1 /* SPDX-License-Identifier: LGPL-2.1+ */
9 #include <linux/loop.h>
13 #include <selinux/selinux.h>
20 #include <sys/mount.h>
21 #include <sys/personality.h>
22 #include <sys/prctl.h>
23 #include <sys/types.h>
28 #include "sd-daemon.h"
31 #include "alloc-util.h"
33 #include "base-filesystem.h"
34 #include "blkid-util.h"
35 #include "btrfs-util.h"
36 #include "bus-error.h"
39 #include "capability-util.h"
40 #include "cgroup-util.h"
42 #include "cpu-set-util.h"
43 #include "dev-setup.h"
44 #include "dissect-image.h"
49 #include "format-util.h"
52 #include "hexdecoct.h"
53 #include "hostname-util.h"
54 #include "id128-util.h"
56 #include "loop-util.h"
57 #include "loopback-setup.h"
58 #include "machine-image.h"
62 #include "mount-util.h"
63 #include "mountpoint-util.h"
64 #include "netlink-util.h"
65 #include "nspawn-cgroup.h"
66 #include "nspawn-def.h"
67 #include "nspawn-expose-ports.h"
68 #include "nspawn-mount.h"
69 #include "nspawn-network.h"
70 #include "nspawn-patch-uid.h"
71 #include "nspawn-register.h"
72 #include "nspawn-seccomp.h"
73 #include "nspawn-settings.h"
74 #include "nspawn-setuid.h"
75 #include "nspawn-stub-pid1.h"
78 #include "parse-util.h"
79 #include "path-util.h"
80 #include "pretty-print.h"
81 #include "process-util.h"
83 #include "random-util.h"
84 #include "raw-clone.h"
85 #include "rlimit-util.h"
87 #include "selinux-util.h"
88 #include "signal-util.h"
89 #include "socket-util.h"
90 #include "stat-util.h"
91 #include "stdio-util.h"
92 #include "string-table.h"
93 #include "string-util.h"
95 #include "terminal-util.h"
96 #include "tmpfile-util.h"
97 #include "umask-util.h"
98 #include "user-util.h"
102 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
104 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
107 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
108 * nspawn_notify_socket_path is relative to the container
109 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
110 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
112 #define EXIT_FORCE_RESTART 133
114 typedef enum ContainerStatus
{
115 CONTAINER_TERMINATED
,
119 static char *arg_directory
= NULL
;
120 static char *arg_template
= NULL
;
121 static char *arg_chdir
= NULL
;
122 static char *arg_pivot_root_new
= NULL
;
123 static char *arg_pivot_root_old
= NULL
;
124 static char *arg_user
= NULL
;
125 static sd_id128_t arg_uuid
= {};
126 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
127 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
128 static const char *arg_selinux_context
= NULL
;
129 static const char *arg_selinux_apifs_context
= NULL
;
130 static const char *arg_slice
= NULL
;
131 static bool arg_private_network
= false;
132 static bool arg_read_only
= false;
133 static StartMode arg_start_mode
= START_PID1
;
134 static bool arg_ephemeral
= false;
135 static LinkJournal arg_link_journal
= LINK_AUTO
;
136 static bool arg_link_journal_try
= false;
137 static uint64_t arg_caps_retain
=
138 (1ULL << CAP_AUDIT_CONTROL
) |
139 (1ULL << CAP_AUDIT_WRITE
) |
140 (1ULL << CAP_CHOWN
) |
141 (1ULL << CAP_DAC_OVERRIDE
) |
142 (1ULL << CAP_DAC_READ_SEARCH
) |
143 (1ULL << CAP_FOWNER
) |
144 (1ULL << CAP_FSETID
) |
145 (1ULL << CAP_IPC_OWNER
) |
147 (1ULL << CAP_LEASE
) |
148 (1ULL << CAP_LINUX_IMMUTABLE
) |
149 (1ULL << CAP_MKNOD
) |
150 (1ULL << CAP_NET_BIND_SERVICE
) |
151 (1ULL << CAP_NET_BROADCAST
) |
152 (1ULL << CAP_NET_RAW
) |
153 (1ULL << CAP_SETFCAP
) |
154 (1ULL << CAP_SETGID
) |
155 (1ULL << CAP_SETPCAP
) |
156 (1ULL << CAP_SETUID
) |
157 (1ULL << CAP_SYS_ADMIN
) |
158 (1ULL << CAP_SYS_BOOT
) |
159 (1ULL << CAP_SYS_CHROOT
) |
160 (1ULL << CAP_SYS_NICE
) |
161 (1ULL << CAP_SYS_PTRACE
) |
162 (1ULL << CAP_SYS_RESOURCE
) |
163 (1ULL << CAP_SYS_TTY_CONFIG
);
164 static CustomMount
*arg_custom_mounts
= NULL
;
165 static size_t arg_n_custom_mounts
= 0;
166 static char **arg_setenv
= NULL
;
167 static bool arg_quiet
= false;
168 static bool arg_register
= true;
169 static bool arg_keep_unit
= false;
170 static char **arg_network_interfaces
= NULL
;
171 static char **arg_network_macvlan
= NULL
;
172 static char **arg_network_ipvlan
= NULL
;
173 static bool arg_network_veth
= false;
174 static char **arg_network_veth_extra
= NULL
;
175 static char *arg_network_bridge
= NULL
;
176 static char *arg_network_zone
= NULL
;
177 static char *arg_network_namespace_path
= NULL
;
178 static unsigned long arg_personality
= PERSONALITY_INVALID
;
179 static char *arg_image
= NULL
;
180 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
181 static ExposePort
*arg_expose_ports
= NULL
;
182 static char **arg_property
= NULL
;
183 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
184 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
185 static bool arg_userns_chown
= false;
186 static int arg_kill_signal
= 0;
187 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
188 static SettingsMask arg_settings_mask
= 0;
189 static int arg_settings_trusted
= -1;
190 static char **arg_parameters
= NULL
;
191 static const char *arg_container_service_name
= "systemd-nspawn";
192 static bool arg_notify_ready
= false;
193 static bool arg_use_cgns
= true;
194 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
195 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
196 static void *arg_root_hash
= NULL
;
197 static size_t arg_root_hash_size
= 0;
198 static char **arg_syscall_whitelist
= NULL
;
199 static char **arg_syscall_blacklist
= NULL
;
200 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
201 static bool arg_no_new_privileges
= false;
202 static int arg_oom_score_adjust
= 0;
203 static bool arg_oom_score_adjust_set
= false;
204 static cpu_set_t
*arg_cpuset
= NULL
;
205 static unsigned arg_cpuset_ncpus
= 0;
206 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
207 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
209 static int help(void) {
210 _cleanup_free_
char *link
= NULL
;
213 (void) pager_open(false);
215 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
219 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
220 "Spawn a command or OS in a light-weight container.\n\n"
221 " -h --help Show this help\n"
222 " --version Print version string\n"
223 " -q --quiet Do not show status information\n"
224 " -D --directory=PATH Root directory for the container\n"
225 " --template=PATH Initialize root directory from template directory,\n"
227 " -x --ephemeral Run container with snapshot of root directory, and\n"
228 " remove it after exit\n"
229 " -i --image=PATH File system device or disk image for the container\n"
230 " --root-hash=HASH Specify verity root hash\n"
231 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
232 " -b --boot Boot up full system (i.e. invoke init)\n"
233 " --chdir=PATH Set working directory in the container\n"
234 " --pivot-root=PATH[:PATH]\n"
235 " Pivot root to given directory in the container\n"
236 " -u --user=USER Run the command under specified user or uid\n"
237 " -M --machine=NAME Set the machine name for the container\n"
238 " --hostname=NAME Override the hostname for the container\n"
239 " --uuid=UUID Set a specific machine UUID for the container\n"
240 " -S --slice=SLICE Place the container in the specified slice\n"
241 " --property=NAME=VALUE Set scope unit property\n"
242 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
243 " --private-users[=UIDBASE[:NUIDS]]\n"
244 " Similar, but with user configured UID/GID range\n"
245 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
246 " --private-network Disable network in container\n"
247 " --network-interface=INTERFACE\n"
248 " Assign an existing network interface to the\n"
250 " --network-macvlan=INTERFACE\n"
251 " Create a macvlan network interface based on an\n"
252 " existing network interface to the container\n"
253 " --network-ipvlan=INTERFACE\n"
254 " Create a ipvlan network interface based on an\n"
255 " existing network interface to the container\n"
256 " -n --network-veth Add a virtual Ethernet connection between host\n"
258 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
259 " Add an additional virtual Ethernet link between\n"
260 " host and container\n"
261 " --network-bridge=INTERFACE\n"
262 " Add a virtual Ethernet connection to the container\n"
263 " and attach it to an existing bridge on the host\n"
264 " --network-zone=NAME Similar, but attach the new interface to an\n"
265 " an automatically managed bridge interface\n"
266 " --network-namespace-path=PATH\n"
267 " Set network namespace to the one represented by\n"
268 " the specified kernel namespace file node\n"
269 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
270 " Expose a container IP port on the host\n"
271 " -Z --selinux-context=SECLABEL\n"
272 " Set the SELinux security context to be used by\n"
273 " processes in the container\n"
274 " -L --selinux-apifs-context=SECLABEL\n"
275 " Set the SELinux security context to be used by\n"
276 " API/tmpfs file systems in the container\n"
277 " --capability=CAP In addition to the default, retain specified\n"
279 " --drop-capability=CAP Drop the specified capability from the default set\n"
280 " --system-call-filter=LIST|~LIST\n"
281 " Permit/prohibit specific system calls\n"
282 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
283 " --oom-score-adjust=VALUE\n"
284 " Adjust the OOM score value for the payload\n"
285 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
286 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
287 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
288 " host, try-guest, try-host\n"
289 " -j Equivalent to --link-journal=try-guest\n"
290 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
291 " --timezone=MODE Select mode of /etc/localtime initialization\n"
292 " --read-only Mount the root directory read-only\n"
293 " --bind=PATH[:PATH[:OPTIONS]]\n"
294 " Bind mount a file or directory from the host into\n"
296 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
297 " Similar, but creates a read-only bind mount\n"
298 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
299 " --overlay=PATH[:PATH...]:PATH\n"
300 " Create an overlay mount from the host to \n"
302 " --overlay-ro=PATH[:PATH...]:PATH\n"
303 " Similar, but creates a read-only overlay mount\n"
304 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
305 " --register=BOOLEAN Register container as machine\n"
306 " --keep-unit Do not register a scope for the machine, reuse\n"
307 " the service unit nspawn is running in\n"
308 " --volatile[=MODE] Run the system in volatile mode\n"
309 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
310 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
311 "\nSee the %s for details.\n"
312 , program_invocation_short_name
319 static int custom_mount_check_all(void) {
322 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
323 CustomMount
*m
= &arg_custom_mounts
[i
];
325 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
326 if (arg_userns_chown
)
327 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
328 "--private-users-chown may not be combined with custom root mounts.");
329 else if (arg_uid_shift
== UID_INVALID
)
330 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
331 "--private-users with automatic UID shift may not be combined with custom root mounts.");
338 static int detect_unified_cgroup_hierarchy_from_environment(void) {
342 /* Allow the user to control whether the unified hierarchy is used */
343 e
= getenv("UNIFIED_CGROUP_HIERARCHY");
345 r
= parse_boolean(e
);
347 return log_error_errno(r
, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
349 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
351 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
357 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
360 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
361 * image actually supports. */
362 r
= cg_all_unified();
364 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
366 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
367 * routine only detects 231, so we'll have a false negative here for 230. */
368 r
= systemd_installation_has_version(directory
, 230);
370 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
372 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
374 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
375 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
376 /* Mixed cgroup hierarchy support was added in 233 */
377 r
= systemd_installation_has_version(directory
, 233);
379 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
381 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
383 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
385 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
387 log_debug("Using %s hierarchy for container.",
388 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
389 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
394 static void parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
397 r
= getenv_bool(name
);
401 log_warning_errno(r
, "Failed to parse %s from environment, defaulting to false.", name
);
402 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
405 static void parse_mount_settings_env(void) {
409 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
411 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
412 else if (r
!= -ENXIO
)
413 log_warning_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP, ignoring: %m");
415 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
419 if (streq(e
, "network")) {
420 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
424 r
= parse_boolean(e
);
426 log_warning_errno(r
, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
430 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
431 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
434 static int parse_argv(int argc
, char *argv
[]) {
451 ARG_NETWORK_INTERFACE
,
456 ARG_NETWORK_VETH_EXTRA
,
457 ARG_NETWORK_NAMESPACE_PATH
,
467 ARG_PRIVATE_USERS_CHOWN
,
470 ARG_SYSTEM_CALL_FILTER
,
473 ARG_NO_NEW_PRIVILEGES
,
474 ARG_OOM_SCORE_ADJUST
,
480 static const struct option options
[] = {
481 { "help", no_argument
, NULL
, 'h' },
482 { "version", no_argument
, NULL
, ARG_VERSION
},
483 { "directory", required_argument
, NULL
, 'D' },
484 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
485 { "ephemeral", no_argument
, NULL
, 'x' },
486 { "user", required_argument
, NULL
, 'u' },
487 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
488 { "as-pid2", no_argument
, NULL
, 'a' },
489 { "boot", no_argument
, NULL
, 'b' },
490 { "uuid", required_argument
, NULL
, ARG_UUID
},
491 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
492 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
493 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
494 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
495 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
496 { "bind", required_argument
, NULL
, ARG_BIND
},
497 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
498 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
499 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
500 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
501 { "machine", required_argument
, NULL
, 'M' },
502 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
503 { "slice", required_argument
, NULL
, 'S' },
504 { "setenv", required_argument
, NULL
, 'E' },
505 { "selinux-context", required_argument
, NULL
, 'Z' },
506 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
507 { "quiet", no_argument
, NULL
, 'q' },
508 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
509 { "register", required_argument
, NULL
, ARG_REGISTER
},
510 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
511 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
512 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
513 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
514 { "network-veth", no_argument
, NULL
, 'n' },
515 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
516 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
517 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
518 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
519 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
520 { "image", required_argument
, NULL
, 'i' },
521 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
522 { "port", required_argument
, NULL
, 'p' },
523 { "property", required_argument
, NULL
, ARG_PROPERTY
},
524 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
525 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
526 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
527 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
528 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
529 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
530 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
531 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
532 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
533 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
534 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
535 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
536 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
537 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
543 uint64_t plus
= 0, minus
= 0;
544 bool mask_all_settings
= false, mask_no_settings
= false;
549 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options
, NULL
)) >= 0)
559 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
565 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
571 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
577 arg_ephemeral
= true;
578 arg_settings_mask
|= SETTING_EPHEMERAL
;
582 r
= free_and_strdup(&arg_user
, optarg
);
586 arg_settings_mask
|= SETTING_USER
;
589 case ARG_NETWORK_ZONE
: {
592 j
= strappend("vz-", optarg
);
596 if (!ifname_valid(j
)) {
597 log_error("Network zone name not valid: %s", j
);
602 free_and_replace(arg_network_zone
, j
);
604 arg_network_veth
= true;
605 arg_private_network
= true;
606 arg_settings_mask
|= SETTING_NETWORK
;
610 case ARG_NETWORK_BRIDGE
:
612 if (!ifname_valid(optarg
))
613 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
614 "Bridge interface name not valid: %s", optarg
);
616 r
= free_and_strdup(&arg_network_bridge
, optarg
);
622 arg_network_veth
= true;
623 arg_private_network
= true;
624 arg_settings_mask
|= SETTING_NETWORK
;
627 case ARG_NETWORK_VETH_EXTRA
:
628 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
630 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
632 arg_private_network
= true;
633 arg_settings_mask
|= SETTING_NETWORK
;
636 case ARG_NETWORK_INTERFACE
:
637 if (!ifname_valid(optarg
))
638 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
639 "Network interface name not valid: %s", optarg
);
641 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
644 arg_private_network
= true;
645 arg_settings_mask
|= SETTING_NETWORK
;
648 case ARG_NETWORK_MACVLAN
:
650 if (!ifname_valid(optarg
))
651 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
652 "MACVLAN network interface name not valid: %s", optarg
);
654 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
657 arg_private_network
= true;
658 arg_settings_mask
|= SETTING_NETWORK
;
661 case ARG_NETWORK_IPVLAN
:
663 if (!ifname_valid(optarg
))
664 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
665 "IPVLAN network interface name not valid: %s", optarg
);
667 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
671 case ARG_PRIVATE_NETWORK
:
672 arg_private_network
= true;
673 arg_settings_mask
|= SETTING_NETWORK
;
676 case ARG_NETWORK_NAMESPACE_PATH
:
677 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
684 if (arg_start_mode
== START_PID2
)
685 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
686 "--boot and --as-pid2 may not be combined.");
688 arg_start_mode
= START_BOOT
;
689 arg_settings_mask
|= SETTING_START_MODE
;
693 if (arg_start_mode
== START_BOOT
)
694 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
695 "--boot and --as-pid2 may not be combined.");
697 arg_start_mode
= START_PID2
;
698 arg_settings_mask
|= SETTING_START_MODE
;
702 r
= sd_id128_from_string(optarg
, &arg_uuid
);
704 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
706 if (sd_id128_is_null(arg_uuid
))
707 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
708 "Machine UUID may not be all zeroes.");
710 arg_settings_mask
|= SETTING_MACHINE_ID
;
719 arg_machine
= mfree(arg_machine
);
721 if (!machine_name_is_valid(optarg
))
722 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
723 "Invalid machine name: %s", optarg
);
725 r
= free_and_strdup(&arg_machine
, optarg
);
733 arg_hostname
= mfree(arg_hostname
);
735 if (!hostname_is_valid(optarg
, false))
736 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
737 "Invalid hostname: %s", optarg
);
739 r
= free_and_strdup(&arg_hostname
, optarg
);
744 arg_settings_mask
|= SETTING_HOSTNAME
;
748 arg_selinux_context
= optarg
;
752 arg_selinux_apifs_context
= optarg
;
756 arg_read_only
= true;
757 arg_settings_mask
|= SETTING_READ_ONLY
;
761 case ARG_DROP_CAPABILITY
: {
764 _cleanup_free_
char *t
= NULL
;
766 r
= extract_first_word(&p
, &t
, ",", 0);
768 return log_error_errno(r
, "Failed to parse capability %s.", t
);
773 if (streq(t
, "all")) {
774 if (c
== ARG_CAPABILITY
)
775 plus
= (uint64_t) -1;
777 minus
= (uint64_t) -1;
779 r
= capability_from_name(t
);
781 return log_error_errno(r
, "Failed to parse capability %s.", t
);
783 if (c
== ARG_CAPABILITY
)
790 arg_settings_mask
|= SETTING_CAPABILITY
;
794 case ARG_NO_NEW_PRIVILEGES
:
795 r
= parse_boolean(optarg
);
797 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
799 arg_no_new_privileges
= r
;
800 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
804 arg_link_journal
= LINK_GUEST
;
805 arg_link_journal_try
= true;
806 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
809 case ARG_LINK_JOURNAL
:
810 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
812 log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
816 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
821 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
823 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
825 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
829 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
831 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
833 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
838 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
839 if (r
== -EADDRNOTAVAIL
)
840 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
842 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
844 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
850 if (!env_assignment_is_valid(optarg
))
851 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
852 "Environment variable assignment '%s' is not valid.", optarg
);
854 n
= strv_env_set(arg_setenv
, optarg
);
858 strv_free_and_replace(arg_setenv
, n
);
859 arg_settings_mask
|= SETTING_ENVIRONMENT
;
867 case ARG_SHARE_SYSTEM
:
868 /* We don't officially support this anymore, except for compat reasons. People should use the
869 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
870 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
871 arg_clone_ns_flags
= 0;
875 r
= parse_boolean(optarg
);
877 log_error("Failed to parse --register= argument: %s", optarg
);
885 arg_keep_unit
= true;
888 case ARG_PERSONALITY
:
890 arg_personality
= personality_from_string(optarg
);
891 if (arg_personality
== PERSONALITY_INVALID
)
892 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
893 "Unknown or unsupported personality '%s'.", optarg
);
895 arg_settings_mask
|= SETTING_PERSONALITY
;
901 arg_volatile_mode
= VOLATILE_YES
;
902 else if (streq(optarg
, "help")) {
903 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
908 m
= volatile_mode_from_string(optarg
);
910 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
911 "Failed to parse --volatile= argument: %s", optarg
);
913 arg_volatile_mode
= m
;
916 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
920 r
= expose_port_parse(&arg_expose_ports
, optarg
);
922 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
924 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
926 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
930 if (strv_extend(&arg_property
, optarg
) < 0)
935 case ARG_PRIVATE_USERS
: {
940 else if (!in_charset(optarg
, DIGITS
))
941 /* do *not* parse numbers as booleans */
942 boolean
= parse_boolean(optarg
);
944 if (boolean
== false) {
945 /* no: User namespacing off */
946 arg_userns_mode
= USER_NAMESPACE_NO
;
947 arg_uid_shift
= UID_INVALID
;
948 arg_uid_range
= UINT32_C(0x10000);
949 } else if (boolean
== true) {
950 /* yes: User namespacing on, UID range is read from root dir */
951 arg_userns_mode
= USER_NAMESPACE_FIXED
;
952 arg_uid_shift
= UID_INVALID
;
953 arg_uid_range
= UINT32_C(0x10000);
954 } else if (streq(optarg
, "pick")) {
955 /* pick: User namespacing on, UID range is picked randomly */
956 arg_userns_mode
= USER_NAMESPACE_PICK
;
957 arg_uid_shift
= UID_INVALID
;
958 arg_uid_range
= UINT32_C(0x10000);
960 _cleanup_free_
char *buffer
= NULL
;
961 const char *range
, *shift
;
963 /* anything else: User namespacing on, UID range is explicitly configured */
965 range
= strchr(optarg
, ':');
967 buffer
= strndup(optarg
, range
- optarg
);
973 r
= safe_atou32(range
, &arg_uid_range
);
975 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
979 r
= parse_uid(shift
, &arg_uid_shift
);
981 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
983 arg_userns_mode
= USER_NAMESPACE_FIXED
;
986 if (arg_uid_range
<= 0)
987 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
988 "UID range cannot be 0.");
990 arg_settings_mask
|= SETTING_USERNS
;
995 if (userns_supported()) {
996 arg_userns_mode
= USER_NAMESPACE_PICK
;
997 arg_uid_shift
= UID_INVALID
;
998 arg_uid_range
= UINT32_C(0x10000);
1000 arg_settings_mask
|= SETTING_USERNS
;
1005 case ARG_PRIVATE_USERS_CHOWN
:
1006 arg_userns_chown
= true;
1008 arg_settings_mask
|= SETTING_USERNS
;
1011 case ARG_KILL_SIGNAL
:
1012 if (streq(optarg
, "help")) {
1013 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1017 arg_kill_signal
= signal_from_string(optarg
);
1018 if (arg_kill_signal
< 0)
1019 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1020 "Cannot parse signal: %s", optarg
);
1022 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1027 /* no → do not read files
1028 * yes → read files, do not override cmdline, trust only subset
1029 * override → read files, override cmdline, trust only subset
1030 * trusted → read files, do not override cmdline, trust all
1033 r
= parse_boolean(optarg
);
1035 if (streq(optarg
, "trusted")) {
1036 mask_all_settings
= false;
1037 mask_no_settings
= false;
1038 arg_settings_trusted
= true;
1040 } else if (streq(optarg
, "override")) {
1041 mask_all_settings
= false;
1042 mask_no_settings
= true;
1043 arg_settings_trusted
= -1;
1045 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1048 mask_all_settings
= false;
1049 mask_no_settings
= false;
1050 arg_settings_trusted
= -1;
1053 mask_all_settings
= true;
1054 mask_no_settings
= false;
1055 arg_settings_trusted
= false;
1061 if (!path_is_absolute(optarg
))
1062 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1063 "Working directory %s is not an absolute path.", optarg
);
1065 r
= free_and_strdup(&arg_chdir
, optarg
);
1069 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1072 case ARG_PIVOT_ROOT
:
1073 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1075 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1077 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1080 case ARG_NOTIFY_READY
:
1081 r
= parse_boolean(optarg
);
1083 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1084 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1085 arg_notify_ready
= r
;
1086 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1089 case ARG_ROOT_HASH
: {
1093 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1095 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1096 if (l
< sizeof(sd_id128_t
)) {
1097 log_error("Root hash must be at least 128bit long: %s", optarg
);
1102 free(arg_root_hash
);
1104 arg_root_hash_size
= l
;
1108 case ARG_SYSTEM_CALL_FILTER
: {
1112 negative
= optarg
[0] == '~';
1113 items
= negative
? optarg
+ 1 : optarg
;
1116 _cleanup_free_
char *word
= NULL
;
1118 r
= extract_first_word(&items
, &word
, NULL
, 0);
1124 return log_error_errno(r
, "Failed to parse system call filter: %m");
1127 r
= strv_extend(&arg_syscall_blacklist
, word
);
1129 r
= strv_extend(&arg_syscall_whitelist
, word
);
1134 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1143 if (streq(optarg
, "help")) {
1144 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1148 eq
= strchr(optarg
, '=');
1150 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1151 "--rlimit= expects an '=' assignment.");
1153 name
= strndup(optarg
, eq
- optarg
);
1157 rl
= rlimit_from_string_harder(name
);
1159 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1160 "Unknown resource limit: %s", name
);
1162 if (!arg_rlimit
[rl
]) {
1163 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1164 if (!arg_rlimit
[rl
])
1168 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1170 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1172 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1176 case ARG_OOM_SCORE_ADJUST
:
1177 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1179 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1181 arg_oom_score_adjust_set
= true;
1182 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1185 case ARG_CPU_AFFINITY
: {
1186 _cleanup_cpu_free_ cpu_set_t
*cpuset
= NULL
;
1188 r
= parse_cpu_set(optarg
, &cpuset
);
1190 return log_error_errno(r
, "Failed to parse CPU affinity mask: %s", optarg
);
1193 CPU_FREE(arg_cpuset
);
1195 arg_cpuset
= TAKE_PTR(cpuset
);
1196 arg_cpuset_ncpus
= r
;
1197 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1201 case ARG_RESOLV_CONF
:
1202 if (streq(optarg
, "help")) {
1203 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1207 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1208 if (arg_resolv_conf
< 0)
1209 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1210 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1212 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1216 if (streq(optarg
, "help")) {
1217 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1221 arg_timezone
= timezone_mode_from_string(optarg
);
1222 if (arg_timezone
< 0)
1223 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1224 "Failed to parse /etc/localtime mode: %s", optarg
);
1226 arg_settings_mask
|= SETTING_TIMEZONE
;
1233 assert_not_reached("Unhandled option");
1236 /* If --network-namespace-path is given with any other network-related option,
1237 * we need to error out, to avoid conflicts between different network options. */
1238 if (arg_network_namespace_path
&&
1239 (arg_network_interfaces
|| arg_network_macvlan
||
1240 arg_network_ipvlan
|| arg_network_veth_extra
||
1241 arg_network_bridge
|| arg_network_zone
||
1242 arg_network_veth
|| arg_private_network
))
1243 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1244 "--network-namespace-path cannot be combined with other network options.");
1246 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
1247 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
1248 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
1249 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
1251 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1252 arg_mount_settings
|= MOUNT_USE_USERNS
;
1254 if (arg_private_network
)
1255 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1257 parse_mount_settings_env();
1259 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1260 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1261 arg_register
= false;
1262 if (arg_start_mode
!= START_PID1
)
1263 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1264 "--boot cannot be used without namespacing.");
1267 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1268 arg_userns_chown
= true;
1270 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1271 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1272 * The latter is not technically a user session, but we don't need to labour the point. */
1273 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1274 "--keep-unit --register=yes may not be used when invoked from a user session.");
1276 if (arg_directory
&& arg_image
)
1277 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1278 "--directory= and --image= may not be combined.");
1280 if (arg_template
&& arg_image
)
1281 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1282 "--template= and --image= may not be combined.");
1284 if (arg_ephemeral
&& arg_template
&& !arg_directory
) {
1285 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1286 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1287 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1290 arg_directory
= TAKE_PTR(arg_template
);
1293 if (arg_template
&& !(arg_directory
|| arg_machine
))
1294 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1295 "--template= needs --directory= or --machine=.");
1297 if (arg_ephemeral
&& arg_template
)
1298 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1299 "--ephemeral and --template= may not be combined.");
1301 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1302 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1303 "--ephemeral and --link-journal= may not be combined.");
1305 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1306 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
1307 "--private-users= is not supported, kernel compiled without user namespace support.");
1309 if (arg_userns_chown
&& arg_read_only
)
1310 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1311 "--read-only and --private-users-chown may not be combined.");
1313 if (arg_network_bridge
&& arg_network_zone
)
1314 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1315 "--network-bridge= and --network-zone= may not be combined.");
1317 if (argc
> optind
) {
1318 arg_parameters
= strv_copy(argv
+ optind
);
1319 if (!arg_parameters
)
1322 arg_settings_mask
|= SETTING_START_MODE
;
1325 /* Load all settings from .nspawn files */
1326 if (mask_no_settings
)
1327 arg_settings_mask
= 0;
1329 /* Don't load any settings from .nspawn files */
1330 if (mask_all_settings
)
1331 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1333 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? 1ULL << CAP_NET_ADMIN
: 0)) & ~minus
;
1335 r
= cg_unified_flush();
1337 return log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
1339 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1341 arg_container_service_name
= e
;
1343 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1345 arg_use_cgns
= cg_ns_supported();
1349 r
= custom_mount_check_all();
1356 static int verify_arguments(void) {
1357 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1358 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1359 "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1361 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1362 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1363 "Cannot combine --private-users with read-write mounts.");
1365 if (arg_volatile_mode
!= VOLATILE_NO
&& arg_read_only
)
1366 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1367 "Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1369 if (arg_expose_ports
&& !arg_private_network
)
1370 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1371 "Cannot use --port= without private networking.");
1374 if (arg_expose_ports
)
1375 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
1376 "--port= is not supported, compiled without libiptc support.");
1379 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1380 arg_kill_signal
= SIGRTMIN
+3;
1385 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1388 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1391 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1394 if (uid
!= UID_INVALID
) {
1395 uid
+= arg_uid_shift
;
1397 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1401 if (gid
!= GID_INVALID
) {
1402 gid
+= (gid_t
) arg_uid_shift
;
1404 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1408 if (lchown(p
, uid
, gid
) < 0)
1414 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1418 q
= prefix_roota(root
, path
);
1419 r
= mkdir_errno_wrapper(q
, mode
);
1425 return userns_lchown(q
, uid
, gid
);
1428 static const char *timezone_from_path(const char *path
) {
1429 return PATH_STARTSWITH_SET(
1431 "../usr/share/zoneinfo/",
1432 "/usr/share/zoneinfo/");
1435 static int setup_timezone(const char *dest
) {
1436 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1437 const char *where
, *check
;
1443 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1444 r
= readlink_malloc("/etc/localtime", &p
);
1445 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1446 m
= arg_read_only
&& arg_volatile_mode
!= VOLATILE_YES
? TIMEZONE_OFF
: TIMEZONE_DELETE
;
1447 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1448 m
= arg_read_only
&& arg_volatile_mode
!= VOLATILE_YES
? TIMEZONE_BIND
: TIMEZONE_COPY
;
1450 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1451 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1455 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1458 } else if (arg_timezone
== TIMEZONE_AUTO
)
1459 m
= arg_read_only
&& arg_volatile_mode
!= VOLATILE_YES
? TIMEZONE_BIND
: TIMEZONE_SYMLINK
;
1465 if (m
== TIMEZONE_OFF
)
1468 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1470 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1474 where
= strjoina(etc
, "/localtime");
1478 case TIMEZONE_DELETE
:
1479 if (unlink(where
) < 0)
1480 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1484 case TIMEZONE_SYMLINK
: {
1485 _cleanup_free_
char *q
= NULL
;
1486 const char *z
, *what
;
1488 z
= timezone_from_path(p
);
1490 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1494 r
= readlink_malloc(where
, &q
);
1495 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1496 return 0; /* Already pointing to the right place? Then do nothing .. */
1498 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1499 r
= chase_symlinks(check
, dest
, 0, NULL
);
1501 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1503 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1504 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1505 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1509 what
= strjoina("../usr/share/zoneinfo/", z
);
1510 if (symlink(what
, where
) < 0) {
1511 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1512 errno
, "Failed to correct timezone of container, ignoring: %m");
1522 case TIMEZONE_BIND
: {
1523 _cleanup_free_
char *resolved
= NULL
;
1526 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1528 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1532 if (found
== 0) /* missing? */
1533 (void) touch(resolved
);
1535 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1537 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1543 /* If mounting failed, try to copy */
1544 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, COPY_REFLINK
|COPY_REPLACE
);
1546 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1547 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1554 assert_not_reached("unexpected mode");
1557 /* Fix permissions of the symlink or file copy we just created */
1558 r
= userns_lchown(where
, 0, 0);
1560 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1565 static int have_resolv_conf(const char *path
) {
1568 if (access(path
, F_OK
) < 0) {
1569 if (errno
== ENOENT
)
1572 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1578 static int resolved_listening(void) {
1579 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1580 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1581 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1584 /* Check if resolved is listening */
1586 r
= sd_bus_open_system(&bus
);
1588 return log_debug_errno(r
, "Failed to open system bus: %m");
1590 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1592 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1596 r
= sd_bus_get_property_string(bus
,
1597 "org.freedesktop.resolve1",
1598 "/org/freedesktop/resolve1",
1599 "org.freedesktop.resolve1.Manager",
1602 &dns_stub_listener_mode
);
1604 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1606 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1609 static int setup_resolv_conf(const char *dest
) {
1610 _cleanup_free_
char *etc
= NULL
;
1611 const char *where
, *what
;
1617 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1618 if (arg_private_network
)
1619 m
= RESOLV_CONF_OFF
;
1620 else if (have_resolv_conf(STATIC_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1621 m
= arg_read_only
&& arg_volatile_mode
!= VOLATILE_YES
? RESOLV_CONF_BIND_STATIC
: RESOLV_CONF_COPY_STATIC
;
1622 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1623 m
= arg_read_only
&& arg_volatile_mode
!= VOLATILE_YES
? RESOLV_CONF_BIND_HOST
: RESOLV_CONF_COPY_HOST
;
1625 m
= arg_read_only
&& arg_volatile_mode
!= VOLATILE_YES
? RESOLV_CONF_OFF
: RESOLV_CONF_DELETE
;
1627 m
= arg_resolv_conf
;
1629 if (m
== RESOLV_CONF_OFF
)
1632 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1634 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1638 where
= strjoina(etc
, "/resolv.conf");
1640 if (m
== RESOLV_CONF_DELETE
) {
1641 if (unlink(where
) < 0)
1642 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1647 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_COPY_STATIC
))
1648 what
= STATIC_RESOLV_CONF
;
1650 what
= "/etc/resolv.conf";
1652 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
)) {
1653 _cleanup_free_
char *resolved
= NULL
;
1656 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1658 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1662 if (found
== 0) /* missing? */
1663 (void) touch(resolved
);
1665 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1667 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1670 /* If that didn't work, let's copy the file */
1671 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, COPY_REFLINK
);
1673 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1674 * resolved or something similar runs inside and the symlink points there.
1676 * If the disk image is read-only, there's also no point in complaining.
1678 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
) && IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1679 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1683 r
= userns_lchown(where
, 0, 0);
1685 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1690 static int setup_boot_id(void) {
1691 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1692 _cleanup_free_
char *path
= NULL
;
1693 sd_id128_t rnd
= SD_ID128_NULL
;
1697 /* Generate a new randomized boot ID, so that each boot-up of
1698 * the container gets a new one */
1700 r
= tempfn_random_child(NULL
, "proc-sys-kernel-random-boot-id", &path
);
1702 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1704 r
= sd_id128_randomize(&rnd
);
1706 return log_error_errno(r
, "Failed to generate random boot id: %m");
1708 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1710 return log_error_errno(r
, "Failed to write boot id: %m");
1712 from
= TAKE_PTR(path
);
1713 to
= "/proc/sys/kernel/random/boot_id";
1715 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1719 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1722 static int copy_devnodes(const char *dest
) {
1723 static const char devnodes
[] =
1734 _cleanup_umask_ mode_t u
;
1740 /* Create /dev/net, so that we can create /dev/net/tun in it */
1741 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1742 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1744 NULSTR_FOREACH(d
, devnodes
) {
1745 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1748 from
= strappend("/dev/", d
);
1752 to
= prefix_root(dest
, from
);
1756 if (stat(from
, &st
) < 0) {
1758 if (errno
!= ENOENT
)
1759 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1761 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1762 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
1763 "%s is not a char or block device, cannot copy.", from
);
1765 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
1767 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1768 /* Explicitly warn the user when /dev is already populated. */
1769 if (errno
== EEXIST
)
1770 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
);
1772 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
1774 /* Some systems abusively restrict mknod but allow bind mounts. */
1777 return log_error_errno(r
, "touch (%s) failed: %m", to
);
1778 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
1780 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
1783 r
= userns_lchown(to
, 0, 0);
1785 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
1787 dn
= strjoin("/dev/", S_ISCHR(st
.st_mode
) ? "char" : "block");
1791 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
1793 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
1795 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
1798 prefixed
= prefix_root(dest
, sl
);
1802 t
= strjoin("../", d
);
1806 if (symlink(t
, prefixed
) < 0)
1807 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
1814 static int setup_pts(const char *dest
) {
1815 _cleanup_free_
char *options
= NULL
;
1820 if (arg_selinux_apifs_context
)
1821 (void) asprintf(&options
,
1822 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
1823 arg_uid_shift
+ TTY_GID
,
1824 arg_selinux_apifs_context
);
1827 (void) asprintf(&options
,
1828 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
1829 arg_uid_shift
+ TTY_GID
);
1834 /* Mount /dev/pts itself */
1835 p
= prefix_roota(dest
, "/dev/pts");
1836 r
= mkdir_errno_wrapper(p
, 0755);
1838 return log_error_errno(r
, "Failed to create /dev/pts: %m");
1840 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
1843 r
= userns_lchown(p
, 0, 0);
1845 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
1847 /* Create /dev/ptmx symlink */
1848 p
= prefix_roota(dest
, "/dev/ptmx");
1849 if (symlink("pts/ptmx", p
) < 0)
1850 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
1851 r
= userns_lchown(p
, 0, 0);
1853 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
1855 /* And fix /dev/pts/ptmx ownership */
1856 p
= prefix_roota(dest
, "/dev/pts/ptmx");
1857 r
= userns_lchown(p
, 0, 0);
1859 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
1864 static int setup_dev_console(const char *dest
, const char *console
) {
1865 _cleanup_umask_ mode_t u
;
1874 r
= chmod_and_chown(console
, 0600, arg_uid_shift
, arg_uid_shift
);
1876 return log_error_errno(r
, "Failed to correct access mode for TTY: %m");
1878 /* We need to bind mount the right tty to /dev/console since
1879 * ptys can only exist on pts file systems. To have something
1880 * to bind mount things on we create a empty regular file. */
1882 to
= prefix_roota(dest
, "/dev/console");
1885 return log_error_errno(r
, "touch() for /dev/console failed: %m");
1887 return mount_verbose(LOG_ERR
, console
, to
, NULL
, MS_BIND
, NULL
);
1890 static int setup_keyring(void) {
1891 key_serial_t keyring
;
1893 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
1894 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
1895 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
1896 * these system calls let's make sure we don't leak anything into the container. */
1898 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
1899 if (keyring
== -1) {
1900 if (errno
== ENOSYS
)
1901 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
1902 else if (IN_SET(errno
, EACCES
, EPERM
))
1903 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
1905 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
1911 static int setup_kmsg(int kmsg_socket
) {
1912 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1913 _cleanup_free_
char *fifo
= NULL
;
1914 _cleanup_close_
int fd
= -1;
1915 _cleanup_umask_ mode_t u
;
1919 assert(kmsg_socket
>= 0);
1923 /* We create the kmsg FIFO as as temporary file in /tmp, but immediately delete it after bind mounting it to
1924 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
1925 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
1926 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
1928 r
= tempfn_random_child(NULL
, "proc-kmsg", &fifo
);
1930 return log_error_errno(r
, "Failed to generate kmsg path: %m");
1932 if (mkfifo(fifo
, 0600) < 0)
1933 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
1935 from
= TAKE_PTR(fifo
);
1938 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1942 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
1944 return log_error_errno(errno
, "Failed to open fifo: %m");
1946 /* Store away the fd in the socket, so that it stays open as long as we run the child */
1947 r
= send_one_fd(kmsg_socket
, fd
, 0);
1949 return log_error_errno(r
, "Failed to send FIFO fd: %m");
1954 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
1955 union in_addr_union
*exposed
= userdata
;
1961 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
1965 static int setup_hostname(void) {
1968 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
1971 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
1973 return log_error_errno(r
, "Failed to set hostname: %m");
1978 static int setup_journal(const char *directory
) {
1980 _cleanup_free_
char *d
= NULL
;
1983 char id
[33], *dirname
;
1986 /* Don't link journals in ephemeral mode */
1990 if (arg_link_journal
== LINK_NO
)
1993 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
1995 r
= sd_id128_get_machine(&this_id
);
1997 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
1999 if (sd_id128_equal(arg_uuid
, this_id
)) {
2000 log_full(try ? LOG_WARNING
: LOG_ERR
,
2001 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2007 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2008 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2010 bool ignore
= r
== -EROFS
&& try;
2011 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2012 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2013 return ignore
? 0 : r
;
2017 (void) sd_id128_to_string(arg_uuid
, id
);
2019 p
= strjoina("/var/log/journal/", id
);
2020 q
= prefix_roota(directory
, p
);
2022 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2026 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2027 "%s: already a mount point, refusing to use for journal", p
);
2030 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2034 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2035 "%s: already a mount point, refusing to use for journal", q
);
2038 r
= readlink_and_make_absolute(p
, &d
);
2040 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2043 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2045 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2050 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2051 } else if (r
== -EINVAL
) {
2053 if (arg_link_journal
== LINK_GUEST
&&
2056 if (errno
== ENOTDIR
) {
2057 log_error("%s already exists and is neither a symlink nor a directory", p
);
2060 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2062 } else if (r
!= -ENOENT
)
2063 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2065 if (arg_link_journal
== LINK_GUEST
) {
2067 if (symlink(q
, p
) < 0) {
2069 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2072 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2075 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2077 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2081 if (arg_link_journal
== LINK_HOST
) {
2082 /* don't create parents here — if the host doesn't have
2083 * permanent journal set up, don't force it here */
2085 r
= mkdir_errno_wrapper(p
, 0755);
2086 if (r
< 0 && r
!= -EEXIST
) {
2088 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2091 return log_error_errno(r
, "Failed to create %s: %m", p
);
2094 } else if (access(p
, F_OK
) < 0)
2097 if (dir_is_empty(q
) == 0)
2098 log_warning("%s is not empty, proceeding anyway.", q
);
2100 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2102 return log_error_errno(r
, "Failed to create %s: %m", q
);
2104 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2106 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2111 static int drop_capabilities(void) {
2112 return capability_bounding_set_drop(arg_caps_retain
, false);
2115 static int reset_audit_loginuid(void) {
2116 _cleanup_free_
char *p
= NULL
;
2119 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2122 r
= read_one_line_file("/proc/self/loginuid", &p
);
2126 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2128 /* Already reset? */
2129 if (streq(p
, "4294967295"))
2132 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2135 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2136 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2137 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2138 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2139 "using systemd-nspawn. Sleeping for 5s... (%m)");
2147 static int setup_propagate(const char *root
) {
2151 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2152 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2153 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2154 (void) mkdir_p(p
, 0600);
2156 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2158 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2160 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2162 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2164 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2166 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2168 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2169 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2173 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2177 /* machined will MS_MOVE into that directory, and that's only
2178 * supported for non-shared mounts. */
2179 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2182 static int setup_machine_id(const char *directory
) {
2183 const char *etc_machine_id
;
2187 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2188 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2189 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2190 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2191 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2192 * container behaves nicely). */
2194 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2196 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2198 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2199 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2201 if (sd_id128_is_null(arg_uuid
)) {
2202 r
= sd_id128_randomize(&arg_uuid
);
2204 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2207 if (sd_id128_is_null(id
))
2208 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2209 "Machine ID in container image is zero, refusing.");
2217 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2222 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2225 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2226 if (r
== -EOPNOTSUPP
)
2227 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.");
2229 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2231 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2233 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2235 log_debug("Patched directory tree to match UID/GID range.");
2242 * < 0 : wait_for_terminate() failed to get the state of the
2243 * container, the container was terminated by a signal, or
2244 * failed for an unknown reason. No change is made to the
2245 * container argument.
2246 * > 0 : The program executed in the container terminated with an
2247 * error. The exit code of the program executed in the
2248 * container is returned. The container argument has been set
2249 * to CONTAINER_TERMINATED.
2250 * 0 : The container is being rebooted, has been shut down or exited
2251 * successfully. The container argument has been set to either
2252 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2254 * That is, success is indicated by a return value of zero, and an
2255 * error is indicated by a non-zero value.
2257 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2261 r
= wait_for_terminate(pid
, &status
);
2263 return log_warning_errno(r
, "Failed to wait for container: %m");
2265 switch (status
.si_code
) {
2268 if (status
.si_status
== 0)
2269 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2271 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2273 *container
= CONTAINER_TERMINATED
;
2274 return status
.si_status
;
2277 if (status
.si_status
== SIGINT
) {
2278 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2279 *container
= CONTAINER_TERMINATED
;
2282 } else if (status
.si_status
== SIGHUP
) {
2283 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2284 *container
= CONTAINER_REBOOTED
;
2290 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2291 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2294 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2295 "Container %s failed due to unknown reason.", arg_machine
);
2299 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2302 pid
= PTR_TO_PID(userdata
);
2304 if (kill(pid
, arg_kill_signal
) >= 0) {
2305 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2306 sd_event_source_set_userdata(s
, NULL
);
2311 sd_event_exit(sd_event_source_get_event(s
), 0);
2315 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2321 pid
= PTR_TO_PID(userdata
);
2326 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2327 return log_error_errno(errno
, "Failed to waitid(): %m");
2328 if (si
.si_pid
== 0) /* No pending children. */
2330 if (si
.si_pid
== pid
) {
2331 /* The main process we care for has exited. Return from
2332 * signal handler but leave the zombie. */
2333 sd_event_exit(sd_event_source_get_event(s
), 0);
2337 /* Reap all other children. */
2338 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2344 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2349 pid
= PTR_TO_PID(userdata
);
2351 if (arg_kill_signal
> 0) {
2352 log_info("Container termination requested. Attempting to halt container.");
2353 (void) kill(pid
, arg_kill_signal
);
2355 log_info("Container termination requested. Exiting.");
2356 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2362 static int determine_names(void) {
2365 if (arg_template
&& !arg_directory
&& arg_machine
) {
2367 /* If --template= was specified then we should not
2368 * search for a machine, but instead create a new one
2369 * in /var/lib/machine. */
2371 arg_directory
= strjoin("/var/lib/machines/", arg_machine
);
2376 if (!arg_image
&& !arg_directory
) {
2378 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2380 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2382 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2384 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2386 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2387 r
= free_and_strdup(&arg_image
, i
->path
);
2389 r
= free_and_strdup(&arg_directory
, i
->path
);
2394 arg_read_only
= arg_read_only
|| i
->read_only
;
2396 r
= safe_getcwd(&arg_directory
);
2398 return log_error_errno(r
, "Failed to determine current directory: %m");
2401 if (!arg_directory
&& !arg_image
) {
2402 log_error("Failed to determine path, please use -D or -i.");
2408 if (arg_directory
&& path_equal(arg_directory
, "/"))
2409 arg_machine
= gethostname_malloc();
2414 arg_machine
= strdup(basename(arg_image
));
2416 /* Truncate suffix if there is one */
2417 e
= endswith(arg_machine
, ".raw");
2421 arg_machine
= strdup(basename(arg_directory
));
2426 hostname_cleanup(arg_machine
);
2427 if (!machine_name_is_valid(arg_machine
)) {
2428 log_error("Failed to determine machine name automatically, please use -M.");
2432 if (arg_ephemeral
) {
2435 /* Add a random suffix when this is an
2436 * ephemeral machine, so that we can run many
2437 * instances at once without manually having
2438 * to specify -M each time. */
2440 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2451 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2460 r
= chase_symlinks(*p
, NULL
, flags
, &chased
);
2462 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2464 free_and_replace(*p
, chased
);
2465 return r
; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2468 static int determine_uid_shift(const char *directory
) {
2471 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2476 if (arg_uid_shift
== UID_INVALID
) {
2479 r
= stat(directory
, &st
);
2481 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2483 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2485 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2486 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2487 "UID and GID base of %s don't match.", directory
);
2489 arg_uid_range
= UINT32_C(0x10000);
2492 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2493 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2494 "UID base too high for UID range.");
2499 static int inner_child(
2501 const char *directory
,
2507 _cleanup_free_
char *home
= NULL
;
2510 const char *envp
[] = {
2511 "PATH=" DEFAULT_PATH_COMPAT
,
2512 NULL
, /* container */
2517 NULL
, /* container_uuid */
2518 NULL
, /* LISTEN_FDS */
2519 NULL
, /* LISTEN_PID */
2520 NULL
, /* NOTIFY_SOCKET */
2523 const char *exec_target
;
2524 _cleanup_strv_free_
char **env_use
= NULL
;
2527 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2528 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2529 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2530 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2531 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2532 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2535 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2536 * unshare(). See below. */
2540 assert(kmsg_socket
>= 0);
2542 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2543 /* Tell the parent, that it now can write the UID map. */
2544 (void) barrier_place(barrier
); /* #1 */
2546 /* Wait until the parent wrote the UID map */
2547 if (!barrier_place_and_sync(barrier
)) /* #2 */
2548 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2549 "Parent died too early");
2552 r
= reset_uid_gid();
2554 return log_error_errno(r
, "Couldn't become new root: %m");
2557 arg_mount_settings
| MOUNT_IN_USERNS
,
2559 arg_selinux_apifs_context
);
2563 if (!arg_network_namespace_path
&& arg_private_network
) {
2564 r
= unshare(CLONE_NEWNET
);
2566 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2568 /* Tell the parent that it can setup network interfaces. */
2569 (void) barrier_place(barrier
); /* #3 */
2572 r
= mount_sysfs(NULL
, arg_mount_settings
);
2576 /* Wait until we are cgroup-ified, so that we
2577 * can mount the right cgroup path writable */
2578 if (!barrier_place_and_sync(barrier
)) /* #4 */
2579 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2580 "Parent died too early");
2582 if (arg_use_cgns
&& cg_ns_supported()) {
2583 r
= unshare(CLONE_NEWCGROUP
);
2585 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2588 arg_unified_cgroup_hierarchy
,
2589 arg_userns_mode
!= USER_NAMESPACE_NO
,
2592 arg_selinux_apifs_context
,
2597 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2602 r
= setup_boot_id();
2606 r
= setup_kmsg(kmsg_socket
);
2609 kmsg_socket
= safe_close(kmsg_socket
);
2612 return log_error_errno(errno
, "setsid() failed: %m");
2614 if (arg_private_network
)
2617 if (arg_expose_ports
) {
2618 r
= expose_port_send_rtnl(rtnl_socket
);
2621 rtnl_socket
= safe_close(rtnl_socket
);
2624 if (arg_oom_score_adjust_set
) {
2625 r
= set_oom_score_adjust(arg_oom_score_adjust
);
2627 return log_error_errno(r
, "Failed to adjust OOM score: %m");
2631 if (sched_setaffinity(0, CPU_ALLOC_SIZE(arg_cpuset_ncpus
), arg_cpuset
) < 0)
2632 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
2634 r
= drop_capabilities();
2636 return log_error_errno(r
, "drop_capabilities() failed: %m");
2638 (void) setup_hostname();
2640 if (arg_personality
!= PERSONALITY_INVALID
) {
2641 r
= safe_personality(arg_personality
);
2643 return log_error_errno(r
, "personality() failed: %m");
2644 } else if (secondary
) {
2645 r
= safe_personality(PER_LINUX32
);
2647 return log_error_errno(r
, "personality() failed: %m");
2651 if (arg_selinux_context
)
2652 if (setexeccon(arg_selinux_context
) < 0)
2653 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
2656 r
= change_uid_gid(arg_user
, &home
);
2660 if (arg_no_new_privileges
)
2661 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
2662 return log_error_errno(errno
, "Failed to disable new privileges: %m");
2664 /* LXC sets container=lxc, so follow the scheme here */
2665 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
2667 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
2671 if ((asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
? home
: "/root") < 0) ||
2672 (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
? arg_user
: "root") < 0) ||
2673 (asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0))
2676 assert(!sd_id128_is_null(arg_uuid
));
2678 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
2681 if (fdset_size(fds
) > 0) {
2682 r
= fdset_cloexec(fds
, false);
2684 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
2686 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
2687 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
2690 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
2693 env_use
= strv_env_merge(2, envp
, arg_setenv
);
2697 /* Let the parent know that we are ready and
2698 * wait until the parent is ready with the
2700 if (!barrier_place_and_sync(barrier
)) /* #5 */
2701 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2702 "Parent died too early");
2705 if (chdir(arg_chdir
) < 0)
2706 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
2708 if (arg_start_mode
== START_PID2
) {
2709 r
= stub_pid1(arg_uuid
);
2714 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
2715 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
2716 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
2718 log_set_open_when_needed(true);
2720 (void) fdset_close_others(fds
);
2722 if (arg_start_mode
== START_BOOT
) {
2726 /* Automatically search for the init system */
2728 m
= strv_length(arg_parameters
);
2729 a
= newa(char*, m
+ 2);
2730 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
2733 a
[0] = (char*) "/usr/lib/systemd/systemd";
2734 execve(a
[0], a
, env_use
);
2736 a
[0] = (char*) "/lib/systemd/systemd";
2737 execve(a
[0], a
, env_use
);
2739 a
[0] = (char*) "/sbin/init";
2740 execve(a
[0], a
, env_use
);
2742 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2743 } else if (!strv_isempty(arg_parameters
)) {
2744 const char *dollar_path
;
2746 exec_target
= arg_parameters
[0];
2748 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
2750 dollar_path
= strv_env_get(env_use
, "PATH");
2752 if (putenv((char*) dollar_path
) != 0)
2753 return log_error_errno(errno
, "Failed to update $PATH: %m");
2756 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
2759 /* If we cannot change the directory, we'll end up in /, that is expected. */
2760 (void) chdir(home
?: "/root");
2762 execle("/bin/bash", "-bash", NULL
, env_use
);
2763 execle("/bin/sh", "-sh", NULL
, env_use
);
2765 exec_target
= "/bin/bash, /bin/sh";
2768 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
2771 static int setup_sd_notify_child(void) {
2772 _cleanup_close_
int fd
= -1;
2773 union sockaddr_union sa
= {
2774 .un
.sun_family
= AF_UNIX
,
2775 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
2779 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2781 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
2783 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
2784 (void) sockaddr_un_unlink(&sa
.un
);
2786 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
2788 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
2790 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
2792 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
2794 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
2796 return log_error_errno(r
, "SO_PASSCRED failed: %m");
2801 static int outer_child(
2803 const char *directory
,
2804 const char *console
,
2805 DissectedImage
*dissected_image
,
2813 int uid_shift_socket
,
2814 int unified_cgroup_hierarchy_socket
,
2818 _cleanup_close_
int fd
= -1;
2819 int r
, which_failed
;
2823 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
2824 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
2825 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
2826 * initializations a second child (the "inner" one) is forked off it, and it exits. */
2831 assert(pid_socket
>= 0);
2832 assert(uuid_socket
>= 0);
2833 assert(notify_socket
>= 0);
2834 assert(kmsg_socket
>= 0);
2836 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
2837 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
2842 terminal
= open_terminal(console
, O_RDWR
);
2844 return log_error_errno(terminal
, "Failed to open console: %m");
2846 /* Make sure we can continue logging to the original stderr, even if stderr points elsewhere now */
2847 r
= log_dup_console();
2849 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2851 r
= rearrange_stdio(terminal
, terminal
, terminal
); /* invalidates 'terminal' on success and failure */
2853 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2856 r
= reset_audit_loginuid();
2860 /* Mark everything as slave, so that we still
2861 * receive mounts from the real root, but don't
2862 * propagate mounts to the real root. */
2863 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
2867 if (dissected_image
) {
2868 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
2869 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
2870 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
2871 * makes sure ESP partitions and userns are compatible. */
2873 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
2874 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
2875 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0)|
2876 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
2881 r
= determine_uid_shift(directory
);
2885 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2886 /* Let the parent know which UID shift we read from the image */
2887 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
2889 return log_error_errno(errno
, "Failed to send UID shift: %m");
2890 if (l
!= sizeof(arg_uid_shift
))
2891 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2892 "Short write while sending UID shift.");
2894 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
2895 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2896 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2897 * not it will pick a different one, and send it back to us. */
2899 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
2901 return log_error_errno(errno
, "Failed to recv UID shift: %m");
2902 if (l
!= sizeof(arg_uid_shift
))
2903 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2904 "Short read while receiving UID shift.");
2907 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
2908 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
2911 if (dissected_image
) {
2912 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
2913 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
2914 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
2919 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
2920 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
2922 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
2926 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
2928 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
2929 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
2930 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2931 "Short write while sending cgroup mode.");
2933 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
2936 /* Turn directory into bind mount */
2937 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
2941 r
= setup_pivot_root(
2944 arg_pivot_root_old
);
2951 arg_userns_mode
!= USER_NAMESPACE_NO
,
2954 arg_selinux_context
);
2958 r
= setup_volatile_state(
2961 arg_userns_mode
!= USER_NAMESPACE_NO
,
2964 arg_selinux_context
);
2968 /* Mark everything as shared so our mounts get propagated down. This is
2969 * required to make new bind mounts available in systemd services
2970 * inside the containter that create a new mount namespace.
2971 * See https://github.com/systemd/systemd/issues/3860
2972 * Further submounts (such as /dev) done after this will inherit the
2973 * shared propagation mode. */
2974 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
2978 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
2982 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
2986 if (arg_read_only
) {
2987 r
= bind_remount_recursive(directory
, true, NULL
);
2989 return log_error_errno(r
, "Failed to make tree read-only: %m");
2992 r
= mount_all(directory
,
2995 arg_selinux_apifs_context
);
2999 r
= copy_devnodes(directory
);
3003 dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3005 r
= setup_pts(directory
);
3009 r
= setup_propagate(directory
);
3013 r
= setup_dev_console(directory
, console
);
3017 r
= setup_keyring();
3021 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
3025 r
= setup_timezone(directory
);
3029 r
= setup_resolv_conf(directory
);
3033 r
= setup_machine_id(directory
);
3037 r
= setup_journal(directory
);
3044 arg_n_custom_mounts
,
3045 arg_userns_mode
!= USER_NAMESPACE_NO
,
3048 arg_selinux_apifs_context
);
3052 if (!arg_use_cgns
|| !cg_ns_supported()) {
3055 arg_unified_cgroup_hierarchy
,
3056 arg_userns_mode
!= USER_NAMESPACE_NO
,
3059 arg_selinux_apifs_context
,
3065 r
= mount_move_root(directory
);
3067 return log_error_errno(r
, "Failed to move root directory: %m");
3069 fd
= setup_sd_notify_child();
3073 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3075 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3077 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3078 arg_clone_ns_flags
|
3079 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3081 return log_error_errno(errno
, "Failed to fork inner child: %m");
3083 pid_socket
= safe_close(pid_socket
);
3084 uuid_socket
= safe_close(uuid_socket
);
3085 notify_socket
= safe_close(notify_socket
);
3086 uid_shift_socket
= safe_close(uid_shift_socket
);
3088 /* The inner child has all namespaces that are
3089 * requested, so that we all are owned by the user if
3090 * user namespaces are turned on. */
3092 if (arg_network_namespace_path
) {
3093 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3095 return log_error_errno(r
, "Failed to join network namespace: %m");
3098 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, fds
);
3100 _exit(EXIT_FAILURE
);
3102 _exit(EXIT_SUCCESS
);
3105 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3107 return log_error_errno(errno
, "Failed to send PID: %m");
3108 if (l
!= sizeof(pid
))
3109 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3110 "Short write while sending PID.");
3112 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3114 return log_error_errno(errno
, "Failed to send machine ID: %m");
3115 if (l
!= sizeof(arg_uuid
))
3116 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3117 "Short write while sending machine ID.");
3119 l
= send_one_fd(notify_socket
, fd
, 0);
3121 return log_error_errno(errno
, "Failed to send notify fd: %m");
3123 pid_socket
= safe_close(pid_socket
);
3124 uuid_socket
= safe_close(uuid_socket
);
3125 notify_socket
= safe_close(notify_socket
);
3126 kmsg_socket
= safe_close(kmsg_socket
);
3127 rtnl_socket
= safe_close(rtnl_socket
);
3128 netns_fd
= safe_close(netns_fd
);
3133 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3134 bool tried_hashed
= false;
3135 unsigned n_tries
= 100;
3140 assert(ret_lock_file
);
3141 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3142 assert(arg_uid_range
== 0x10000U
);
3146 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3149 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3150 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3155 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3157 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3160 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3161 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3162 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3167 /* Make some superficial checks whether the range is currently known in the user database */
3168 if (getpwuid(candidate
))
3170 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3172 if (getgrgid(candidate
))
3174 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3177 *ret_lock_file
= lf
;
3178 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3183 if (arg_machine
&& !tried_hashed
) {
3184 /* Try to hash the base from the container name */
3186 static const uint8_t hash_key
[] = {
3187 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3188 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3191 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3193 tried_hashed
= true;
3195 random_bytes(&candidate
, sizeof(candidate
));
3197 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3198 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3202 static int setup_uid_map(pid_t pid
) {
3203 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3208 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3209 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3210 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3212 return log_error_errno(r
, "Failed to write UID map: %m");
3214 /* We always assign the same UID and GID ranges */
3215 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3216 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3218 return log_error_errno(r
, "Failed to write GID map: %m");
3223 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3224 char buf
[NOTIFY_BUFFER_MAX
+1];
3226 struct iovec iovec
= {
3228 .iov_len
= sizeof(buf
)-1,
3231 struct cmsghdr cmsghdr
;
3232 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
3233 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
3235 struct msghdr msghdr
= {
3238 .msg_control
= &control
,
3239 .msg_controllen
= sizeof(control
),
3241 struct cmsghdr
*cmsg
;
3242 struct ucred
*ucred
= NULL
;
3244 pid_t inner_child_pid
;
3245 _cleanup_strv_free_
char **tags
= NULL
;
3249 inner_child_pid
= PTR_TO_PID(userdata
);
3251 if (revents
!= EPOLLIN
) {
3252 log_warning("Got unexpected poll event for notify fd.");
3256 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3258 if (IN_SET(errno
, EAGAIN
, EINTR
))
3261 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
3263 cmsg_close_all(&msghdr
);
3265 CMSG_FOREACH(cmsg
, &msghdr
) {
3266 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
3267 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
3268 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
3270 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
3274 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3275 log_debug("Received notify message without valid credentials. Ignoring.");
3279 if ((size_t) n
>= sizeof(buf
)) {
3280 log_warning("Received notify message exceeded maximum size. Ignoring.");
3285 tags
= strv_split(buf
, "\n\r");
3289 if (strv_find(tags
, "READY=1"))
3290 sd_notifyf(false, "READY=1\n");
3292 p
= strv_find_startswith(tags
, "STATUS=");
3294 sd_notifyf(false, "STATUS=Container running: %s", p
);
3299 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3302 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3304 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3306 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3311 static int merge_settings(Settings
*settings
, const char *path
) {
3317 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3318 * that this steals the fields of the Settings* structure, and hence modifies it. */
3320 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3321 settings
->start_mode
>= 0) {
3322 arg_start_mode
= settings
->start_mode
;
3323 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3326 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3327 arg_ephemeral
= settings
->ephemeral
;
3329 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3330 settings
->pivot_root_new
) {
3331 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3332 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3335 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3336 settings
->working_directory
)
3337 free_and_replace(arg_chdir
, settings
->working_directory
);
3339 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3340 settings
->environment
)
3341 strv_free_and_replace(arg_setenv
, settings
->environment
);
3343 if ((arg_settings_mask
& SETTING_USER
) == 0 &&
3345 free_and_replace(arg_user
, settings
->user
);
3347 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3350 plus
= settings
->capability
;
3351 if (settings_private_network(settings
))
3352 plus
|= (1ULL << CAP_NET_ADMIN
);
3354 if (!arg_settings_trusted
&& plus
!= 0) {
3355 if (settings
->capability
!= 0)
3356 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3358 arg_caps_retain
|= plus
;
3360 arg_caps_retain
&= ~settings
->drop_capability
;
3363 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3364 settings
->kill_signal
> 0)
3365 arg_kill_signal
= settings
->kill_signal
;
3367 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3368 settings
->personality
!= PERSONALITY_INVALID
)
3369 arg_personality
= settings
->personality
;
3371 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3372 !sd_id128_is_null(settings
->machine_id
)) {
3374 if (!arg_settings_trusted
)
3375 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3377 arg_uuid
= settings
->machine_id
;
3380 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3381 settings
->read_only
>= 0)
3382 arg_read_only
= settings
->read_only
;
3384 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3385 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3386 arg_volatile_mode
= settings
->volatile_mode
;
3388 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3389 settings
->n_custom_mounts
> 0) {
3391 if (!arg_settings_trusted
)
3392 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3394 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3395 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3396 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3397 settings
->n_custom_mounts
= 0;
3401 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3402 (settings
->private_network
>= 0 ||
3403 settings
->network_veth
>= 0 ||
3404 settings
->network_bridge
||
3405 settings
->network_zone
||
3406 settings
->network_interfaces
||
3407 settings
->network_macvlan
||
3408 settings
->network_ipvlan
||
3409 settings
->network_veth_extra
)) {
3411 if (!arg_settings_trusted
)
3412 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3414 arg_network_veth
= settings_network_veth(settings
);
3415 arg_private_network
= settings_private_network(settings
);
3417 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3418 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3419 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3420 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3422 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3423 free_and_replace(arg_network_zone
, settings
->network_zone
);
3427 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3428 settings
->expose_ports
) {
3430 if (!arg_settings_trusted
)
3431 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3433 expose_port_free_all(arg_expose_ports
);
3434 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3438 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3439 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3441 if (!arg_settings_trusted
)
3442 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3444 arg_userns_mode
= settings
->userns_mode
;
3445 arg_uid_shift
= settings
->uid_shift
;
3446 arg_uid_range
= settings
->uid_range
;
3447 arg_userns_chown
= settings
->userns_chown
;
3451 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3452 arg_notify_ready
= settings
->notify_ready
;
3454 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3456 if (!arg_settings_trusted
&& !strv_isempty(arg_syscall_whitelist
))
3457 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3459 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3460 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3464 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3465 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3468 if (!settings
->rlimit
[rl
])
3471 if (!arg_settings_trusted
) {
3472 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3476 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3479 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3481 free_and_replace(arg_hostname
, settings
->hostname
);
3483 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3484 settings
->no_new_privileges
>= 0)
3485 arg_no_new_privileges
= settings
->no_new_privileges
;
3487 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3488 settings
->oom_score_adjust_set
) {
3490 if (!arg_settings_trusted
)
3491 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
3493 arg_oom_score_adjust
= settings
->oom_score_adjust
;
3494 arg_oom_score_adjust_set
= true;
3498 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
3501 if (!arg_settings_trusted
)
3502 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
3505 CPU_FREE(arg_cpuset
);
3506 arg_cpuset
= TAKE_PTR(settings
->cpuset
);
3507 arg_cpuset_ncpus
= settings
->cpuset_ncpus
;
3511 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
3512 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
3513 arg_resolv_conf
= settings
->resolv_conf
;
3515 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
3516 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
3518 if (!arg_settings_trusted
)
3519 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
3521 arg_link_journal
= settings
->link_journal
;
3522 arg_link_journal_try
= settings
->link_journal_try
;
3526 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
3527 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
3528 arg_timezone
= settings
->timezone
;
3533 static int load_settings(void) {
3534 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
3535 _cleanup_fclose_
FILE *f
= NULL
;
3536 _cleanup_free_
char *p
= NULL
;
3540 /* If all settings are masked, there's no point in looking for
3541 * the settings file */
3542 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
3545 fn
= strjoina(arg_machine
, ".nspawn");
3547 /* We first look in the admin's directories in /etc and /run */
3548 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3549 _cleanup_free_
char *j
= NULL
;
3551 j
= strjoin(i
, "/", fn
);
3559 /* By default, we trust configuration from /etc and /run */
3560 if (arg_settings_trusted
< 0)
3561 arg_settings_trusted
= true;
3566 if (errno
!= ENOENT
)
3567 return log_error_errno(errno
, "Failed to open %s: %m", j
);
3571 /* After that, let's look for a file next to the
3572 * actual image we shall boot. */
3575 p
= file_in_same_dir(arg_image
, fn
);
3578 } else if (arg_directory
) {
3579 p
= file_in_same_dir(arg_directory
, fn
);
3586 if (!f
&& errno
!= ENOENT
)
3587 return log_error_errno(errno
, "Failed to open %s: %m", p
);
3589 /* By default, we do not trust configuration from /var/lib/machines */
3590 if (arg_settings_trusted
< 0)
3591 arg_settings_trusted
= false;
3598 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
3600 r
= settings_load(f
, p
, &settings
);
3604 return merge_settings(settings
, p
);
3607 static int run(int master
,
3608 const char* console
,
3609 DissectedImage
*dissected_image
,
3613 char veth_name
[IFNAMSIZ
], bool *veth_created
,
3614 union in_addr_union
*exposed
,
3615 pid_t
*pid
, int *ret
) {
3617 static const struct sigaction sa
= {
3618 .sa_handler
= nop_signal_handler
,
3619 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
3622 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
3623 _cleanup_close_
int etc_passwd_lock
= -1;
3624 _cleanup_close_pair_
int
3625 kmsg_socket_pair
[2] = { -1, -1 },
3626 rtnl_socket_pair
[2] = { -1, -1 },
3627 pid_socket_pair
[2] = { -1, -1 },
3628 uuid_socket_pair
[2] = { -1, -1 },
3629 notify_socket_pair
[2] = { -1, -1 },
3630 uid_shift_socket_pair
[2] = { -1, -1 },
3631 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
3633 _cleanup_close_
int notify_socket
= -1;
3634 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
3635 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
3636 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
3637 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
3638 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
3639 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
3640 ContainerStatus container_status
= 0;
3645 _cleanup_close_
int netns_fd
= -1;
3647 assert_se(sigemptyset(&mask_chld
) == 0);
3648 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
3650 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3651 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3652 * check with getpwuid() if the specific user already exists. Note that /etc might be
3653 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3654 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3655 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3658 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
3659 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
3660 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
3663 r
= barrier_create(&barrier
);
3665 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
3667 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
3668 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
3670 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
3671 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
3673 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
3674 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
3676 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
3677 return log_error_errno(errno
, "Failed to create id socket pair: %m");
3679 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
3680 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
3682 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3683 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
3684 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
3686 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
3687 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
3688 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
3690 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3691 * parent's blocking calls and give it a chance to call wait() and terminate. */
3692 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
3694 return log_error_errno(errno
, "Failed to change the signal mask: %m");
3696 r
= sigaction(SIGCHLD
, &sa
, NULL
);
3698 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
3700 if (arg_network_namespace_path
) {
3701 netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
3703 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
3705 r
= fd_is_network_ns(netns_fd
);
3707 log_debug_errno(r
, "Cannot determine if passed network namespace path '%s' really refers to a network namespace, assuming it does.", arg_network_namespace_path
);
3709 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
3711 log_error("Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
3716 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
3718 return log_error_errno(errno
, "clone() failed%s: %m",
3720 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3723 /* The outer child only has a file system namespace. */
3724 barrier_set_role(&barrier
, BARRIER_CHILD
);
3726 master
= safe_close(master
);
3728 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
3729 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
3730 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
3731 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
3732 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
3733 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
3734 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
3736 (void) reset_all_signal_handlers();
3737 (void) reset_signal_mask();
3739 r
= outer_child(&barrier
,
3746 uuid_socket_pair
[1],
3747 notify_socket_pair
[1],
3748 kmsg_socket_pair
[1],
3749 rtnl_socket_pair
[1],
3750 uid_shift_socket_pair
[1],
3751 unified_cgroup_hierarchy_socket_pair
[1],
3755 _exit(EXIT_FAILURE
);
3757 _exit(EXIT_SUCCESS
);
3760 barrier_set_role(&barrier
, BARRIER_PARENT
);
3762 fds
= fdset_free(fds
);
3764 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
3765 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
3766 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
3767 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
3768 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
3769 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
3770 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
3772 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3773 /* The child just let us know the UID shift it might have read from the image. */
3774 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
3776 return log_error_errno(errno
, "Failed to read UID shift: %m");
3777 if (l
!= sizeof arg_uid_shift
) {
3778 log_error("Short read while reading UID shift.");
3782 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3783 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3784 * image, but if that's already in use, pick a new one, and report back to the child,
3785 * which one we now picked. */
3787 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
3789 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
3791 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
3793 return log_error_errno(errno
, "Failed to send UID shift: %m");
3794 if (l
!= sizeof arg_uid_shift
) {
3795 log_error("Short write while writing UID shift.");
3801 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3802 /* The child let us know the support cgroup mode it might have read from the image. */
3803 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
3805 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
3806 if (l
!= sizeof(arg_unified_cgroup_hierarchy
)) {
3807 log_error("Short read while reading cgroup mode (%zu bytes).%s",
3808 l
, l
== 0 ? " The child is most likely dead." : "");
3813 /* Wait for the outer child. */
3814 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
3817 if (r
!= EXIT_SUCCESS
)
3820 /* And now retrieve the PID of the inner child. */
3821 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
3823 return log_error_errno(errno
, "Failed to read inner child PID: %m");
3824 if (l
!= sizeof *pid
) {
3825 log_error("Short read while reading inner child PID.");
3829 /* We also retrieve container UUID in case it was generated by outer child */
3830 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
3832 return log_error_errno(errno
, "Failed to read container machine ID: %m");
3833 if (l
!= sizeof(arg_uuid
)) {
3834 log_error("Short read while reading container machined ID.");
3838 /* We also retrieve the socket used for notifications generated by outer child */
3839 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
3840 if (notify_socket
< 0)
3841 return log_error_errno(notify_socket
,
3842 "Failed to receive notification socket from the outer child: %m");
3844 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
3846 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3847 if (!barrier_place_and_sync(&barrier
)) { /* #1 */
3848 log_error("Child died too early.");
3852 r
= setup_uid_map(*pid
);
3856 (void) barrier_place(&barrier
); /* #2 */
3859 if (arg_private_network
) {
3860 if (!arg_network_namespace_path
) {
3861 /* Wait until the child has unshared its network namespace. */
3862 if (!barrier_place_and_sync(&barrier
)) { /* #3 */
3863 log_error("Child died too early");
3868 r
= move_network_interfaces(*pid
, arg_network_interfaces
);
3872 if (arg_network_veth
) {
3873 r
= setup_veth(arg_machine
, *pid
, veth_name
,
3874 arg_network_bridge
|| arg_network_zone
);
3880 if (arg_network_bridge
) {
3881 /* Add the interface to a bridge */
3882 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
3887 } else if (arg_network_zone
) {
3888 /* Add the interface to a bridge, possibly creating it */
3889 r
= setup_bridge(veth_name
, arg_network_zone
, true);
3897 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
3901 /* We created the primary and extra veth links now; let's remember this, so that we know to
3902 remove them later on. Note that we don't bother with removing veth links that were created
3903 here when their setup failed half-way, because in that case the kernel should be able to
3904 remove them on its own, since they cannot be referenced by anything yet. */
3905 *veth_created
= true;
3907 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
3911 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
3916 if (arg_register
|| !arg_keep_unit
) {
3917 r
= sd_bus_default_system(&bus
);
3919 return log_error_errno(r
, "Failed to open system bus: %m");
3921 r
= sd_bus_set_close_on_exit(bus
, false);
3923 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
3926 if (!arg_keep_unit
) {
3927 /* When a new scope is created for this container, then we'll be registered as its controller, in which
3928 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
3929 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
3931 r
= sd_bus_match_signal_async(
3934 "org.freedesktop.systemd1",
3936 "org.freedesktop.systemd1.Scope",
3938 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
3940 return log_error_errno(r
, "Failed to request RequestStop match: %m");
3944 r
= register_machine(
3952 arg_custom_mounts
, arg_n_custom_mounts
,
3956 arg_container_service_name
);
3960 } else if (!arg_keep_unit
) {
3966 arg_custom_mounts
, arg_n_custom_mounts
,
3972 } else if (arg_slice
|| arg_property
)
3973 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
3975 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
3979 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
3983 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
3987 /* Notify the child that the parent is ready with all
3988 * its setup (including cgroup-ification), and that
3989 * the child can now hand over control to the code to
3990 * run inside the container. */
3991 (void) barrier_place(&barrier
); /* #4 */
3993 /* Block SIGCHLD here, before notifying child.
3994 * process_pty() will handle it with the other signals. */
3995 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
3997 /* Reset signal to default */
3998 r
= default_signals(SIGCHLD
, -1);
4000 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4002 r
= sd_event_new(&event
);
4004 return log_error_errno(r
, "Failed to get default event source: %m");
4006 (void) sd_event_set_watchdog(event
, true);
4009 r
= sd_bus_attach_event(bus
, event
, 0);
4011 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4014 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4018 /* Let the child know that we are ready and wait that the child is completely ready now. */
4019 if (!barrier_place_and_sync(&barrier
)) { /* #5 */
4020 log_error("Child died too early.");
4024 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4025 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4026 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4029 "STATUS=Container running.\n"
4030 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4031 if (!arg_notify_ready
)
4032 (void) sd_notify(false, "READY=1\n");
4034 if (arg_kill_signal
> 0) {
4035 /* Try to kill the init system on SIGINT or SIGTERM */
4036 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4037 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4039 /* Immediately exit */
4040 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4041 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4044 /* Exit when the child exits */
4045 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4047 if (arg_expose_ports
) {
4048 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4052 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4055 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4057 r
= pty_forward_new(event
, master
,
4058 PTY_FORWARD_IGNORE_VHANGUP
| (interactive
? 0 : PTY_FORWARD_READ_ONLY
),
4061 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4063 r
= sd_event_loop(event
);
4065 return log_error_errno(r
, "Failed to run event loop: %m");
4067 pty_forward_get_last_char(forward
, &last_char
);
4069 forward
= pty_forward_free(forward
);
4071 if (!arg_quiet
&& last_char
!= '\n')
4074 /* Kill if it is not dead yet anyway */
4077 terminate_machine(bus
, arg_machine
);
4078 else if (!arg_keep_unit
)
4079 terminate_scope(bus
, arg_machine
);
4082 /* Normally redundant, but better safe than sorry */
4083 (void) kill(*pid
, SIGKILL
);
4085 r
= wait_for_container(*pid
, &container_status
);
4089 /* We failed to wait for the container, or the container exited abnormally. */
4091 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4092 /* r > 0 → The container exited with a non-zero status.
4093 * As a special case, we need to replace 133 with a different value,
4094 * because 133 is special-cased in the service file to reboot the container.
4095 * otherwise → The container exited with zero status and a reboot was not requested.
4097 if (r
== EXIT_FORCE_RESTART
)
4098 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4100 return 0; /* finito */
4103 /* CONTAINER_REBOOTED, loop again */
4105 if (arg_keep_unit
) {
4106 /* Special handling if we are running as a service: instead of simply
4107 * restarting the machine we want to restart the entire service, so let's
4108 * inform systemd about this with the special exit code 133. The service
4109 * file uses RestartForceExitStatus=133 so that this results in a full
4110 * nspawn restart. This is necessary since we might have cgroup parameters
4111 * set we want to have flushed out. */
4112 *ret
= EXIT_FORCE_RESTART
;
4113 return 0; /* finito */
4116 expose_port_flush(arg_expose_ports
, exposed
);
4118 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4119 *veth_created
= false;
4120 return 1; /* loop again */
4123 static int initialize_rlimits(void) {
4124 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4125 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4126 * container execution environments. */
4128 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4129 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4130 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4131 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4132 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4133 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4134 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4135 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4136 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4137 [RLIMIT_NICE
] = { 0, 0 },
4138 [RLIMIT_NOFILE
] = { 1024, 4096 },
4139 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4140 [RLIMIT_RTPRIO
] = { 0, 0 },
4141 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4142 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4144 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4145 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4146 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4147 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4148 * that PID 1 changes a number of other resource limits during early initialization which is why we
4149 * don't read the other limits from PID 1 but prefer the static table above. */
4154 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4155 /* Let's only fill in what the user hasn't explicitly configured anyway */
4156 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4157 const struct rlimit
*v
;
4158 struct rlimit buffer
;
4160 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4161 /* For these two let's read the limits off PID 1. See above for an explanation. */
4163 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4164 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4168 v
= kernel_defaults
+ rl
;
4170 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4171 if (!arg_rlimit
[rl
])
4175 if (DEBUG_LOGGING
) {
4176 _cleanup_free_
char *k
= NULL
;
4178 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4179 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4186 int main(int argc
, char *argv
[]) {
4187 _cleanup_free_
char *console
= NULL
;
4188 _cleanup_close_
int master
= -1;
4189 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4190 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4191 char veth_name
[IFNAMSIZ
] = "";
4192 bool secondary
= false, remove_directory
= false, remove_image
= false;
4194 union in_addr_union exposed
= {};
4195 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4196 bool interactive
, veth_created
= false, remove_tmprootdir
= false;
4197 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4198 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4199 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4200 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4202 log_parse_environment();
4205 /* Make sure rename_process() in the stub init process can work */
4209 r
= parse_argv(argc
, argv
);
4217 r
= initialize_rlimits();
4221 r
= determine_names();
4225 r
= load_settings();
4229 r
= verify_arguments();
4233 r
= detect_unified_cgroup_hierarchy_from_environment();
4237 n_fd_passed
= sd_listen_fds(false);
4238 if (n_fd_passed
> 0) {
4239 r
= fdset_new_listen_fds(&fds
, false);
4241 log_error_errno(r
, "Failed to collect file descriptors: %m");
4246 /* The "default" umask. This is appropriate for most file and directory
4247 * operations performed by nspawn, and is the umask that will be used for
4248 * the child. Functions like copy_devnodes() change the umask temporarily. */
4251 if (arg_directory
) {
4254 if (path_equal(arg_directory
, "/") && !arg_ephemeral
) {
4255 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4260 if (arg_ephemeral
) {
4261 _cleanup_free_
char *np
= NULL
;
4263 r
= chase_symlinks_and_update(&arg_directory
, 0);
4267 /* If the specified path is a mount point we
4268 * generate the new snapshot immediately
4269 * inside it under a random name. However if
4270 * the specified is not a mount point we
4271 * create the new snapshot in the parent
4272 * directory, just next to it. */
4273 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4275 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4279 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4281 r
= tempfn_random(arg_directory
, "machine.", &np
);
4283 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4287 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4289 log_error_errno(r
, "Failed to lock %s: %m", np
);
4293 r
= btrfs_subvol_snapshot(arg_directory
, np
,
4294 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4295 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4296 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4297 BTRFS_SNAPSHOT_RECURSIVE
|
4298 BTRFS_SNAPSHOT_QUOTA
);
4300 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
4304 free_and_replace(arg_directory
, np
);
4306 remove_directory
= true;
4309 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
4313 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4315 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
4319 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
4324 r
= chase_symlinks_and_update(&arg_template
, 0);
4328 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
4329 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4330 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4331 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4332 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
4333 BTRFS_SNAPSHOT_RECURSIVE
|
4334 BTRFS_SNAPSHOT_QUOTA
);
4336 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4337 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
4339 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
4342 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4343 "Populated %s from template %s.", arg_directory
, arg_template
);
4347 if (arg_start_mode
== START_BOOT
) {
4350 if (arg_pivot_root_new
)
4351 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4355 if (path_is_os_tree(p
) <= 0) {
4356 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
4363 if (arg_pivot_root_new
)
4364 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4368 q
= strjoina(p
, "/usr/");
4370 if (laccess(q
, F_OK
) < 0) {
4371 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
4379 assert(!arg_template
);
4381 r
= chase_symlinks_and_update(&arg_image
, 0);
4385 if (arg_ephemeral
) {
4386 _cleanup_free_
char *np
= NULL
;
4388 r
= tempfn_random(arg_image
, "machine.", &np
);
4390 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
4394 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4396 r
= log_error_errno(r
, "Failed to create image lock: %m");
4400 r
= copy_file(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600, FS_NOCOW_FL
, COPY_REFLINK
);
4402 r
= log_error_errno(r
, "Failed to copy image file: %m");
4406 free_and_replace(arg_image
, np
);
4408 remove_image
= true;
4410 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4412 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
4416 r
= log_error_errno(r
, "Failed to create image lock: %m");
4420 if (!arg_root_hash
) {
4421 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
4423 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
4429 if (!mkdtemp(tmprootdir
)) {
4430 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
4434 remove_tmprootdir
= true;
4436 arg_directory
= strdup(tmprootdir
);
4437 if (!arg_directory
) {
4442 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, &loop
);
4444 log_error_errno(r
, "Failed to set up loopback block device: %m");
4448 r
= dissect_image_and_warn(
4451 arg_root_hash
, arg_root_hash_size
,
4452 DISSECT_IMAGE_REQUIRE_ROOT
,
4455 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4456 log_notice("Note that the disk image needs to\n"
4457 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4458 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4459 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4460 " d) or contain a file system without a partition table\n"
4461 "in order to be bootable with systemd-nspawn.");
4467 if (!arg_root_hash
&& dissected_image
->can_verity
)
4468 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
4470 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
4474 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4475 if (remove_image
&& unlink(arg_image
) >= 0)
4476 remove_image
= false;
4479 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
4484 isatty(STDIN_FILENO
) > 0 &&
4485 isatty(STDOUT_FILENO
) > 0;
4487 master
= posix_openpt(O_RDWR
|O_NOCTTY
|O_CLOEXEC
|O_NONBLOCK
);
4489 r
= log_error_errno(errno
, "Failed to acquire pseudo tty: %m");
4493 r
= ptsname_malloc(master
, &console
);
4495 r
= log_error_errno(r
, "Failed to determine tty name: %m");
4499 if (arg_selinux_apifs_context
) {
4500 r
= mac_selinux_apply(console
, arg_selinux_apifs_context
);
4505 if (unlockpt(master
) < 0) {
4506 r
= log_error_errno(errno
, "Failed to unlock tty: %m");
4511 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4512 arg_machine
, arg_image
?: arg_directory
);
4514 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
4516 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
4517 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
4525 interactive
, secondary
,
4527 veth_name
, &veth_created
,
4536 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
4537 "STOPPING=1\nSTATUS=Terminating...");
4540 (void) kill(pid
, SIGKILL
);
4542 /* Try to flush whatever is still queued in the pty */
4544 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
4545 master
= safe_close(master
);
4549 (void) wait_for_terminate(pid
, NULL
);
4553 if (remove_directory
&& arg_directory
) {
4556 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
4558 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
4561 if (remove_image
&& arg_image
) {
4562 if (unlink(arg_image
) < 0)
4563 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
4566 if (remove_tmprootdir
) {
4567 if (rmdir(tmprootdir
) < 0)
4568 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
4574 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
4575 (void) rm_rf(p
, REMOVE_ROOT
);
4578 expose_port_flush(arg_expose_ports
, &exposed
);
4581 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4582 (void) remove_bridge(arg_network_zone
);
4584 free(arg_directory
);
4590 free(arg_pivot_root_new
);
4591 free(arg_pivot_root_old
);
4593 strv_free(arg_setenv
);
4594 free(arg_network_bridge
);
4595 strv_free(arg_network_interfaces
);
4596 strv_free(arg_network_macvlan
);
4597 strv_free(arg_network_ipvlan
);
4598 strv_free(arg_network_veth_extra
);
4599 strv_free(arg_parameters
);
4600 free(arg_network_zone
);
4601 free(arg_network_namespace_path
);
4602 strv_free(arg_property
);
4603 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4604 expose_port_free_all(arg_expose_ports
);
4605 free(arg_root_hash
);
4606 rlimit_free_all(arg_rlimit
);
4607 strv_free(arg_syscall_whitelist
);
4608 strv_free(arg_syscall_blacklist
);
4609 arg_cpuset
= cpu_set_mfree(arg_cpuset
);
4611 return r
< 0 ? EXIT_FAILURE
: ret
;