota(dest
, "/dev/console");
1561 return log_error_errno(r
, "touch() for /dev/console failed: %m");
1563 return mount_verbose(LOG_ERR
, console
, to
, NULL
, MS_BIND
, NULL
);
1566 static int setup_kmsg(const char *dest
, int kmsg_socket
) {
1567 const char *from
, *to
;
1568 _cleanup_umask_ mode_t u
;
1571 assert(kmsg_socket
>= 0);
1575 /* We create the kmsg FIFO as /run/kmsg, but immediately
1576 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1577 * on the reading side behave very similar to /proc/kmsg,
1578 * their writing side behaves differently from /dev/kmsg in
1579 * that writing blocks when nothing is reading. In order to
1580 * avoid any problems with containers deadlocking due to this
1581 * we simply make /dev/kmsg unavailable to the container. */
1582 from
= prefix_roota(dest
, "/run/kmsg");
1583 to
= prefix_roota(dest
, "/proc/kmsg");
1585 if (mkfifo(from
, 0600) < 0)
1586 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
1587 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1591 fd
= open(from
, O_RDWR
|O_NDELAY
|O_CLOEXEC
);
1593 return log_error_errno(errno
, "Failed to open fifo: %m");
1595 /* Store away the fd in the socket, so that it stays open as
1596 * long as we run the child */
1597 r
= send_one_fd(kmsg_socket
, fd
, 0);
1601 return log_error_errno(r
, "Failed to send FIFO fd: %m");
1603 /* And now make the FIFO unavailable as /run/kmsg... */
1604 (void) unlink(from
);
1609 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
1610 union in_addr_union
*exposed
= userdata
;
1616 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
1620 static int setup_hostname(void) {
1622 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
1625 if (sethostname_idempotent(arg_machine
) < 0)
1631 static int setup_journal(const char *directory
) {
1633 _cleanup_free_
char *d
= NULL
;
1639 /* Don't link journals in ephemeral mode */
1643 if (arg_link_journal
== LINK_NO
)
1646 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
1648 r
= sd_id128_get_machine(&this_id
);
1650 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
1652 if (sd_id128_equal(arg_uuid
, this_id
)) {
1653 log_full(try ? LOG_WARNING
: LOG_ERR
,
1654 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
1660 r
= userns_mkdir(directory
, "/var", 0755, 0, 0);
1662 return log_error_errno(r
, "Failed to create /var: %m");
1664 r
= userns_mkdir(directory
, "/var/log", 0755, 0, 0);
1666 return log_error_errno(r
, "Failed to create /var/log: %m");
1668 r
= userns_mkdir(directory
, "/var/log/journal", 0755, 0, 0);
1670 return log_error_errno(r
, "Failed to create /var/log/journal: %m");
1672 (void) sd_id128_to_string(arg_uuid
, id
);
1674 p
= strjoina("/var/log/journal/", id
);
1675 q
= prefix_roota(directory
, p
);
1677 if (path_is_mount_point(p
, NULL
, 0) > 0) {
1681 log_error("%s: already a mount point, refusing to use for journal", p
);
1685 if (path_is_mount_point(q
, NULL
, 0) > 0) {
1689 log_error("%s: already a mount point, refusing to use for journal", q
);
1693 r
= readlink_and_make_absolute(p
, &d
);
1695 if ((arg_link_journal
== LINK_GUEST
||
1696 arg_link_journal
== LINK_AUTO
) &&
1699 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
1701 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
1706 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
1707 } else if (r
== -EINVAL
) {
1709 if (arg_link_journal
== LINK_GUEST
&&
1712 if (errno
== ENOTDIR
) {
1713 log_error("%s already exists and is neither a symlink nor a directory", p
);
1716 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
1718 } else if (r
!= -ENOENT
)
1719 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
1721 if (arg_link_journal
== LINK_GUEST
) {
1723 if (symlink(q
, p
) < 0) {
1725 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
1728 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
1731 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
1733 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
1737 if (arg_link_journal
== LINK_HOST
) {
1738 /* don't create parents here — if the host doesn't have
1739 * permanent journal set up, don't force it here */
1741 if (mkdir(p
, 0755) < 0 && errno
!= EEXIST
) {
1743 log_debug_errno(errno
, "Failed to create %s, skipping journal setup: %m", p
);
1746 return log_error_errno(errno
, "Failed to create %s: %m", p
);
1749 } else if (access(p
, F_OK
) < 0)
1752 if (dir_is_empty(q
) == 0)
1753 log_warning("%s is not empty, proceeding anyway.", q
);
1755 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
1757 return log_error_errno(r
, "Failed to create %s: %m", q
);
1759 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
1761 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
1766 static int drop_capabilities(void) {
1767 return capability_bounding_set_drop(arg_caps_retain
, false);
1770 static int reset_audit_loginuid(void) {
1771 _cleanup_free_
char *p
= NULL
;
1774 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
1777 r
= read_one_line_file("/proc/self/loginuid", &p
);
1781 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
1783 /* Already reset? */
1784 if (streq(p
, "4294967295"))
1787 r
= write_string_file("/proc/self/loginuid", "4294967295", 0);
1790 "Failed to reset audit login UID. This probably means that your kernel is too\n"
1791 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1792 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1793 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1794 "using systemd-nspawn. Sleeping for 5s... (%m)");
1803 static int setup_propagate(const char *root
) {
1807 (void) mkdir_p("/run/systemd/nspawn/", 0755);
1808 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
1809 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
1810 (void) mkdir_p(p
, 0600);
1812 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
1814 return log_error_errno(r
, "Failed to create /run/systemd: %m");
1816 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
1818 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
1820 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
1822 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
1824 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
1825 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
1829 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
1833 /* machined will MS_MOVE into that directory, and that's only
1834 * supported for non-shared mounts. */
1835 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
1838 static int setup_machine_id(const char *directory
) {
1839 const char *etc_machine_id
;
1843 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
1844 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
1845 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
1846 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
1847 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
1848 * container behaves nicely). */
1850 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
1852 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
1854 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
1855 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
1857 if (sd_id128_is_null(arg_uuid
)) {
1858 r
= sd_id128_randomize(&arg_uuid
);
1860 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
1863 if (sd_id128_is_null(id
)) {
1864 log_error("Machine ID in container image is zero, refusing.");
1874 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
1879 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
1882 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
1883 if (r
== -EOPNOTSUPP
)
1884 return log_error_errno(r
, "Automatic UID/GID adjusting is only supported for UID/GID ranges starting at multiples of 2^16 with a range of 2^16.");
1886 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
1888 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
1890 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
1892 log_debug("Patched directory tree to match UID/GID range.");
1899 * < 0 : wait_for_terminate() failed to get the state of the
1900 * container, the container was terminated by a signal, or
1901 * failed for an unknown reason. No change is made to the
1902 * container argument.
1903 * > 0 : The program executed in the container terminated with an
1904 * error. The exit code of the program executed in the
1905 * container is returned. The container argument has been set
1906 * to CONTAINER_TERMINATED.
1907 * 0 : The container is being rebooted, has been shut down or exited
1908 * successfully. The container argument has been set to either
1909 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
1911 * That is, success is indicated by a return value of zero, and an
1912 * error is indicated by a non-zero value.
1914 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
1918 r
= wait_for_terminate(pid
, &status
);
1920 return log_warning_errno(r
, "Failed to wait for container: %m");
1922 switch (status
.si_code
) {
1925 if (status
.si_status
== 0)
1926 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
1928 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
1930 *container
= CONTAINER_TERMINATED
;
1931 return status
.si_status
;
1934 if (status
.si_status
== SIGINT
) {
1935 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
1936 *container
= CONTAINER_TERMINATED
;
1939 } else if (status
.si_status
== SIGHUP
) {
1940 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
1941 *container
= CONTAINER_REBOOTED
;
1948 log_error("Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
1952 log_error("Container %s failed due to unknown reason.", arg_machine
);
1957 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
1960 pid
= PTR_TO_PID(userdata
);
1962 if (kill(pid
, arg_kill_signal
) >= 0) {
1963 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
1964 sd_event_source_set_userdata(s
, NULL
);
1969 sd_event_exit(sd_event_source_get_event(s
), 0);
1973 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
1976 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
1977 return log_error_errno(errno
, "Failed to waitid(): %m");
1978 if (si
.si_pid
== 0) /* No pending children. */
1980 if (si
.si_pid
== PTR_TO_PID(userdata
)) {
1981 /* The main process we care for has exited. Return from
1982 * signal handler but leave the zombie. */
1983 sd_event_exit(sd_event_source_get_event(s
), 0);
1986 /* Reap all other children. */
1987 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
1993 static int determine_names(void) {
1996 if (arg_template
&& !arg_directory
&& arg_machine
) {
1998 /* If --template= was specified then we should not
1999 * search for a machine, but instead create a new one
2000 * in /var/lib/machine. */
2002 arg_directory
= strjoin("/var/lib/machines/", arg_machine
);
2007 if (!arg_image
&& !arg_directory
) {
2009 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2011 r
= image_find(arg_machine
, &i
);
2013 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2015 log_error("No image for machine '%s': %m", arg_machine
);
2019 if (i
->type
== IMAGE_RAW
)
2020 r
= free_and_strdup(&arg_image
, i
->path
);
2022 r
= free_and_strdup(&arg_directory
, i
->path
);
2027 arg_read_only
= arg_read_only
|| i
->read_only
;
2029 arg_directory
= get_current_dir_name();
2031 if (!arg_directory
&& !arg_image
) {
2032 log_error("Failed to determine path, please use -D or -i.");
2039 if (arg_directory
&& path_equal(arg_directory
, "/"))
2040 arg_machine
= gethostname_malloc();
2045 arg_machine
= strdup(basename(arg_image
));
2047 /* Truncate suffix if there is one */
2048 e
= endswith(arg_machine
, ".raw");
2052 arg_machine
= strdup(basename(arg_directory
));
2057 hostname_cleanup(arg_machine
);
2058 if (!machine_name_is_valid(arg_machine
)) {
2059 log_error("Failed to determine machine name automatically, please use -M.");
2063 if (arg_ephemeral
) {
2066 /* Add a random suffix when this is an
2067 * ephemeral machine, so that we can run many
2068 * instances at once without manually having
2069 * to specify -M each time. */
2071 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2082 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2091 r
= chase_symlinks(*p
, NULL
, flags
, &chased
);
2093 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2101 static int determine_uid_shift(const char *directory
) {
2104 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2109 if (arg_uid_shift
== UID_INVALID
) {
2112 r
= stat(directory
, &st
);
2114 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2116 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2118 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000))) {
2119 log_error("UID and GID base of %s don't match.", directory
);
2123 arg_uid_range
= UINT32_C(0x10000);
2126 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
) {
2127 log_error("UID base too high for UID range.");
2134 static int inner_child(
2136 const char *directory
,
2142 _cleanup_free_
char *home
= NULL
;
2145 const char *envp
[] = {
2146 "PATH=" DEFAULT_PATH_SPLIT_USR
,
2147 NULL
, /* container */
2152 NULL
, /* container_uuid */
2153 NULL
, /* LISTEN_FDS */
2154 NULL
, /* LISTEN_PID */
2155 NULL
, /* NOTIFY_SOCKET */
2158 const char *exec_target
;
2160 _cleanup_strv_free_
char **env_use
= NULL
;
2165 assert(kmsg_socket
>= 0);
2167 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2168 /* Tell the parent, that it now can write the UID map. */
2169 (void) barrier_place(barrier
); /* #1 */
2171 /* Wait until the parent wrote the UID map */
2172 if (!barrier_place_and_sync(barrier
)) { /* #2 */
2173 log_error("Parent died too early");
2178 r
= reset_uid_gid();
2180 return log_error_errno(r
, "Couldn't become new root: %m");
2183 arg_mount_settings
| MOUNT_IN_USERNS
,
2186 arg_selinux_apifs_context
);
2191 r
= mount_sysfs(NULL
, arg_mount_settings
);
2195 /* Wait until we are cgroup-ified, so that we
2196 * can mount the right cgroup path writable */
2197 if (!barrier_place_and_sync(barrier
)) { /* #3 */
2198 log_error("Parent died too early");
2202 if (arg_use_cgns
&& cg_ns_supported()) {
2203 r
= unshare(CLONE_NEWCGROUP
);
2205 return log_error_errno(errno
, "Failed to unshare cgroup namespace");
2208 arg_unified_cgroup_hierarchy
,
2209 arg_userns_mode
!= USER_NAMESPACE_NO
,
2212 arg_selinux_apifs_context
,
2217 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2222 r
= setup_boot_id(NULL
);
2226 r
= setup_kmsg(NULL
, kmsg_socket
);
2229 kmsg_socket
= safe_close(kmsg_socket
);
2234 return log_error_errno(errno
, "setsid() failed: %m");
2236 if (arg_private_network
)
2239 if (arg_expose_ports
) {
2240 r
= expose_port_send_rtnl(rtnl_socket
);
2243 rtnl_socket
= safe_close(rtnl_socket
);
2246 r
= drop_capabilities();
2248 return log_error_errno(r
, "drop_capabilities() failed: %m");
2252 if (arg_personality
!= PERSONALITY_INVALID
) {
2253 if (personality(arg_personality
) < 0)
2254 return log_error_errno(errno
, "personality() failed: %m");
2255 } else if (secondary
) {
2256 if (personality(PER_LINUX32
) < 0)
2257 return log_error_errno(errno
, "personality() failed: %m");
2261 if (arg_selinux_context
)
2262 if (setexeccon(arg_selinux_context
) < 0)
2263 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
2266 r
= change_uid_gid(arg_user
, &home
);
2270 /* LXC sets container=lxc, so follow the scheme here */
2271 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
2273 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
2277 if ((asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
? home
: "/root") < 0) ||
2278 (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
? arg_user
: "root") < 0) ||
2279 (asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0))
2282 assert(!sd_id128_is_null(arg_uuid
));
2284 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
2287 if (fdset_size(fds
) > 0) {
2288 r
= fdset_cloexec(fds
, false);
2290 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
2292 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
2293 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
2296 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
2299 env_use
= strv_env_merge(2, envp
, arg_setenv
);
2303 /* Let the parent know that we are ready and
2304 * wait until the parent is ready with the
2306 if (!barrier_place_and_sync(barrier
)) { /* #4 */
2307 log_error("Parent died too early");
2312 if (chdir(arg_chdir
) < 0)
2313 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
2315 if (arg_start_mode
== START_PID2
) {
2316 r
= stub_pid1(arg_uuid
);
2321 /* Now, explicitly close the log, so that we
2322 * then can close all remaining fds. Closing
2323 * the log explicitly first has the benefit
2324 * that the logging subsystem knows about it,
2325 * and is thus ready to be reopened should we
2326 * need it again. Note that the other fds
2327 * closed here are at least the locking and
2330 (void) fdset_close_others(fds
);
2332 if (arg_start_mode
== START_BOOT
) {
2336 /* Automatically search for the init system */
2338 m
= strv_length(arg_parameters
);
2339 a
= newa(char*, m
+ 2);
2340 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
2343 a
[0] = (char*) "/usr/lib/systemd/systemd";
2344 execve(a
[0], a
, env_use
);
2346 a
[0] = (char*) "/lib/systemd/systemd";
2347 execve(a
[0], a
, env_use
);
2349 a
[0] = (char*) "/sbin/init";
2350 execve(a
[0], a
, env_use
);
2352 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2353 } else if (!strv_isempty(arg_parameters
)) {
2354 exec_target
= arg_parameters
[0];
2355 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
2358 /* If we cannot change the directory, we'll end up in /, that is expected. */
2359 (void) chdir(home
?: "/root");
2361 execle("/bin/bash", "-bash", NULL
, env_use
);
2362 execle("/bin/sh", "-sh", NULL
, env_use
);
2364 exec_target
= "/bin/bash, /bin/sh";
2369 return log_error_errno(r
, "execv(%s) failed: %m", exec_target
);
2372 static int setup_sd_notify_child(void) {
2373 static const int one
= 1;
2375 union sockaddr_union sa
= {
2376 .sa
.sa_family
= AF_UNIX
,
2380 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2382 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
2384 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
2385 (void) unlink(NSPAWN_NOTIFY_SOCKET_PATH
);
2387 strncpy(sa
.un
.sun_path
, NSPAWN_NOTIFY_SOCKET_PATH
, sizeof(sa
.un
.sun_path
)-1);
2388 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
2391 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
2394 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
2397 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
2400 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
2403 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
2409 static int outer_child(
2411 const char *directory
,
2412 const char *console
,
2413 DissectedImage
*dissected_image
,
2421 int uid_shift_socket
,
2427 _cleanup_close_
int fd
= -1;
2432 assert(pid_socket
>= 0);
2433 assert(uuid_socket
>= 0);
2434 assert(notify_socket
>= 0);
2435 assert(kmsg_socket
>= 0);
2437 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
2438 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
2441 close_nointr(STDIN_FILENO
);
2442 close_nointr(STDOUT_FILENO
);
2443 close_nointr(STDERR_FILENO
);
2445 r
= open_terminal(console
, O_RDWR
);
2446 if (r
!= STDIN_FILENO
) {
2452 return log_error_errno(r
, "Failed to open console: %m");
2455 if (dup2(STDIN_FILENO
, STDOUT_FILENO
) != STDOUT_FILENO
||
2456 dup2(STDIN_FILENO
, STDERR_FILENO
) != STDERR_FILENO
)
2457 return log_error_errno(errno
, "Failed to duplicate console: %m");
2460 r
= reset_audit_loginuid();
2464 /* Mark everything as slave, so that we still
2465 * receive mounts from the real root, but don't
2466 * propagate mounts to the real root. */
2467 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
2471 if (dissected_image
) {
2472 r
= dissected_image_mount(dissected_image
, directory
, DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
2477 r
= determine_uid_shift(directory
);
2481 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2482 /* Let the parent know which UID shift we read from the image */
2483 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
2485 return log_error_errno(errno
, "Failed to send UID shift: %m");
2486 if (l
!= sizeof(arg_uid_shift
)) {
2487 log_error("Short write while sending UID shift.");
2491 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
2492 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2493 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2494 * not it will pick a different one, and send it back to us. */
2496 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
2498 return log_error_errno(errno
, "Failed to recv UID shift: %m");
2499 if (l
!= sizeof(arg_uid_shift
)) {
2500 log_error("Short read while receiving UID shift.");
2505 log_info("Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
2508 /* Turn directory into bind mount */
2509 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
2513 r
= setup_pivot_root(
2516 arg_pivot_root_old
);
2523 arg_userns_mode
!= USER_NAMESPACE_NO
,
2526 arg_selinux_context
);
2530 r
= setup_volatile_state(
2533 arg_userns_mode
!= USER_NAMESPACE_NO
,
2536 arg_selinux_context
);
2540 /* Mark everything as shared so our mounts get propagated down. This is
2541 * required to make new bind mounts available in systemd services
2542 * inside the containter that create a new mount namespace.
2543 * See https://github.com/systemd/systemd/issues/3860
2544 * Further submounts (such as /dev) done after this will inherit the
2545 * shared propagation mode. */
2546 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
2550 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
2554 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
2558 if (arg_read_only
) {
2559 r
= bind_remount_recursive(directory
, true, NULL
);
2561 return log_error_errno(r
, "Failed to make tree read-only: %m");
2564 r
= mount_all(directory
,
2568 arg_selinux_apifs_context
);
2572 r
= copy_devnodes(directory
);
2576 dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
2578 r
= setup_pts(directory
);
2582 r
= setup_propagate(directory
);
2586 r
= setup_dev_console(directory
, console
);
2590 r
= setup_seccomp(arg_caps_retain
);
2594 r
= setup_timezone(directory
);
2598 r
= setup_resolv_conf(directory
);
2602 r
= setup_machine_id(directory
);
2606 r
= setup_journal(directory
);
2613 arg_n_custom_mounts
,
2614 arg_userns_mode
!= USER_NAMESPACE_NO
,
2617 arg_selinux_apifs_context
);
2621 if (!arg_use_cgns
|| !cg_ns_supported()) {
2624 arg_unified_cgroup_hierarchy
,
2625 arg_userns_mode
!= USER_NAMESPACE_NO
,
2628 arg_selinux_apifs_context
,
2634 r
= mount_move_root(directory
);
2636 return log_error_errno(r
, "Failed to move root directory: %m");
2638 fd
= setup_sd_notify_child();
2642 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
2643 arg_clone_ns_flags
|
2644 (arg_private_network
? CLONE_NEWNET
: 0) |
2645 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
2647 return log_error_errno(errno
, "Failed to fork inner child: %m");
2649 pid_socket
= safe_close(pid_socket
);
2650 uuid_socket
= safe_close(uuid_socket
);
2651 notify_socket
= safe_close(notify_socket
);
2652 uid_shift_socket
= safe_close(uid_shift_socket
);
2654 /* The inner child has all namespaces that are
2655 * requested, so that we all are owned by the user if
2656 * user namespaces are turned on. */
2658 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, fds
);
2660 _exit(EXIT_FAILURE
);
2662 _exit(EXIT_SUCCESS
);
2665 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
2667 return log_error_errno(errno
, "Failed to send PID: %m");
2668 if (l
!= sizeof(pid
)) {
2669 log_error("Short write while sending PID.");
2673 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
2675 return log_error_errno(errno
, "Failed to send machine ID: %m");
2676 if (l
!= sizeof(arg_uuid
)) {
2677 log_error("Short write while sending machine ID.");
2681 l
= send_one_fd(notify_socket
, fd
, 0);
2683 return log_error_errno(errno
, "Failed to send notify fd: %m");
2685 pid_socket
= safe_close(pid_socket
);
2686 uuid_socket
= safe_close(uuid_socket
);
2687 notify_socket
= safe_close(notify_socket
);
2688 kmsg_socket
= safe_close(kmsg_socket
);
2689 rtnl_socket
= safe_close(rtnl_socket
);
2694 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
2695 unsigned n_tries
= 100;
2700 assert(ret_lock_file
);
2701 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
2702 assert(arg_uid_range
== 0x10000U
);
2706 (void) mkdir("/run/systemd/nspawn-uid", 0755);
2709 char lock_path
[strlen("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
2710 _cleanup_release_lock_file_ LockFile lf
= LOCK_FILE_INIT
;
2715 if (candidate
< UID_SHIFT_PICK_MIN
|| candidate
> UID_SHIFT_PICK_MAX
)
2717 if ((candidate
& UINT32_C(0xFFFF)) != 0)
2720 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
2721 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
2722 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
2727 /* Make some superficial checks whether the range is currently known in the user database */
2728 if (getpwuid(candidate
))
2730 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
2732 if (getgrgid(candidate
))
2734 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
2737 *ret_lock_file
= lf
;
2738 lf
= (struct LockFile
) LOCK_FILE_INIT
;
2743 random_bytes(&candidate
, sizeof(candidate
));
2744 candidate
= (candidate
% (UID_SHIFT_PICK_MAX
- UID_SHIFT_PICK_MIN
)) + UID_SHIFT_PICK_MIN
;
2745 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
2749 static int setup_uid_map(pid_t pid
) {
2750 char uid_map
[strlen("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
2755 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
2756 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
2757 r
= write_string_file(uid_map
, line
, 0);
2759 return log_error_errno(r
, "Failed to write UID map: %m");
2761 /* We always assign the same UID and GID ranges */
2762 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
2763 r
= write_string_file(uid_map
, line
, 0);
2765 return log_error_errno(r
, "Failed to write GID map: %m");
2770 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2771 char buf
[NOTIFY_BUFFER_MAX
+1];
2773 struct iovec iovec
= {
2775 .iov_len
= sizeof(buf
)-1,
2778 struct cmsghdr cmsghdr
;
2779 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2780 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2782 struct msghdr msghdr
= {
2785 .msg_control
= &control
,
2786 .msg_controllen
= sizeof(control
),
2788 struct cmsghdr
*cmsg
;
2789 struct ucred
*ucred
= NULL
;
2791 pid_t inner_child_pid
;
2792 _cleanup_strv_free_
char **tags
= NULL
;
2796 inner_child_pid
= PTR_TO_PID(userdata
);
2798 if (revents
!= EPOLLIN
) {
2799 log_warning("Got unexpected poll event for notify fd.");
2803 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
2805 if (errno
== EAGAIN
|| errno
== EINTR
)
2808 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
2810 cmsg_close_all(&msghdr
);
2812 CMSG_FOREACH(cmsg
, &msghdr
) {
2813 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2814 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2815 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2817 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2821 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
2822 log_warning("Received notify message without valid credentials. Ignoring.");
2826 if ((size_t) n
>= sizeof(buf
)) {
2827 log_warning("Received notify message exceeded maximum size. Ignoring.");
2832 tags
= strv_split(buf
, "\n\r");
2836 if (strv_find(tags
, "READY=1"))
2837 sd_notifyf(false, "READY=1\n");
2839 p
= strv_find_startswith(tags
, "STATUS=");
2841 sd_notifyf(false, "STATUS=Container running: %s", p
);
2846 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
2849 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
2851 return log_error_errno(r
, "Failed to allocate notify event source: %m");
2853 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
2858 static int load_settings(void) {
2859 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
2860 _cleanup_fclose_
FILE *f
= NULL
;
2861 _cleanup_free_
char *p
= NULL
;
2865 /* If all settings are masked, there's no point in looking for
2866 * the settings file */
2867 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
2870 fn
= strjoina(arg_machine
, ".nspawn");
2872 /* We first look in the admin's directories in /etc and /run */
2873 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
2874 _cleanup_free_
char *j
= NULL
;
2876 j
= strjoin(i
, "/", fn
);
2885 /* By default, we trust configuration from /etc and /run */
2886 if (arg_settings_trusted
< 0)
2887 arg_settings_trusted
= true;
2892 if (errno
!= ENOENT
)
2893 return log_error_errno(errno
, "Failed to open %s: %m", j
);
2897 /* After that, let's look for a file next to the
2898 * actual image we shall boot. */
2901 p
= file_in_same_dir(arg_image
, fn
);
2904 } else if (arg_directory
) {
2905 p
= file_in_same_dir(arg_directory
, fn
);
2912 if (!f
&& errno
!= ENOENT
)
2913 return log_error_errno(errno
, "Failed to open %s: %m", p
);
2915 /* By default, we do not trust configuration from /var/lib/machines */
2916 if (arg_settings_trusted
< 0)
2917 arg_settings_trusted
= false;
2924 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
2926 r
= settings_load(f
, p
, &settings
);
2930 /* Copy over bits from the settings, unless they have been
2931 * explicitly masked by command line switches. */
2933 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
2934 settings
->start_mode
>= 0) {
2935 arg_start_mode
= settings
->start_mode
;
2937 strv_free(arg_parameters
);
2938 arg_parameters
= settings
->parameters
;
2939 settings
->parameters
= NULL
;
2942 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
2943 settings
->pivot_root_new
) {
2944 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
2945 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
2948 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
2949 settings
->working_directory
) {
2951 arg_chdir
= settings
->working_directory
;
2952 settings
->working_directory
= NULL
;
2955 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
2956 settings
->environment
) {
2957 strv_free(arg_setenv
);
2958 arg_setenv
= settings
->environment
;
2959 settings
->environment
= NULL
;
2962 if ((arg_settings_mask
& SETTING_USER
) == 0 &&
2965 arg_user
= settings
->user
;
2966 settings
->user
= NULL
;
2969 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
2972 plus
= settings
->capability
;
2973 if (settings_private_network(settings
))
2974 plus
|= (1ULL << CAP_NET_ADMIN
);
2976 if (!arg_settings_trusted
&& plus
!= 0) {
2977 if (settings
->capability
!= 0)
2978 log_warning("Ignoring Capability= setting, file %s is not trusted.", p
);
2980 arg_caps_retain
|= plus
;
2982 arg_caps_retain
&= ~settings
->drop_capability
;
2985 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
2986 settings
->kill_signal
> 0)
2987 arg_kill_signal
= settings
->kill_signal
;
2989 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
2990 settings
->personality
!= PERSONALITY_INVALID
)
2991 arg_personality
= settings
->personality
;
2993 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
2994 !sd_id128_is_null(settings
->machine_id
)) {
2996 if (!arg_settings_trusted
)
2997 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p
);
2999 arg_uuid
= settings
->machine_id
;
3002 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3003 settings
->read_only
>= 0)
3004 arg_read_only
= settings
->read_only
;
3006 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3007 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3008 arg_volatile_mode
= settings
->volatile_mode
;
3010 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3011 settings
->n_custom_mounts
> 0) {
3013 if (!arg_settings_trusted
)
3014 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p
);
3016 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3017 arg_custom_mounts
= settings
->custom_mounts
;
3018 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3020 settings
->custom_mounts
= NULL
;
3021 settings
->n_custom_mounts
= 0;
3025 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3026 (settings
->private_network
>= 0 ||
3027 settings
->network_veth
>= 0 ||
3028 settings
->network_bridge
||
3029 settings
->network_zone
||
3030 settings
->network_interfaces
||
3031 settings
->network_macvlan
||
3032 settings
->network_ipvlan
||
3033 settings
->network_veth_extra
)) {
3035 if (!arg_settings_trusted
)
3036 log_warning("Ignoring network settings, file %s is not trusted.", p
);
3038 arg_network_veth
= settings_network_veth(settings
);
3039 arg_private_network
= settings_private_network(settings
);
3041 strv_free(arg_network_interfaces
);
3042 arg_network_interfaces
= settings
->network_interfaces
;
3043 settings
->network_interfaces
= NULL
;
3045 strv_free(arg_network_macvlan
);
3046 arg_network_macvlan
= settings
->network_macvlan
;
3047 settings
->network_macvlan
= NULL
;
3049 strv_free(arg_network_ipvlan
);
3050 arg_network_ipvlan
= settings
->network_ipvlan
;
3051 settings
->network_ipvlan
= NULL
;
3053 strv_free(arg_network_veth_extra
);
3054 arg_network_veth_extra
= settings
->network_veth_extra
;
3055 settings
->network_veth_extra
= NULL
;
3057 free(arg_network_bridge
);
3058 arg_network_bridge
= settings
->network_bridge
;
3059 settings
->network_bridge
= NULL
;
3061 free(arg_network_zone
);
3062 arg_network_zone
= settings
->network_zone
;
3063 settings
->network_zone
= NULL
;
3067 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3068 settings
->expose_ports
) {
3070 if (!arg_settings_trusted
)
3071 log_warning("Ignoring Port= setting, file %s is not trusted.", p
);
3073 expose_port_free_all(arg_expose_ports
);
3074 arg_expose_ports
= settings
->expose_ports
;
3075 settings
->expose_ports
= NULL
;
3079 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3080 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3082 if (!arg_settings_trusted
)
3083 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", p
);
3085 arg_userns_mode
= settings
->userns_mode
;
3086 arg_uid_shift
= settings
->uid_shift
;
3087 arg_uid_range
= settings
->uid_range
;
3088 arg_userns_chown
= settings
->userns_chown
;
3092 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3093 arg_notify_ready
= settings
->notify_ready
;
3098 static int run(int master
,
3099 const char* console
,
3100 DissectedImage
*dissected_image
,
3104 char veth_name
[IFNAMSIZ
], bool *veth_created
,
3105 union in_addr_union
*exposed
,
3106 pid_t
*pid
, int *ret
) {
3108 static const struct sigaction sa
= {
3109 .sa_handler
= nop_signal_handler
,
3110 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
3113 _cleanup_release_lock_file_ LockFile uid_shift_lock
= LOCK_FILE_INIT
;
3114 _cleanup_close_
int etc_passwd_lock
= -1;
3115 _cleanup_close_pair_
int
3116 kmsg_socket_pair
[2] = { -1, -1 },
3117 rtnl_socket_pair
[2] = { -1, -1 },
3118 pid_socket_pair
[2] = { -1, -1 },
3119 uuid_socket_pair
[2] = { -1, -1 },
3120 notify_socket_pair
[2] = { -1, -1 },
3121 uid_shift_socket_pair
[2] = { -1, -1 };
3122 _cleanup_close_
int notify_socket
= -1;
3123 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
3124 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
3125 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
3126 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
3127 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
3128 ContainerStatus container_status
= 0;
3134 assert_se(sigemptyset(&mask_chld
) == 0);
3135 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
3137 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3138 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3139 * check with getpwuid() if the specific user already exists. Note that /etc might be
3140 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3141 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3142 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3145 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
3146 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
3147 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
3150 r
= barrier_create(&barrier
);
3152 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
3154 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
3155 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
3157 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
3158 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
3160 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
3161 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
3163 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
3164 return log_error_errno(errno
, "Failed to create id socket pair: %m");
3166 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
3167 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
3169 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3170 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
3171 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
3173 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3174 * parent's blocking calls and give it a chance to call wait() and terminate. */
3175 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
3177 return log_error_errno(errno
, "Failed to change the signal mask: %m");
3179 r
= sigaction(SIGCHLD
, &sa
, NULL
);
3181 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
3183 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
3185 return log_error_errno(errno
, "clone() failed%s: %m",
3187 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3190 /* The outer child only has a file system namespace. */
3191 barrier_set_role(&barrier
, BARRIER_CHILD
);
3193 master
= safe_close(master
);
3195 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
3196 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
3197 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
3198 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
3199 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
3200 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
3202 (void) reset_all_signal_handlers();
3203 (void) reset_signal_mask();
3205 r
= outer_child(&barrier
,
3212 uuid_socket_pair
[1],
3213 notify_socket_pair
[1],
3214 kmsg_socket_pair
[1],
3215 rtnl_socket_pair
[1],
3216 uid_shift_socket_pair
[1],
3219 _exit(EXIT_FAILURE
);
3221 _exit(EXIT_SUCCESS
);
3224 barrier_set_role(&barrier
, BARRIER_PARENT
);
3226 fds
= fdset_free(fds
);
3228 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
3229 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
3230 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
3231 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
3232 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
3233 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
3235 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3236 /* The child just let us know the UID shift it might have read from the image. */
3237 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
3239 return log_error_errno(errno
, "Failed to read UID shift: %m");
3240 if (l
!= sizeof arg_uid_shift
) {
3241 log_error("Short read while reading UID shift.");
3245 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3246 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3247 * image, but if that's already in use, pick a new one, and report back to the child,
3248 * which one we now picked. */
3250 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
3252 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
3254 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
3256 return log_error_errno(errno
, "Failed to send UID shift: %m");
3257 if (l
!= sizeof arg_uid_shift
) {
3258 log_error("Short write while writing UID shift.");
3264 /* Wait for the outer child. */
3265 r
= wait_for_terminate_and_warn("namespace helper", *pid
, NULL
);
3267 return r
< 0 ? r
: -EIO
;
3269 /* And now retrieve the PID of the inner child. */
3270 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
3272 return log_error_errno(errno
, "Failed to read inner child PID: %m");
3273 if (l
!= sizeof *pid
) {
3274 log_error("Short read while reading inner child PID.");
3278 /* We also retrieve container UUID in case it was generated by outer child */
3279 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
3281 return log_error_errno(errno
, "Failed to read container machine ID: %m");
3282 if (l
!= sizeof(arg_uuid
)) {
3283 log_error("Short read while reading container machined ID.");
3287 /* We also retrieve the socket used for notifications generated by outer child */
3288 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
3289 if (notify_socket
< 0)
3290 return log_error_errno(notify_socket
,
3291 "Failed to receive notification socket from the outer child: %m");
3293 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
3295 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3296 if (!barrier_place_and_sync(&barrier
)) { /* #1 */
3297 log_error("Child died too early.");
3301 r
= setup_uid_map(*pid
);
3305 (void) barrier_place(&barrier
); /* #2 */
3308 if (arg_private_network
) {
3310 r
= move_network_interfaces(*pid
, arg_network_interfaces
);
3314 if (arg_network_veth
) {
3315 r
= setup_veth(arg_machine
, *pid
, veth_name
,
3316 arg_network_bridge
|| arg_network_zone
);
3322 if (arg_network_bridge
) {
3323 /* Add the interface to a bridge */
3324 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
3329 } else if (arg_network_zone
) {
3330 /* Add the interface to a bridge, possibly creating it */
3331 r
= setup_bridge(veth_name
, arg_network_zone
, true);
3339 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
3343 /* We created the primary and extra veth links now; let's remember this, so that we know to
3344 remove them later on. Note that we don't bother with removing veth links that were created
3345 here when their setup failed half-way, because in that case the kernel should be able to
3346 remove them on its own, since they cannot be referenced by anything yet. */
3347 *veth_created
= true;
3349 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
3353 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
3359 r
= register_machine(
3366 arg_custom_mounts
, arg_n_custom_mounts
,
3370 arg_container_service_name
);
3375 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
3379 if (arg_keep_unit
) {
3380 r
= create_subcgroup(*pid
, arg_unified_cgroup_hierarchy
);
3385 r
= chown_cgroup(*pid
, arg_uid_shift
);
3389 /* Notify the child that the parent is ready with all
3390 * its setup (including cgroup-ification), and that
3391 * the child can now hand over control to the code to
3392 * run inside the container. */
3393 (void) barrier_place(&barrier
); /* #3 */
3395 /* Block SIGCHLD here, before notifying child.
3396 * process_pty() will handle it with the other signals. */
3397 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
3399 /* Reset signal to default */
3400 r
= default_signals(SIGCHLD
, -1);
3402 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
3404 r
= sd_event_new(&event
);
3406 return log_error_errno(r
, "Failed to get default event source: %m");
3408 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
3412 /* Let the child know that we are ready and wait that the child is completely ready now. */
3413 if (!barrier_place_and_sync(&barrier
)) { /* #4 */
3414 log_error("Child died too early.");
3418 /* At this point we have made use of the UID we picked, and thus nss-mymachines
3419 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
3420 etc_passwd_lock
= safe_close(etc_passwd_lock
);
3423 "STATUS=Container running.\n"
3424 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
3425 if (!arg_notify_ready
)
3426 sd_notify(false, "READY=1\n");
3428 if (arg_kill_signal
> 0) {
3429 /* Try to kill the init system on SIGINT or SIGTERM */
3430 sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
3431 sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
3433 /* Immediately exit */
3434 sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
3435 sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
3438 /* Exit when the child exits */
3439 sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
3441 if (arg_expose_ports
) {
3442 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
3446 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
3449 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
3451 r
= pty_forward_new(event
, master
,
3452 PTY_FORWARD_IGNORE_VHANGUP
| (interactive
? 0 : PTY_FORWARD_READ_ONLY
),
3455 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
3457 r
= sd_event_loop(event
);
3459 return log_error_errno(r
, "Failed to run event loop: %m");
3461 pty_forward_get_last_char(forward
, &last_char
);
3463 forward
= pty_forward_free(forward
);
3465 if (!arg_quiet
&& last_char
!= '\n')
3468 /* Kill if it is not dead yet anyway */
3469 if (arg_register
&& !arg_keep_unit
)
3470 terminate_machine(*pid
);
3472 /* Normally redundant, but better safe than sorry */
3473 (void) kill(*pid
, SIGKILL
);
3475 r
= wait_for_container(*pid
, &container_status
);
3479 /* We failed to wait for the container, or the container exited abnormally. */
3481 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
3482 /* r > 0 → The container exited with a non-zero status.
3483 * As a special case, we need to replace 133 with a different value,
3484 * because 133 is special-cased in the service file to reboot the container.
3485 * otherwise → The container exited with zero status and a reboot was not requested.
3487 if (r
== EXIT_FORCE_RESTART
)
3488 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
3490 return 0; /* finito */
3493 /* CONTAINER_REBOOTED, loop again */
3495 if (arg_keep_unit
) {
3496 /* Special handling if we are running as a service: instead of simply
3497 * restarting the machine we want to restart the entire service, so let's
3498 * inform systemd about this with the special exit code 133. The service
3499 * file uses RestartForceExitStatus=133 so that this results in a full
3500 * nspawn restart. This is necessary since we might have cgroup parameters
3501 * set we want to have flushed out. */
3502 *ret
= EXIT_FORCE_RESTART
;
3503 return 0; /* finito */
3506 expose_port_flush(arg_expose_ports
, exposed
);
3508 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
3509 *veth_created
= false;
3510 return 1; /* loop again */
3513 int main(int argc
, char *argv
[]) {
3515 _cleanup_free_
char *console
= NULL
;
3516 _cleanup_close_
int master
= -1;
3517 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3518 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
3519 char veth_name
[IFNAMSIZ
] = "";
3520 bool secondary
= false, remove_directory
= false, remove_image
= false;
3522 union in_addr_union exposed
= {};
3523 _cleanup_release_lock_file_ LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
3524 bool interactive
, veth_created
= false, remove_tmprootdir
= false;
3525 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
3526 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
3527 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
3528 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3530 log_parse_environment();
3533 r
= cg_unified_flush();
3535 return log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
3537 /* Make sure rename_process() in the stub init process can work */
3541 r
= parse_argv(argc
, argv
);
3545 if (geteuid() != 0) {
3546 log_error("Need to be root.");
3550 r
= determine_names();
3554 r
= load_settings();
3558 r
= verify_arguments();
3562 n_fd_passed
= sd_listen_fds(false);
3563 if (n_fd_passed
> 0) {
3564 r
= fdset_new_listen_fds(&fds
, false);
3566 log_error_errno(r
, "Failed to collect file descriptors: %m");
3571 if (arg_directory
) {
3574 if (path_equal(arg_directory
, "/") && !arg_ephemeral
) {
3575 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3580 if (arg_ephemeral
) {
3581 _cleanup_free_
char *np
= NULL
;
3583 r
= chase_symlinks_and_update(&arg_directory
, 0);
3587 /* If the specified path is a mount point we
3588 * generate the new snapshot immediately
3589 * inside it under a random name. However if
3590 * the specified is not a mount point we
3591 * create the new snapshot in the parent
3592 * directory, just next to it. */
3593 r
= path_is_mount_point(arg_directory
, NULL
, 0);
3595 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
3599 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
3601 r
= tempfn_random(arg_directory
, "machine.", &np
);
3603 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
3607 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3609 log_error_errno(r
, "Failed to lock %s: %m", np
);
3613 r
= btrfs_subvol_snapshot(arg_directory
, np
,
3614 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
3615 BTRFS_SNAPSHOT_FALLBACK_COPY
|
3616 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
3617 BTRFS_SNAPSHOT_RECURSIVE
|
3618 BTRFS_SNAPSHOT_QUOTA
);
3620 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
3624 free(arg_directory
);
3628 remove_directory
= true;
3631 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
3635 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3637 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
3641 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
3646 r
= chase_symlinks_and_update(&arg_template
, 0);
3650 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
3651 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
3652 BTRFS_SNAPSHOT_FALLBACK_COPY
|
3653 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
3654 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
3655 BTRFS_SNAPSHOT_RECURSIVE
|
3656 BTRFS_SNAPSHOT_QUOTA
);
3659 log_info("Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
3661 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
3665 log_info("Populated %s from template %s.", arg_directory
, arg_template
);
3670 if (arg_start_mode
== START_BOOT
) {
3671 if (path_is_os_tree(arg_directory
) <= 0) {
3672 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory
);
3679 p
= strjoina(arg_directory
, "/usr/");
3680 if (laccess(p
, F_OK
) < 0) {
3681 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", arg_directory
);
3689 assert(!arg_template
);
3691 r
= chase_symlinks_and_update(&arg_image
, 0);
3695 if (arg_ephemeral
) {
3696 _cleanup_free_
char *np
= NULL
;
3698 r
= tempfn_random(arg_image
, "machine.", &np
);
3700 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
3704 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3706 r
= log_error_errno(r
, "Failed to create image lock: %m");
3710 r
= copy_file(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600, FS_NOCOW_FL
, COPY_REFLINK
);
3712 r
= log_error_errno(r
, "Failed to copy image file: %m");
3720 remove_image
= true;
3722 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3724 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
3728 r
= log_error_errno(r
, "Failed to create image lock: %m");
3732 if (!arg_root_hash
) {
3733 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
3735 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
3741 if (!mkdtemp(tmprootdir
)) {
3742 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
3746 remove_tmprootdir
= true;
3748 arg_directory
= strdup(tmprootdir
);
3749 if (!arg_directory
) {
3754 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, &loop
);
3756 log_error_errno(r
, "Failed to set up loopback block device: %m");
3762 arg_root_hash
, arg_root_hash_size
,
3763 DISSECT_IMAGE_REQUIRE_ROOT
,
3766 log_error_errno(r
, "Could not find a suitable file system or partition table in image: %s", arg_image
);
3768 log_notice("Note that the disk image needs to\n"
3769 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
3770 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
3771 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
3772 " d) or contain a file system without a partition table\n"
3773 "in order to be bootable with systemd-nspawn.");
3776 if (r
== -EADDRNOTAVAIL
) {
3777 log_error_errno(r
, "No root partition for specified root hash found.");
3780 if (r
== -EOPNOTSUPP
) {
3781 log_error_errno(r
, "--image= is not supported, compiled without blkid support.");
3785 log_error_errno(r
, "Failed to dissect image: %m");
3789 if (!arg_root_hash
&& dissected_image
->can_verity
)
3790 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
3792 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
3796 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
3797 if (remove_image
&& unlink(arg_image
) >= 0)
3798 remove_image
= false;
3801 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
3805 r
= detect_unified_cgroup_hierarchy(arg_directory
);
3810 isatty(STDIN_FILENO
) > 0 &&
3811 isatty(STDOUT_FILENO
) > 0;
3813 master
= posix_openpt(O_RDWR
|O_NOCTTY
|O_CLOEXEC
|O_NDELAY
);
3815 r
= log_error_errno(errno
, "Failed to acquire pseudo tty: %m");
3819 r
= ptsname_malloc(master
, &console
);
3821 r
= log_error_errno(r
, "Failed to determine tty name: %m");
3825 if (arg_selinux_apifs_context
) {
3826 r
= mac_selinux_apply(console
, arg_selinux_apifs_context
);
3831 if (unlockpt(master
) < 0) {
3832 r
= log_error_errno(errno
, "Failed to unlock tty: %m");
3837 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
3838 arg_machine
, arg_image
?: arg_directory
);
3840 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
3842 if (prctl(PR_SET_CHILD_SUBREAPER
, 1) < 0) {
3843 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
3851 interactive
, secondary
,
3853 veth_name
, &veth_created
,
3862 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
3863 "STOPPING=1\nSTATUS=Terminating...");
3866 (void) kill(pid
, SIGKILL
);
3868 /* Try to flush whatever is still queued in the pty */
3870 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
3871 master
= safe_close(master
);
3875 (void) wait_for_terminate(pid
, NULL
);
3877 if (remove_directory
&& arg_directory
) {
3880 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
3882 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
3885 if (remove_image
&& arg_image
) {
3886 if (unlink(arg_image
) < 0)
3887 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
3890 if (remove_tmprootdir
) {
3891 if (rmdir(tmprootdir
) < 0)
3892 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
3898 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
3899 (void) rm_rf(p
, REMOVE_ROOT
);
3902 expose_port_flush(arg_expose_ports
, &exposed
);
3905 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
3906 (void) remove_bridge(arg_network_zone
);
3908 free(arg_directory
);
3913 free(arg_pivot_root_new
);
3914 free(arg_pivot_root_old
);
3916 strv_free(arg_setenv
);
3917 free(arg_network_bridge
);
3918 strv_free(arg_network_interfaces
);
3919 strv_free(arg_network_macvlan
);
3920 strv_free(arg_network_ipvlan
);
3921 strv_free(arg_network_veth_extra
);
3922 strv_free(arg_parameters
);
3923 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3924 expose_port_free_all(arg_expose_ports
);
3925 free(arg_root_hash
);
3927 return r
< 0 ? EXIT_FAILURE
: ret
;