1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include <linux/loop.h>
7 #include <selinux/selinux.h>
11 #include <sys/ioctl.h>
12 #include <sys/mount.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
15 #include <sys/types.h>
20 #include <linux/fs.h> /* Must be included after <sys/mount.h> */
23 #include "sd-daemon.h"
26 #include "alloc-util.h"
27 #include "ether-addr-util.h"
29 #include "base-filesystem.h"
30 #include "blkid-util.h"
31 #include "btrfs-util.h"
33 #include "bus-error.h"
34 #include "bus-locator.h"
37 #include "capability-util.h"
38 #include "cgroup-util.h"
40 #include "common-signal.h"
42 #include "cpu-set-util.h"
43 #include "creds-util.h"
44 #include "dev-setup.h"
45 #include "discover-image.h"
46 #include "dissect-image.h"
52 #include "format-util.h"
55 #include "hexdecoct.h"
56 #include "hostname-setup.h"
57 #include "hostname-util.h"
58 #include "id128-util.h"
61 #include "loop-util.h"
62 #include "loopback-setup.h"
63 #include "machine-credential.h"
65 #include "main-func.h"
66 #include "missing_sched.h"
68 #include "mount-util.h"
69 #include "mountpoint-util.h"
70 #include "namespace-util.h"
71 #include "netlink-util.h"
72 #include "nspawn-bind-user.h"
73 #include "nspawn-cgroup.h"
74 #include "nspawn-def.h"
75 #include "nspawn-expose-ports.h"
76 #include "nspawn-mount.h"
77 #include "nspawn-network.h"
78 #include "nspawn-oci.h"
79 #include "nspawn-patch-uid.h"
80 #include "nspawn-register.h"
81 #include "nspawn-seccomp.h"
82 #include "nspawn-settings.h"
83 #include "nspawn-setuid.h"
84 #include "nspawn-stub-pid1.h"
85 #include "nspawn-util.h"
87 #include "nulstr-util.h"
90 #include "parse-argument.h"
91 #include "parse-util.h"
92 #include "pretty-print.h"
93 #include "process-util.h"
95 #include "random-util.h"
96 #include "raw-clone.h"
97 #include "resolve-util.h"
98 #include "rlimit-util.h"
100 #include "seccomp-util.h"
101 #include "selinux-util.h"
102 #include "signal-util.h"
103 #include "socket-util.h"
104 #include "stat-util.h"
105 #include "stdio-util.h"
106 #include "string-table.h"
107 #include "string-util.h"
109 #include "sysctl-util.h"
110 #include "terminal-util.h"
111 #include "tmpfile-util.h"
112 #include "umask-util.h"
113 #include "unit-name.h"
114 #include "user-util.h"
117 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
119 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
121 #define EXIT_FORCE_RESTART 133
123 typedef enum ContainerStatus
{
124 CONTAINER_TERMINATED
,
128 static char *arg_directory
= NULL
;
129 static char *arg_template
= NULL
;
130 static char *arg_chdir
= NULL
;
131 static char *arg_pivot_root_new
= NULL
;
132 static char *arg_pivot_root_old
= NULL
;
133 static char *arg_user
= NULL
;
134 static uid_t arg_uid
= UID_INVALID
;
135 static gid_t arg_gid
= GID_INVALID
;
136 static gid_t
* arg_supplementary_gids
= NULL
;
137 static size_t arg_n_supplementary_gids
= 0;
138 static sd_id128_t arg_uuid
= {};
139 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
140 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
141 static const char *arg_selinux_context
= NULL
;
142 static const char *arg_selinux_apifs_context
= NULL
;
143 static char *arg_slice
= NULL
;
144 static bool arg_private_network
= false;
145 static bool arg_read_only
= false;
146 static StartMode arg_start_mode
= START_PID1
;
147 static bool arg_ephemeral
= false;
148 static LinkJournal arg_link_journal
= LINK_AUTO
;
149 static bool arg_link_journal_try
= false;
150 static uint64_t arg_caps_retain
=
151 (1ULL << CAP_AUDIT_CONTROL
) |
152 (1ULL << CAP_AUDIT_WRITE
) |
153 (1ULL << CAP_CHOWN
) |
154 (1ULL << CAP_DAC_OVERRIDE
) |
155 (1ULL << CAP_DAC_READ_SEARCH
) |
156 (1ULL << CAP_FOWNER
) |
157 (1ULL << CAP_FSETID
) |
158 (1ULL << CAP_IPC_OWNER
) |
160 (1ULL << CAP_LEASE
) |
161 (1ULL << CAP_LINUX_IMMUTABLE
) |
162 (1ULL << CAP_MKNOD
) |
163 (1ULL << CAP_NET_BIND_SERVICE
) |
164 (1ULL << CAP_NET_BROADCAST
) |
165 (1ULL << CAP_NET_RAW
) |
166 (1ULL << CAP_SETFCAP
) |
167 (1ULL << CAP_SETGID
) |
168 (1ULL << CAP_SETPCAP
) |
169 (1ULL << CAP_SETUID
) |
170 (1ULL << CAP_SYS_ADMIN
) |
171 (1ULL << CAP_SYS_BOOT
) |
172 (1ULL << CAP_SYS_CHROOT
) |
173 (1ULL << CAP_SYS_NICE
) |
174 (1ULL << CAP_SYS_PTRACE
) |
175 (1ULL << CAP_SYS_RESOURCE
) |
176 (1ULL << CAP_SYS_TTY_CONFIG
);
177 static uint64_t arg_caps_ambient
= 0;
178 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
179 static CustomMount
*arg_custom_mounts
= NULL
;
180 static size_t arg_n_custom_mounts
= 0;
181 static char **arg_setenv
= NULL
;
182 static bool arg_quiet
= false;
183 static bool arg_register
= true;
184 static bool arg_keep_unit
= false;
185 static char **arg_network_interfaces
= NULL
;
186 static char **arg_network_macvlan
= NULL
;
187 static char **arg_network_ipvlan
= NULL
;
188 static bool arg_network_veth
= false;
189 static char **arg_network_veth_extra
= NULL
;
190 static char *arg_network_bridge
= NULL
;
191 static char *arg_network_zone
= NULL
;
192 static char *arg_network_namespace_path
= NULL
;
193 struct ether_addr arg_network_provided_mac
= {};
194 static PagerFlags arg_pager_flags
= 0;
195 static unsigned long arg_personality
= PERSONALITY_INVALID
;
196 static char *arg_image
= NULL
;
197 static char *arg_oci_bundle
= NULL
;
198 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
199 static ExposePort
*arg_expose_ports
= NULL
;
200 static char **arg_property
= NULL
;
201 static sd_bus_message
*arg_property_message
= NULL
;
202 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
203 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
204 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
205 static int arg_kill_signal
= 0;
206 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
207 static SettingsMask arg_settings_mask
= 0;
208 static int arg_settings_trusted
= -1;
209 static char **arg_parameters
= NULL
;
210 static const char *arg_container_service_name
= "systemd-nspawn";
211 static bool arg_notify_ready
= false;
212 static bool arg_use_cgns
= true;
213 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
214 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
215 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
216 static char **arg_syscall_allow_list
= NULL
;
217 static char **arg_syscall_deny_list
= NULL
;
219 static scmp_filter_ctx arg_seccomp
= NULL
;
221 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
222 static bool arg_no_new_privileges
= false;
223 static int arg_oom_score_adjust
= 0;
224 static bool arg_oom_score_adjust_set
= false;
225 static CPUSet arg_cpu_set
= {};
226 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
227 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
228 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
229 static DeviceNode
* arg_extra_nodes
= NULL
;
230 static size_t arg_n_extra_nodes
= 0;
231 static char **arg_sysctl
= NULL
;
232 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
233 static MachineCredentialContext arg_credentials
= {};
234 static char **arg_bind_user
= NULL
;
235 static bool arg_suppress_sync
= false;
236 static char *arg_settings_filename
= NULL
;
237 static Architecture arg_architecture
= _ARCHITECTURE_INVALID
;
238 static ImagePolicy
*arg_image_policy
= NULL
;
239 static char *arg_background
= NULL
;
241 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
260 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
262 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
263 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
264 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
265 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
266 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
268 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
270 STATIC_DESTRUCTOR_REGISTER(arg_credentials
, machine_credential_context_done
);
271 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
272 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
273 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
274 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
275 STATIC_DESTRUCTOR_REGISTER(arg_image_policy
, image_policy_freep
);
276 STATIC_DESTRUCTOR_REGISTER(arg_background
, freep
);
278 static int handle_arg_console(const char *arg
) {
279 if (streq(arg
, "help")) {
288 if (streq(arg
, "interactive"))
289 arg_console_mode
= CONSOLE_INTERACTIVE
;
290 else if (streq(arg
, "read-only"))
291 arg_console_mode
= CONSOLE_READ_ONLY
;
292 else if (streq(arg
, "passive"))
293 arg_console_mode
= CONSOLE_PASSIVE
;
294 else if (streq(arg
, "pipe")) {
295 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
296 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
297 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
298 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
299 "Proceeding anyway.");
301 arg_console_mode
= CONSOLE_PIPE
;
302 } else if (streq(arg
, "autopipe")) {
303 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
304 arg_console_mode
= CONSOLE_INTERACTIVE
;
306 arg_console_mode
= CONSOLE_PIPE
;
308 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
310 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
314 static int help(void) {
315 _cleanup_free_
char *link
= NULL
;
318 pager_open(arg_pager_flags
);
320 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
324 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
325 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
326 " -h --help Show this help\n"
327 " --version Print version string\n"
328 " -q --quiet Do not show status information\n"
329 " --no-pager Do not pipe output into a pager\n"
330 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
332 " -D --directory=PATH Root directory for the container\n"
333 " --template=PATH Initialize root directory from template directory,\n"
335 " -x --ephemeral Run container with snapshot of root directory, and\n"
336 " remove it after exit\n"
337 " -i --image=PATH Root file system disk image (or device node) for\n"
339 " --image-policy=POLICY Specify disk image dissection policy\n"
340 " --oci-bundle=PATH OCI bundle directory\n"
341 " --read-only Mount the root directory read-only\n"
342 " --volatile[=MODE] Run the system in volatile mode\n"
343 " --root-hash=HASH Specify verity root hash for root disk image\n"
344 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
345 " as a DER encoded PKCS7, either as a path to a file\n"
346 " or as an ASCII base64 encoded string prefixed by\n"
348 " --verity-data=PATH Specify hash device for verity\n"
349 " --pivot-root=PATH[:PATH]\n"
350 " Pivot root to given directory in the container\n"
351 "\n%3$sExecution:%4$s\n"
352 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
353 " -b --boot Boot up full system (i.e. invoke init)\n"
354 " --chdir=PATH Set working directory in the container\n"
355 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
356 " -u --user=USER Run the command under specified user or UID\n"
357 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
358 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
359 " --suppress-sync=BOOLEAN\n"
360 " Suppress any form of disk data synchronization\n"
361 "\n%3$sSystem Identity:%4$s\n"
362 " -M --machine=NAME Set the machine name for the container\n"
363 " --hostname=NAME Override the hostname for the container\n"
364 " --uuid=UUID Set a specific machine UUID for the container\n"
365 "\n%3$sProperties:%4$s\n"
366 " -S --slice=SLICE Place the container in the specified slice\n"
367 " --property=NAME=VALUE Set scope unit property\n"
368 " --register=BOOLEAN Register container as machine\n"
369 " --keep-unit Do not register a scope for the machine, reuse\n"
370 " the service unit nspawn is running in\n"
371 "\n%3$sUser Namespacing:%4$s\n"
372 " --private-users=no Run without user namespacing\n"
373 " --private-users=yes|pick|identity\n"
374 " Run within user namespace, autoselect UID/GID range\n"
375 " --private-users=UIDBASE[:NUIDS]\n"
376 " Similar, but with user configured UID/GID range\n"
377 " --private-users-ownership=MODE\n"
378 " Adjust ('chown') or map ('map') OS tree ownership\n"
379 " to private UID/GID range\n"
380 " -U Equivalent to --private-users=pick and\n"
381 " --private-users-ownership=auto\n"
382 "\n%3$sNetworking:%4$s\n"
383 " --private-network Disable network in container\n"
384 " --network-interface=HOSTIF[:CONTAINERIF]\n"
385 " Assign an existing network interface to the\n"
387 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
388 " Create a macvlan network interface based on an\n"
389 " existing network interface to the container\n"
390 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
391 " Create an ipvlan network interface based on an\n"
392 " existing network interface to the container\n"
393 " -n --network-veth Add a virtual Ethernet connection between host\n"
395 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
396 " Add an additional virtual Ethernet link between\n"
397 " host and container\n"
398 " --network-bridge=INTERFACE\n"
399 " Add a virtual Ethernet connection to the container\n"
400 " and attach it to an existing bridge on the host\n"
401 " --network-zone=NAME Similar, but attach the new interface to an\n"
402 " an automatically managed bridge interface\n"
403 " --network-namespace-path=PATH\n"
404 " Set network namespace to the one represented by\n"
405 " the specified kernel namespace file node\n"
406 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
407 " Expose a container IP port on the host\n"
408 "\n%3$sSecurity:%4$s\n"
409 " --capability=CAP In addition to the default, retain specified\n"
411 " --drop-capability=CAP Drop the specified capability from the default set\n"
412 " --ambient-capability=CAP\n"
413 " Sets the specified capability for the started\n"
414 " process. Not useful if booting a machine.\n"
415 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
416 " --system-call-filter=LIST|~LIST\n"
417 " Permit/prohibit specific system calls\n"
418 " -Z --selinux-context=SECLABEL\n"
419 " Set the SELinux security context to be used by\n"
420 " processes in the container\n"
421 " -L --selinux-apifs-context=SECLABEL\n"
422 " Set the SELinux security context to be used by\n"
423 " API/tmpfs file systems in the container\n"
424 "\n%3$sResources:%4$s\n"
425 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
426 " --oom-score-adjust=VALUE\n"
427 " Adjust the OOM score value for the payload\n"
428 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
429 " --personality=ARCH Pick personality for this container\n"
430 "\n%3$sIntegration:%4$s\n"
431 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
432 " --timezone=MODE Select mode of /etc/localtime initialization\n"
433 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
434 " host, try-guest, try-host\n"
435 " -j Equivalent to --link-journal=try-guest\n"
436 "\n%3$sMounts:%4$s\n"
437 " --bind=PATH[:PATH[:OPTIONS]]\n"
438 " Bind mount a file or directory from the host into\n"
440 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
441 " Similar, but creates a read-only bind mount\n"
442 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
444 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
445 " --overlay=PATH[:PATH...]:PATH\n"
446 " Create an overlay mount from the host to \n"
448 " --overlay-ro=PATH[:PATH...]:PATH\n"
449 " Similar, but creates a read-only overlay mount\n"
450 " --bind-user=NAME Bind user from host to container\n"
451 "\n%3$sInput/Output:%4$s\n"
452 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
453 " set up for the container.\n"
454 " -P --pipe Equivalent to --console=pipe\n"
455 " --background=COLOR Set ANSI color for background\n"
456 "\n%3$sCredentials:%4$s\n"
457 " --set-credential=ID:VALUE\n"
458 " Pass a credential with literal value to container.\n"
459 " --load-credential=ID:PATH\n"
460 " Load credential to pass to container from file or\n"
461 " AF_UNIX stream socket.\n"
462 "\nSee the %2$s for details.\n",
463 program_invocation_short_name
,
473 static int custom_mount_check_all(void) {
476 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
477 CustomMount
*m
= &arg_custom_mounts
[i
];
479 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
480 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
481 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
482 "--private-users-ownership=own may not be combined with custom root mounts.");
483 if (arg_uid_shift
== UID_INVALID
)
484 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
485 "--private-users with automatic UID shift may not be combined with custom root mounts.");
492 static int detect_unified_cgroup_hierarchy_from_environment(void) {
493 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
496 /* Allow the user to control whether the unified hierarchy is used */
500 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
501 var
= "UNIFIED_CGROUP_HIERARCHY";
506 r
= parse_boolean(e
);
508 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
510 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
512 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
518 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
521 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
522 * in the image actually supports. */
523 r
= cg_all_unified();
525 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
527 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
528 * routine only detects 231, so we'll have a false negative here for 230. */
529 r
= systemd_installation_has_version(directory
, "230");
531 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
533 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
535 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
536 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
537 /* Mixed cgroup hierarchy support was added in 233 */
538 r
= systemd_installation_has_version(directory
, "233");
540 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
542 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
544 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
546 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
548 log_debug("Using %s hierarchy for container.",
549 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
550 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
555 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
560 _cleanup_free_
char *t
= NULL
;
562 r
= extract_first_word(&spec
, &t
, ",", 0);
564 return log_error_errno(r
, "Failed to parse capability %s.", t
);
568 if (streq(t
, "help")) {
569 for (int i
= 0; i
< capability_list_length(); i
++) {
572 name
= capability_to_name(i
);
583 r
= capability_from_name(t
);
585 return log_error_errno(r
, "Failed to parse capability %s.", t
);
592 return 1; /* continue */
595 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
598 r
= getenv_bool(name
);
602 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
604 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
605 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
609 static int parse_mount_settings_env(void) {
613 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
614 if (r
< 0 && r
!= -ENXIO
)
615 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
617 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
619 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
620 if (streq_ptr(e
, "network"))
621 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
624 r
= parse_boolean(e
);
626 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
628 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
629 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
635 static int parse_environment(void) {
639 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
642 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
645 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
648 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
652 r
= parse_mount_settings_env();
656 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
657 * even if it is supported. If not supported, it has no effect. */
658 if (!cg_ns_supported())
659 arg_use_cgns
= false;
661 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
664 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
668 arg_use_cgns
= r
> 0;
669 arg_settings_mask
|= SETTING_USE_CGNS
;
673 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
675 arg_container_service_name
= e
;
677 e
= getenv("SYSTEMD_NSPAWN_NETWORK_MAC");
679 r
= parse_ether_addr(e
, &arg_network_provided_mac
);
681 return log_error_errno(r
, "Failed to parse provided MAC address via environment variable");
684 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
686 arg_suppress_sync
= r
;
687 else if (r
!= -ENXIO
)
688 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
690 return detect_unified_cgroup_hierarchy_from_environment();
693 static int parse_argv(int argc
, char *argv
[]) {
700 ARG_AMBIENT_CAPABILITY
,
712 ARG_NETWORK_INTERFACE
,
717 ARG_NETWORK_VETH_EXTRA
,
718 ARG_NETWORK_NAMESPACE_PATH
,
728 ARG_PRIVATE_USERS_CHOWN
,
729 ARG_PRIVATE_USERS_OWNERSHIP
,
734 ARG_SYSTEM_CALL_FILTER
,
737 ARG_NO_NEW_PRIVILEGES
,
738 ARG_OOM_SCORE_ADJUST
,
754 static const struct option options
[] = {
755 { "help", no_argument
, NULL
, 'h' },
756 { "version", no_argument
, NULL
, ARG_VERSION
},
757 { "directory", required_argument
, NULL
, 'D' },
758 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
759 { "ephemeral", no_argument
, NULL
, 'x' },
760 { "user", required_argument
, NULL
, 'u' },
761 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
762 { "as-pid2", no_argument
, NULL
, 'a' },
763 { "boot", no_argument
, NULL
, 'b' },
764 { "uuid", required_argument
, NULL
, ARG_UUID
},
765 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
766 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
767 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
768 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
769 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
770 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
771 { "bind", required_argument
, NULL
, ARG_BIND
},
772 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
773 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
774 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
775 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
776 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
777 { "machine", required_argument
, NULL
, 'M' },
778 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
779 { "slice", required_argument
, NULL
, 'S' },
780 { "setenv", required_argument
, NULL
, 'E' },
781 { "selinux-context", required_argument
, NULL
, 'Z' },
782 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
783 { "quiet", no_argument
, NULL
, 'q' },
784 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
785 { "register", required_argument
, NULL
, ARG_REGISTER
},
786 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
787 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
788 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
789 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
790 { "network-veth", no_argument
, NULL
, 'n' },
791 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
792 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
793 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
794 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
795 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
796 { "image", required_argument
, NULL
, 'i' },
797 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
798 { "port", required_argument
, NULL
, 'p' },
799 { "property", required_argument
, NULL
, ARG_PROPERTY
},
800 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
801 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
802 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
803 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
804 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
805 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
806 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
807 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
808 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
809 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
810 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
811 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
812 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
813 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
814 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
815 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
816 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
817 { "console", required_argument
, NULL
, ARG_CONSOLE
},
818 { "pipe", no_argument
, NULL
, ARG_PIPE
},
819 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
820 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
821 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
822 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
823 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
824 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
825 { "image-policy", required_argument
, NULL
, ARG_IMAGE_POLICY
},
826 { "background", required_argument
, NULL
, ARG_BACKGROUND
},
831 uint64_t plus
= 0, minus
= 0;
832 bool mask_all_settings
= false, mask_no_settings
= false;
837 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
838 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
840 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
850 r
= parse_path_argument(optarg
, false, &arg_directory
);
854 arg_settings_mask
|= SETTING_DIRECTORY
;
858 r
= parse_path_argument(optarg
, false, &arg_template
);
862 arg_settings_mask
|= SETTING_DIRECTORY
;
866 r
= parse_path_argument(optarg
, false, &arg_image
);
870 arg_settings_mask
|= SETTING_DIRECTORY
;
874 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
881 arg_ephemeral
= true;
882 arg_settings_mask
|= SETTING_EPHEMERAL
;
886 r
= free_and_strdup(&arg_user
, optarg
);
890 arg_settings_mask
|= SETTING_USER
;
893 case ARG_NETWORK_ZONE
: {
894 _cleanup_free_
char *j
= NULL
;
896 j
= strjoin("vz-", optarg
);
900 if (!ifname_valid(j
))
901 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
902 "Network zone name not valid: %s", j
);
904 free_and_replace(arg_network_zone
, j
);
906 arg_network_veth
= true;
907 arg_private_network
= true;
908 arg_settings_mask
|= SETTING_NETWORK
;
912 case ARG_NETWORK_BRIDGE
:
914 if (!ifname_valid(optarg
))
915 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
916 "Bridge interface name not valid: %s", optarg
);
918 r
= free_and_strdup(&arg_network_bridge
, optarg
);
924 arg_network_veth
= true;
925 arg_private_network
= true;
926 arg_settings_mask
|= SETTING_NETWORK
;
929 case ARG_NETWORK_VETH_EXTRA
:
930 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
932 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
934 arg_private_network
= true;
935 arg_settings_mask
|= SETTING_NETWORK
;
938 case ARG_NETWORK_INTERFACE
:
939 r
= interface_pair_parse(&arg_network_interfaces
, optarg
);
943 arg_private_network
= true;
944 arg_settings_mask
|= SETTING_NETWORK
;
947 case ARG_NETWORK_MACVLAN
:
948 r
= macvlan_pair_parse(&arg_network_macvlan
, optarg
);
952 arg_private_network
= true;
953 arg_settings_mask
|= SETTING_NETWORK
;
956 case ARG_NETWORK_IPVLAN
:
957 r
= ipvlan_pair_parse(&arg_network_ipvlan
, optarg
);
962 case ARG_PRIVATE_NETWORK
:
963 arg_private_network
= true;
964 arg_settings_mask
|= SETTING_NETWORK
;
967 case ARG_NETWORK_NAMESPACE_PATH
:
968 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
972 arg_settings_mask
|= SETTING_NETWORK
;
976 if (arg_start_mode
== START_PID2
)
977 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
978 "--boot and --as-pid2 may not be combined.");
980 arg_start_mode
= START_BOOT
;
981 arg_settings_mask
|= SETTING_START_MODE
;
985 if (arg_start_mode
== START_BOOT
)
986 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
987 "--boot and --as-pid2 may not be combined.");
989 arg_start_mode
= START_PID2
;
990 arg_settings_mask
|= SETTING_START_MODE
;
994 r
= id128_from_string_nonzero(optarg
, &arg_uuid
);
996 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
997 "Machine UUID may not be all zeroes.");
999 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
1001 arg_settings_mask
|= SETTING_MACHINE_ID
;
1005 _cleanup_free_
char *mangled
= NULL
;
1007 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1011 free_and_replace(arg_slice
, mangled
);
1012 arg_settings_mask
|= SETTING_SLICE
;
1017 if (isempty(optarg
))
1018 arg_machine
= mfree(arg_machine
);
1020 if (!hostname_is_valid(optarg
, 0))
1021 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1022 "Invalid machine name: %s", optarg
);
1024 r
= free_and_strdup(&arg_machine
, optarg
);
1031 if (isempty(optarg
))
1032 arg_hostname
= mfree(arg_hostname
);
1034 if (!hostname_is_valid(optarg
, 0))
1035 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1036 "Invalid hostname: %s", optarg
);
1038 r
= free_and_strdup(&arg_hostname
, optarg
);
1043 arg_settings_mask
|= SETTING_HOSTNAME
;
1047 arg_selinux_context
= optarg
;
1051 arg_selinux_apifs_context
= optarg
;
1055 arg_read_only
= true;
1056 arg_settings_mask
|= SETTING_READ_ONLY
;
1059 case ARG_AMBIENT_CAPABILITY
: {
1061 r
= parse_capability_spec(optarg
, &m
);
1064 arg_caps_ambient
|= m
;
1065 arg_settings_mask
|= SETTING_CAPABILITY
;
1068 case ARG_CAPABILITY
:
1069 case ARG_DROP_CAPABILITY
: {
1071 r
= parse_capability_spec(optarg
, &m
);
1075 if (c
== ARG_CAPABILITY
)
1079 arg_settings_mask
|= SETTING_CAPABILITY
;
1082 case ARG_NO_NEW_PRIVILEGES
:
1083 r
= parse_boolean(optarg
);
1085 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1087 arg_no_new_privileges
= r
;
1088 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1092 arg_link_journal
= LINK_GUEST
;
1093 arg_link_journal_try
= true;
1094 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1097 case ARG_LINK_JOURNAL
:
1098 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1100 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1102 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1107 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1109 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1111 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1115 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1117 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1119 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1123 case ARG_OVERLAY_RO
:
1124 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1125 if (r
== -EADDRNOTAVAIL
)
1126 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1128 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1130 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1133 case ARG_INACCESSIBLE
:
1134 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1136 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1138 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1142 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1144 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1146 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1153 case ARG_SHARE_SYSTEM
:
1154 /* We don't officially support this anymore, except for compat reasons. People should use the
1155 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1156 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1157 arg_clone_ns_flags
= 0;
1161 r
= parse_boolean(optarg
);
1163 log_error("Failed to parse --register= argument: %s", optarg
);
1171 arg_keep_unit
= true;
1174 case ARG_PERSONALITY
:
1176 arg_personality
= personality_from_string(optarg
);
1177 if (arg_personality
== PERSONALITY_INVALID
)
1178 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1179 "Unknown or unsupported personality '%s'.", optarg
);
1181 arg_settings_mask
|= SETTING_PERSONALITY
;
1187 arg_volatile_mode
= VOLATILE_YES
;
1188 else if (streq(optarg
, "help")) {
1189 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1194 m
= volatile_mode_from_string(optarg
);
1196 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1197 "Failed to parse --volatile= argument: %s", optarg
);
1199 arg_volatile_mode
= m
;
1202 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1206 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1208 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1210 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1212 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1216 if (strv_extend(&arg_property
, optarg
) < 0)
1221 case ARG_PRIVATE_USERS
: {
1226 else if (!in_charset(optarg
, DIGITS
))
1227 /* do *not* parse numbers as booleans */
1228 boolean
= parse_boolean(optarg
);
1233 /* no: User namespacing off */
1234 arg_userns_mode
= USER_NAMESPACE_NO
;
1235 arg_uid_shift
= UID_INVALID
;
1236 arg_uid_range
= UINT32_C(0x10000);
1237 } else if (boolean
> 0) {
1238 /* yes: User namespacing on, UID range is read from root dir */
1239 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1240 arg_uid_shift
= UID_INVALID
;
1241 arg_uid_range
= UINT32_C(0x10000);
1242 } else if (streq(optarg
, "pick")) {
1243 /* pick: User namespacing on, UID range is picked randomly */
1244 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1245 * implied by USER_NAMESPACE_PICK
1247 arg_uid_shift
= UID_INVALID
;
1248 arg_uid_range
= UINT32_C(0x10000);
1250 } else if (streq(optarg
, "identity")) {
1251 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1252 * itself, i.e. we don't actually map anything, but do take benefit of
1253 * isolation of capability sets. */
1254 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1256 arg_uid_range
= UINT32_C(0x10000);
1258 _cleanup_free_
char *buffer
= NULL
;
1259 const char *range
, *shift
;
1261 /* anything else: User namespacing on, UID range is explicitly configured */
1263 range
= strchr(optarg
, ':');
1265 buffer
= strndup(optarg
, range
- optarg
);
1271 r
= safe_atou32(range
, &arg_uid_range
);
1273 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1277 r
= parse_uid(shift
, &arg_uid_shift
);
1279 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1281 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1283 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1284 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1287 arg_settings_mask
|= SETTING_USERNS
;
1292 if (userns_supported()) {
1293 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1294 * implied by USER_NAMESPACE_PICK
1296 arg_uid_shift
= UID_INVALID
;
1297 arg_uid_range
= UINT32_C(0x10000);
1299 arg_settings_mask
|= SETTING_USERNS
;
1304 case ARG_PRIVATE_USERS_CHOWN
:
1305 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1307 arg_settings_mask
|= SETTING_USERNS
;
1310 case ARG_PRIVATE_USERS_OWNERSHIP
:
1311 if (streq(optarg
, "help")) {
1312 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1316 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1317 if (arg_userns_ownership
< 0)
1318 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1320 arg_settings_mask
|= SETTING_USERNS
;
1323 case ARG_KILL_SIGNAL
:
1324 if (streq(optarg
, "help")) {
1325 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1329 arg_kill_signal
= signal_from_string(optarg
);
1330 if (arg_kill_signal
< 0)
1331 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1333 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1338 /* no → do not read files
1339 * yes → read files, do not override cmdline, trust only subset
1340 * override → read files, override cmdline, trust only subset
1341 * trusted → read files, do not override cmdline, trust all
1344 r
= parse_boolean(optarg
);
1346 if (streq(optarg
, "trusted")) {
1347 mask_all_settings
= false;
1348 mask_no_settings
= false;
1349 arg_settings_trusted
= true;
1351 } else if (streq(optarg
, "override")) {
1352 mask_all_settings
= false;
1353 mask_no_settings
= true;
1354 arg_settings_trusted
= -1;
1356 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1359 mask_all_settings
= false;
1360 mask_no_settings
= false;
1361 arg_settings_trusted
= -1;
1364 mask_all_settings
= true;
1365 mask_no_settings
= false;
1366 arg_settings_trusted
= false;
1372 if (!path_is_absolute(optarg
))
1373 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1374 "Working directory %s is not an absolute path.", optarg
);
1376 r
= free_and_strdup(&arg_chdir
, optarg
);
1380 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1383 case ARG_PIVOT_ROOT
:
1384 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1386 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1388 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1391 case ARG_NOTIFY_READY
:
1392 r
= parse_boolean(optarg
);
1394 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1395 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1396 arg_notify_ready
= r
;
1397 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1400 case ARG_ROOT_HASH
: {
1401 _cleanup_free_
void *k
= NULL
;
1404 r
= unhexmem(optarg
, &k
, &l
);
1406 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1407 if (l
< sizeof(sd_id128_t
))
1408 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128-bit long: %s", optarg
);
1410 free_and_replace(arg_verity_settings
.root_hash
, k
);
1411 arg_verity_settings
.root_hash_size
= l
;
1415 case ARG_ROOT_HASH_SIG
: {
1420 if ((value
= startswith(optarg
, "base64:"))) {
1421 r
= unbase64mem(value
, &p
, &l
);
1423 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1426 r
= read_full_file(optarg
, (char**) &p
, &l
);
1428 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1431 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1432 arg_verity_settings
.root_hash_sig_size
= l
;
1436 case ARG_VERITY_DATA
:
1437 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1442 case ARG_SYSTEM_CALL_FILTER
: {
1446 negative
= optarg
[0] == '~';
1447 items
= negative
? optarg
+ 1 : optarg
;
1450 _cleanup_free_
char *word
= NULL
;
1452 r
= extract_first_word(&items
, &word
, NULL
, 0);
1458 return log_error_errno(r
, "Failed to parse system call filter: %m");
1461 r
= strv_extend(&arg_syscall_deny_list
, word
);
1463 r
= strv_extend(&arg_syscall_allow_list
, word
);
1468 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1474 _cleanup_free_
char *name
= NULL
;
1477 if (streq(optarg
, "help")) {
1478 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1482 eq
= strchr(optarg
, '=');
1484 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1485 "--rlimit= expects an '=' assignment.");
1487 name
= strndup(optarg
, eq
- optarg
);
1491 rl
= rlimit_from_string_harder(name
);
1493 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1495 if (!arg_rlimit
[rl
]) {
1496 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1497 if (!arg_rlimit
[rl
])
1501 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1503 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1505 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1509 case ARG_OOM_SCORE_ADJUST
:
1510 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1512 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1514 arg_oom_score_adjust_set
= true;
1515 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1518 case ARG_CPU_AFFINITY
: {
1521 r
= parse_cpu_set(optarg
, &cpuset
);
1523 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1525 cpu_set_reset(&arg_cpu_set
);
1526 arg_cpu_set
= cpuset
;
1527 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1531 case ARG_RESOLV_CONF
:
1532 if (streq(optarg
, "help")) {
1533 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1537 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1538 if (arg_resolv_conf
< 0)
1539 return log_error_errno(arg_resolv_conf
,
1540 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1542 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1546 if (streq(optarg
, "help")) {
1547 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1551 arg_timezone
= timezone_mode_from_string(optarg
);
1552 if (arg_timezone
< 0)
1553 return log_error_errno(arg_timezone
,
1554 "Failed to parse /etc/localtime mode: %s", optarg
);
1556 arg_settings_mask
|= SETTING_TIMEZONE
;
1560 r
= handle_arg_console(optarg
);
1567 r
= handle_arg_console("pipe");
1573 arg_pager_flags
|= PAGER_DISABLE
;
1576 case ARG_SET_CREDENTIAL
:
1577 r
= machine_credential_set(&arg_credentials
, optarg
);
1581 arg_settings_mask
|= SETTING_CREDENTIALS
;
1584 case ARG_LOAD_CREDENTIAL
:
1585 r
= machine_credential_load(&arg_credentials
, optarg
);
1589 arg_settings_mask
|= SETTING_CREDENTIALS
;
1593 if (!valid_user_group_name(optarg
, 0))
1594 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1596 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1599 arg_settings_mask
|= SETTING_BIND_USER
;
1602 case ARG_SUPPRESS_SYNC
:
1603 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1607 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1610 case ARG_IMAGE_POLICY
:
1611 r
= parse_image_policy_argument(optarg
, &arg_image_policy
);
1616 case ARG_BACKGROUND
:
1617 r
= free_and_strdup_warn(&arg_background
, optarg
);
1626 assert_not_reached();
1629 if (argc
> optind
) {
1630 strv_free(arg_parameters
);
1631 arg_parameters
= strv_copy(argv
+ optind
);
1632 if (!arg_parameters
)
1635 arg_settings_mask
|= SETTING_START_MODE
;
1638 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1639 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1640 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1641 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1643 arg_directory
= TAKE_PTR(arg_template
);
1645 arg_caps_retain
|= plus
;
1646 arg_caps_retain
|= arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0;
1647 arg_caps_retain
&= ~minus
;
1649 /* Make sure to parse environment before we reset the settings mask below */
1650 r
= parse_environment();
1654 /* Load all settings from .nspawn files */
1655 if (mask_no_settings
)
1656 arg_settings_mask
= 0;
1658 /* Don't load any settings from .nspawn files */
1659 if (mask_all_settings
)
1660 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1665 static int verify_arguments(void) {
1668 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1669 /* If we are running the stub init in the container, we don't need to look at what the init
1670 * in the container supports, because we are not using it. Let's immediately pick the right
1671 * setting based on the host system configuration.
1673 * We only do this, if the user didn't use an environment variable to override the detection.
1676 r
= cg_all_unified();
1678 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1680 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1681 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1682 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1684 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1687 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1688 arg_mount_settings
|= MOUNT_USE_USERNS
;
1690 if (arg_private_network
)
1691 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1693 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1694 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1695 arg_register
= false;
1696 if (arg_start_mode
!= START_PID1
)
1697 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1700 if (arg_userns_ownership
< 0)
1701 arg_userns_ownership
=
1702 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1703 USER_NAMESPACE_OWNERSHIP_OFF
;
1705 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1706 arg_kill_signal
= SIGRTMIN
+3;
1708 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 */
1709 arg_read_only
= true;
1711 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1712 arg_read_only
= true;
1714 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1715 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1716 * The latter is not technically a user session, but we don't need to labour the point. */
1717 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1719 if (arg_directory
&& arg_image
)
1720 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1722 if (arg_template
&& arg_image
)
1723 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1725 if (arg_template
&& !(arg_directory
|| arg_machine
))
1726 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1728 if (arg_ephemeral
&& arg_template
)
1729 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1731 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1732 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1734 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1735 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1737 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1738 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1739 "--read-only and --private-users-ownership=chown may not be combined.");
1741 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1742 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1743 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1744 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1745 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1747 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1748 * we need to error out, to avoid conflicts between different network options. */
1749 if (arg_network_namespace_path
&&
1750 (arg_network_interfaces
|| arg_network_macvlan
||
1751 arg_network_ipvlan
|| arg_network_veth_extra
||
1752 arg_network_bridge
|| arg_network_zone
||
1754 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1756 if (arg_network_bridge
&& arg_network_zone
)
1757 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1758 "--network-bridge= and --network-zone= may not be combined.");
1760 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1761 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1763 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1764 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1766 if (arg_expose_ports
&& !arg_private_network
)
1767 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1769 if (arg_caps_ambient
) {
1770 if (arg_caps_ambient
== UINT64_MAX
)
1771 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1773 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1774 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1776 if (arg_start_mode
== START_BOOT
)
1777 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1780 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1781 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1783 /* Drop duplicate --bind-user= entries */
1784 strv_uniq(arg_bind_user
);
1786 r
= custom_mount_check_all();
1793 static int verify_network_interfaces_initialized(void) {
1795 r
= test_network_interfaces_initialized(arg_network_interfaces
);
1799 r
= test_network_interfaces_initialized(arg_network_macvlan
);
1803 r
= test_network_interfaces_initialized(arg_network_ipvlan
);
1810 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1813 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1816 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1819 if (uid
!= UID_INVALID
) {
1820 uid
+= arg_uid_shift
;
1822 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1826 if (gid
!= GID_INVALID
) {
1827 gid
+= (gid_t
) arg_uid_shift
;
1829 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1833 return RET_NERRNO(lchown(p
, uid
, gid
));
1836 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1840 q
= prefix_roota(root
, path
);
1841 r
= RET_NERRNO(mkdir(q
, mode
));
1847 return userns_lchown(q
, uid
, gid
);
1850 static const char *timezone_from_path(const char *path
) {
1851 return PATH_STARTSWITH_SET(
1853 "../usr/share/zoneinfo/",
1854 "/usr/share/zoneinfo/");
1857 static bool etc_writable(void) {
1858 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1861 static int setup_timezone(const char *dest
) {
1862 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1863 const char *where
, *check
;
1869 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1870 r
= readlink_malloc("/etc/localtime", &p
);
1871 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1872 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1873 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1874 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1876 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1877 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1881 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1884 } else if (arg_timezone
== TIMEZONE_AUTO
)
1885 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1891 if (m
== TIMEZONE_OFF
)
1894 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1896 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1900 where
= strjoina(etc
, "/localtime");
1904 case TIMEZONE_DELETE
:
1905 if (unlink(where
) < 0)
1906 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1910 case TIMEZONE_SYMLINK
: {
1911 _cleanup_free_
char *q
= NULL
;
1912 const char *z
, *what
;
1914 z
= timezone_from_path(p
);
1916 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1920 r
= readlink_malloc(where
, &q
);
1921 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1922 return 0; /* Already pointing to the right place? Then do nothing .. */
1924 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1925 r
= chase(check
, dest
, 0, NULL
, NULL
);
1927 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1929 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1930 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1931 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1935 what
= strjoina("../usr/share/zoneinfo/", z
);
1936 if (symlink(what
, where
) < 0) {
1937 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1938 errno
, "Failed to correct timezone of container, ignoring: %m");
1948 case TIMEZONE_BIND
: {
1949 _cleanup_free_
char *resolved
= NULL
;
1952 found
= chase(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1954 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1958 if (found
== 0) /* missing? */
1959 (void) touch(resolved
);
1961 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1963 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1969 /* If mounting failed, try to copy */
1970 r
= copy_file_atomic("/etc/localtime", where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
1972 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1973 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1980 assert_not_reached();
1983 /* Fix permissions of the symlink or file copy we just created */
1984 r
= userns_lchown(where
, 0, 0);
1986 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1991 static int have_resolv_conf(const char *path
) {
1994 if (access(path
, F_OK
) < 0) {
1995 if (errno
== ENOENT
)
1998 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
2004 static int resolved_listening(void) {
2005 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2006 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
2007 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
2010 /* Check if resolved is listening */
2012 r
= sd_bus_open_system(&bus
);
2014 return log_debug_errno(r
, "Failed to open system bus: %m");
2016 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2018 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2022 r
= bus_get_property_string(bus
, bus_resolve_mgr
, "DNSStubListener", &error
, &dns_stub_listener_mode
);
2024 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2026 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2029 static int setup_resolv_conf(const char *dest
) {
2030 _cleanup_free_
char *etc
= NULL
;
2031 const char *where
, *what
;
2037 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2038 if (arg_private_network
)
2039 m
= RESOLV_CONF_OFF
;
2040 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2041 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2042 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2043 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2045 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2048 m
= arg_resolv_conf
;
2050 if (m
== RESOLV_CONF_OFF
)
2053 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2055 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2059 where
= strjoina(etc
, "/resolv.conf");
2061 if (m
== RESOLV_CONF_DELETE
) {
2062 if (unlink(where
) < 0)
2063 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2068 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2069 what
= PRIVATE_STATIC_RESOLV_CONF
;
2070 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2071 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2072 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2073 what
= PRIVATE_STUB_RESOLV_CONF
;
2075 what
= "/etc/resolv.conf";
2077 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2078 _cleanup_free_
char *resolved
= NULL
;
2081 found
= chase(where
, dest
, CHASE_NONEXISTENT
|CHASE_NOFOLLOW
, &resolved
, NULL
);
2083 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2087 if (found
== 0) /* missing? */
2088 (void) touch(resolved
);
2090 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2092 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2094 /* If that didn't work, let's copy the file */
2097 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2098 r
= copy_file_atomic(what
, where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2100 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, COPY_REFLINK
);
2102 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2103 * resolved or something similar runs inside and the symlink points there.
2105 * If the disk image is read-only, there's also no point in complaining.
2107 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2108 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2109 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2113 r
= userns_lchown(where
, 0, 0);
2115 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2120 static int setup_boot_id(void) {
2121 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2122 _cleanup_free_
char *path
= NULL
;
2123 sd_id128_t rnd
= SD_ID128_NULL
;
2127 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2129 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2131 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2133 r
= sd_id128_randomize(&rnd
);
2135 return log_error_errno(r
, "Failed to generate random boot id: %m");
2137 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2139 return log_error_errno(r
, "Failed to write boot id: %m");
2141 from
= TAKE_PTR(path
);
2142 to
= "/proc/sys/kernel/random/boot_id";
2144 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2148 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2151 static int copy_devnodes(const char *dest
) {
2152 static const char devnodes
[] =
2165 BLOCK_WITH_UMASK(0000);
2167 /* Create /dev/net, so that we can create /dev/net/tun in it */
2168 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2169 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2171 NULSTR_FOREACH(d
, devnodes
) {
2172 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2175 from
= path_join("/dev/", d
);
2179 to
= path_join(dest
, from
);
2183 if (stat(from
, &st
) < 0) {
2185 if (errno
!= ENOENT
)
2186 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2188 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2189 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2190 "%s is not a char or block device, cannot copy.", from
);
2192 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2194 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2195 /* Explicitly warn the user when /dev is already populated. */
2196 if (errno
== EEXIST
)
2197 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
);
2199 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2201 /* Some systems abusively restrict mknod but allow bind mounts. */
2204 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2205 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2207 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2210 r
= userns_lchown(to
, 0, 0);
2212 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2214 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2218 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2220 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2222 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2225 prefixed
= path_join(dest
, sl
);
2229 t
= path_join("..", d
);
2233 if (symlink(t
, prefixed
) < 0)
2234 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2241 static int make_extra_nodes(const char *dest
) {
2245 BLOCK_WITH_UMASK(0000);
2247 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2248 _cleanup_free_
char *path
= NULL
;
2249 DeviceNode
*n
= arg_extra_nodes
+ i
;
2251 path
= path_join(dest
, n
->path
);
2255 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2256 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2258 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2260 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2266 static int setup_pts(const char *dest
) {
2267 _cleanup_free_
char *options
= NULL
;
2272 if (arg_selinux_apifs_context
)
2273 (void) asprintf(&options
,
2274 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2275 arg_uid_shift
+ TTY_GID
,
2276 arg_selinux_apifs_context
);
2279 (void) asprintf(&options
,
2280 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2281 arg_uid_shift
+ TTY_GID
);
2286 /* Mount /dev/pts itself */
2287 p
= prefix_roota(dest
, "/dev/pts");
2288 r
= RET_NERRNO(mkdir(p
, 0755));
2290 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2292 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2295 r
= userns_lchown(p
, 0, 0);
2297 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2299 /* Create /dev/ptmx symlink */
2300 p
= prefix_roota(dest
, "/dev/ptmx");
2301 if (symlink("pts/ptmx", p
) < 0)
2302 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2303 r
= userns_lchown(p
, 0, 0);
2305 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2307 /* And fix /dev/pts/ptmx ownership */
2308 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2309 r
= userns_lchown(p
, 0, 0);
2311 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2316 static int setup_stdio_as_dev_console(void) {
2317 _cleanup_close_
int terminal
= -EBADF
;
2320 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2321 * explicitly, if we are configured to. */
2322 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2324 return log_error_errno(terminal
, "Failed to open console: %m");
2326 /* Make sure we can continue logging to the original stderr, even if
2327 * stderr points elsewhere now */
2328 r
= log_dup_console();
2330 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2332 /* invalidates 'terminal' on success and failure */
2333 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2336 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2341 static int setup_dev_console(const char *console
) {
2342 _cleanup_free_
char *p
= NULL
;
2345 /* Create /dev/console symlink */
2346 r
= path_make_relative("/dev", console
, &p
);
2348 return log_error_errno(r
, "Failed to create relative path: %m");
2350 if (symlink(p
, "/dev/console") < 0)
2351 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2356 static int setup_keyring(void) {
2357 key_serial_t keyring
;
2359 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2360 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2361 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2362 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2363 * into the container. */
2365 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2366 if (keyring
== -1) {
2367 if (errno
== ENOSYS
)
2368 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2369 else if (ERRNO_IS_PRIVILEGE(errno
))
2370 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2372 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2378 int make_run_host(const char *root
) {
2383 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2385 return log_error_errno(r
, "Failed to create /run/host/: %m");
2390 static int setup_credentials(const char *root
) {
2394 if (arg_credentials
.n_credentials
== 0)
2397 r
= make_run_host(root
);
2401 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2403 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2405 q
= prefix_roota(root
, "/run/host/credentials");
2406 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2410 FOREACH_ARRAY(cred
, arg_credentials
.credentials
, arg_credentials
.n_credentials
) {
2411 _cleanup_free_
char *j
= NULL
;
2412 _cleanup_close_
int fd
= -EBADF
;
2414 j
= path_join(q
, cred
->id
);
2418 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2420 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2422 r
= loop_write(fd
, cred
->data
, cred
->size
);
2424 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2426 if (fchmod(fd
, 0400) < 0)
2427 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2429 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2430 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2431 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2435 if (chmod(q
, 0500) < 0)
2436 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2438 r
= userns_lchown(q
, 0, 0);
2442 /* Make both mount and superblock read-only now */
2443 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2447 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2450 static int setup_kmsg(int fd_inner_socket
) {
2451 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2452 _cleanup_free_
char *fifo
= NULL
;
2453 _cleanup_close_
int fd
= -EBADF
;
2456 assert(fd_inner_socket
>= 0);
2458 BLOCK_WITH_UMASK(0000);
2460 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2461 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2462 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2463 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2465 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2467 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2469 if (mkfifo(fifo
, 0600) < 0)
2470 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2472 from
= TAKE_PTR(fifo
);
2474 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2478 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2480 return log_error_errno(errno
, "Failed to open fifo: %m");
2482 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2483 r
= send_one_fd(fd_inner_socket
, fd
, 0);
2485 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2491 union in_addr_union address4
;
2492 union in_addr_union address6
;
2493 struct FirewallContext
*fw_ctx
;
2496 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2497 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2502 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2503 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2507 static int setup_hostname(void) {
2510 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2513 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2515 return log_error_errno(r
, "Failed to set hostname: %m");
2520 static int setup_journal(const char *directory
) {
2521 _cleanup_free_
char *d
= NULL
;
2527 /* Don't link journals in ephemeral mode */
2531 if (arg_link_journal
== LINK_NO
)
2534 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2536 r
= sd_id128_get_machine(&this_id
);
2538 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2540 if (sd_id128_equal(arg_uuid
, this_id
)) {
2541 log_full(try ? LOG_WARNING
: LOG_ERR
,
2542 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2548 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2549 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2551 bool ignore
= r
== -EROFS
&& try;
2552 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2553 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2554 return ignore
? 0 : r
;
2558 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2559 q
= prefix_roota(directory
, p
);
2561 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2565 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2566 "%s: already a mount point, refusing to use for journal", p
);
2569 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2573 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2574 "%s: already a mount point, refusing to use for journal", q
);
2577 r
= readlink_and_make_absolute(p
, &d
);
2579 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2582 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2584 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2589 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2590 } else if (r
== -EINVAL
) {
2592 if (arg_link_journal
== LINK_GUEST
&&
2595 if (errno
== ENOTDIR
) {
2596 log_error("%s already exists and is neither a symlink nor a directory", p
);
2599 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2601 } else if (r
!= -ENOENT
)
2602 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2604 if (arg_link_journal
== LINK_GUEST
) {
2606 if (symlink(q
, p
) < 0) {
2608 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2611 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2614 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2616 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2620 if (arg_link_journal
== LINK_HOST
) {
2621 /* don't create parents here — if the host doesn't have
2622 * permanent journal set up, don't force it here */
2624 r
= RET_NERRNO(mkdir(p
, 0755));
2625 if (r
< 0 && r
!= -EEXIST
) {
2627 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2630 return log_error_errno(r
, "Failed to create %s: %m", p
);
2633 } else if (access(p
, F_OK
) < 0)
2636 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2637 log_warning("%s is not empty, proceeding anyway.", q
);
2639 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2641 return log_error_errno(r
, "Failed to create %s: %m", q
);
2643 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2645 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2650 static int drop_capabilities(uid_t uid
) {
2651 CapabilityQuintet q
;
2653 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2654 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2655 * arg_caps_retain. */
2657 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2658 q
= arg_full_capabilities
;
2660 if (q
.bounding
== UINT64_MAX
)
2661 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2663 if (q
.effective
== UINT64_MAX
)
2664 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2666 if (q
.inheritable
== UINT64_MAX
)
2667 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2669 if (q
.permitted
== UINT64_MAX
)
2670 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2672 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2673 q
.ambient
= arg_caps_ambient
;
2675 if (capability_quintet_mangle(&q
))
2676 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2679 q
= (CapabilityQuintet
) {
2680 .bounding
= arg_caps_retain
,
2681 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2682 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2683 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2684 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2687 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2688 * in order to maintain the same behavior as systemd < 242. */
2689 if (capability_quintet_mangle(&q
))
2690 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2691 "Some capabilities will not be set because they are not in the current bounding set.");
2695 return capability_quintet_enforce(&q
);
2698 static int reset_audit_loginuid(void) {
2699 _cleanup_free_
char *p
= NULL
;
2702 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2705 r
= read_one_line_file("/proc/self/loginuid", &p
);
2709 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2711 /* Already reset? */
2712 if (streq(p
, "4294967295"))
2715 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2718 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2719 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2720 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2721 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2722 "using systemd-nspawn. Sleeping for 5s... (%m)");
2730 static int mount_tunnel_dig(const char *root
) {
2734 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2735 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2736 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2737 (void) mkdir_p(p
, 0600);
2739 r
= make_run_host(root
);
2743 r
= userns_mkdir(root
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2745 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2747 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
2748 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2752 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2759 static int mount_tunnel_open(void) {
2762 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2769 static int setup_machine_id(const char *directory
) {
2772 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2773 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2774 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2775 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2776 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2777 * container behaves nicely). */
2779 r
= id128_get_machine(directory
, &arg_uuid
);
2780 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) {
2781 /* If the file is missing, empty, or uninitialized, we don't mind */
2782 if (sd_id128_is_null(arg_uuid
)) {
2783 r
= sd_id128_randomize(&arg_uuid
);
2785 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2788 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2793 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2798 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2801 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2802 if (r
== -EOPNOTSUPP
)
2803 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.");
2805 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2807 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2809 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2811 log_debug("Patched directory tree to match UID/GID range.");
2818 * < 0 : wait_for_terminate() failed to get the state of the
2819 * container, the container was terminated by a signal, or
2820 * failed for an unknown reason. No change is made to the
2821 * container argument.
2822 * > 0 : The program executed in the container terminated with an
2823 * error. The exit code of the program executed in the
2824 * container is returned. The container argument has been set
2825 * to CONTAINER_TERMINATED.
2826 * 0 : The container is being rebooted, has been shut down or exited
2827 * successfully. The container argument has been set to either
2828 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2830 * That is, success is indicated by a return value of zero, and an
2831 * error is indicated by a non-zero value.
2833 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2837 r
= wait_for_terminate(pid
, &status
);
2839 return log_warning_errno(r
, "Failed to wait for container: %m");
2841 switch (status
.si_code
) {
2844 if (status
.si_status
== 0)
2845 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2847 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2849 *container
= CONTAINER_TERMINATED
;
2850 return status
.si_status
;
2853 if (status
.si_status
== SIGINT
) {
2854 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2855 *container
= CONTAINER_TERMINATED
;
2858 } else if (status
.si_status
== SIGHUP
) {
2859 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2860 *container
= CONTAINER_REBOOTED
;
2866 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2867 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2870 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2871 "Container %s failed due to unknown reason.", arg_machine
);
2875 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2878 pid
= PTR_TO_PID(userdata
);
2880 if (kill(pid
, arg_kill_signal
) >= 0) {
2881 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2882 sd_event_source_set_userdata(s
, NULL
);
2887 sd_event_exit(sd_event_source_get_event(s
), 0);
2891 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2897 pid
= PTR_TO_PID(userdata
);
2902 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2903 return log_error_errno(errno
, "Failed to waitid(): %m");
2904 if (si
.si_pid
== 0) /* No pending children. */
2906 if (si
.si_pid
== pid
) {
2907 /* The main process we care for has exited. Return from
2908 * signal handler but leave the zombie. */
2909 sd_event_exit(sd_event_source_get_event(s
), 0);
2913 /* Reap all other children. */
2914 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2920 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2925 pid
= PTR_TO_PID(userdata
);
2927 if (arg_kill_signal
> 0) {
2928 log_info("Container termination requested. Attempting to halt container.");
2929 (void) kill(pid
, arg_kill_signal
);
2931 log_info("Container termination requested. Exiting.");
2932 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2938 static int pick_paths(void) {
2941 if (arg_directory
) {
2942 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2943 PickFilter filter
= pick_filter_image_dir
;
2945 filter
.architecture
= arg_architecture
;
2947 r
= path_pick_update_warn(
2950 PICK_ARCHITECTURE
|PICK_TRIES
,
2953 /* Accept ENOENT here so that the --template= logic can work */
2957 arg_architecture
= result
.architecture
;
2961 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2962 PickFilter filter
= pick_filter_image_raw
;
2964 filter
.architecture
= arg_architecture
;
2966 r
= path_pick_update_warn(
2969 PICK_ARCHITECTURE
|PICK_TRIES
,
2974 arg_architecture
= result
.architecture
;
2978 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2979 PickFilter filter
= pick_filter_image_dir
;
2981 filter
.architecture
= arg_architecture
;
2983 r
= path_pick_update_warn(
2991 arg_architecture
= result
.architecture
;
2997 static int determine_names(void) {
3000 if (arg_template
&& !arg_directory
&& arg_machine
) {
3002 /* If --template= was specified then we should not search for a machine, but instead create a
3003 * new one in /var/lib/machine. */
3005 arg_directory
= path_join("/var/lib/machines", arg_machine
);
3010 if (!arg_image
&& !arg_directory
) {
3012 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3014 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3016 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3018 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3020 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3021 r
= free_and_strdup(&arg_image
, i
->path
);
3023 r
= free_and_strdup(&arg_directory
, i
->path
);
3028 arg_read_only
= arg_read_only
|| i
->read_only
;
3030 r
= safe_getcwd(&arg_directory
);
3032 return log_error_errno(r
, "Failed to determine current directory: %m");
3035 if (!arg_directory
&& !arg_image
)
3036 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3040 if (arg_directory
&& path_equal(arg_directory
, "/"))
3041 arg_machine
= gethostname_malloc();
3042 else if (arg_image
) {
3045 r
= path_extract_filename(arg_image
, &arg_machine
);
3047 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_image
);
3049 /* Truncate suffix if there is one */
3050 e
= endswith(arg_machine
, ".raw");
3054 r
= path_extract_filename(arg_directory
, &arg_machine
);
3056 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_directory
);
3059 hostname_cleanup(arg_machine
);
3060 if (!hostname_is_valid(arg_machine
, 0))
3061 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3063 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3064 * to match fixed config file names. */
3065 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3066 if (!arg_settings_filename
)
3069 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3070 * instances at once without manually having to specify -M each time. */
3072 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3075 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3076 if (!arg_settings_filename
)
3083 static int chase_and_update(char **p
, unsigned flags
) {
3092 r
= chase(*p
, NULL
, flags
, &chased
, NULL
);
3094 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3096 return free_and_replace(*p
, chased
);
3099 static int determine_uid_shift(const char *directory
) {
3101 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3106 if (arg_uid_shift
== UID_INVALID
) {
3109 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3111 if (stat(directory
, &st
) < 0)
3112 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3114 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3116 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3117 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3118 "UID and GID base of %s don't match.", directory
);
3120 arg_uid_range
= UINT32_C(0x10000);
3122 if (arg_uid_shift
!= 0) {
3123 /* If the image is shifted already, then we'll fall back to classic chowning, for
3124 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3126 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3127 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3128 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3129 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3130 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3131 "UID base of %s is not zero, UID mapping not supported.", directory
);
3135 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3136 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3141 static unsigned long effective_clone_ns_flags(void) {
3142 unsigned long flags
= arg_clone_ns_flags
;
3144 if (arg_private_network
)
3145 flags
|= CLONE_NEWNET
;
3147 flags
|= CLONE_NEWCGROUP
;
3148 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3149 flags
|= CLONE_NEWUSER
;
3154 static int patch_sysctl(void) {
3156 /* This table is inspired by runc's sysctl() function */
3157 static const struct {
3160 unsigned long clone_flags
;
3162 { "kernel.hostname", false, CLONE_NEWUTS
},
3163 { "kernel.domainname", false, CLONE_NEWUTS
},
3164 { "kernel.msgmax", false, CLONE_NEWIPC
},
3165 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3166 { "kernel.msgmni", false, CLONE_NEWIPC
},
3167 { "kernel.sem", false, CLONE_NEWIPC
},
3168 { "kernel.shmall", false, CLONE_NEWIPC
},
3169 { "kernel.shmmax", false, CLONE_NEWIPC
},
3170 { "kernel.shmmni", false, CLONE_NEWIPC
},
3171 { "fs.mqueue.", true, CLONE_NEWIPC
},
3172 { "net.", true, CLONE_NEWNET
},
3175 unsigned long flags
;
3178 flags
= effective_clone_ns_flags();
3180 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3184 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3186 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3189 if (safe_sysctl
[i
].prefix
)
3190 good
= startswith(*k
, safe_sysctl
[i
].key
);
3192 good
= streq(*k
, safe_sysctl
[i
].key
);
3199 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3201 r
= sysctl_write(*k
, *v
);
3203 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3209 static int inner_child(
3211 int fd_inner_socket
,
3213 char **os_release_pairs
) {
3215 _cleanup_free_
char *home
= NULL
;
3218 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3219 NULL
, /* container */
3224 NULL
, /* container_uuid */
3225 NULL
, /* LISTEN_FDS */
3226 NULL
, /* LISTEN_PID */
3227 NULL
, /* NOTIFY_SOCKET */
3228 NULL
, /* CREDENTIALS_DIRECTORY */
3232 const char *exec_target
;
3233 _cleanup_strv_free_
char **env_use
= NULL
;
3234 int r
, which_failed
;
3236 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3237 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3238 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3239 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3240 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3241 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3244 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3245 * unshare(). See below. */
3248 assert(fd_inner_socket
>= 0);
3250 log_debug("Inner child is initializing.");
3252 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3253 /* Tell the parent, that it now can write the UID map. */
3254 (void) barrier_place(barrier
); /* #1 */
3256 /* Wait until the parent wrote the UID map */
3257 if (!barrier_place_and_sync(barrier
)) /* #2 */
3258 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3260 /* Become the new root user inside our namespace */
3261 r
= reset_uid_gid();
3263 return log_error_errno(r
, "Couldn't become new root: %m");
3265 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3266 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3267 * propagation, but simply create new peer groups for all our mounts). */
3268 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3274 arg_mount_settings
| MOUNT_IN_USERNS
,
3276 arg_selinux_apifs_context
);
3280 if (!arg_network_namespace_path
&& arg_private_network
) {
3281 r
= unshare(CLONE_NEWNET
);
3283 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3285 /* Tell the parent that it can setup network interfaces. */
3286 (void) barrier_place(barrier
); /* #3 */
3289 r
= mount_sysfs(NULL
, arg_mount_settings
);
3293 /* Wait until we are cgroup-ified, so that we
3294 * can mount the right cgroup path writable */
3295 if (!barrier_place_and_sync(barrier
)) /* #4 */
3296 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3297 "Parent died too early");
3300 r
= unshare(CLONE_NEWCGROUP
);
3302 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3305 arg_unified_cgroup_hierarchy
,
3306 arg_userns_mode
!= USER_NAMESPACE_NO
,
3309 arg_selinux_apifs_context
,
3312 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3316 r
= setup_boot_id();
3320 r
= setup_kmsg(fd_inner_socket
);
3327 arg_n_custom_mounts
,
3330 arg_selinux_apifs_context
,
3331 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3336 return log_error_errno(errno
, "setsid() failed: %m");
3338 if (arg_private_network
)
3339 (void) loopback_setup();
3341 if (arg_expose_ports
) {
3342 r
= expose_port_send_rtnl(fd_inner_socket
);
3347 if (arg_console_mode
!= CONSOLE_PIPE
) {
3348 _cleanup_close_
int master
= -EBADF
;
3349 _cleanup_free_
char *console
= NULL
;
3351 /* Allocate a pty and make it available as /dev/console. */
3352 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3354 return log_error_errno(master
, "Failed to allocate a pty: %m");
3356 r
= setup_dev_console(console
);
3358 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3360 r
= send_one_fd(fd_inner_socket
, master
, 0);
3362 return log_error_errno(r
, "Failed to send master fd: %m");
3364 r
= setup_stdio_as_dev_console();
3373 if (arg_oom_score_adjust_set
) {
3374 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3376 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3379 if (arg_cpu_set
.set
)
3380 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3381 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3383 (void) setup_hostname();
3385 if (arg_personality
!= PERSONALITY_INVALID
) {
3386 r
= safe_personality(arg_personality
);
3388 return log_error_errno(r
, "personality() failed: %m");
3389 #ifdef ARCHITECTURE_SECONDARY
3390 } else if (arg_architecture
== ARCHITECTURE_SECONDARY
) {
3391 r
= safe_personality(PER_LINUX32
);
3393 return log_error_errno(r
, "personality() failed: %m");
3395 } else if (!arg_quiet
&& arg_architecture
>= 0 && arg_architecture
!= native_architecture())
3396 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3397 "invocation with qemu userspace emulator (or equivalent) in effect.",
3398 architecture_to_string(arg_architecture
));
3400 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3402 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3407 if (is_seccomp_available()) {
3408 r
= seccomp_load(arg_seccomp
);
3409 if (ERRNO_IS_NEG_SECCOMP_FATAL(r
))
3410 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3412 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3417 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3422 if (arg_suppress_sync
) {
3424 r
= seccomp_suppress_sync();
3426 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3428 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3433 if (arg_selinux_context
)
3434 if (setexeccon(arg_selinux_context
) < 0)
3435 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3438 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3439 * if we need to later on. */
3440 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3441 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3443 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3444 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3446 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3450 r
= drop_capabilities(getuid());
3452 return log_error_errno(r
, "Dropping capabilities failed: %m");
3454 if (arg_no_new_privileges
)
3455 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3456 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3458 /* LXC sets container=lxc, so follow the scheme here */
3459 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3461 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3465 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3466 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3469 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3470 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3471 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
?: "root") < 0)
3474 assert(!sd_id128_is_null(arg_uuid
));
3476 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3479 if (!fdset_isempty(fds
)) {
3480 r
= fdset_cloexec(fds
, false);
3482 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3484 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3485 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3488 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3491 if (arg_credentials
.n_credentials
> 0) {
3492 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3498 if (arg_start_mode
!= START_BOOT
) {
3499 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3505 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3509 /* Let the parent know that we are ready and wait until the parent is ready with the setup, too... */
3510 if (!barrier_place_and_sync(barrier
)) /* #5 */
3511 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3514 if (chdir(arg_chdir
) < 0)
3515 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3517 if (arg_start_mode
== START_PID2
) {
3518 r
= stub_pid1(arg_uuid
);
3523 if (arg_console_mode
!= CONSOLE_PIPE
) {
3524 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3525 * are configured for that. Acquire it as controlling tty. */
3526 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3527 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3530 log_debug("Inner child completed, invoking payload.");
3532 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3533 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3534 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3536 log_set_open_when_needed(true);
3537 log_settle_target();
3539 (void) fdset_close_others(fds
);
3541 if (arg_start_mode
== START_BOOT
) {
3545 /* Automatically search for the init system */
3547 m
= strv_length(arg_parameters
);
3548 a
= newa(char*, m
+ 2);
3549 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3552 FOREACH_STRING(init
,
3553 "/usr/lib/systemd/systemd",
3554 "/lib/systemd/systemd",
3556 a
[0] = (char*) init
;
3557 execve(a
[0], a
, env_use
);
3560 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3561 } else if (!strv_isempty(arg_parameters
)) {
3562 const char *dollar_path
;
3564 exec_target
= arg_parameters
[0];
3566 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3568 dollar_path
= strv_env_get(env_use
, "PATH");
3570 if (setenv("PATH", dollar_path
, 1) < 0)
3571 return log_error_errno(errno
, "Failed to update $PATH: %m");
3574 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3577 /* If we cannot change the directory, we'll end up in /, that is expected. */
3578 (void) chdir(home
?: "/root");
3580 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3581 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3582 execle("/bin/bash", "-bash", NULL
, env_use
);
3583 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3584 execle("/bin/sh", "-sh", NULL
, env_use
);
3586 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3589 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3592 static int setup_notify_child(void) {
3593 _cleanup_close_
int fd
= -EBADF
;
3594 static const union sockaddr_union sa
= {
3595 .un
.sun_family
= AF_UNIX
,
3596 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3600 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3602 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3604 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3605 (void) sockaddr_un_unlink(&sa
.un
);
3607 WITH_UMASK(0577) { /* only set "w" bit, which is all that's necessary for connecting from the container */
3608 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3610 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3613 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3615 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3617 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3619 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3624 static int setup_unix_export_dir_outside(char **ret
) {
3629 _cleanup_free_
char *p
= NULL
;
3630 p
= path_join("/run/systemd/nspawn/unix-export", arg_machine
);
3634 r
= path_is_mount_point(p
, /* root= */ NULL
, 0);
3636 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Mount point '%s' exists already, refusing.", p
);
3637 if (r
< 0 && r
!= -ENOENT
)
3638 return log_error_errno(r
, "Failed to detect if '%s' is a mount point: %m", p
);
3640 r
= mkdir_p(p
, 0755);
3642 return log_error_errno(r
, "Failed to create '%s': %m", p
);
3644 _cleanup_(rmdir_and_freep
) char *q
= TAKE_PTR(p
);
3646 /* Mount the "unix export" directory really tiny, just 64 inodes. We mark the superblock writable
3647 * (since the container shall bind sockets into it). */
3648 r
= mount_nofollow_verbose(
3653 MS_NODEV
|MS_NOEXEC
|MS_NOSUID
|ms_nosymfollow_supported(),
3654 "size=4M,nr_inodes=64,mode=0755");
3658 _cleanup_(umount_and_rmdir_and_freep
) char *w
= TAKE_PTR(q
);
3660 /* After creating the superblock we change the bind mount to be read-only. This means that the fs
3661 * itself is writable, but not through the mount accessible from the host. */
3662 r
= mount_nofollow_verbose(
3667 MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
|ms_nosymfollow_supported(),
3668 /* options= */ NULL
);
3676 static int setup_unix_export_host_inside(const char *directory
, const char *unix_export_path
) {
3680 assert(unix_export_path
);
3682 r
= make_run_host(directory
);
3686 _cleanup_free_
char *p
= path_join(directory
, "run/host/unix-export");
3690 if (mkdir(p
, 0755) < 0)
3691 return log_error_errno(errno
, "Failed to create '%s': %m", p
);
3693 r
= mount_nofollow_verbose(
3699 /* options= */ NULL
);
3703 r
= mount_nofollow_verbose(
3708 MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
|ms_nosymfollow_supported(),
3709 /* options= */ NULL
);
3713 r
= userns_lchown(p
, 0, 0);
3715 return log_error_errno(r
, "Failed to chown '%s': %m", p
);
3720 static int outer_child(
3722 const char *directory
,
3723 DissectedImage
*dissected_image
,
3724 int fd_outer_socket
,
3725 int fd_inner_socket
,
3728 const char *unix_export_path
) {
3730 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3731 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3732 _cleanup_close_
int fd
= -EBADF
, mntns_fd
= -EBADF
;
3739 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3740 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3741 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3742 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3743 * forked off it, and it exits. */
3747 assert(fd_outer_socket
>= 0);
3748 assert(fd_inner_socket
>= 0);
3750 log_debug("Outer child is initializing.");
3752 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3754 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3756 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3757 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3759 r
= reset_audit_loginuid();
3763 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3764 * mounts to the real root. */
3765 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3769 if (dissected_image
) {
3770 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3771 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3772 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3773 * right place right away. This makes sure ESP partitions and userns are compatible. */
3775 r
= dissected_image_mount_and_warn(
3780 /* userns_fd= */ -EBADF
,
3781 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3782 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3783 DISSECT_IMAGE_USR_NO_ROOT
|
3784 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3785 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3790 r
= determine_uid_shift(directory
);
3794 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3795 r
= namespace_open(0,
3796 /* ret_pidns_fd = */ NULL
,
3798 /* ret_netns_fd = */ NULL
,
3799 /* ret_userns_fd = */ NULL
,
3800 /* ret_root_fd = */ NULL
);
3802 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3804 l
= send_one_fd(fd_outer_socket
, mntns_fd
, 0);
3806 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3807 mntns_fd
= safe_close(mntns_fd
);
3809 /* Let the parent know which UID shift we read from the image */
3810 l
= send(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3812 return log_error_errno(errno
, "Failed to send UID shift: %m");
3813 if (l
!= sizeof(arg_uid_shift
))
3814 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3815 "Short write while sending UID shift.");
3817 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3818 /* When we are supposed to pick the UID shift, the parent will check now whether the
3819 * UID shift we just read from the image is available. If yes, it will send the UID
3820 * shift back to us, if not it will pick a different one, and send it back to us. */
3822 l
= recv(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3824 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3825 if (l
!= sizeof(arg_uid_shift
))
3826 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3827 "Short read while receiving UID shift.");
3830 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3831 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3834 if (path_equal(directory
, "/")) {
3835 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3836 * place, so that we can make changes to its mount structure (for example, to implement
3837 * --volatile=) without this interfering with our ability to access files such as
3838 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3839 * (instead of a temporary directory, since we are living in our own mount namespace here
3840 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3841 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3843 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3847 directory
= "/run/systemd/nspawn-root";
3850 /* Make sure we always have a mount that we can move to root later on. */
3851 r
= make_mount_point(directory
);
3855 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3856 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3857 * we'll live in our own little world from now on, and propagation from the host may only happen via
3858 * the mount tunnel dir, or not at all. */
3859 r
= mount_follow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_PRIVATE
|MS_REC
, NULL
);
3863 r
= setup_pivot_root(
3866 arg_pivot_root_old
);
3870 r
= setup_volatile_mode(
3874 arg_selinux_apifs_context
);
3878 r
= bind_user_prepare(
3883 &arg_custom_mounts
, &arg_n_custom_mounts
,
3884 &bind_user_context
);
3888 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3889 /* Send the user maps we determined to the parent, so that it installs it in our user
3890 * namespace UID map table */
3892 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3894 bind_user_context
->data
[i
].payload_user
->uid
,
3895 bind_user_context
->data
[i
].host_user
->uid
,
3896 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3897 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3900 l
= send(fd_outer_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3902 return log_error_errno(errno
, "Failed to send user UID map: %m");
3903 if (l
!= sizeof(map
))
3904 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3905 "Short write while sending user UID map.");
3912 arg_n_custom_mounts
,
3915 arg_selinux_apifs_context
,
3920 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3921 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3922 arg_uid_shift
!= 0) {
3923 _cleanup_free_
char *usr_subtree
= NULL
;
3927 dirs
[i
++] = (char*) directory
;
3929 if (dissected_image
&& dissected_image
->partitions
[PARTITION_USR
].found
) {
3930 usr_subtree
= path_join(directory
, "/usr");
3934 dirs
[i
++] = usr_subtree
;
3939 r
= remount_idmap(dirs
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3940 if (r
== -EINVAL
|| ERRNO_IS_NEG_NOT_SUPPORTED(r
)) {
3941 /* This might fail because the kernel or file system doesn't support idmapping. We
3942 * can't really distinguish this nicely, nor do we have any guarantees about the
3943 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3944 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3945 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3946 "ID mapped mounts are apparently not available, sorry.");
3948 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3949 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3951 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3953 log_debug("ID mapped mounts available, making use of them.");
3958 if (dissected_image
) {
3959 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3960 r
= dissected_image_mount(
3965 /* userns_fd= */ -EBADF
,
3966 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3967 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3968 DISSECT_IMAGE_USR_NO_ROOT
|
3969 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3970 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3972 return log_error_errno(r
, "File system check for image failed: %m");
3974 return log_error_errno(r
, "Failed to mount image file system: %m");
3977 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3978 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3980 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3984 l
= send(fd_outer_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3986 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3987 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3988 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3989 "Short write while sending cgroup mode.");
3992 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3996 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
4000 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
4001 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
4002 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
4004 return log_error_errno(r
, "Failed to make tree read-only: %m");
4007 r
= mount_all(directory
,
4010 arg_selinux_apifs_context
);
4014 r
= copy_devnodes(directory
);
4018 r
= make_extra_nodes(directory
);
4022 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
4024 p
= prefix_roota(directory
, "/run/host");
4025 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
4027 r
= setup_unix_export_host_inside(directory
, unix_export_path
);
4031 r
= setup_pts(directory
);
4035 r
= mount_tunnel_dig(directory
);
4039 r
= setup_keyring();
4043 r
= setup_credentials(directory
);
4047 r
= bind_user_setup(bind_user_context
, directory
);
4054 arg_n_custom_mounts
,
4057 arg_selinux_apifs_context
,
4058 MOUNT_NON_ROOT_ONLY
);
4062 r
= setup_timezone(directory
);
4066 r
= setup_resolv_conf(directory
);
4070 r
= setup_machine_id(directory
);
4074 r
= setup_journal(directory
);
4078 /* The same stuff as the $container env var, but nicely readable for the entire payload */
4079 p
= prefix_roota(directory
, "/run/host/container-manager");
4080 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
|WRITE_STRING_FILE_MODE_0444
);
4082 /* The same stuff as the $container_uuid env var */
4083 p
= prefix_roota(directory
, "/run/host/container-uuid");
4084 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
|WRITE_STRING_FILE_MODE_0444
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
4086 if (!arg_use_cgns
) {
4089 arg_unified_cgroup_hierarchy
,
4090 arg_userns_mode
!= USER_NAMESPACE_NO
,
4093 arg_selinux_apifs_context
,
4099 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
4100 * mounts available in systemd services inside the container that create a new mount namespace. See
4101 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
4102 * will inherit the shared propagation mode.
4104 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
4105 * directory mount to root later on.
4106 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
4108 r
= mount_switch_root(directory
, MS_SHARED
);
4110 return log_error_errno(r
, "Failed to move root directory: %m");
4112 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
4113 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
4115 r
= mount_tunnel_open();
4119 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4120 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4121 * requires that a fully visible instance is already present in the target mount
4122 * namespace. Mount one here so the inner child can mount its own instances. Later
4123 * we umount the temporary instances created here before we actually exec the
4124 * payload. Since the rootfs is shared the umount will propagate into the container.
4125 * Note, the inner child wouldn't be able to unmount the instances on its own since
4126 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4128 r
= pin_fully_visible_fs();
4133 fd
= setup_notify_child();
4137 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
4138 arg_clone_ns_flags
|
4139 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
4141 return log_error_errno(errno
, "Failed to fork inner child: %m");
4143 fd_outer_socket
= safe_close(fd_outer_socket
);
4145 /* The inner child has all namespaces that are requested, so that we all are owned by the
4146 * user if user namespaces are turned on. */
4148 if (arg_network_namespace_path
) {
4149 r
= namespace_enter(/* pidns_fd = */ -EBADF
,
4150 /* mntns_fd = */ -EBADF
,
4152 /* userns_fd = */ -EBADF
,
4153 /* root_fd = */ -EBADF
);
4155 return log_error_errno(r
, "Failed to join network namespace: %m");
4158 r
= inner_child(barrier
, fd_inner_socket
, fds
, os_release_pairs
);
4160 _exit(EXIT_FAILURE
);
4162 _exit(EXIT_SUCCESS
);
4165 l
= send(fd_outer_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4167 return log_error_errno(errno
, "Failed to send PID: %m");
4168 if (l
!= sizeof(pid
))
4169 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4170 "Short write while sending PID.");
4172 l
= send(fd_outer_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4174 return log_error_errno(errno
, "Failed to send machine ID: %m");
4175 if (l
!= sizeof(arg_uuid
))
4176 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4177 "Short write while sending machine ID.");
4179 l
= send_one_fd(fd_outer_socket
, fd
, 0);
4181 return log_error_errno(l
, "Failed to send notify fd: %m");
4183 fd_outer_socket
= safe_close(fd_outer_socket
);
4184 fd_inner_socket
= safe_close(fd_inner_socket
);
4185 netns_fd
= safe_close(netns_fd
);
4190 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4191 bool tried_hashed
= false;
4192 unsigned n_tries
= 100;
4197 assert(ret_lock_file
);
4198 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4199 assert(arg_uid_range
== 0x10000U
);
4203 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4206 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4207 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4212 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4214 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4217 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4218 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4219 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4224 /* Make some superficial checks whether the range is currently known in the user database */
4225 if (getpwuid_malloc(candidate
, /* ret= */ NULL
) >= 0)
4227 if (getpwuid_malloc(candidate
+ UINT32_C(0xFFFE), /* ret= */ NULL
) >= 0)
4229 if (getgrgid_malloc(candidate
, /* ret= */ NULL
) >= 0)
4231 if (getgrgid_malloc(candidate
+ UINT32_C(0xFFFE), /* ret= */ NULL
) >= 0)
4234 *ret_lock_file
= lf
;
4235 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4240 if (arg_machine
&& !tried_hashed
) {
4241 /* Try to hash the base from the container name */
4243 static const uint8_t hash_key
[] = {
4244 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4245 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4248 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4250 tried_hashed
= true;
4252 random_bytes(&candidate
, sizeof(candidate
));
4254 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4255 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4259 static int add_one_uid_map(
4261 uid_t container_uid
,
4265 return strextendf(p
,
4266 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4267 container_uid
, host_uid
, range
);
4270 static int make_uid_map_string(
4271 const uid_t bind_user_uid
[],
4272 size_t n_bind_user_uid
,
4276 _cleanup_free_
char *s
= NULL
;
4277 uid_t previous_uid
= 0;
4280 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4281 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4284 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4285 * quadruplet, consisting of host and container UID + GID. */
4287 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4288 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4289 host_uid
= bind_user_uid
[i
*4+offset
+1];
4291 assert(previous_uid
<= payload_uid
);
4292 assert(payload_uid
< arg_uid_range
);
4294 /* Add a range to close the gap to previous entry */
4295 if (payload_uid
> previous_uid
) {
4296 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4301 /* Map this specific user */
4302 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4306 previous_uid
= payload_uid
+ 1;
4309 /* And add a range to close the gap to finish the range */
4310 if (arg_uid_range
> previous_uid
) {
4311 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4322 static int setup_uid_map(
4324 const uid_t bind_user_uid
[],
4325 size_t n_bind_user_uid
) {
4327 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4328 _cleanup_free_
char *s
= NULL
;
4333 /* Build the UID map string */
4334 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4337 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4338 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4340 return log_error_errno(r
, "Failed to write UID map: %m");
4342 /* And now build the GID map string */
4344 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4347 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4348 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4350 return log_error_errno(r
, "Failed to write GID map: %m");
4355 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4356 char buf
[NOTIFY_BUFFER_MAX
+1];
4358 struct iovec iovec
= {
4360 .iov_len
= sizeof(buf
)-1,
4362 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4363 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4364 struct msghdr msghdr
= {
4367 .msg_control
= &control
,
4368 .msg_controllen
= sizeof(control
),
4370 struct ucred
*ucred
;
4372 pid_t inner_child_pid
;
4373 _cleanup_strv_free_
char **tags
= NULL
;
4378 inner_child_pid
= PTR_TO_PID(userdata
);
4380 if (revents
!= EPOLLIN
) {
4381 log_warning("Got unexpected poll event for notify fd.");
4385 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4386 if (ERRNO_IS_NEG_TRANSIENT(n
))
4388 else if (n
== -EXFULL
) {
4389 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4392 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4394 cmsg_close_all(&msghdr
);
4396 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4397 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4398 log_debug("Received notify message without valid credentials. Ignoring.");
4402 if ((size_t) n
>= sizeof(buf
)) {
4403 log_warning("Received notify message exceeded maximum size. Ignoring.");
4408 tags
= strv_split(buf
, "\n\r");
4412 if (strv_contains(tags
, "READY=1")) {
4413 r
= sd_notify(false, "READY=1\n");
4415 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4418 p
= strv_find_startswith(tags
, "STATUS=");
4420 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4425 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4428 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4430 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4432 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4437 static int merge_settings(Settings
*settings
, const char *path
) {
4443 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4444 * that this steals the fields of the Settings* structure, and hence modifies it. */
4446 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4447 settings
->start_mode
>= 0) {
4448 arg_start_mode
= settings
->start_mode
;
4449 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4452 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4453 settings
->ephemeral
>= 0)
4454 arg_ephemeral
= settings
->ephemeral
;
4456 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4459 if (!arg_settings_trusted
)
4460 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4462 free_and_replace(arg_directory
, settings
->root
);
4465 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4466 settings
->pivot_root_new
) {
4467 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4468 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4471 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4472 settings
->working_directory
)
4473 free_and_replace(arg_chdir
, settings
->working_directory
);
4475 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4476 settings
->environment
)
4477 strv_free_and_replace(arg_setenv
, settings
->environment
);
4479 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4482 free_and_replace(arg_user
, settings
->user
);
4484 if (uid_is_valid(settings
->uid
))
4485 arg_uid
= settings
->uid
;
4486 if (gid_is_valid(settings
->gid
))
4487 arg_gid
= settings
->gid
;
4488 if (settings
->n_supplementary_gids
> 0) {
4489 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4490 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4494 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4495 uint64_t plus
, minus
;
4496 uint64_t network_minus
= 0;
4499 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4500 * Settings structure */
4502 plus
= settings
->capability
;
4503 minus
= settings
->drop_capability
;
4505 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4506 settings_network_configured(settings
)) {
4507 if (settings_private_network(settings
))
4508 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4510 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4513 if (!arg_settings_trusted
&& plus
!= 0) {
4514 if (settings
->capability
!= 0)
4515 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4517 arg_caps_retain
&= ~network_minus
;
4518 arg_caps_retain
|= plus
;
4521 arg_caps_retain
&= ~minus
;
4523 /* Copy the full capabilities over too */
4524 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4525 if (!arg_settings_trusted
)
4526 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4528 arg_full_capabilities
= settings
->full_capabilities
;
4531 ambient
= settings
->ambient_capability
;
4532 if (!arg_settings_trusted
&& ambient
!= 0)
4533 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4535 arg_caps_ambient
|= ambient
;
4538 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4539 settings
->kill_signal
> 0)
4540 arg_kill_signal
= settings
->kill_signal
;
4542 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4543 settings
->personality
!= PERSONALITY_INVALID
)
4544 arg_personality
= settings
->personality
;
4546 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4547 !sd_id128_is_null(settings
->machine_id
)) {
4549 if (!arg_settings_trusted
)
4550 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4552 arg_uuid
= settings
->machine_id
;
4555 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4556 settings
->read_only
>= 0)
4557 arg_read_only
= settings
->read_only
;
4559 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4560 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4561 arg_volatile_mode
= settings
->volatile_mode
;
4563 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4564 settings
->n_custom_mounts
> 0) {
4566 if (!arg_settings_trusted
)
4567 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4569 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4570 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4571 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4572 settings
->n_custom_mounts
= 0;
4576 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4577 settings_network_configured(settings
)) {
4579 if (!arg_settings_trusted
)
4580 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4582 arg_network_veth
= settings_network_veth(settings
);
4583 arg_private_network
= settings_private_network(settings
);
4585 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4586 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4587 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4588 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4590 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4591 free_and_replace(arg_network_zone
, settings
->network_zone
);
4593 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4597 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4598 settings
->expose_ports
) {
4600 if (!arg_settings_trusted
)
4601 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4603 expose_port_free_all(arg_expose_ports
);
4604 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4608 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4609 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4611 if (!arg_settings_trusted
)
4612 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4614 arg_userns_mode
= settings
->userns_mode
;
4615 arg_uid_shift
= settings
->uid_shift
;
4616 arg_uid_range
= settings
->uid_range
;
4617 arg_userns_ownership
= settings
->userns_ownership
;
4621 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4622 !strv_isempty(settings
->bind_user
))
4623 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4625 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4626 settings
->notify_ready
>= 0)
4627 arg_notify_ready
= settings
->notify_ready
;
4629 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4631 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4632 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4633 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4635 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4636 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4641 if (settings
->seccomp
) {
4642 if (!arg_settings_trusted
)
4643 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4645 seccomp_release(arg_seccomp
);
4646 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4652 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4653 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4656 if (!settings
->rlimit
[rl
])
4659 if (!arg_settings_trusted
) {
4660 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4664 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4667 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4669 free_and_replace(arg_hostname
, settings
->hostname
);
4671 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4672 settings
->no_new_privileges
>= 0)
4673 arg_no_new_privileges
= settings
->no_new_privileges
;
4675 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4676 settings
->oom_score_adjust_set
) {
4678 if (!arg_settings_trusted
)
4679 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4681 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4682 arg_oom_score_adjust_set
= true;
4686 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4687 settings
->cpu_set
.set
) {
4689 if (!arg_settings_trusted
)
4690 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4692 cpu_set_reset(&arg_cpu_set
);
4693 arg_cpu_set
= TAKE_STRUCT(settings
->cpu_set
);
4697 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4698 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4699 arg_resolv_conf
= settings
->resolv_conf
;
4701 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4702 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4704 if (!arg_settings_trusted
)
4705 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4707 arg_link_journal
= settings
->link_journal
;
4708 arg_link_journal_try
= settings
->link_journal_try
;
4712 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4713 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4714 arg_timezone
= settings
->timezone
;
4716 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4719 if (!arg_settings_trusted
)
4720 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4722 free_and_replace(arg_slice
, settings
->slice
);
4725 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4726 settings
->use_cgns
>= 0) {
4728 if (!arg_settings_trusted
)
4729 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4731 arg_use_cgns
= settings
->use_cgns
;
4734 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4735 settings
->clone_ns_flags
!= ULONG_MAX
) {
4737 if (!arg_settings_trusted
)
4738 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4740 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4743 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4744 settings
->console_mode
>= 0) {
4746 if (!arg_settings_trusted
)
4747 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4749 arg_console_mode
= settings
->console_mode
;
4752 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4753 settings
->suppress_sync
>= 0)
4754 arg_suppress_sync
= settings
->suppress_sync
;
4756 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4757 * don't consult arg_settings_mask for them. */
4759 sd_bus_message_unref(arg_property_message
);
4760 arg_property_message
= TAKE_PTR(settings
->properties
);
4762 arg_console_width
= settings
->console_width
;
4763 arg_console_height
= settings
->console_height
;
4765 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4766 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4767 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4768 settings
->n_extra_nodes
= 0;
4773 static int load_settings(void) {
4774 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4775 _cleanup_fclose_
FILE *f
= NULL
;
4776 _cleanup_free_
char *p
= NULL
;
4782 /* If all settings are masked, there's no point in looking for
4783 * the settings file */
4784 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4787 /* We first look in the admin's directories in /etc and /run */
4788 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4789 _cleanup_free_
char *j
= NULL
;
4791 j
= path_join(i
, arg_settings_filename
);
4799 /* By default, we trust configuration from /etc and /run */
4800 if (arg_settings_trusted
< 0)
4801 arg_settings_trusted
= true;
4806 if (errno
!= ENOENT
)
4807 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4811 /* After that, let's look for a file next to the
4812 * actual image we shall boot. */
4815 r
= file_in_same_dir(arg_image
, arg_settings_filename
, &p
);
4817 return log_error_errno(r
, "Failed to generate settings path from image path: %m");
4818 } else if (arg_directory
) {
4819 r
= file_in_same_dir(arg_directory
, arg_settings_filename
, &p
);
4820 if (r
< 0 && r
!= -EADDRNOTAVAIL
) /* if directory is root fs, don't complain */
4821 return log_error_errno(r
, "Failed to generate settings path from directory path: %m");
4826 if (!f
&& errno
!= ENOENT
)
4827 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4829 /* By default, we do not trust configuration from /var/lib/machines */
4830 if (arg_settings_trusted
< 0)
4831 arg_settings_trusted
= false;
4838 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4840 r
= settings_load(f
, p
, &settings
);
4844 return merge_settings(settings
, p
);
4847 static int load_oci_bundle(void) {
4848 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4851 if (!arg_oci_bundle
)
4854 /* By default let's trust OCI bundles */
4855 if (arg_settings_trusted
< 0)
4856 arg_settings_trusted
= true;
4858 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4862 return merge_settings(settings
, arg_oci_bundle
);
4865 static int run_container(
4866 DissectedImage
*dissected_image
,
4868 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4869 struct ExposeArgs
*expose_args
,
4870 int *master
, pid_t
*pid
, int *ret
) {
4872 static const struct sigaction sa
= {
4873 .sa_handler
= nop_signal_handler
,
4874 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4877 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4878 _cleanup_close_
int etc_passwd_lock
= -EBADF
;
4879 _cleanup_close_pair_
int
4880 fd_inner_socket_pair
[2] = EBADF_PAIR
,
4881 fd_outer_socket_pair
[2] = EBADF_PAIR
;
4883 _cleanup_close_
int notify_socket
= -EBADF
, mntns_fd
= -EBADF
, fd_kmsg_fifo
= -EBADF
;
4884 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4885 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4886 _cleanup_(umount_and_rmdir_and_freep
) char *unix_export_host_dir
= NULL
;
4887 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4888 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4889 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4890 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4891 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4892 size_t n_bind_user_uid
= 0;
4893 ContainerStatus container_status
= 0;
4897 _cleanup_close_
int child_netns_fd
= -EBADF
;
4899 assert_se(sigemptyset(&mask_chld
) == 0);
4900 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4902 /* Set up the unix export host directory on the host first */
4903 r
= setup_unix_export_dir_outside(&unix_export_host_dir
);
4907 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4908 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4909 * check with getpwuid() if the specific user already exists. Note that /etc might be
4910 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4911 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4912 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4915 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4916 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4917 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4920 r
= barrier_create(&barrier
);
4922 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4924 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_inner_socket_pair
) < 0)
4925 return log_error_errno(errno
, "Failed to create inner socket pair: %m");
4927 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_outer_socket_pair
) < 0)
4928 return log_error_errno(errno
, "Failed to create outer socket pair: %m");
4930 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4931 * parent's blocking calls and give it a chance to call wait() and terminate. */
4932 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4934 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4936 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4938 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4940 if (arg_network_namespace_path
) {
4941 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4942 if (child_netns_fd
< 0)
4943 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4945 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4947 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
);
4949 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4951 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4952 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4955 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4957 return log_error_errno(errno
, "clone() failed%s: %m",
4959 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4962 /* The outer child only has a file system namespace. */
4963 barrier_set_role(&barrier
, BARRIER_CHILD
);
4965 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
4966 fd_outer_socket_pair
[0] = safe_close(fd_outer_socket_pair
[0]);
4968 (void) reset_all_signal_handlers();
4969 (void) reset_signal_mask();
4971 r
= outer_child(&barrier
,
4974 fd_outer_socket_pair
[1],
4975 fd_inner_socket_pair
[1],
4978 unix_export_host_dir
);
4980 _exit(EXIT_FAILURE
);
4982 _exit(EXIT_SUCCESS
);
4985 barrier_set_role(&barrier
, BARRIER_PARENT
);
4989 fd_inner_socket_pair
[1] = safe_close(fd_inner_socket_pair
[1]);
4990 fd_outer_socket_pair
[1] = safe_close(fd_outer_socket_pair
[1]);
4992 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4993 mntns_fd
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4995 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4997 /* The child just let us know the UID shift it might have read from the image. */
4998 l
= recv(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
5000 return log_error_errno(errno
, "Failed to read UID shift: %m");
5001 if (l
!= sizeof arg_uid_shift
)
5002 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
5004 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
5005 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
5006 * image, but if that's already in use, pick a new one, and report back to the child,
5007 * which one we now picked. */
5009 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
5011 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
5013 l
= send(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
5015 return log_error_errno(errno
, "Failed to send UID shift: %m");
5016 if (l
!= sizeof arg_uid_shift
)
5017 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
5020 n_bind_user_uid
= strv_length(arg_bind_user
);
5021 if (n_bind_user_uid
> 0) {
5022 /* Right after the UID shift, we'll receive the list of UID mappings for the
5023 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
5025 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
5029 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
5030 l
= recv(fd_outer_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
5032 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
5033 if (l
!= sizeof(uid_t
)*4)
5034 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
5035 SYNTHETIC_ERRNO(EIO
),
5036 "Short read while reading bind user UID pairs.");
5041 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
5042 /* The child let us know the support cgroup mode it might have read from the image. */
5043 l
= recv(fd_outer_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
5045 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
5046 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
5047 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
5048 l
, l
== 0 ? " The child is most likely dead." : "");
5051 /* Wait for the outer child. */
5052 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
5055 if (r
!= EXIT_SUCCESS
)
5058 /* And now retrieve the PID of the inner child. */
5059 l
= recv(fd_outer_socket_pair
[0], pid
, sizeof *pid
, 0);
5061 return log_error_errno(errno
, "Failed to read inner child PID: %m");
5062 if (l
!= sizeof *pid
)
5063 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
5065 /* We also retrieve container UUID in case it was generated by outer child */
5066 l
= recv(fd_outer_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
5068 return log_error_errno(errno
, "Failed to read container machine ID: %m");
5069 if (l
!= sizeof(arg_uuid
))
5070 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
5072 /* We also retrieve the socket used for notifications generated by outer child */
5073 notify_socket
= receive_one_fd(fd_outer_socket_pair
[0], 0);
5074 if (notify_socket
< 0)
5075 return log_error_errno(notify_socket
,
5076 "Failed to receive notification socket from the outer child: %m");
5078 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
5080 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5081 if (!barrier_place_and_sync(&barrier
)) /* #1 */
5082 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5084 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
5088 (void) barrier_place(&barrier
); /* #2 */
5091 if (arg_private_network
) {
5092 if (!arg_network_namespace_path
) {
5093 /* Wait until the child has unshared its network namespace. */
5094 if (!barrier_place_and_sync(&barrier
)) /* #3 */
5095 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
5098 if (child_netns_fd
< 0) {
5099 /* Make sure we have an open file descriptor to the child's network
5100 * namespace so it stays alive even if the child exits. */
5101 r
= namespace_open(*pid
,
5102 /* ret_pidns_fd = */ NULL
,
5103 /* ret_mntns_fd = */ NULL
,
5105 /* ret_userns_fd = */ NULL
,
5106 /* ret_root_fd = */ NULL
);
5108 return log_error_errno(r
, "Failed to open child network namespace: %m");
5111 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
5115 if (arg_network_veth
) {
5116 r
= setup_veth(arg_machine
, *pid
, veth_name
,
5117 arg_network_bridge
|| arg_network_zone
, &arg_network_provided_mac
);
5123 if (arg_network_bridge
) {
5124 /* Add the interface to a bridge */
5125 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
5130 } else if (arg_network_zone
) {
5131 /* Add the interface to a bridge, possibly creating it */
5132 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5140 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5144 /* We created the primary and extra veth links now; let's remember this, so that we know to
5145 remove them later on. Note that we don't bother with removing veth links that were created
5146 here when their setup failed half-way, because in that case the kernel should be able to
5147 remove them on its own, since they cannot be referenced by anything yet. */
5148 *veth_created
= true;
5150 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5154 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5159 if (arg_register
|| !arg_keep_unit
) {
5160 r
= sd_bus_default_system(&bus
);
5162 return log_error_errno(r
, "Failed to open system bus: %m");
5164 r
= sd_bus_set_close_on_exit(bus
, false);
5166 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5169 if (!arg_keep_unit
) {
5170 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5171 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5172 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5174 r
= sd_bus_match_signal_async(
5177 "org.freedesktop.systemd1",
5179 "org.freedesktop.systemd1.Scope",
5181 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5183 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5187 r
= register_machine(
5195 arg_custom_mounts
, arg_n_custom_mounts
,
5198 arg_property_message
,
5200 arg_container_service_name
,
5205 } else if (!arg_keep_unit
) {
5211 arg_custom_mounts
, arg_n_custom_mounts
,
5214 arg_property_message
,
5215 /* allow_pidfds= */ true,
5220 } else if (arg_slice
|| arg_property
)
5221 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5223 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5227 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5231 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5235 /* Notify the child that the parent is ready with all
5236 * its setup (including cgroup-ification), and that
5237 * the child can now hand over control to the code to
5238 * run inside the container. */
5239 (void) barrier_place(&barrier
); /* #4 */
5241 /* Block SIGCHLD here, before notifying child.
5242 * process_pty() will handle it with the other signals. */
5243 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5245 /* Reset signal to default */
5246 r
= default_signals(SIGCHLD
);
5248 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5250 r
= sd_event_new(&event
);
5252 return log_error_errno(r
, "Failed to get default event source: %m");
5254 (void) sd_event_set_watchdog(event
, true);
5257 r
= sd_bus_attach_event(bus
, event
, 0);
5259 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5262 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5266 /* Wait that the child is completely ready now, and has mounted their own copies of procfs and so on,
5267 * before we take the fully visible instances away. */
5268 if (!barrier_sync(&barrier
)) /* #5.1 */
5269 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5271 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5272 r
= wipe_fully_visible_fs(mntns_fd
);
5275 mntns_fd
= safe_close(mntns_fd
);
5278 /* And now let the child know that we completed removing the procfs instances, and it can start the
5280 if (!barrier_place(&barrier
)) /* #5.2 */
5281 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5283 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5284 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5285 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5287 (void) sd_notifyf(false,
5288 "STATUS=Container running.\n"
5289 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5290 if (!arg_notify_ready
) {
5291 r
= sd_notify(false, "READY=1\n");
5293 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5296 if (arg_kill_signal
> 0) {
5297 /* Try to kill the init system on SIGINT or SIGTERM */
5298 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5299 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5301 /* Immediately exit */
5302 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5303 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5306 (void) sd_event_add_signal(event
, NULL
, SIGRTMIN
+18, sigrtmin18_handler
, NULL
);
5308 r
= sd_event_add_memory_pressure(event
, NULL
, NULL
, NULL
);
5310 log_debug_errno(r
, "Failed allocate memory pressure event source, ignoring: %m");
5312 /* Exit when the child exits */
5313 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5315 /* Retrieve the kmsg fifo allocated by inner child */
5316 fd_kmsg_fifo
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5317 if (fd_kmsg_fifo
< 0)
5318 return log_error_errno(fd_kmsg_fifo
, "Failed to receive kmsg fifo from inner child: %m");
5320 if (arg_expose_ports
) {
5321 r
= expose_port_watch_rtnl(event
, fd_inner_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5325 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5326 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5329 if (arg_console_mode
!= CONSOLE_PIPE
) {
5330 _cleanup_close_
int fd
= -EBADF
;
5331 PTYForwardFlags flags
= 0;
5333 /* Retrieve the master pty allocated by inner child */
5334 fd
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5336 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5338 switch (arg_console_mode
) {
5340 case CONSOLE_READ_ONLY
:
5341 flags
|= PTY_FORWARD_READ_ONLY
;
5345 case CONSOLE_INTERACTIVE
:
5346 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5348 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5350 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5352 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5353 (void) pty_forward_set_width_height(
5356 arg_console_height
);
5358 if (!arg_background
) {
5359 _cleanup_free_
char *bg
= NULL
;
5361 r
= terminal_tint_color(220 /* blue */, &bg
);
5363 log_debug_errno(r
, "Failed to determine terminal background color, not tinting.");
5365 (void) pty_forward_set_background_color(forward
, bg
);
5366 } else if (!isempty(arg_background
))
5367 (void) pty_forward_set_background_color(forward
, arg_background
);
5372 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5375 *master
= TAKE_FD(fd
);
5378 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
5380 r
= sd_event_loop(event
);
5382 return log_error_errno(r
, "Failed to run event loop: %m");
5387 (void) pty_forward_get_last_char(forward
, &last_char
);
5388 forward
= pty_forward_free(forward
);
5390 if (!arg_quiet
&& last_char
!= '\n')
5394 /* Kill if it is not dead yet anyway */
5395 if (!arg_register
&& !arg_keep_unit
&& bus
)
5396 terminate_scope(bus
, arg_machine
);
5398 /* Normally redundant, but better safe than sorry */
5399 (void) kill(*pid
, SIGKILL
);
5401 fd_kmsg_fifo
= safe_close(fd_kmsg_fifo
);
5403 if (arg_private_network
) {
5404 r
= move_back_network_interfaces(child_netns_fd
, arg_network_interfaces
);
5409 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5411 /* Tell machined that we are gone. */
5413 (void) unregister_machine(bus
, arg_machine
);
5416 /* We failed to wait for the container, or the container exited abnormally. */
5418 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5419 /* r > 0 → The container exited with a non-zero status.
5420 * As a special case, we need to replace 133 with a different value,
5421 * because 133 is special-cased in the service file to reboot the container.
5422 * otherwise → The container exited with zero status and a reboot was not requested.
5424 if (r
== EXIT_FORCE_RESTART
)
5425 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5427 return 0; /* finito */
5430 /* CONTAINER_REBOOTED, loop again */
5432 if (arg_keep_unit
) {
5433 /* Special handling if we are running as a service: instead of simply
5434 * restarting the machine we want to restart the entire service, so let's
5435 * inform systemd about this with the special exit code 133. The service
5436 * file uses RestartForceExitStatus=133 so that this results in a full
5437 * nspawn restart. This is necessary since we might have cgroup parameters
5438 * set we want to have flushed out. */
5439 *ret
= EXIT_FORCE_RESTART
;
5440 return 0; /* finito */
5443 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5444 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5446 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5447 *veth_created
= false;
5448 return 1; /* loop again */
5451 static int initialize_rlimits(void) {
5452 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5453 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5454 * container execution environments. */
5456 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5457 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5458 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5459 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5460 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5461 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5462 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5463 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5464 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5465 [RLIMIT_NICE
] = { 0, 0 },
5466 [RLIMIT_NOFILE
] = { 1024, 4096 },
5467 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5468 [RLIMIT_RTPRIO
] = { 0, 0 },
5469 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5470 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5472 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5473 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5474 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5475 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5476 * that PID 1 changes a number of other resource limits during early initialization which is why we
5477 * don't read the other limits from PID 1 but prefer the static table above. */
5482 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5483 /* Let's only fill in what the user hasn't explicitly configured anyway */
5484 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5485 const struct rlimit
*v
;
5486 struct rlimit buffer
;
5488 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5489 /* For these two let's read the limits off PID 1. See above for an explanation. */
5491 r
= pid_getrlimit(1, rl
, &buffer
);
5493 return log_error_errno(r
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5496 } else if (rl
== RLIMIT_NOFILE
) {
5497 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5498 * userspace. Given that nspawn containers are often run without our PID 1,
5499 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5500 * so that container userspace gets similar resources as host userspace
5502 buffer
= kernel_defaults
[rl
];
5503 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5506 v
= kernel_defaults
+ rl
;
5508 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5509 if (!arg_rlimit
[rl
])
5513 if (DEBUG_LOGGING
) {
5514 _cleanup_free_
char *k
= NULL
;
5516 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5517 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5524 static int cant_be_in_netns(void) {
5525 _cleanup_close_
int fd
= -EBADF
;
5529 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5530 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5533 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5536 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5538 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5540 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5541 if (r
== -ENOENT
|| ERRNO_IS_NEG_DISCONNECT(r
))
5542 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5543 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5545 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5547 r
= getpeercred(fd
, &ucred
);
5549 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5551 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5553 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5555 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5556 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5560 static int run(int argc
, char *argv
[]) {
5561 bool remove_directory
= false, remove_image
= false, veth_created
= false, remove_tmprootdir
= false;
5562 _cleanup_close_
int master
= -EBADF
;
5563 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5564 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5565 char veth_name
[IFNAMSIZ
] = "";
5566 struct ExposeArgs expose_args
= {};
5567 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5568 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5569 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5570 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5571 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5574 log_parse_environment();
5577 r
= parse_argv(argc
, argv
);
5581 if (geteuid() != 0) {
5582 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5583 argc
>= 2 ? "Need to be root." :
5584 "Need to be root (and some arguments are usually required).\nHint: try --help");
5588 r
= cant_be_in_netns();
5592 r
= initialize_rlimits();
5596 r
= load_oci_bundle();
5604 r
= determine_names();
5608 r
= load_settings();
5612 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
5613 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
5615 if (!arg_private_network
&& arg_userns_mode
!= USER_NAMESPACE_NO
&& arg_uid_shift
> 0)
5616 arg_caps_retain
&= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE
);
5620 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5624 r
= verify_arguments();
5628 r
= resolve_network_interface_names(arg_network_interfaces
);
5632 r
= verify_network_interfaces_initialized();
5636 /* Reapply environment settings. */
5637 (void) detect_unified_cgroup_hierarchy_from_environment();
5639 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5640 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5641 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5642 (void) ignore_signals(SIGPIPE
);
5644 n_fd_passed
= sd_listen_fds(false);
5645 if (n_fd_passed
> 0) {
5646 r
= fdset_new_listen_fds(&fds
, false);
5648 log_error_errno(r
, "Failed to collect file descriptors: %m");
5653 /* The "default" umask. This is appropriate for most file and directory
5654 * operations performed by nspawn, and is the umask that will be used for
5655 * the child. Functions like copy_devnodes() change the umask temporarily. */
5658 if (arg_directory
) {
5661 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5662 * /var from the host will propagate into container dynamically (because bad things happen if
5663 * two systems write to the same /var). Let's allow it for the special cases where /var is
5664 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5665 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5666 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5667 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5671 if (arg_ephemeral
) {
5672 _cleanup_free_
char *np
= NULL
;
5674 r
= chase_and_update(&arg_directory
, 0);
5678 /* If the specified path is a mount point we generate the new snapshot immediately
5679 * inside it under a random name. However if the specified is not a mount point we
5680 * create the new snapshot in the parent directory, just next to it. */
5681 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5683 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5687 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5689 r
= tempfn_random(arg_directory
, "machine.", &np
);
5691 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5695 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5696 * only owned by us and no one else. */
5697 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5699 log_error_errno(r
, "Failed to lock %s: %m", np
);
5704 BLOCK_SIGNALS(SIGINT
);
5705 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_directory
, AT_FDCWD
, np
,
5706 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5707 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5708 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5709 BTRFS_SNAPSHOT_RECURSIVE
|
5710 BTRFS_SNAPSHOT_QUOTA
|
5711 BTRFS_SNAPSHOT_SIGINT
);
5714 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5718 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5722 free_and_replace(arg_directory
, np
);
5723 remove_directory
= true;
5725 r
= chase_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5729 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5731 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5735 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5740 r
= chase_and_update(&arg_template
, 0);
5745 BLOCK_SIGNALS(SIGINT
);
5746 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_template
, AT_FDCWD
, arg_directory
,
5747 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5748 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5749 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5750 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5751 BTRFS_SNAPSHOT_RECURSIVE
|
5752 BTRFS_SNAPSHOT_QUOTA
|
5753 BTRFS_SNAPSHOT_SIGINT
);
5756 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5757 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5758 else if (r
== -EINTR
) {
5759 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5762 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5765 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5766 "Populated %s from template %s.", arg_directory
, arg_template
);
5770 if (arg_start_mode
== START_BOOT
) {
5771 _cleanup_free_
char *b
= NULL
;
5773 int check_os_release
, is_os_tree
;
5775 if (arg_pivot_root_new
) {
5776 b
= path_join(arg_directory
, arg_pivot_root_new
);
5786 check_os_release
= getenv_bool("SYSTEMD_NSPAWN_CHECK_OS_RELEASE");
5787 if (check_os_release
< 0 && check_os_release
!= -ENXIO
) {
5788 r
= log_error_errno(check_os_release
, "Failed to parse $SYSTEMD_NSPAWN_CHECK_OS_RELEASE: %m");
5792 is_os_tree
= path_is_os_tree(p
);
5793 if (is_os_tree
== 0 && check_os_release
== 0)
5794 log_debug("Directory %s is missing an os-release file, continuing anyway.", p
);
5795 else if (is_os_tree
<= 0) {
5796 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5797 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5801 _cleanup_free_
char *p
= NULL
;
5803 if (arg_pivot_root_new
)
5804 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5806 p
= path_join(arg_directory
, "/usr/");
5812 if (laccess(p
, F_OK
) < 0) {
5813 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5814 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5820 DissectImageFlags dissect_image_flags
=
5821 DISSECT_IMAGE_GENERIC_ROOT
|
5822 DISSECT_IMAGE_REQUIRE_ROOT
|
5823 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5824 DISSECT_IMAGE_USR_NO_ROOT
|
5825 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5826 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5828 assert(!arg_template
);
5830 r
= chase_and_update(&arg_image
, 0);
5834 if (arg_ephemeral
) {
5835 _cleanup_free_
char *np
= NULL
;
5837 r
= tempfn_random(arg_image
, "machine.", &np
);
5839 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5843 /* Always take an exclusive lock on our own ephemeral copy. */
5844 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5846 log_error_errno(r
, "Failed to create image lock: %m");
5851 BLOCK_SIGNALS(SIGINT
);
5852 r
= copy_file_full(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600,
5853 FS_NOCOW_FL
, FS_NOCOW_FL
,
5854 COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
,
5858 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5862 r
= log_error_errno(r
, "Failed to copy image file: %m");
5866 free_and_replace(arg_image
, np
);
5867 remove_image
= true;
5869 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5871 log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5875 log_error_errno(r
, "Failed to create image lock: %m");
5879 r
= verity_settings_load(
5880 &arg_verity_settings
,
5881 arg_image
, NULL
, NULL
);
5883 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5887 if (arg_verity_settings
.data_path
)
5888 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5891 if (!mkdtemp(tmprootdir
)) {
5892 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5896 remove_tmprootdir
= true;
5898 arg_directory
= strdup(tmprootdir
);
5899 if (!arg_directory
) {
5904 r
= loop_device_make_by_path(
5906 arg_read_only
? O_RDONLY
: O_RDWR
,
5907 /* sector_size= */ UINT32_MAX
,
5908 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5912 log_error_errno(r
, "Failed to set up loopback block device: %m");
5916 r
= dissect_loop_device_and_warn(
5918 &arg_verity_settings
,
5919 /* mount_options=*/ NULL
,
5920 arg_image_policy
?: &image_policy_container
,
5921 dissect_image_flags
,
5924 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5925 log_notice("Note that the disk image needs to\n"
5926 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5927 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5928 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5929 " d) or contain a file system without a partition table\n"
5930 "in order to be bootable with systemd-nspawn.");
5936 r
= dissected_image_load_verity_sig_partition(
5939 &arg_verity_settings
);
5943 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5944 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5945 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5947 r
= dissected_image_decrypt_interactively(
5950 &arg_verity_settings
,
5955 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5956 if (remove_image
&& unlink(arg_image
) >= 0)
5957 remove_image
= false;
5959 if (arg_architecture
< 0)
5960 arg_architecture
= dissected_image_architecture(dissected_image
);
5963 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5967 if (arg_console_mode
< 0)
5968 arg_console_mode
= isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
) ?
5969 CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5971 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5975 const char *t
= arg_image
?: arg_directory
;
5976 _cleanup_free_
char *u
= NULL
;
5977 (void) terminal_urlify_path(t
, t
, &u
);
5979 log_info("%s %sSpawning container %s on %s.%s\n"
5980 "%s %sPress %sCtrl-]%s three times within 1s to kill container.%s",
5981 special_glyph(SPECIAL_GLYPH_LIGHT_SHADE
), ansi_grey(), arg_machine
, u
?: t
, ansi_normal(),
5982 special_glyph(SPECIAL_GLYPH_LIGHT_SHADE
), ansi_grey(), ansi_highlight(), ansi_grey(), ansi_normal());
5985 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, SIGRTMIN
+18, -1) >= 0);
5987 r
= make_reaper_process(true);
5989 log_error_errno(r
, "Failed to become subreaper: %m");
5993 if (arg_expose_ports
) {
5994 r
= fw_ctx_new(&fw_ctx
);
5996 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5999 expose_args
.fw_ctx
= fw_ctx
;
6002 r
= run_container(dissected_image
,
6004 veth_name
, &veth_created
,
6005 &expose_args
, &master
,
6012 (void) sd_notify(false,
6013 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
6014 "STOPPING=1\nSTATUS=Terminating...");
6017 (void) kill(pid
, SIGKILL
);
6019 /* Try to flush whatever is still queued in the pty */
6021 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
6022 master
= safe_close(master
);
6026 (void) wait_for_terminate(pid
, NULL
);
6030 if (remove_directory
&& arg_directory
) {
6033 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
6035 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
6038 if (remove_image
&& arg_image
) {
6039 if (unlink(arg_image
) < 0)
6040 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
6043 if (remove_tmprootdir
) {
6044 if (rmdir(tmprootdir
) < 0)
6045 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
6051 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
6052 (void) rm_rf(p
, REMOVE_ROOT
);
6054 p
= strjoina("/run/systemd/nspawn/unix-export/", arg_machine
);
6055 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
6059 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
6060 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
6063 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
6064 (void) remove_bridge(arg_network_zone
);
6066 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
6067 expose_port_free_all(arg_expose_ports
);
6068 rlimit_free_all(arg_rlimit
);
6069 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
6077 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);