1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <linux/loop.h>
14 #include <selinux/selinux.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"
60 #include "main-func.h"
63 #include "mount-util.h"
64 #include "mountpoint-util.h"
65 #include "namespace-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-def.h"
69 #include "nspawn-expose-ports.h"
70 #include "nspawn-mount.h"
71 #include "nspawn-network.h"
72 #include "nspawn-oci.h"
73 #include "nspawn-patch-uid.h"
74 #include "nspawn-register.h"
75 #include "nspawn-seccomp.h"
76 #include "nspawn-settings.h"
77 #include "nspawn-setuid.h"
78 #include "nspawn-stub-pid1.h"
79 #include "nulstr-util.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "pretty-print.h"
85 #include "process-util.h"
87 #include "random-util.h"
88 #include "raw-clone.h"
89 #include "rlimit-util.h"
92 #include "seccomp-util.h"
94 #include "selinux-util.h"
95 #include "signal-util.h"
96 #include "socket-util.h"
97 #include "stat-util.h"
98 #include "stdio-util.h"
99 #include "string-table.h"
100 #include "string-util.h"
102 #include "sysctl-util.h"
103 #include "terminal-util.h"
104 #include "tmpfile-util.h"
105 #include "umask-util.h"
106 #include "user-util.h"
110 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
112 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
115 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
116 * nspawn_notify_socket_path is relative to the container
117 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
120 #define EXIT_FORCE_RESTART 133
122 typedef enum ContainerStatus
{
123 CONTAINER_TERMINATED
,
127 static char *arg_directory
= NULL
;
128 static char *arg_template
= NULL
;
129 static char *arg_chdir
= NULL
;
130 static char *arg_pivot_root_new
= NULL
;
131 static char *arg_pivot_root_old
= NULL
;
132 static char *arg_user
= NULL
;
133 static uid_t arg_uid
= UID_INVALID
;
134 static gid_t arg_gid
= GID_INVALID
;
135 static gid_t
* arg_supplementary_gids
= NULL
;
136 static size_t arg_n_supplementary_gids
= 0;
137 static sd_id128_t arg_uuid
= {};
138 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
139 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
140 static const char *arg_selinux_context
= NULL
;
141 static const char *arg_selinux_apifs_context
= NULL
;
142 static char *arg_slice
= NULL
;
143 static bool arg_private_network
= false;
144 static bool arg_read_only
= false;
145 static StartMode arg_start_mode
= START_PID1
;
146 static bool arg_ephemeral
= false;
147 static LinkJournal arg_link_journal
= LINK_AUTO
;
148 static bool arg_link_journal_try
= false;
149 static uint64_t arg_caps_retain
=
150 (1ULL << CAP_AUDIT_CONTROL
) |
151 (1ULL << CAP_AUDIT_WRITE
) |
152 (1ULL << CAP_CHOWN
) |
153 (1ULL << CAP_DAC_OVERRIDE
) |
154 (1ULL << CAP_DAC_READ_SEARCH
) |
155 (1ULL << CAP_FOWNER
) |
156 (1ULL << CAP_FSETID
) |
157 (1ULL << CAP_IPC_OWNER
) |
159 (1ULL << CAP_LEASE
) |
160 (1ULL << CAP_LINUX_IMMUTABLE
) |
161 (1ULL << CAP_MKNOD
) |
162 (1ULL << CAP_NET_BIND_SERVICE
) |
163 (1ULL << CAP_NET_BROADCAST
) |
164 (1ULL << CAP_NET_RAW
) |
165 (1ULL << CAP_SETFCAP
) |
166 (1ULL << CAP_SETGID
) |
167 (1ULL << CAP_SETPCAP
) |
168 (1ULL << CAP_SETUID
) |
169 (1ULL << CAP_SYS_ADMIN
) |
170 (1ULL << CAP_SYS_BOOT
) |
171 (1ULL << CAP_SYS_CHROOT
) |
172 (1ULL << CAP_SYS_NICE
) |
173 (1ULL << CAP_SYS_PTRACE
) |
174 (1ULL << CAP_SYS_RESOURCE
) |
175 (1ULL << CAP_SYS_TTY_CONFIG
);
176 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
177 static CustomMount
*arg_custom_mounts
= NULL
;
178 static size_t arg_n_custom_mounts
= 0;
179 static char **arg_setenv
= NULL
;
180 static bool arg_quiet
= false;
181 static bool arg_register
= true;
182 static bool arg_keep_unit
= false;
183 static char **arg_network_interfaces
= NULL
;
184 static char **arg_network_macvlan
= NULL
;
185 static char **arg_network_ipvlan
= NULL
;
186 static bool arg_network_veth
= false;
187 static char **arg_network_veth_extra
= NULL
;
188 static char *arg_network_bridge
= NULL
;
189 static char *arg_network_zone
= NULL
;
190 static char *arg_network_namespace_path
= NULL
;
191 static PagerFlags arg_pager_flags
= 0;
192 static unsigned long arg_personality
= PERSONALITY_INVALID
;
193 static char *arg_image
= NULL
;
194 static char *arg_oci_bundle
= NULL
;
195 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
196 static ExposePort
*arg_expose_ports
= NULL
;
197 static char **arg_property
= NULL
;
198 static sd_bus_message
*arg_property_message
= NULL
;
199 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
200 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
201 static bool arg_userns_chown
= false;
202 static int arg_kill_signal
= 0;
203 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
204 static SettingsMask arg_settings_mask
= 0;
205 static int arg_settings_trusted
= -1;
206 static char **arg_parameters
= NULL
;
207 static const char *arg_container_service_name
= "systemd-nspawn";
208 static bool arg_notify_ready
= false;
209 static bool arg_use_cgns
= true;
210 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
211 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
212 static void *arg_root_hash
= NULL
;
213 static size_t arg_root_hash_size
= 0;
214 static char **arg_syscall_whitelist
= NULL
;
215 static char **arg_syscall_blacklist
= NULL
;
217 static scmp_filter_ctx arg_seccomp
= NULL
;
219 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
220 static bool arg_no_new_privileges
= false;
221 static int arg_oom_score_adjust
= 0;
222 static bool arg_oom_score_adjust_set
= false;
223 static cpu_set_t
*arg_cpuset
= NULL
;
224 static unsigned arg_cpuset_ncpus
= 0;
225 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
226 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
227 static unsigned arg_console_width
= (unsigned) -1, arg_console_height
= (unsigned) -1;
228 static DeviceNode
* arg_extra_nodes
= NULL
;
229 static size_t arg_n_extra_nodes
= 0;
230 static char **arg_sysctl
= NULL
;
231 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
233 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
255 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_root_hash
, freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist
, strv_freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist
, strv_freep
);
260 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
262 STATIC_DESTRUCTOR_REGISTER(arg_cpuset
, CPU_FREEp
);
263 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
265 static int help(void) {
266 _cleanup_free_
char *link
= NULL
;
269 (void) pager_open(arg_pager_flags
);
271 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
275 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
276 "Spawn a command or OS in a light-weight container.\n\n"
277 " -h --help Show this help\n"
278 " --version Print version string\n"
279 " -q --quiet Do not show status information\n"
280 " --no-pager Do not pipe output into a pager\n"
281 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
283 " -D --directory=PATH Root directory for the container\n"
284 " --template=PATH Initialize root directory from template directory,\n"
286 " -x --ephemeral Run container with snapshot of root directory, and\n"
287 " remove it after exit\n"
288 " -i --image=PATH File system device or disk image for the container\n"
289 " --oci-bundle=PATH OCI bundle directory\n"
290 " --read-only Mount the root directory read-only\n"
291 " --volatile[=MODE] Run the system in volatile mode\n"
292 " --root-hash=HASH Specify verity root hash\n"
293 " --pivot-root=PATH[:PATH]\n"
294 " Pivot root to given directory in the container\n\n"
295 "%3$sExecution:%4$s\n"
296 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
297 " -b --boot Boot up full system (i.e. invoke init)\n"
298 " --chdir=PATH Set working directory in the container\n"
299 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
300 " -u --user=USER Run the command under specified user or UID\n"
301 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
302 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
303 "%3$sSystem Identity:%4$s\n"
304 " -M --machine=NAME Set the machine name for the container\n"
305 " --hostname=NAME Override the hostname for the container\n"
306 " --uuid=UUID Set a specific machine UUID for the container\n\n"
307 "%3$sProperties:%4$s\n"
308 " -S --slice=SLICE Place the container in the specified slice\n"
309 " --property=NAME=VALUE Set scope unit property\n"
310 " --register=BOOLEAN Register container as machine\n"
311 " --keep-unit Do not register a scope for the machine, reuse\n"
312 " the service unit nspawn is running in\n\n"
313 "%3$sUser Namespacing:%4$s\n"
314 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
315 " --private-users[=UIDBASE[:NUIDS]]\n"
316 " Similar, but with user configured UID/GID range\n"
317 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
318 "%3$sNetworking:%4$s\n"
319 " --private-network Disable network in container\n"
320 " --network-interface=INTERFACE\n"
321 " Assign an existing network interface to the\n"
323 " --network-macvlan=INTERFACE\n"
324 " Create a macvlan network interface based on an\n"
325 " existing network interface to the container\n"
326 " --network-ipvlan=INTERFACE\n"
327 " Create a ipvlan network interface based on an\n"
328 " existing network interface to the container\n"
329 " -n --network-veth Add a virtual Ethernet connection between host\n"
331 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
332 " Add an additional virtual Ethernet link between\n"
333 " host and container\n"
334 " --network-bridge=INTERFACE\n"
335 " Add a virtual Ethernet connection to the container\n"
336 " and attach it to an existing bridge on the host\n"
337 " --network-zone=NAME Similar, but attach the new interface to an\n"
338 " an automatically managed bridge interface\n"
339 " --network-namespace-path=PATH\n"
340 " Set network namespace to the one represented by\n"
341 " the specified kernel namespace file node\n"
342 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
343 " Expose a container IP port on the host\n\n"
344 "%3$sSecurity:%4$s\n"
345 " --capability=CAP In addition to the default, retain specified\n"
347 " --drop-capability=CAP Drop the specified capability from the default set\n"
348 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
349 " --system-call-filter=LIST|~LIST\n"
350 " Permit/prohibit specific system calls\n"
351 " -Z --selinux-context=SECLABEL\n"
352 " Set the SELinux security context to be used by\n"
353 " processes in the container\n"
354 " -L --selinux-apifs-context=SECLABEL\n"
355 " Set the SELinux security context to be used by\n"
356 " API/tmpfs file systems in the container\n\n"
357 "%3$sResources:%4$s\n"
358 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
359 " --oom-score-adjust=VALUE\n"
360 " Adjust the OOM score value for the payload\n"
361 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
362 " --personality=ARCH Pick personality for this container\n\n"
363 "%3$sIntegration:%4$s\n"
364 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
365 " --timezone=MODE Select mode of /etc/localtime initialization\n"
366 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
367 " host, try-guest, try-host\n"
368 " -j Equivalent to --link-journal=try-guest\n\n"
370 " --bind=PATH[:PATH[:OPTIONS]]\n"
371 " Bind mount a file or directory from the host into\n"
373 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
374 " Similar, but creates a read-only bind mount\n"
375 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
377 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
378 " --overlay=PATH[:PATH...]:PATH\n"
379 " Create an overlay mount from the host to \n"
381 " --overlay-ro=PATH[:PATH...]:PATH\n"
382 " Similar, but creates a read-only overlay mount\n\n"
383 "%3$sInput/Output:%4$s\n"
384 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
385 " set up for the container.\n"
386 " -P --pipe Equivalent to --console=pipe\n"
387 "\nSee the %2$s for details.\n"
388 , program_invocation_short_name
390 , ansi_underline(), ansi_normal());
395 static int custom_mount_check_all(void) {
398 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
399 CustomMount
*m
= &arg_custom_mounts
[i
];
401 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
402 if (arg_userns_chown
)
403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
404 "--private-users-chown may not be combined with custom root mounts.");
405 else if (arg_uid_shift
== UID_INVALID
)
406 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
407 "--private-users with automatic UID shift may not be combined with custom root mounts.");
414 static int detect_unified_cgroup_hierarchy_from_environment(void) {
418 /* Allow the user to control whether the unified hierarchy is used */
419 e
= getenv("UNIFIED_CGROUP_HIERARCHY");
421 r
= parse_boolean(e
);
423 return log_error_errno(r
, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
425 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
427 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
433 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
436 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
437 * image actually supports. */
438 r
= cg_all_unified();
440 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
442 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
443 * routine only detects 231, so we'll have a false negative here for 230. */
444 r
= systemd_installation_has_version(directory
, 230);
446 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
448 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
450 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
451 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
452 /* Mixed cgroup hierarchy support was added in 233 */
453 r
= systemd_installation_has_version(directory
, 233);
455 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
457 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
459 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
461 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
463 log_debug("Using %s hierarchy for container.",
464 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
465 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
470 static void parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
473 r
= getenv_bool(name
);
477 log_warning_errno(r
, "Failed to parse %s from environment, defaulting to false.", name
);
479 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
480 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
483 static void parse_mount_settings_env(void) {
487 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
489 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
490 else if (r
!= -ENXIO
)
491 log_warning_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP, ignoring: %m");
493 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
497 if (streq(e
, "network")) {
498 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
502 r
= parse_boolean(e
);
504 log_warning_errno(r
, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
508 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
509 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
512 static void parse_environment(void) {
516 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
517 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
518 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
519 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
521 parse_mount_settings_env();
523 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
524 * even if it is supported. If not supported, it has no effect. */
525 if (!cg_ns_supported())
526 arg_use_cgns
= false;
528 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
531 log_warning_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS, ignoring: %m");
535 arg_use_cgns
= r
> 0;
536 arg_settings_mask
|= SETTING_USE_CGNS
;
540 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
542 arg_container_service_name
= e
;
544 detect_unified_cgroup_hierarchy_from_environment();
547 static int parse_argv(int argc
, char *argv
[]) {
565 ARG_NETWORK_INTERFACE
,
570 ARG_NETWORK_VETH_EXTRA
,
571 ARG_NETWORK_NAMESPACE_PATH
,
581 ARG_PRIVATE_USERS_CHOWN
,
584 ARG_SYSTEM_CALL_FILTER
,
587 ARG_NO_NEW_PRIVILEGES
,
588 ARG_OOM_SCORE_ADJUST
,
598 static const struct option options
[] = {
599 { "help", no_argument
, NULL
, 'h' },
600 { "version", no_argument
, NULL
, ARG_VERSION
},
601 { "directory", required_argument
, NULL
, 'D' },
602 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
603 { "ephemeral", no_argument
, NULL
, 'x' },
604 { "user", required_argument
, NULL
, 'u' },
605 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
606 { "as-pid2", no_argument
, NULL
, 'a' },
607 { "boot", no_argument
, NULL
, 'b' },
608 { "uuid", required_argument
, NULL
, ARG_UUID
},
609 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
610 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
611 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
612 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
613 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
614 { "bind", required_argument
, NULL
, ARG_BIND
},
615 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
616 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
617 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
618 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
619 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
620 { "machine", required_argument
, NULL
, 'M' },
621 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
622 { "slice", required_argument
, NULL
, 'S' },
623 { "setenv", required_argument
, NULL
, 'E' },
624 { "selinux-context", required_argument
, NULL
, 'Z' },
625 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
626 { "quiet", no_argument
, NULL
, 'q' },
627 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
628 { "register", required_argument
, NULL
, ARG_REGISTER
},
629 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
630 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
631 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
632 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
633 { "network-veth", no_argument
, NULL
, 'n' },
634 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
635 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
636 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
637 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
638 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
639 { "image", required_argument
, NULL
, 'i' },
640 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
641 { "port", required_argument
, NULL
, 'p' },
642 { "property", required_argument
, NULL
, ARG_PROPERTY
},
643 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
644 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
645 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
646 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
647 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
648 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
649 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
650 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
651 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
652 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
653 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
654 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
655 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
656 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
657 { "console", required_argument
, NULL
, ARG_CONSOLE
},
658 { "pipe", no_argument
, NULL
, ARG_PIPE
},
659 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
660 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
666 uint64_t plus
= 0, minus
= 0;
667 bool mask_all_settings
= false, mask_no_settings
= false;
672 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
682 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
686 arg_settings_mask
|= SETTING_DIRECTORY
;
690 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
694 arg_settings_mask
|= SETTING_DIRECTORY
;
698 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
702 arg_settings_mask
|= SETTING_DIRECTORY
;
706 r
= parse_path_argument_and_warn(optarg
, false, &arg_oci_bundle
);
713 arg_ephemeral
= true;
714 arg_settings_mask
|= SETTING_EPHEMERAL
;
718 r
= free_and_strdup(&arg_user
, optarg
);
722 arg_settings_mask
|= SETTING_USER
;
725 case ARG_NETWORK_ZONE
: {
728 j
= strappend("vz-", optarg
);
732 if (!ifname_valid(j
)) {
733 log_error("Network zone name not valid: %s", j
);
738 free_and_replace(arg_network_zone
, j
);
740 arg_network_veth
= true;
741 arg_private_network
= true;
742 arg_settings_mask
|= SETTING_NETWORK
;
746 case ARG_NETWORK_BRIDGE
:
748 if (!ifname_valid(optarg
))
749 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
750 "Bridge interface name not valid: %s", optarg
);
752 r
= free_and_strdup(&arg_network_bridge
, optarg
);
758 arg_network_veth
= true;
759 arg_private_network
= true;
760 arg_settings_mask
|= SETTING_NETWORK
;
763 case ARG_NETWORK_VETH_EXTRA
:
764 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
766 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
768 arg_private_network
= true;
769 arg_settings_mask
|= SETTING_NETWORK
;
772 case ARG_NETWORK_INTERFACE
:
773 if (!ifname_valid(optarg
))
774 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
775 "Network interface name not valid: %s", optarg
);
777 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
780 arg_private_network
= true;
781 arg_settings_mask
|= SETTING_NETWORK
;
784 case ARG_NETWORK_MACVLAN
:
786 if (!ifname_valid(optarg
))
787 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
788 "MACVLAN network interface name not valid: %s", optarg
);
790 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
793 arg_private_network
= true;
794 arg_settings_mask
|= SETTING_NETWORK
;
797 case ARG_NETWORK_IPVLAN
:
799 if (!ifname_valid(optarg
))
800 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
801 "IPVLAN network interface name not valid: %s", optarg
);
803 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
807 case ARG_PRIVATE_NETWORK
:
808 arg_private_network
= true;
809 arg_settings_mask
|= SETTING_NETWORK
;
812 case ARG_NETWORK_NAMESPACE_PATH
:
813 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
817 arg_settings_mask
|= SETTING_NETWORK
;
821 if (arg_start_mode
== START_PID2
)
822 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
823 "--boot and --as-pid2 may not be combined.");
825 arg_start_mode
= START_BOOT
;
826 arg_settings_mask
|= SETTING_START_MODE
;
830 if (arg_start_mode
== START_BOOT
)
831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
832 "--boot and --as-pid2 may not be combined.");
834 arg_start_mode
= START_PID2
;
835 arg_settings_mask
|= SETTING_START_MODE
;
839 r
= sd_id128_from_string(optarg
, &arg_uuid
);
841 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
843 if (sd_id128_is_null(arg_uuid
))
844 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
845 "Machine UUID may not be all zeroes.");
847 arg_settings_mask
|= SETTING_MACHINE_ID
;
851 r
= free_and_strdup(&arg_slice
, optarg
);
855 arg_settings_mask
|= SETTING_SLICE
;
860 arg_machine
= mfree(arg_machine
);
862 if (!machine_name_is_valid(optarg
))
863 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
864 "Invalid machine name: %s", optarg
);
866 r
= free_and_strdup(&arg_machine
, optarg
);
874 arg_hostname
= mfree(arg_hostname
);
876 if (!hostname_is_valid(optarg
, false))
877 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
878 "Invalid hostname: %s", optarg
);
880 r
= free_and_strdup(&arg_hostname
, optarg
);
885 arg_settings_mask
|= SETTING_HOSTNAME
;
889 arg_selinux_context
= optarg
;
893 arg_selinux_apifs_context
= optarg
;
897 arg_read_only
= true;
898 arg_settings_mask
|= SETTING_READ_ONLY
;
902 case ARG_DROP_CAPABILITY
: {
905 _cleanup_free_
char *t
= NULL
;
907 r
= extract_first_word(&p
, &t
, ",", 0);
909 return log_error_errno(r
, "Failed to parse capability %s.", t
);
913 if (streq(t
, "all")) {
914 if (c
== ARG_CAPABILITY
)
915 plus
= (uint64_t) -1;
917 minus
= (uint64_t) -1;
919 r
= capability_from_name(t
);
921 return log_error_errno(r
, "Failed to parse capability %s.", t
);
923 if (c
== ARG_CAPABILITY
)
930 arg_settings_mask
|= SETTING_CAPABILITY
;
934 case ARG_NO_NEW_PRIVILEGES
:
935 r
= parse_boolean(optarg
);
937 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
939 arg_no_new_privileges
= r
;
940 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
944 arg_link_journal
= LINK_GUEST
;
945 arg_link_journal_try
= true;
946 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
949 case ARG_LINK_JOURNAL
:
950 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
952 log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
956 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
961 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
963 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
965 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
969 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
971 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
973 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
978 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
979 if (r
== -EADDRNOTAVAIL
)
980 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
982 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
984 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
987 case ARG_INACCESSIBLE
:
988 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
990 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
992 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
998 if (!env_assignment_is_valid(optarg
))
999 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1000 "Environment variable assignment '%s' is not valid.", optarg
);
1002 n
= strv_env_set(arg_setenv
, optarg
);
1006 strv_free_and_replace(arg_setenv
, n
);
1007 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1015 case ARG_SHARE_SYSTEM
:
1016 /* We don't officially support this anymore, except for compat reasons. People should use the
1017 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1018 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1019 arg_clone_ns_flags
= 0;
1023 r
= parse_boolean(optarg
);
1025 log_error("Failed to parse --register= argument: %s", optarg
);
1033 arg_keep_unit
= true;
1036 case ARG_PERSONALITY
:
1038 arg_personality
= personality_from_string(optarg
);
1039 if (arg_personality
== PERSONALITY_INVALID
)
1040 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1041 "Unknown or unsupported personality '%s'.", optarg
);
1043 arg_settings_mask
|= SETTING_PERSONALITY
;
1049 arg_volatile_mode
= VOLATILE_YES
;
1050 else if (streq(optarg
, "help")) {
1051 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1056 m
= volatile_mode_from_string(optarg
);
1058 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1059 "Failed to parse --volatile= argument: %s", optarg
);
1061 arg_volatile_mode
= m
;
1064 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1068 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1070 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1072 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1074 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1078 if (strv_extend(&arg_property
, optarg
) < 0)
1083 case ARG_PRIVATE_USERS
: {
1088 else if (!in_charset(optarg
, DIGITS
))
1089 /* do *not* parse numbers as booleans */
1090 boolean
= parse_boolean(optarg
);
1092 if (boolean
== false) {
1093 /* no: User namespacing off */
1094 arg_userns_mode
= USER_NAMESPACE_NO
;
1095 arg_uid_shift
= UID_INVALID
;
1096 arg_uid_range
= UINT32_C(0x10000);
1097 } else if (boolean
== true) {
1098 /* yes: User namespacing on, UID range is read from root dir */
1099 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1100 arg_uid_shift
= UID_INVALID
;
1101 arg_uid_range
= UINT32_C(0x10000);
1102 } else if (streq(optarg
, "pick")) {
1103 /* pick: User namespacing on, UID range is picked randomly */
1104 arg_userns_mode
= USER_NAMESPACE_PICK
;
1105 arg_uid_shift
= UID_INVALID
;
1106 arg_uid_range
= UINT32_C(0x10000);
1108 _cleanup_free_
char *buffer
= NULL
;
1109 const char *range
, *shift
;
1111 /* anything else: User namespacing on, UID range is explicitly configured */
1113 range
= strchr(optarg
, ':');
1115 buffer
= strndup(optarg
, range
- optarg
);
1121 r
= safe_atou32(range
, &arg_uid_range
);
1123 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1127 r
= parse_uid(shift
, &arg_uid_shift
);
1129 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1131 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1134 if (arg_uid_range
<= 0)
1135 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1136 "UID range cannot be 0.");
1138 arg_settings_mask
|= SETTING_USERNS
;
1143 if (userns_supported()) {
1144 arg_userns_mode
= USER_NAMESPACE_PICK
;
1145 arg_uid_shift
= UID_INVALID
;
1146 arg_uid_range
= UINT32_C(0x10000);
1148 arg_settings_mask
|= SETTING_USERNS
;
1153 case ARG_PRIVATE_USERS_CHOWN
:
1154 arg_userns_chown
= true;
1156 arg_settings_mask
|= SETTING_USERNS
;
1159 case ARG_KILL_SIGNAL
:
1160 if (streq(optarg
, "help")) {
1161 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1165 arg_kill_signal
= signal_from_string(optarg
);
1166 if (arg_kill_signal
< 0)
1167 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1168 "Cannot parse signal: %s", optarg
);
1170 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1175 /* no → do not read files
1176 * yes → read files, do not override cmdline, trust only subset
1177 * override → read files, override cmdline, trust only subset
1178 * trusted → read files, do not override cmdline, trust all
1181 r
= parse_boolean(optarg
);
1183 if (streq(optarg
, "trusted")) {
1184 mask_all_settings
= false;
1185 mask_no_settings
= false;
1186 arg_settings_trusted
= true;
1188 } else if (streq(optarg
, "override")) {
1189 mask_all_settings
= false;
1190 mask_no_settings
= true;
1191 arg_settings_trusted
= -1;
1193 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1196 mask_all_settings
= false;
1197 mask_no_settings
= false;
1198 arg_settings_trusted
= -1;
1201 mask_all_settings
= true;
1202 mask_no_settings
= false;
1203 arg_settings_trusted
= false;
1209 if (!path_is_absolute(optarg
))
1210 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1211 "Working directory %s is not an absolute path.", optarg
);
1213 r
= free_and_strdup(&arg_chdir
, optarg
);
1217 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1220 case ARG_PIVOT_ROOT
:
1221 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1223 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1225 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1228 case ARG_NOTIFY_READY
:
1229 r
= parse_boolean(optarg
);
1231 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1232 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1233 arg_notify_ready
= r
;
1234 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1237 case ARG_ROOT_HASH
: {
1241 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1243 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1244 if (l
< sizeof(sd_id128_t
)) {
1245 log_error("Root hash must be at least 128bit long: %s", optarg
);
1250 free(arg_root_hash
);
1252 arg_root_hash_size
= l
;
1256 case ARG_SYSTEM_CALL_FILTER
: {
1260 negative
= optarg
[0] == '~';
1261 items
= negative
? optarg
+ 1 : optarg
;
1264 _cleanup_free_
char *word
= NULL
;
1266 r
= extract_first_word(&items
, &word
, NULL
, 0);
1272 return log_error_errno(r
, "Failed to parse system call filter: %m");
1275 r
= strv_extend(&arg_syscall_blacklist
, word
);
1277 r
= strv_extend(&arg_syscall_whitelist
, word
);
1282 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1291 if (streq(optarg
, "help")) {
1292 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1296 eq
= strchr(optarg
, '=');
1298 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1299 "--rlimit= expects an '=' assignment.");
1301 name
= strndup(optarg
, eq
- optarg
);
1305 rl
= rlimit_from_string_harder(name
);
1307 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1308 "Unknown resource limit: %s", name
);
1310 if (!arg_rlimit
[rl
]) {
1311 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1312 if (!arg_rlimit
[rl
])
1316 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1318 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1320 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1324 case ARG_OOM_SCORE_ADJUST
:
1325 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1327 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1329 arg_oom_score_adjust_set
= true;
1330 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1333 case ARG_CPU_AFFINITY
: {
1334 _cleanup_cpu_free_ cpu_set_t
*cpuset
= NULL
;
1336 r
= parse_cpu_set(optarg
, &cpuset
);
1338 return log_error_errno(r
, "Failed to parse CPU affinity mask: %s", optarg
);
1341 CPU_FREE(arg_cpuset
);
1343 arg_cpuset
= TAKE_PTR(cpuset
);
1344 arg_cpuset_ncpus
= r
;
1345 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1349 case ARG_RESOLV_CONF
:
1350 if (streq(optarg
, "help")) {
1351 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1355 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1356 if (arg_resolv_conf
< 0)
1357 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1358 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1360 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1364 if (streq(optarg
, "help")) {
1365 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1369 arg_timezone
= timezone_mode_from_string(optarg
);
1370 if (arg_timezone
< 0)
1371 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1372 "Failed to parse /etc/localtime mode: %s", optarg
);
1374 arg_settings_mask
|= SETTING_TIMEZONE
;
1378 if (streq(optarg
, "interactive"))
1379 arg_console_mode
= CONSOLE_INTERACTIVE
;
1380 else if (streq(optarg
, "read-only"))
1381 arg_console_mode
= CONSOLE_READ_ONLY
;
1382 else if (streq(optarg
, "passive"))
1383 arg_console_mode
= CONSOLE_PASSIVE
;
1384 else if (streq(optarg
, "pipe"))
1385 arg_console_mode
= CONSOLE_PIPE
;
1386 else if (streq(optarg
, "help"))
1387 puts("interactive\n"
1392 log_error("Unknown console mode: %s", optarg
);
1396 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
1401 arg_console_mode
= CONSOLE_PIPE
;
1402 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
1406 arg_pager_flags
|= PAGER_DISABLE
;
1413 assert_not_reached("Unhandled option");
1416 if (argc
> optind
) {
1417 strv_free(arg_parameters
);
1418 arg_parameters
= strv_copy(argv
+ optind
);
1419 if (!arg_parameters
)
1422 arg_settings_mask
|= SETTING_START_MODE
;
1425 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1426 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1427 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1428 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1430 arg_directory
= TAKE_PTR(arg_template
);
1432 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1434 /* Make sure to parse environment before we reset the settings mask below */
1435 parse_environment();
1437 /* Load all settings from .nspawn files */
1438 if (mask_no_settings
)
1439 arg_settings_mask
= 0;
1441 /* Don't load any settings from .nspawn files */
1442 if (mask_all_settings
)
1443 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1448 static int verify_arguments(void) {
1451 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1452 arg_mount_settings
|= MOUNT_USE_USERNS
;
1454 if (arg_private_network
)
1455 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1457 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1458 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1459 arg_register
= false;
1460 if (arg_start_mode
!= START_PID1
)
1461 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1464 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1465 arg_userns_chown
= true;
1467 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1468 arg_kill_signal
= SIGRTMIN
+3;
1470 if (arg_volatile_mode
!= VOLATILE_NO
) /* Make sure all file systems contained in the image are mounted read-only if we are in volatile mode */
1471 arg_read_only
= true;
1473 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1474 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1475 * The latter is not technically a user session, but we don't need to labour the point. */
1476 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1478 if (arg_directory
&& arg_image
)
1479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1481 if (arg_template
&& arg_image
)
1482 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1484 if (arg_template
&& !(arg_directory
|| arg_machine
))
1485 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1487 if (arg_ephemeral
&& arg_template
)
1488 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1490 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1491 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1493 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1494 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1496 if (arg_userns_chown
&& arg_read_only
)
1497 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1498 "--read-only and --private-users-chown may not be combined.");
1500 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1501 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1502 * copy-up (in case of overlay) making the entire excercise pointless. */
1503 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1504 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1506 /* If --network-namespace-path is given with any other network-related option, we need to error out,
1507 * to avoid conflicts between different network options. */
1508 if (arg_network_namespace_path
&&
1509 (arg_network_interfaces
|| arg_network_macvlan
||
1510 arg_network_ipvlan
|| arg_network_veth_extra
||
1511 arg_network_bridge
|| arg_network_zone
||
1512 arg_network_veth
|| arg_private_network
))
1513 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1515 if (arg_network_bridge
&& arg_network_zone
)
1516 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1517 "--network-bridge= and --network-zone= may not be combined.");
1519 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1520 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1522 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1523 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1525 if (arg_expose_ports
&& !arg_private_network
)
1526 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1529 if (arg_expose_ports
)
1530 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1533 r
= custom_mount_check_all();
1540 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1543 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1546 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1549 if (uid
!= UID_INVALID
) {
1550 uid
+= arg_uid_shift
;
1552 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1556 if (gid
!= GID_INVALID
) {
1557 gid
+= (gid_t
) arg_uid_shift
;
1559 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1563 if (lchown(p
, uid
, gid
) < 0)
1569 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1573 q
= prefix_roota(root
, path
);
1574 r
= mkdir_errno_wrapper(q
, mode
);
1580 return userns_lchown(q
, uid
, gid
);
1583 static const char *timezone_from_path(const char *path
) {
1584 return PATH_STARTSWITH_SET(
1586 "../usr/share/zoneinfo/",
1587 "/usr/share/zoneinfo/");
1590 static bool etc_writable(void) {
1591 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1594 static int setup_timezone(const char *dest
) {
1595 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1596 const char *where
, *check
;
1602 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1603 r
= readlink_malloc("/etc/localtime", &p
);
1604 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1605 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1606 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1607 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1609 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1610 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1614 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1617 } else if (arg_timezone
== TIMEZONE_AUTO
)
1618 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1624 if (m
== TIMEZONE_OFF
)
1627 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1629 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1633 where
= strjoina(etc
, "/localtime");
1637 case TIMEZONE_DELETE
:
1638 if (unlink(where
) < 0)
1639 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1643 case TIMEZONE_SYMLINK
: {
1644 _cleanup_free_
char *q
= NULL
;
1645 const char *z
, *what
;
1647 z
= timezone_from_path(p
);
1649 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1653 r
= readlink_malloc(where
, &q
);
1654 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1655 return 0; /* Already pointing to the right place? Then do nothing .. */
1657 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1658 r
= chase_symlinks(check
, dest
, 0, NULL
);
1660 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1662 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1663 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1664 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1668 what
= strjoina("../usr/share/zoneinfo/", z
);
1669 if (symlink(what
, where
) < 0) {
1670 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1671 errno
, "Failed to correct timezone of container, ignoring: %m");
1681 case TIMEZONE_BIND
: {
1682 _cleanup_free_
char *resolved
= NULL
;
1685 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1687 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1691 if (found
== 0) /* missing? */
1692 (void) touch(resolved
);
1694 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1696 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1702 /* If mounting failed, try to copy */
1703 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, COPY_REFLINK
|COPY_REPLACE
);
1705 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1706 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1713 assert_not_reached("unexpected mode");
1716 /* Fix permissions of the symlink or file copy we just created */
1717 r
= userns_lchown(where
, 0, 0);
1719 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1724 static int have_resolv_conf(const char *path
) {
1727 if (access(path
, F_OK
) < 0) {
1728 if (errno
== ENOENT
)
1731 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1737 static int resolved_listening(void) {
1738 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1739 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1740 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1743 /* Check if resolved is listening */
1745 r
= sd_bus_open_system(&bus
);
1747 return log_debug_errno(r
, "Failed to open system bus: %m");
1749 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1751 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1755 r
= sd_bus_get_property_string(bus
,
1756 "org.freedesktop.resolve1",
1757 "/org/freedesktop/resolve1",
1758 "org.freedesktop.resolve1.Manager",
1761 &dns_stub_listener_mode
);
1763 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1765 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1768 static int setup_resolv_conf(const char *dest
) {
1769 _cleanup_free_
char *etc
= NULL
;
1770 const char *where
, *what
;
1776 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1777 if (arg_private_network
)
1778 m
= RESOLV_CONF_OFF
;
1779 else if (have_resolv_conf(STATIC_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1780 m
= etc_writable() ? RESOLV_CONF_COPY_STATIC
: RESOLV_CONF_BIND_STATIC
;
1781 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1782 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1784 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1786 m
= arg_resolv_conf
;
1788 if (m
== RESOLV_CONF_OFF
)
1791 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1793 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1797 where
= strjoina(etc
, "/resolv.conf");
1799 if (m
== RESOLV_CONF_DELETE
) {
1800 if (unlink(where
) < 0)
1801 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1806 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_COPY_STATIC
))
1807 what
= STATIC_RESOLV_CONF
;
1809 what
= "/etc/resolv.conf";
1811 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
)) {
1812 _cleanup_free_
char *resolved
= NULL
;
1815 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1817 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1821 if (found
== 0) /* missing? */
1822 (void) touch(resolved
);
1824 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1826 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1829 /* If that didn't work, let's copy the file */
1830 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, COPY_REFLINK
);
1832 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1833 * resolved or something similar runs inside and the symlink points there.
1835 * If the disk image is read-only, there's also no point in complaining.
1837 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
,
1838 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1842 r
= userns_lchown(where
, 0, 0);
1844 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1849 static int setup_boot_id(void) {
1850 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1851 _cleanup_free_
char *path
= NULL
;
1852 sd_id128_t rnd
= SD_ID128_NULL
;
1856 /* Generate a new randomized boot ID, so that each boot-up of
1857 * the container gets a new one */
1859 r
= tempfn_random_child(NULL
, "proc-sys-kernel-random-boot-id", &path
);
1861 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1863 r
= sd_id128_randomize(&rnd
);
1865 return log_error_errno(r
, "Failed to generate random boot id: %m");
1867 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1869 return log_error_errno(r
, "Failed to write boot id: %m");
1871 from
= TAKE_PTR(path
);
1872 to
= "/proc/sys/kernel/random/boot_id";
1874 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1878 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1881 static int copy_devnodes(const char *dest
) {
1882 static const char devnodes
[] =
1891 _cleanup_umask_ mode_t u
;
1899 /* Create /dev/net, so that we can create /dev/net/tun in it */
1900 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1901 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1903 NULSTR_FOREACH(d
, devnodes
) {
1904 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1907 from
= strappend("/dev/", d
);
1911 to
= prefix_root(dest
, from
);
1915 if (stat(from
, &st
) < 0) {
1917 if (errno
!= ENOENT
)
1918 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1920 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1921 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
1922 "%s is not a char or block device, cannot copy.", from
);
1924 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
1926 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1927 /* Explicitly warn the user when /dev is already populated. */
1928 if (errno
== EEXIST
)
1929 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
);
1931 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
1933 /* Some systems abusively restrict mknod but allow bind mounts. */
1936 return log_error_errno(r
, "touch (%s) failed: %m", to
);
1937 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
1939 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
1942 r
= userns_lchown(to
, 0, 0);
1944 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
1946 dn
= strjoin("/dev/", S_ISCHR(st
.st_mode
) ? "char" : "block");
1950 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
1952 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
1954 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
1957 prefixed
= prefix_root(dest
, sl
);
1961 t
= strjoin("../", d
);
1965 if (symlink(t
, prefixed
) < 0)
1966 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
1973 static int make_extra_nodes(const char *dest
) {
1974 _cleanup_umask_ mode_t u
;
1980 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
1981 _cleanup_free_
char *path
= NULL
;
1982 DeviceNode
*n
= arg_extra_nodes
+ i
;
1984 path
= prefix_root(dest
, n
->path
);
1988 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
1989 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
1991 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
1993 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
1999 static int setup_pts(const char *dest
) {
2000 _cleanup_free_
char *options
= NULL
;
2005 if (arg_selinux_apifs_context
)
2006 (void) asprintf(&options
,
2007 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2008 arg_uid_shift
+ TTY_GID
,
2009 arg_selinux_apifs_context
);
2012 (void) asprintf(&options
,
2013 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2014 arg_uid_shift
+ TTY_GID
);
2019 /* Mount /dev/pts itself */
2020 p
= prefix_roota(dest
, "/dev/pts");
2021 r
= mkdir_errno_wrapper(p
, 0755);
2023 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2025 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2028 r
= userns_lchown(p
, 0, 0);
2030 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2032 /* Create /dev/ptmx symlink */
2033 p
= prefix_roota(dest
, "/dev/ptmx");
2034 if (symlink("pts/ptmx", p
) < 0)
2035 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2036 r
= userns_lchown(p
, 0, 0);
2038 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2040 /* And fix /dev/pts/ptmx ownership */
2041 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2042 r
= userns_lchown(p
, 0, 0);
2044 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2049 static int setup_dev_console(const char *dest
, const char *console
) {
2050 _cleanup_umask_ mode_t u
;
2061 r
= chmod_and_chown(console
, 0600, arg_uid_shift
, arg_uid_shift
);
2063 return log_error_errno(r
, "Failed to correct access mode for TTY: %m");
2065 /* We need to bind mount the right tty to /dev/console since
2066 * ptys can only exist on pts file systems. To have something
2067 * to bind mount things on we create a empty regular file. */
2069 to
= prefix_roota(dest
, "/dev/console");
2072 return log_error_errno(r
, "touch() for /dev/console failed: %m");
2074 return mount_verbose(LOG_ERR
, console
, to
, NULL
, MS_BIND
, NULL
);
2077 static int setup_keyring(void) {
2078 key_serial_t keyring
;
2080 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2081 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2082 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2083 * these system calls let's make sure we don't leak anything into the container. */
2085 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2086 if (keyring
== -1) {
2087 if (errno
== ENOSYS
)
2088 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2089 else if (IN_SET(errno
, EACCES
, EPERM
))
2090 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2092 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2098 static int setup_kmsg(int kmsg_socket
) {
2099 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2100 _cleanup_free_
char *fifo
= NULL
;
2101 _cleanup_close_
int fd
= -1;
2102 _cleanup_umask_ mode_t u
;
2106 assert(kmsg_socket
>= 0);
2110 /* We create the kmsg FIFO as as temporary file in /tmp, but immediately delete it after bind mounting it to
2111 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2112 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2113 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2115 r
= tempfn_random_child(NULL
, "proc-kmsg", &fifo
);
2117 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2119 if (mkfifo(fifo
, 0600) < 0)
2120 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2122 from
= TAKE_PTR(fifo
);
2125 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2129 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2131 return log_error_errno(errno
, "Failed to open fifo: %m");
2133 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2134 r
= send_one_fd(kmsg_socket
, fd
, 0);
2136 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2141 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2142 union in_addr_union
*exposed
= userdata
;
2148 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2152 static int setup_hostname(void) {
2155 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2158 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2160 return log_error_errno(r
, "Failed to set hostname: %m");
2165 static int setup_journal(const char *directory
) {
2166 _cleanup_free_
char *d
= NULL
;
2167 const char *dirname
, *p
, *q
;
2173 /* Don't link journals in ephemeral mode */
2177 if (arg_link_journal
== LINK_NO
)
2180 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2182 r
= sd_id128_get_machine(&this_id
);
2184 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2186 if (sd_id128_equal(arg_uuid
, this_id
)) {
2187 log_full(try ? LOG_WARNING
: LOG_ERR
,
2188 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2194 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2195 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2197 bool ignore
= r
== -EROFS
&& try;
2198 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2199 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2200 return ignore
? 0 : r
;
2204 (void) sd_id128_to_string(arg_uuid
, id
);
2206 p
= strjoina("/var/log/journal/", id
);
2207 q
= prefix_roota(directory
, p
);
2209 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2213 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2214 "%s: already a mount point, refusing to use for journal", p
);
2217 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2221 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2222 "%s: already a mount point, refusing to use for journal", q
);
2225 r
= readlink_and_make_absolute(p
, &d
);
2227 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2230 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2232 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2237 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2238 } else if (r
== -EINVAL
) {
2240 if (arg_link_journal
== LINK_GUEST
&&
2243 if (errno
== ENOTDIR
) {
2244 log_error("%s already exists and is neither a symlink nor a directory", p
);
2247 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2249 } else if (r
!= -ENOENT
)
2250 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2252 if (arg_link_journal
== LINK_GUEST
) {
2254 if (symlink(q
, p
) < 0) {
2256 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2259 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2262 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2264 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2268 if (arg_link_journal
== LINK_HOST
) {
2269 /* don't create parents here — if the host doesn't have
2270 * permanent journal set up, don't force it here */
2272 r
= mkdir_errno_wrapper(p
, 0755);
2273 if (r
< 0 && r
!= -EEXIST
) {
2275 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2278 return log_error_errno(r
, "Failed to create %s: %m", p
);
2281 } else if (access(p
, F_OK
) < 0)
2284 if (dir_is_empty(q
) == 0)
2285 log_warning("%s is not empty, proceeding anyway.", q
);
2287 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2289 return log_error_errno(r
, "Failed to create %s: %m", q
);
2291 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2293 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2298 static int drop_capabilities(uid_t uid
) {
2299 CapabilityQuintet q
;
2301 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2302 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2303 * arg_caps_retain. */
2305 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2306 q
= arg_full_capabilities
;
2308 if (q
.bounding
== (uint64_t) -1)
2309 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2311 if (q
.effective
== (uint64_t) -1)
2312 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2314 if (q
.inheritable
== (uint64_t) -1)
2315 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2317 if (q
.permitted
== (uint64_t) -1)
2318 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2320 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2323 q
= (CapabilityQuintet
) {
2324 .bounding
= arg_caps_retain
,
2325 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2326 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2327 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2328 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2331 return capability_quintet_enforce(&q
);
2334 static int reset_audit_loginuid(void) {
2335 _cleanup_free_
char *p
= NULL
;
2338 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2341 r
= read_one_line_file("/proc/self/loginuid", &p
);
2345 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2347 /* Already reset? */
2348 if (streq(p
, "4294967295"))
2351 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2354 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2355 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2356 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2357 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2358 "using systemd-nspawn. Sleeping for 5s... (%m)");
2366 static int setup_propagate(const char *root
) {
2370 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2371 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2372 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2373 (void) mkdir_p(p
, 0600);
2375 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2377 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2379 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2381 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2383 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2385 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2387 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2388 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2392 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2396 /* machined will MS_MOVE into that directory, and that's only
2397 * supported for non-shared mounts. */
2398 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2401 static int setup_machine_id(const char *directory
) {
2402 const char *etc_machine_id
;
2406 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2407 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2408 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2409 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2410 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2411 * container behaves nicely). */
2413 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2415 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2417 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2418 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2420 if (sd_id128_is_null(arg_uuid
)) {
2421 r
= sd_id128_randomize(&arg_uuid
);
2423 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2426 if (sd_id128_is_null(id
))
2427 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2428 "Machine ID in container image is zero, refusing.");
2436 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2441 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2444 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2445 if (r
== -EOPNOTSUPP
)
2446 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.");
2448 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2450 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2452 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2454 log_debug("Patched directory tree to match UID/GID range.");
2461 * < 0 : wait_for_terminate() failed to get the state of the
2462 * container, the container was terminated by a signal, or
2463 * failed for an unknown reason. No change is made to the
2464 * container argument.
2465 * > 0 : The program executed in the container terminated with an
2466 * error. The exit code of the program executed in the
2467 * container is returned. The container argument has been set
2468 * to CONTAINER_TERMINATED.
2469 * 0 : The container is being rebooted, has been shut down or exited
2470 * successfully. The container argument has been set to either
2471 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2473 * That is, success is indicated by a return value of zero, and an
2474 * error is indicated by a non-zero value.
2476 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2480 r
= wait_for_terminate(pid
, &status
);
2482 return log_warning_errno(r
, "Failed to wait for container: %m");
2484 switch (status
.si_code
) {
2487 if (status
.si_status
== 0)
2488 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2490 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2492 *container
= CONTAINER_TERMINATED
;
2493 return status
.si_status
;
2496 if (status
.si_status
== SIGINT
) {
2497 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2498 *container
= CONTAINER_TERMINATED
;
2501 } else if (status
.si_status
== SIGHUP
) {
2502 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2503 *container
= CONTAINER_REBOOTED
;
2509 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2510 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2513 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2514 "Container %s failed due to unknown reason.", arg_machine
);
2518 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2521 pid
= PTR_TO_PID(userdata
);
2523 if (kill(pid
, arg_kill_signal
) >= 0) {
2524 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2525 sd_event_source_set_userdata(s
, NULL
);
2530 sd_event_exit(sd_event_source_get_event(s
), 0);
2534 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2540 pid
= PTR_TO_PID(userdata
);
2545 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2546 return log_error_errno(errno
, "Failed to waitid(): %m");
2547 if (si
.si_pid
== 0) /* No pending children. */
2549 if (si
.si_pid
== pid
) {
2550 /* The main process we care for has exited. Return from
2551 * signal handler but leave the zombie. */
2552 sd_event_exit(sd_event_source_get_event(s
), 0);
2556 /* Reap all other children. */
2557 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2563 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2568 pid
= PTR_TO_PID(userdata
);
2570 if (arg_kill_signal
> 0) {
2571 log_info("Container termination requested. Attempting to halt container.");
2572 (void) kill(pid
, arg_kill_signal
);
2574 log_info("Container termination requested. Exiting.");
2575 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2581 static int determine_names(void) {
2584 if (arg_template
&& !arg_directory
&& arg_machine
) {
2586 /* If --template= was specified then we should not
2587 * search for a machine, but instead create a new one
2588 * in /var/lib/machine. */
2590 arg_directory
= strjoin("/var/lib/machines/", arg_machine
);
2595 if (!arg_image
&& !arg_directory
) {
2597 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2599 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2601 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2603 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2605 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2606 r
= free_and_strdup(&arg_image
, i
->path
);
2608 r
= free_and_strdup(&arg_directory
, i
->path
);
2613 arg_read_only
= arg_read_only
|| i
->read_only
;
2615 r
= safe_getcwd(&arg_directory
);
2617 return log_error_errno(r
, "Failed to determine current directory: %m");
2620 if (!arg_directory
&& !arg_image
) {
2621 log_error("Failed to determine path, please use -D or -i.");
2627 if (arg_directory
&& path_equal(arg_directory
, "/"))
2628 arg_machine
= gethostname_malloc();
2633 arg_machine
= strdup(basename(arg_image
));
2635 /* Truncate suffix if there is one */
2636 e
= endswith(arg_machine
, ".raw");
2640 arg_machine
= strdup(basename(arg_directory
));
2645 hostname_cleanup(arg_machine
);
2646 if (!machine_name_is_valid(arg_machine
)) {
2647 log_error("Failed to determine machine name automatically, please use -M.");
2651 if (arg_ephemeral
) {
2654 /* Add a random suffix when this is an
2655 * ephemeral machine, so that we can run many
2656 * instances at once without manually having
2657 * to specify -M each time. */
2659 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2670 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2679 r
= chase_symlinks(*p
, NULL
, flags
, &chased
);
2681 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2683 free_and_replace(*p
, chased
);
2684 return r
; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2687 static int determine_uid_shift(const char *directory
) {
2690 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2695 if (arg_uid_shift
== UID_INVALID
) {
2698 r
= stat(directory
, &st
);
2700 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2702 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2704 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2705 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2706 "UID and GID base of %s don't match.", directory
);
2708 arg_uid_range
= UINT32_C(0x10000);
2711 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2712 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2713 "UID base too high for UID range.");
2718 static unsigned long effective_clone_ns_flags(void) {
2719 unsigned long flags
= arg_clone_ns_flags
;
2721 if (arg_private_network
)
2722 flags
|= CLONE_NEWNET
;
2724 flags
|= CLONE_NEWCGROUP
;
2725 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2726 flags
|= CLONE_NEWUSER
;
2731 static int patch_sysctl(void) {
2733 /* This table is inspired by runc's sysctl() function */
2734 static const struct {
2737 unsigned long clone_flags
;
2739 { "kernel.hostname", false, CLONE_NEWUTS
},
2740 { "kernel.domainname", false, CLONE_NEWUTS
},
2741 { "kernel.msgmax", false, CLONE_NEWIPC
},
2742 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2743 { "kernel.msgmni", false, CLONE_NEWIPC
},
2744 { "kernel.sem", false, CLONE_NEWIPC
},
2745 { "kernel.shmall", false, CLONE_NEWIPC
},
2746 { "kernel.shmmax", false, CLONE_NEWIPC
},
2747 { "kernel.shmmni", false, CLONE_NEWIPC
},
2748 { "fs.mqueue.", true, CLONE_NEWIPC
},
2749 { "net.", true, CLONE_NEWNET
},
2752 unsigned long flags
;
2756 flags
= effective_clone_ns_flags();
2758 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2762 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2764 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2767 if (safe_sysctl
[i
].prefix
)
2768 good
= startswith(*k
, safe_sysctl
[i
].key
);
2770 good
= streq(*k
, safe_sysctl
[i
].key
);
2777 log_error("Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2781 r
= sysctl_write(*k
, *v
);
2783 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2789 static int inner_child(
2791 const char *directory
,
2797 _cleanup_free_
char *home
= NULL
;
2800 const char *envp
[] = {
2801 "PATH=" DEFAULT_PATH_COMPAT
,
2802 NULL
, /* container */
2807 NULL
, /* container_uuid */
2808 NULL
, /* LISTEN_FDS */
2809 NULL
, /* LISTEN_PID */
2810 NULL
, /* NOTIFY_SOCKET */
2813 const char *exec_target
;
2814 _cleanup_strv_free_
char **env_use
= NULL
;
2815 int r
, which_failed
;
2817 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2818 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2819 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2820 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2821 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2822 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2825 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2826 * unshare(). See below. */
2830 assert(kmsg_socket
>= 0);
2832 log_debug("Inner child is initializing.");
2834 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2835 /* Tell the parent, that it now can write the UID map. */
2836 (void) barrier_place(barrier
); /* #1 */
2838 /* Wait until the parent wrote the UID map */
2839 if (!barrier_place_and_sync(barrier
)) /* #2 */
2840 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2841 "Parent died too early");
2844 r
= reset_uid_gid();
2846 return log_error_errno(r
, "Couldn't become new root: %m");
2849 arg_mount_settings
| MOUNT_IN_USERNS
,
2851 arg_selinux_apifs_context
);
2855 if (!arg_network_namespace_path
&& arg_private_network
) {
2856 r
= unshare(CLONE_NEWNET
);
2858 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2860 /* Tell the parent that it can setup network interfaces. */
2861 (void) barrier_place(barrier
); /* #3 */
2864 r
= mount_sysfs(NULL
, arg_mount_settings
);
2868 /* Wait until we are cgroup-ified, so that we
2869 * can mount the right cgroup path writable */
2870 if (!barrier_place_and_sync(barrier
)) /* #4 */
2871 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2872 "Parent died too early");
2875 r
= unshare(CLONE_NEWCGROUP
);
2877 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2880 arg_unified_cgroup_hierarchy
,
2881 arg_userns_mode
!= USER_NAMESPACE_NO
,
2884 arg_selinux_apifs_context
,
2889 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2894 r
= setup_boot_id();
2898 r
= setup_kmsg(kmsg_socket
);
2901 kmsg_socket
= safe_close(kmsg_socket
);
2906 arg_n_custom_mounts
,
2910 arg_selinux_apifs_context
,
2916 return log_error_errno(errno
, "setsid() failed: %m");
2918 if (arg_private_network
)
2921 if (arg_expose_ports
) {
2922 r
= expose_port_send_rtnl(rtnl_socket
);
2925 rtnl_socket
= safe_close(rtnl_socket
);
2932 if (arg_oom_score_adjust_set
) {
2933 r
= set_oom_score_adjust(arg_oom_score_adjust
);
2935 return log_error_errno(r
, "Failed to adjust OOM score: %m");
2939 if (sched_setaffinity(0, CPU_ALLOC_SIZE(arg_cpuset_ncpus
), arg_cpuset
) < 0)
2940 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
2942 (void) setup_hostname();
2944 if (arg_personality
!= PERSONALITY_INVALID
) {
2945 r
= safe_personality(arg_personality
);
2947 return log_error_errno(r
, "personality() failed: %m");
2948 } else if (secondary
) {
2949 r
= safe_personality(PER_LINUX32
);
2951 return log_error_errno(r
, "personality() failed: %m");
2954 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
2956 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
2961 if (is_seccomp_available()) {
2963 r
= seccomp_load(arg_seccomp
);
2964 if (IN_SET(r
, -EPERM
, -EACCES
))
2965 return log_error_errno(r
, "Failed to install seccomp filter: %m");
2967 log_debug_errno(r
, "Failed to install seccomp filter: %m");
2972 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
2978 if (arg_selinux_context
)
2979 if (setexeccon(arg_selinux_context
) < 0)
2980 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
2983 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
2984 * if we need to later on. */
2985 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
2986 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
2988 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
2989 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
2991 r
= change_uid_gid(arg_user
, &home
);
2995 r
= drop_capabilities(getuid());
2997 return log_error_errno(r
, "Dropping capabilities failed: %m");
2999 if (arg_no_new_privileges
)
3000 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3001 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3003 /* LXC sets container=lxc, so follow the scheme here */
3004 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3006 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3010 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3011 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3014 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3015 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3016 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3019 assert(!sd_id128_is_null(arg_uuid
));
3021 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3024 if (fdset_size(fds
) > 0) {
3025 r
= fdset_cloexec(fds
, false);
3027 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3029 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3030 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3033 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3036 env_use
= strv_env_merge(2, envp
, arg_setenv
);
3040 /* Let the parent know that we are ready and
3041 * wait until the parent is ready with the
3043 if (!barrier_place_and_sync(barrier
)) /* #5 */
3044 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3045 "Parent died too early");
3048 if (chdir(arg_chdir
) < 0)
3049 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3051 if (arg_start_mode
== START_PID2
) {
3052 r
= stub_pid1(arg_uuid
);
3057 log_debug("Inner child completed, invoking payload.");
3059 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3060 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3061 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3063 log_set_open_when_needed(true);
3065 (void) fdset_close_others(fds
);
3067 if (arg_start_mode
== START_BOOT
) {
3071 /* Automatically search for the init system */
3073 m
= strv_length(arg_parameters
);
3074 a
= newa(char*, m
+ 2);
3075 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3078 a
[0] = (char*) "/usr/lib/systemd/systemd";
3079 execve(a
[0], a
, env_use
);
3081 a
[0] = (char*) "/lib/systemd/systemd";
3082 execve(a
[0], a
, env_use
);
3084 a
[0] = (char*) "/sbin/init";
3085 execve(a
[0], a
, env_use
);
3087 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3088 } else if (!strv_isempty(arg_parameters
)) {
3089 const char *dollar_path
;
3091 exec_target
= arg_parameters
[0];
3093 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3095 dollar_path
= strv_env_get(env_use
, "PATH");
3097 if (putenv((char*) dollar_path
) != 0)
3098 return log_error_errno(errno
, "Failed to update $PATH: %m");
3101 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3104 /* If we cannot change the directory, we'll end up in /, that is expected. */
3105 (void) chdir(home
?: "/root");
3107 execle("/bin/bash", "-bash", NULL
, env_use
);
3108 execle("/bin/sh", "-sh", NULL
, env_use
);
3110 exec_target
= "/bin/bash, /bin/sh";
3113 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3116 static int setup_sd_notify_child(void) {
3117 _cleanup_close_
int fd
= -1;
3118 union sockaddr_union sa
= {
3119 .un
.sun_family
= AF_UNIX
,
3120 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3124 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3126 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3128 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3129 (void) sockaddr_un_unlink(&sa
.un
);
3131 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3133 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3135 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3137 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3139 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3141 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3146 static int outer_child(
3148 const char *directory
,
3149 const char *console
,
3150 DissectedImage
*dissected_image
,
3157 int uid_shift_socket
,
3158 int unified_cgroup_hierarchy_socket
,
3162 _cleanup_close_
int fd
= -1;
3167 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3168 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3169 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3170 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3174 assert(pid_socket
>= 0);
3175 assert(uuid_socket
>= 0);
3176 assert(notify_socket
>= 0);
3177 assert(kmsg_socket
>= 0);
3179 log_debug("Outer child is initializing.");
3181 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3182 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3184 if (arg_console_mode
!= CONSOLE_PIPE
) {
3189 terminal
= open_terminal(console
, O_RDWR
);
3191 return log_error_errno(terminal
, "Failed to open console: %m");
3193 /* Make sure we can continue logging to the original stderr, even if stderr points elsewhere now */
3194 r
= log_dup_console();
3196 return log_error_errno(r
, "Failed to duplicate stderr: %m");
3198 r
= rearrange_stdio(terminal
, terminal
, terminal
); /* invalidates 'terminal' on success and failure */
3200 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
3203 r
= reset_audit_loginuid();
3207 /* Mark everything as slave, so that we still
3208 * receive mounts from the real root, but don't
3209 * propagate mounts to the real root. */
3210 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3214 if (dissected_image
) {
3215 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3216 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3217 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3218 * makes sure ESP partitions and userns are compatible. */
3220 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3221 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3222 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0)|
3223 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3228 r
= determine_uid_shift(directory
);
3232 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3233 /* Let the parent know which UID shift we read from the image */
3234 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3236 return log_error_errno(errno
, "Failed to send UID shift: %m");
3237 if (l
!= sizeof(arg_uid_shift
))
3238 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3239 "Short write while sending UID shift.");
3241 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3242 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3243 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3244 * not it will pick a different one, and send it back to us. */
3246 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3248 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3249 if (l
!= sizeof(arg_uid_shift
))
3250 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3251 "Short read while receiving UID shift.");
3254 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3255 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3258 if (!dissected_image
) {
3259 /* Turn directory into bind mount */
3260 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3265 r
= setup_pivot_root(
3268 arg_pivot_root_old
);
3272 r
= setup_volatile_mode(
3275 arg_userns_mode
!= USER_NAMESPACE_NO
,
3278 arg_selinux_context
);
3282 if (dissected_image
) {
3283 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3284 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3285 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
3290 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3291 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3293 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3297 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3299 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3300 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3301 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3302 "Short write while sending cgroup mode.");
3304 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3307 /* Mark everything as shared so our mounts get propagated down. This is
3308 * required to make new bind mounts available in systemd services
3309 * inside the containter that create a new mount namespace.
3310 * See https://github.com/systemd/systemd/issues/3860
3311 * Further submounts (such as /dev) done after this will inherit the
3312 * shared propagation mode. */
3313 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3317 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3321 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3325 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
) {
3326 r
= bind_remount_recursive(directory
, true, NULL
);
3328 return log_error_errno(r
, "Failed to make tree read-only: %m");
3331 r
= mount_all(directory
,
3334 arg_selinux_apifs_context
);
3338 r
= copy_devnodes(directory
);
3342 r
= make_extra_nodes(directory
);
3346 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3347 (void) make_inaccessible_nodes(directory
, arg_uid_shift
, arg_uid_shift
);
3349 r
= setup_pts(directory
);
3353 r
= setup_propagate(directory
);
3357 r
= setup_dev_console(directory
, console
);
3361 r
= setup_keyring();
3365 r
= setup_timezone(directory
);
3369 r
= setup_resolv_conf(directory
);
3373 r
= setup_machine_id(directory
);
3377 r
= setup_journal(directory
);
3384 arg_n_custom_mounts
,
3385 arg_userns_mode
!= USER_NAMESPACE_NO
,
3388 arg_selinux_apifs_context
,
3393 if (!arg_use_cgns
) {
3396 arg_unified_cgroup_hierarchy
,
3397 arg_userns_mode
!= USER_NAMESPACE_NO
,
3400 arg_selinux_apifs_context
,
3406 r
= mount_move_root(directory
);
3408 return log_error_errno(r
, "Failed to move root directory: %m");
3410 fd
= setup_sd_notify_child();
3414 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3415 arg_clone_ns_flags
|
3416 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3418 return log_error_errno(errno
, "Failed to fork inner child: %m");
3420 pid_socket
= safe_close(pid_socket
);
3421 uuid_socket
= safe_close(uuid_socket
);
3422 notify_socket
= safe_close(notify_socket
);
3423 uid_shift_socket
= safe_close(uid_shift_socket
);
3425 /* The inner child has all namespaces that are
3426 * requested, so that we all are owned by the user if
3427 * user namespaces are turned on. */
3429 if (arg_network_namespace_path
) {
3430 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3432 return log_error_errno(r
, "Failed to join network namespace: %m");
3435 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, fds
);
3437 _exit(EXIT_FAILURE
);
3439 _exit(EXIT_SUCCESS
);
3442 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3444 return log_error_errno(errno
, "Failed to send PID: %m");
3445 if (l
!= sizeof(pid
))
3446 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3447 "Short write while sending PID.");
3449 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3451 return log_error_errno(errno
, "Failed to send machine ID: %m");
3452 if (l
!= sizeof(arg_uuid
))
3453 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3454 "Short write while sending machine ID.");
3456 l
= send_one_fd(notify_socket
, fd
, 0);
3458 return log_error_errno(errno
, "Failed to send notify fd: %m");
3460 pid_socket
= safe_close(pid_socket
);
3461 uuid_socket
= safe_close(uuid_socket
);
3462 notify_socket
= safe_close(notify_socket
);
3463 kmsg_socket
= safe_close(kmsg_socket
);
3464 rtnl_socket
= safe_close(rtnl_socket
);
3465 netns_fd
= safe_close(netns_fd
);
3470 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3471 bool tried_hashed
= false;
3472 unsigned n_tries
= 100;
3477 assert(ret_lock_file
);
3478 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3479 assert(arg_uid_range
== 0x10000U
);
3483 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3486 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3487 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3492 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3494 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3497 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3498 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3499 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3504 /* Make some superficial checks whether the range is currently known in the user database */
3505 if (getpwuid(candidate
))
3507 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3509 if (getgrgid(candidate
))
3511 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3514 *ret_lock_file
= lf
;
3515 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3520 if (arg_machine
&& !tried_hashed
) {
3521 /* Try to hash the base from the container name */
3523 static const uint8_t hash_key
[] = {
3524 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3525 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3528 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3530 tried_hashed
= true;
3532 random_bytes(&candidate
, sizeof(candidate
));
3534 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3535 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3539 static int setup_uid_map(pid_t pid
) {
3540 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3545 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3546 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3547 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3549 return log_error_errno(r
, "Failed to write UID map: %m");
3551 /* We always assign the same UID and GID ranges */
3552 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3553 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3555 return log_error_errno(r
, "Failed to write GID map: %m");
3560 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3561 char buf
[NOTIFY_BUFFER_MAX
+1];
3563 struct iovec iovec
= {
3565 .iov_len
= sizeof(buf
)-1,
3568 struct cmsghdr cmsghdr
;
3569 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
3570 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
3572 struct msghdr msghdr
= {
3575 .msg_control
= &control
,
3576 .msg_controllen
= sizeof(control
),
3578 struct cmsghdr
*cmsg
;
3579 struct ucred
*ucred
= NULL
;
3581 pid_t inner_child_pid
;
3582 _cleanup_strv_free_
char **tags
= NULL
;
3586 inner_child_pid
= PTR_TO_PID(userdata
);
3588 if (revents
!= EPOLLIN
) {
3589 log_warning("Got unexpected poll event for notify fd.");
3593 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3595 if (IN_SET(errno
, EAGAIN
, EINTR
))
3598 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
3600 cmsg_close_all(&msghdr
);
3602 CMSG_FOREACH(cmsg
, &msghdr
) {
3603 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
3604 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
3605 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
3607 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
3611 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3612 log_debug("Received notify message without valid credentials. Ignoring.");
3616 if ((size_t) n
>= sizeof(buf
)) {
3617 log_warning("Received notify message exceeded maximum size. Ignoring.");
3622 tags
= strv_split(buf
, "\n\r");
3626 if (strv_find(tags
, "READY=1"))
3627 (void) sd_notifyf(false, "READY=1\n");
3629 p
= strv_find_startswith(tags
, "STATUS=");
3631 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3636 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3639 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3641 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3643 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3648 static int merge_settings(Settings
*settings
, const char *path
) {
3654 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3655 * that this steals the fields of the Settings* structure, and hence modifies it. */
3657 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3658 settings
->start_mode
>= 0) {
3659 arg_start_mode
= settings
->start_mode
;
3660 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3663 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3664 arg_ephemeral
= settings
->ephemeral
;
3666 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3669 if (!arg_settings_trusted
)
3670 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3672 free_and_replace(arg_directory
, settings
->root
);
3675 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3676 settings
->pivot_root_new
) {
3677 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3678 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3681 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3682 settings
->working_directory
)
3683 free_and_replace(arg_chdir
, settings
->working_directory
);
3685 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3686 settings
->environment
)
3687 strv_free_and_replace(arg_setenv
, settings
->environment
);
3689 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3692 free_and_replace(arg_user
, settings
->user
);
3694 if (uid_is_valid(settings
->uid
))
3695 arg_uid
= settings
->uid
;
3696 if (gid_is_valid(settings
->gid
))
3697 arg_gid
= settings
->gid
;
3698 if (settings
->n_supplementary_gids
> 0) {
3699 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3700 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3704 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3705 uint64_t plus
, minus
;
3707 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3708 * Settings structure */
3710 plus
= settings
->capability
;
3711 minus
= settings
->drop_capability
;
3713 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3714 if (settings_private_network(settings
))
3715 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3717 minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3720 if (!arg_settings_trusted
&& plus
!= 0) {
3721 if (settings
->capability
!= 0)
3722 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3724 arg_caps_retain
|= plus
;
3726 arg_caps_retain
&= ~minus
;
3728 /* Copy the full capabilities over too */
3729 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3730 if (!arg_settings_trusted
)
3731 log_warning("Ignoring capabilitiy settings, file %s is not trusted.", path
);
3733 arg_full_capabilities
= settings
->full_capabilities
;
3737 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3738 settings
->kill_signal
> 0)
3739 arg_kill_signal
= settings
->kill_signal
;
3741 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3742 settings
->personality
!= PERSONALITY_INVALID
)
3743 arg_personality
= settings
->personality
;
3745 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3746 !sd_id128_is_null(settings
->machine_id
)) {
3748 if (!arg_settings_trusted
)
3749 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3751 arg_uuid
= settings
->machine_id
;
3754 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3755 settings
->read_only
>= 0)
3756 arg_read_only
= settings
->read_only
;
3758 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3759 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3760 arg_volatile_mode
= settings
->volatile_mode
;
3762 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3763 settings
->n_custom_mounts
> 0) {
3765 if (!arg_settings_trusted
)
3766 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3768 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3769 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3770 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3771 settings
->n_custom_mounts
= 0;
3775 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3776 (settings
->private_network
>= 0 ||
3777 settings
->network_veth
>= 0 ||
3778 settings
->network_bridge
||
3779 settings
->network_zone
||
3780 settings
->network_interfaces
||
3781 settings
->network_macvlan
||
3782 settings
->network_ipvlan
||
3783 settings
->network_veth_extra
||
3784 settings
->network_namespace_path
)) {
3786 if (!arg_settings_trusted
)
3787 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3789 arg_network_veth
= settings_network_veth(settings
);
3790 arg_private_network
= settings_private_network(settings
);
3792 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3793 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3794 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3795 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3797 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3798 free_and_replace(arg_network_zone
, settings
->network_zone
);
3800 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3804 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3805 settings
->expose_ports
) {
3807 if (!arg_settings_trusted
)
3808 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3810 expose_port_free_all(arg_expose_ports
);
3811 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3815 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3816 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3818 if (!arg_settings_trusted
)
3819 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3821 arg_userns_mode
= settings
->userns_mode
;
3822 arg_uid_shift
= settings
->uid_shift
;
3823 arg_uid_range
= settings
->uid_range
;
3824 arg_userns_chown
= settings
->userns_chown
;
3828 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3829 arg_notify_ready
= settings
->notify_ready
;
3831 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3833 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_whitelist
))
3834 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3836 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3837 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3841 if (!arg_settings_trusted
&& settings
->seccomp
)
3842 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
3844 seccomp_release(arg_seccomp
);
3845 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
3850 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3851 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3854 if (!settings
->rlimit
[rl
])
3857 if (!arg_settings_trusted
) {
3858 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3862 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3865 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3867 free_and_replace(arg_hostname
, settings
->hostname
);
3869 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3870 settings
->no_new_privileges
>= 0)
3871 arg_no_new_privileges
= settings
->no_new_privileges
;
3873 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3874 settings
->oom_score_adjust_set
) {
3876 if (!arg_settings_trusted
)
3877 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
3879 arg_oom_score_adjust
= settings
->oom_score_adjust
;
3880 arg_oom_score_adjust_set
= true;
3884 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
3887 if (!arg_settings_trusted
)
3888 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
3891 CPU_FREE(arg_cpuset
);
3892 arg_cpuset
= TAKE_PTR(settings
->cpuset
);
3893 arg_cpuset_ncpus
= settings
->cpuset_ncpus
;
3897 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
3898 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
3899 arg_resolv_conf
= settings
->resolv_conf
;
3901 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
3902 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
3904 if (!arg_settings_trusted
)
3905 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
3907 arg_link_journal
= settings
->link_journal
;
3908 arg_link_journal_try
= settings
->link_journal_try
;
3912 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
3913 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
3914 arg_timezone
= settings
->timezone
;
3916 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
3919 if (!arg_settings_trusted
)
3920 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
3922 free_and_replace(arg_slice
, settings
->slice
);
3925 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
3926 settings
->use_cgns
>= 0) {
3928 if (!arg_settings_trusted
)
3929 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
3931 arg_use_cgns
= settings
->use_cgns
;
3934 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
3935 settings
->clone_ns_flags
!= (unsigned long) -1) {
3937 if (!arg_settings_trusted
)
3938 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
3940 arg_clone_ns_flags
= settings
->clone_ns_flags
;
3943 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
3944 settings
->console_mode
>= 0) {
3946 if (!arg_settings_trusted
)
3947 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
3949 arg_console_mode
= settings
->console_mode
;
3952 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
3953 * don't consult arg_settings_mask for them. */
3955 sd_bus_message_unref(arg_property_message
);
3956 arg_property_message
= TAKE_PTR(settings
->properties
);
3958 arg_console_width
= settings
->console_width
;
3959 arg_console_height
= settings
->console_height
;
3961 device_node_free_many(arg_extra_nodes
, arg_n_extra_nodes
);
3962 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
3963 arg_n_extra_nodes
= settings
->n_extra_nodes
;
3968 static int load_settings(void) {
3969 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
3970 _cleanup_fclose_
FILE *f
= NULL
;
3971 _cleanup_free_
char *p
= NULL
;
3978 /* If all settings are masked, there's no point in looking for
3979 * the settings file */
3980 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
3983 fn
= strjoina(arg_machine
, ".nspawn");
3985 /* We first look in the admin's directories in /etc and /run */
3986 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3987 _cleanup_free_
char *j
= NULL
;
3989 j
= strjoin(i
, "/", fn
);
3997 /* By default, we trust configuration from /etc and /run */
3998 if (arg_settings_trusted
< 0)
3999 arg_settings_trusted
= true;
4004 if (errno
!= ENOENT
)
4005 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4009 /* After that, let's look for a file next to the
4010 * actual image we shall boot. */
4013 p
= file_in_same_dir(arg_image
, fn
);
4016 } else if (arg_directory
) {
4017 p
= file_in_same_dir(arg_directory
, fn
);
4024 if (!f
&& errno
!= ENOENT
)
4025 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4027 /* By default, we do not trust configuration from /var/lib/machines */
4028 if (arg_settings_trusted
< 0)
4029 arg_settings_trusted
= false;
4036 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4038 r
= settings_load(f
, p
, &settings
);
4042 return merge_settings(settings
, p
);
4045 static int load_oci_bundle(void) {
4046 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4049 if (!arg_oci_bundle
)
4052 /* By default let's trust OCI bundles */
4053 if (arg_settings_trusted
< 0)
4054 arg_settings_trusted
= true;
4056 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4060 return merge_settings(settings
, arg_oci_bundle
);
4063 static int run_container(int master
,
4064 const char* console
,
4065 DissectedImage
*dissected_image
,
4068 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4069 union in_addr_union
*exposed
,
4070 pid_t
*pid
, int *ret
) {
4072 static const struct sigaction sa
= {
4073 .sa_handler
= nop_signal_handler
,
4074 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4077 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4078 _cleanup_close_
int etc_passwd_lock
= -1;
4079 _cleanup_close_pair_
int
4080 kmsg_socket_pair
[2] = { -1, -1 },
4081 rtnl_socket_pair
[2] = { -1, -1 },
4082 pid_socket_pair
[2] = { -1, -1 },
4083 uuid_socket_pair
[2] = { -1, -1 },
4084 notify_socket_pair
[2] = { -1, -1 },
4085 uid_shift_socket_pair
[2] = { -1, -1 },
4086 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4088 _cleanup_close_
int notify_socket
= -1;
4089 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4090 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4091 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4092 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4093 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4094 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4095 ContainerStatus container_status
= 0;
4099 _cleanup_close_
int netns_fd
= -1;
4101 assert_se(sigemptyset(&mask_chld
) == 0);
4102 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4104 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4105 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4106 * check with getpwuid() if the specific user already exists. Note that /etc might be
4107 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4108 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4109 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4112 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4113 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4114 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4117 r
= barrier_create(&barrier
);
4119 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4121 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4122 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4124 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4125 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4127 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4128 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4130 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4131 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4133 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4134 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4136 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4137 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4138 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4140 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4141 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4142 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4144 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4145 * parent's blocking calls and give it a chance to call wait() and terminate. */
4146 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4148 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4150 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4152 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4154 if (arg_network_namespace_path
) {
4155 netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4157 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4159 r
= fd_is_network_ns(netns_fd
);
4161 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
);
4163 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4165 log_error("Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4170 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4172 return log_error_errno(errno
, "clone() failed%s: %m",
4174 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4177 /* The outer child only has a file system namespace. */
4178 barrier_set_role(&barrier
, BARRIER_CHILD
);
4180 master
= safe_close(master
);
4182 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4183 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4184 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4185 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4186 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4187 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4188 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4190 (void) reset_all_signal_handlers();
4191 (void) reset_signal_mask();
4193 r
= outer_child(&barrier
,
4199 uuid_socket_pair
[1],
4200 notify_socket_pair
[1],
4201 kmsg_socket_pair
[1],
4202 rtnl_socket_pair
[1],
4203 uid_shift_socket_pair
[1],
4204 unified_cgroup_hierarchy_socket_pair
[1],
4208 _exit(EXIT_FAILURE
);
4210 _exit(EXIT_SUCCESS
);
4213 barrier_set_role(&barrier
, BARRIER_PARENT
);
4215 fds
= fdset_free(fds
);
4217 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4218 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4219 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4220 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4221 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4222 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4223 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4225 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4226 /* The child just let us know the UID shift it might have read from the image. */
4227 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4229 return log_error_errno(errno
, "Failed to read UID shift: %m");
4230 if (l
!= sizeof arg_uid_shift
) {
4231 log_error("Short read while reading UID shift.");
4235 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4236 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4237 * image, but if that's already in use, pick a new one, and report back to the child,
4238 * which one we now picked. */
4240 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4242 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4244 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4246 return log_error_errno(errno
, "Failed to send UID shift: %m");
4247 if (l
!= sizeof arg_uid_shift
) {
4248 log_error("Short write while writing UID shift.");
4254 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4255 /* The child let us know the support cgroup mode it might have read from the image. */
4256 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4258 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4259 if (l
!= sizeof(arg_unified_cgroup_hierarchy
)) {
4260 log_error("Short read while reading cgroup mode (%zu bytes).%s",
4261 l
, l
== 0 ? " The child is most likely dead." : "");
4266 /* Wait for the outer child. */
4267 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4270 if (r
!= EXIT_SUCCESS
)
4273 /* And now retrieve the PID of the inner child. */
4274 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4276 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4277 if (l
!= sizeof *pid
) {
4278 log_error("Short read while reading inner child PID.");
4282 /* We also retrieve container UUID in case it was generated by outer child */
4283 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4285 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4286 if (l
!= sizeof(arg_uuid
)) {
4287 log_error("Short read while reading container machined ID.");
4291 /* We also retrieve the socket used for notifications generated by outer child */
4292 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4293 if (notify_socket
< 0)
4294 return log_error_errno(notify_socket
,
4295 "Failed to receive notification socket from the outer child: %m");
4297 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4299 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4300 if (!barrier_place_and_sync(&barrier
)) { /* #1 */
4301 log_error("Child died too early.");
4305 r
= setup_uid_map(*pid
);
4309 (void) barrier_place(&barrier
); /* #2 */
4312 if (arg_private_network
) {
4313 if (!arg_network_namespace_path
) {
4314 /* Wait until the child has unshared its network namespace. */
4315 if (!barrier_place_and_sync(&barrier
)) { /* #3 */
4316 log_error("Child died too early");
4321 r
= move_network_interfaces(*pid
, arg_network_interfaces
);
4325 if (arg_network_veth
) {
4326 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4327 arg_network_bridge
|| arg_network_zone
);
4333 if (arg_network_bridge
) {
4334 /* Add the interface to a bridge */
4335 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4340 } else if (arg_network_zone
) {
4341 /* Add the interface to a bridge, possibly creating it */
4342 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4350 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4354 /* We created the primary and extra veth links now; let's remember this, so that we know to
4355 remove them later on. Note that we don't bother with removing veth links that were created
4356 here when their setup failed half-way, because in that case the kernel should be able to
4357 remove them on its own, since they cannot be referenced by anything yet. */
4358 *veth_created
= true;
4360 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4364 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4369 if (arg_register
|| !arg_keep_unit
) {
4370 r
= sd_bus_default_system(&bus
);
4372 return log_error_errno(r
, "Failed to open system bus: %m");
4374 r
= sd_bus_set_close_on_exit(bus
, false);
4376 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4379 if (!arg_keep_unit
) {
4380 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4381 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4382 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4384 r
= sd_bus_match_signal_async(
4387 "org.freedesktop.systemd1",
4389 "org.freedesktop.systemd1.Scope",
4391 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4393 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4397 r
= register_machine(
4405 arg_custom_mounts
, arg_n_custom_mounts
,
4408 arg_property_message
,
4410 arg_container_service_name
);
4414 } else if (!arg_keep_unit
) {
4420 arg_custom_mounts
, arg_n_custom_mounts
,
4423 arg_property_message
);
4427 } else if (arg_slice
|| arg_property
)
4428 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4430 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4434 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4438 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4442 /* Notify the child that the parent is ready with all
4443 * its setup (including cgroup-ification), and that
4444 * the child can now hand over control to the code to
4445 * run inside the container. */
4446 (void) barrier_place(&barrier
); /* #4 */
4448 /* Block SIGCHLD here, before notifying child.
4449 * process_pty() will handle it with the other signals. */
4450 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4452 /* Reset signal to default */
4453 r
= default_signals(SIGCHLD
, -1);
4455 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4457 r
= sd_event_new(&event
);
4459 return log_error_errno(r
, "Failed to get default event source: %m");
4461 (void) sd_event_set_watchdog(event
, true);
4464 r
= sd_bus_attach_event(bus
, event
, 0);
4466 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4469 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4473 /* Let the child know that we are ready and wait that the child is completely ready now. */
4474 if (!barrier_place_and_sync(&barrier
)) { /* #5 */
4475 log_error("Child died too early.");
4479 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4480 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4481 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4483 (void) sd_notifyf(false,
4484 "STATUS=Container running.\n"
4485 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4486 if (!arg_notify_ready
)
4487 (void) sd_notify(false, "READY=1\n");
4489 if (arg_kill_signal
> 0) {
4490 /* Try to kill the init system on SIGINT or SIGTERM */
4491 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4492 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4494 /* Immediately exit */
4495 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4496 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4499 /* Exit when the child exits */
4500 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4502 if (arg_expose_ports
) {
4503 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4507 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4510 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4512 if (IN_SET(arg_console_mode
, CONSOLE_INTERACTIVE
, CONSOLE_READ_ONLY
)) {
4513 assert(master
>= 0);
4515 r
= pty_forward_new(event
, master
,
4516 PTY_FORWARD_IGNORE_VHANGUP
| (arg_console_mode
== CONSOLE_READ_ONLY
? PTY_FORWARD_READ_ONLY
: 0),
4519 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4521 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4522 (void) pty_forward_set_width_height(forward
, arg_console_width
, arg_console_height
);
4525 r
= sd_event_loop(event
);
4527 return log_error_errno(r
, "Failed to run event loop: %m");
4532 (void) pty_forward_get_last_char(forward
, &last_char
);
4533 forward
= pty_forward_free(forward
);
4535 if (!arg_quiet
&& last_char
!= '\n')
4539 /* Kill if it is not dead yet anyway */
4542 terminate_machine(bus
, arg_machine
);
4543 else if (!arg_keep_unit
)
4544 terminate_scope(bus
, arg_machine
);
4547 /* Normally redundant, but better safe than sorry */
4548 (void) kill(*pid
, SIGKILL
);
4550 r
= wait_for_container(*pid
, &container_status
);
4554 /* We failed to wait for the container, or the container exited abnormally. */
4556 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4557 /* r > 0 → The container exited with a non-zero status.
4558 * As a special case, we need to replace 133 with a different value,
4559 * because 133 is special-cased in the service file to reboot the container.
4560 * otherwise → The container exited with zero status and a reboot was not requested.
4562 if (r
== EXIT_FORCE_RESTART
)
4563 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4565 return 0; /* finito */
4568 /* CONTAINER_REBOOTED, loop again */
4570 if (arg_keep_unit
) {
4571 /* Special handling if we are running as a service: instead of simply
4572 * restarting the machine we want to restart the entire service, so let's
4573 * inform systemd about this with the special exit code 133. The service
4574 * file uses RestartForceExitStatus=133 so that this results in a full
4575 * nspawn restart. This is necessary since we might have cgroup parameters
4576 * set we want to have flushed out. */
4577 *ret
= EXIT_FORCE_RESTART
;
4578 return 0; /* finito */
4581 expose_port_flush(arg_expose_ports
, exposed
);
4583 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4584 *veth_created
= false;
4585 return 1; /* loop again */
4588 static int initialize_rlimits(void) {
4589 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4590 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4591 * container execution environments. */
4593 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4594 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4595 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4596 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4597 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4598 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4599 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4600 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4601 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4602 [RLIMIT_NICE
] = { 0, 0 },
4603 [RLIMIT_NOFILE
] = { 1024, 4096 },
4604 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4605 [RLIMIT_RTPRIO
] = { 0, 0 },
4606 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4607 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4609 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4610 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4611 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4612 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4613 * that PID 1 changes a number of other resource limits during early initialization which is why we
4614 * don't read the other limits from PID 1 but prefer the static table above. */
4619 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4620 /* Let's only fill in what the user hasn't explicitly configured anyway */
4621 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4622 const struct rlimit
*v
;
4623 struct rlimit buffer
;
4625 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4626 /* For these two let's read the limits off PID 1. See above for an explanation. */
4628 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4629 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4633 v
= kernel_defaults
+ rl
;
4635 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4636 if (!arg_rlimit
[rl
])
4640 if (DEBUG_LOGGING
) {
4641 _cleanup_free_
char *k
= NULL
;
4643 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4644 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4651 static int run(int argc
, char *argv
[]) {
4652 _cleanup_free_
char *console
= NULL
;
4653 _cleanup_close_
int master
= -1;
4654 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4655 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4656 char veth_name
[IFNAMSIZ
] = "";
4657 bool secondary
= false, remove_directory
= false, remove_image
= false;
4659 union in_addr_union exposed
= {};
4660 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4661 bool veth_created
= false, remove_tmprootdir
= false;
4662 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4663 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4664 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4665 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4667 log_parse_environment();
4670 /* Make sure rename_process() in the stub init process can work */
4674 r
= parse_argv(argc
, argv
);
4682 r
= initialize_rlimits();
4686 r
= load_oci_bundle();
4690 r
= determine_names();
4694 r
= load_settings();
4698 r
= cg_unified_flush();
4700 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4704 r
= verify_arguments();
4708 r
= detect_unified_cgroup_hierarchy_from_environment();
4712 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4713 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4714 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4715 (void) ignore_signals(SIGPIPE
, -1);
4717 n_fd_passed
= sd_listen_fds(false);
4718 if (n_fd_passed
> 0) {
4719 r
= fdset_new_listen_fds(&fds
, false);
4721 log_error_errno(r
, "Failed to collect file descriptors: %m");
4726 /* The "default" umask. This is appropriate for most file and directory
4727 * operations performed by nspawn, and is the umask that will be used for
4728 * the child. Functions like copy_devnodes() change the umask temporarily. */
4731 if (arg_directory
) {
4734 if (path_equal(arg_directory
, "/") && !arg_ephemeral
) {
4735 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4740 if (arg_ephemeral
) {
4741 _cleanup_free_
char *np
= NULL
;
4743 r
= chase_symlinks_and_update(&arg_directory
, 0);
4747 /* If the specified path is a mount point we
4748 * generate the new snapshot immediately
4749 * inside it under a random name. However if
4750 * the specified is not a mount point we
4751 * create the new snapshot in the parent
4752 * directory, just next to it. */
4753 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4755 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4759 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4761 r
= tempfn_random(arg_directory
, "machine.", &np
);
4763 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4767 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4769 log_error_errno(r
, "Failed to lock %s: %m", np
);
4773 r
= btrfs_subvol_snapshot(arg_directory
, np
,
4774 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4775 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4776 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4777 BTRFS_SNAPSHOT_RECURSIVE
|
4778 BTRFS_SNAPSHOT_QUOTA
);
4780 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
4784 free_and_replace(arg_directory
, np
);
4786 remove_directory
= true;
4789 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
4793 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4795 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
4799 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
4804 r
= chase_symlinks_and_update(&arg_template
, 0);
4808 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
4809 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4810 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4811 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4812 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
4813 BTRFS_SNAPSHOT_RECURSIVE
|
4814 BTRFS_SNAPSHOT_QUOTA
);
4816 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4817 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
4819 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
4822 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4823 "Populated %s from template %s.", arg_directory
, arg_template
);
4827 if (arg_start_mode
== START_BOOT
) {
4830 if (arg_pivot_root_new
)
4831 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4835 if (path_is_os_tree(p
) <= 0) {
4836 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
4843 if (arg_pivot_root_new
)
4844 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4848 q
= strjoina(p
, "/usr/");
4850 if (laccess(q
, F_OK
) < 0) {
4851 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
4859 assert(!arg_template
);
4861 r
= chase_symlinks_and_update(&arg_image
, 0);
4865 if (arg_ephemeral
) {
4866 _cleanup_free_
char *np
= NULL
;
4868 r
= tempfn_random(arg_image
, "machine.", &np
);
4870 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
4874 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4876 r
= log_error_errno(r
, "Failed to create image lock: %m");
4880 r
= copy_file(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600, FS_NOCOW_FL
, COPY_REFLINK
|COPY_CRTIME
);
4882 r
= log_error_errno(r
, "Failed to copy image file: %m");
4886 free_and_replace(arg_image
, np
);
4888 remove_image
= true;
4890 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4892 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
4896 r
= log_error_errno(r
, "Failed to create image lock: %m");
4900 if (!arg_root_hash
) {
4901 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
4903 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
4909 if (!mkdtemp(tmprootdir
)) {
4910 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
4914 remove_tmprootdir
= true;
4916 arg_directory
= strdup(tmprootdir
);
4917 if (!arg_directory
) {
4922 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, &loop
);
4924 log_error_errno(r
, "Failed to set up loopback block device: %m");
4928 r
= dissect_image_and_warn(
4931 arg_root_hash
, arg_root_hash_size
,
4932 DISSECT_IMAGE_REQUIRE_ROOT
,
4935 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4936 log_notice("Note that the disk image needs to\n"
4937 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4938 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4939 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4940 " d) or contain a file system without a partition table\n"
4941 "in order to be bootable with systemd-nspawn.");
4947 if (!arg_root_hash
&& dissected_image
->can_verity
)
4948 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
4950 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
4954 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4955 if (remove_image
&& unlink(arg_image
) >= 0)
4956 remove_image
= false;
4959 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
4963 if (arg_console_mode
< 0)
4965 isatty(STDIN_FILENO
) > 0 &&
4966 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
4968 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
4971 if (arg_console_mode
!= CONSOLE_PIPE
) {
4972 master
= posix_openpt(O_RDWR
|O_NOCTTY
|O_CLOEXEC
|O_NONBLOCK
);
4974 r
= log_error_errno(errno
, "Failed to acquire pseudo tty: %m");
4978 r
= ptsname_malloc(master
, &console
);
4980 r
= log_error_errno(r
, "Failed to determine tty name: %m");
4984 if (arg_selinux_apifs_context
) {
4985 r
= mac_selinux_apply(console
, arg_selinux_apifs_context
);
4990 if (unlockpt(master
) < 0) {
4991 r
= log_error_errno(errno
, "Failed to unlock tty: %m");
4997 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4998 arg_machine
, arg_image
?: arg_directory
);
5000 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5002 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5003 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5008 r
= run_container(master
,
5013 veth_name
, &veth_created
,
5021 (void) sd_notify(false,
5022 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5023 "STOPPING=1\nSTATUS=Terminating...");
5026 (void) kill(pid
, SIGKILL
);
5028 /* Try to flush whatever is still queued in the pty */
5030 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5031 master
= safe_close(master
);
5035 (void) wait_for_terminate(pid
, NULL
);
5039 if (remove_directory
&& arg_directory
) {
5042 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5044 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5047 if (remove_image
&& arg_image
) {
5048 if (unlink(arg_image
) < 0)
5049 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5052 if (remove_tmprootdir
) {
5053 if (rmdir(tmprootdir
) < 0)
5054 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5060 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5061 (void) rm_rf(p
, REMOVE_ROOT
);
5064 expose_port_flush(arg_expose_ports
, &exposed
);
5067 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5068 (void) remove_bridge(arg_network_zone
);
5070 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5071 expose_port_free_all(arg_expose_ports
);
5072 rlimit_free_all(arg_rlimit
);
5073 device_node_free_many(arg_extra_nodes
, arg_n_extra_nodes
);
5081 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);