1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
8 #include <linux/loop.h>
10 #include <selinux/selinux.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.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
;
240 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
260 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
262 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
264 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
265 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
267 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
269 STATIC_DESTRUCTOR_REGISTER(arg_credentials
, machine_credential_context_done
);
270 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
271 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
272 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
273 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
274 STATIC_DESTRUCTOR_REGISTER(arg_image_policy
, image_policy_freep
);
276 static int handle_arg_console(const char *arg
) {
277 if (streq(arg
, "help")) {
286 if (streq(arg
, "interactive"))
287 arg_console_mode
= CONSOLE_INTERACTIVE
;
288 else if (streq(arg
, "read-only"))
289 arg_console_mode
= CONSOLE_READ_ONLY
;
290 else if (streq(arg
, "passive"))
291 arg_console_mode
= CONSOLE_PASSIVE
;
292 else if (streq(arg
, "pipe")) {
293 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
294 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
295 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
296 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
297 "Proceeding anyway.");
299 arg_console_mode
= CONSOLE_PIPE
;
300 } else if (streq(arg
, "autopipe")) {
301 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
302 arg_console_mode
= CONSOLE_INTERACTIVE
;
304 arg_console_mode
= CONSOLE_PIPE
;
306 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
308 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
312 static int help(void) {
313 _cleanup_free_
char *link
= NULL
;
316 pager_open(arg_pager_flags
);
318 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
322 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
323 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
324 " -h --help Show this help\n"
325 " --version Print version string\n"
326 " -q --quiet Do not show status information\n"
327 " --no-pager Do not pipe output into a pager\n"
328 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
330 " -D --directory=PATH Root directory for the container\n"
331 " --template=PATH Initialize root directory from template directory,\n"
333 " -x --ephemeral Run container with snapshot of root directory, and\n"
334 " remove it after exit\n"
335 " -i --image=PATH Root file system disk image (or device node) for\n"
337 " --image-policy=POLICY Specify disk image dissection policy\n"
338 " --oci-bundle=PATH OCI bundle directory\n"
339 " --read-only Mount the root directory read-only\n"
340 " --volatile[=MODE] Run the system in volatile mode\n"
341 " --root-hash=HASH Specify verity root hash for root disk image\n"
342 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
343 " as a DER encoded PKCS7, either as a path to a file\n"
344 " or as an ASCII base64 encoded string prefixed by\n"
346 " --verity-data=PATH Specify hash device for verity\n"
347 " --pivot-root=PATH[:PATH]\n"
348 " Pivot root to given directory in the container\n\n"
349 "%3$sExecution:%4$s\n"
350 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
351 " -b --boot Boot up full system (i.e. invoke init)\n"
352 " --chdir=PATH Set working directory in the container\n"
353 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
354 " -u --user=USER Run the command under specified user or UID\n"
355 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
356 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
357 " --suppress-sync=BOOLEAN\n"
358 " Suppress any form of disk data synchronization\n\n"
359 "%3$sSystem Identity:%4$s\n"
360 " -M --machine=NAME Set the machine name for the container\n"
361 " --hostname=NAME Override the hostname for the container\n"
362 " --uuid=UUID Set a specific machine UUID for the container\n\n"
363 "%3$sProperties:%4$s\n"
364 " -S --slice=SLICE Place the container in the specified slice\n"
365 " --property=NAME=VALUE Set scope unit property\n"
366 " --register=BOOLEAN Register container as machine\n"
367 " --keep-unit Do not register a scope for the machine, reuse\n"
368 " the service unit nspawn is running in\n\n"
369 "%3$sUser Namespacing:%4$s\n"
370 " --private-users=no Run without user namespacing\n"
371 " --private-users=yes|pick|identity\n"
372 " Run within user namespace, autoselect UID/GID range\n"
373 " --private-users=UIDBASE[:NUIDS]\n"
374 " Similar, but with user configured UID/GID range\n"
375 " --private-users-ownership=MODE\n"
376 " Adjust ('chown') or map ('map') OS tree ownership\n"
377 " to private UID/GID range\n"
378 " -U Equivalent to --private-users=pick and\n"
379 " --private-users-ownership=auto\n\n"
380 "%3$sNetworking:%4$s\n"
381 " --private-network Disable network in container\n"
382 " --network-interface=HOSTIF[:CONTAINERIF]\n"
383 " Assign an existing network interface to the\n"
385 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
386 " Create a macvlan network interface based on an\n"
387 " existing network interface to the container\n"
388 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
389 " Create an ipvlan network interface based on an\n"
390 " existing network interface to the container\n"
391 " -n --network-veth Add a virtual Ethernet connection between host\n"
393 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
394 " Add an additional virtual Ethernet link between\n"
395 " host and container\n"
396 " --network-bridge=INTERFACE\n"
397 " Add a virtual Ethernet connection to the container\n"
398 " and attach it to an existing bridge on the host\n"
399 " --network-zone=NAME Similar, but attach the new interface to an\n"
400 " an automatically managed bridge interface\n"
401 " --network-namespace-path=PATH\n"
402 " Set network namespace to the one represented by\n"
403 " the specified kernel namespace file node\n"
404 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
405 " Expose a container IP port on the host\n\n"
406 "%3$sSecurity:%4$s\n"
407 " --capability=CAP In addition to the default, retain specified\n"
409 " --drop-capability=CAP Drop the specified capability from the default set\n"
410 " --ambient-capability=CAP\n"
411 " Sets the specified capability for the started\n"
412 " process. Not useful if booting a machine.\n"
413 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
414 " --system-call-filter=LIST|~LIST\n"
415 " Permit/prohibit specific system calls\n"
416 " -Z --selinux-context=SECLABEL\n"
417 " Set the SELinux security context to be used by\n"
418 " processes in the container\n"
419 " -L --selinux-apifs-context=SECLABEL\n"
420 " Set the SELinux security context to be used by\n"
421 " API/tmpfs file systems in the container\n\n"
422 "%3$sResources:%4$s\n"
423 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
424 " --oom-score-adjust=VALUE\n"
425 " Adjust the OOM score value for the payload\n"
426 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
427 " --personality=ARCH Pick personality for this container\n\n"
428 "%3$sIntegration:%4$s\n"
429 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
430 " --timezone=MODE Select mode of /etc/localtime initialization\n"
431 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
432 " host, try-guest, try-host\n"
433 " -j Equivalent to --link-journal=try-guest\n\n"
435 " --bind=PATH[:PATH[:OPTIONS]]\n"
436 " Bind mount a file or directory from the host into\n"
438 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
439 " Similar, but creates a read-only bind mount\n"
440 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
442 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
443 " --overlay=PATH[:PATH...]:PATH\n"
444 " Create an overlay mount from the host to \n"
446 " --overlay-ro=PATH[:PATH...]:PATH\n"
447 " Similar, but creates a read-only overlay mount\n"
448 " --bind-user=NAME Bind user from host to container\n\n"
449 "%3$sInput/Output:%4$s\n"
450 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
451 " set up for the container.\n"
452 " -P --pipe Equivalent to --console=pipe\n\n"
453 "%3$sCredentials:%4$s\n"
454 " --set-credential=ID:VALUE\n"
455 " Pass a credential with literal value to container.\n"
456 " --load-credential=ID:PATH\n"
457 " Load credential to pass to container from file or\n"
458 " AF_UNIX stream socket.\n"
459 "\nSee the %2$s for details.\n",
460 program_invocation_short_name
,
470 static int custom_mount_check_all(void) {
473 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
474 CustomMount
*m
= &arg_custom_mounts
[i
];
476 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
477 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
478 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
479 "--private-users-ownership=own may not be combined with custom root mounts.");
480 if (arg_uid_shift
== UID_INVALID
)
481 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
482 "--private-users with automatic UID shift may not be combined with custom root mounts.");
489 static int detect_unified_cgroup_hierarchy_from_environment(void) {
490 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
493 /* Allow the user to control whether the unified hierarchy is used */
497 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
498 var
= "UNIFIED_CGROUP_HIERARCHY";
503 r
= parse_boolean(e
);
505 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
507 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
509 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
515 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
518 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
519 * in the image actually supports. */
520 r
= cg_all_unified();
522 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
524 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
525 * routine only detects 231, so we'll have a false negative here for 230. */
526 r
= systemd_installation_has_version(directory
, "230");
528 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
530 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
532 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
533 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
534 /* Mixed cgroup hierarchy support was added in 233 */
535 r
= systemd_installation_has_version(directory
, "233");
537 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
539 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
541 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
543 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
545 log_debug("Using %s hierarchy for container.",
546 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
547 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
552 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
557 _cleanup_free_
char *t
= NULL
;
559 r
= extract_first_word(&spec
, &t
, ",", 0);
561 return log_error_errno(r
, "Failed to parse capability %s.", t
);
565 if (streq(t
, "help")) {
566 for (int i
= 0; i
< capability_list_length(); i
++) {
569 name
= capability_to_name(i
);
580 r
= capability_from_name(t
);
582 return log_error_errno(r
, "Failed to parse capability %s.", t
);
589 return 1; /* continue */
592 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
595 r
= getenv_bool(name
);
599 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
601 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
602 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
606 static int parse_mount_settings_env(void) {
610 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
611 if (r
< 0 && r
!= -ENXIO
)
612 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
614 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
616 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
617 if (streq_ptr(e
, "network"))
618 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
621 r
= parse_boolean(e
);
623 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
625 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
626 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
632 static int parse_environment(void) {
636 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
639 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
642 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
645 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
649 r
= parse_mount_settings_env();
653 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
654 * even if it is supported. If not supported, it has no effect. */
655 if (!cg_ns_supported())
656 arg_use_cgns
= false;
658 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
661 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
665 arg_use_cgns
= r
> 0;
666 arg_settings_mask
|= SETTING_USE_CGNS
;
670 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
672 arg_container_service_name
= e
;
674 e
= getenv("SYSTEMD_NSPAWN_NETWORK_MAC");
676 r
= parse_ether_addr(e
, &arg_network_provided_mac
);
678 return log_error_errno(r
, "Failed to parse provided MAC address via environment variable");
681 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
683 arg_suppress_sync
= r
;
684 else if (r
!= -ENXIO
)
685 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
687 return detect_unified_cgroup_hierarchy_from_environment();
690 static int parse_argv(int argc
, char *argv
[]) {
697 ARG_AMBIENT_CAPABILITY
,
709 ARG_NETWORK_INTERFACE
,
714 ARG_NETWORK_VETH_EXTRA
,
715 ARG_NETWORK_NAMESPACE_PATH
,
725 ARG_PRIVATE_USERS_CHOWN
,
726 ARG_PRIVATE_USERS_OWNERSHIP
,
731 ARG_SYSTEM_CALL_FILTER
,
734 ARG_NO_NEW_PRIVILEGES
,
735 ARG_OOM_SCORE_ADJUST
,
750 static const struct option options
[] = {
751 { "help", no_argument
, NULL
, 'h' },
752 { "version", no_argument
, NULL
, ARG_VERSION
},
753 { "directory", required_argument
, NULL
, 'D' },
754 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
755 { "ephemeral", no_argument
, NULL
, 'x' },
756 { "user", required_argument
, NULL
, 'u' },
757 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
758 { "as-pid2", no_argument
, NULL
, 'a' },
759 { "boot", no_argument
, NULL
, 'b' },
760 { "uuid", required_argument
, NULL
, ARG_UUID
},
761 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
762 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
763 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
764 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
765 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
766 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
767 { "bind", required_argument
, NULL
, ARG_BIND
},
768 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
769 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
770 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
771 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
772 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
773 { "machine", required_argument
, NULL
, 'M' },
774 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
775 { "slice", required_argument
, NULL
, 'S' },
776 { "setenv", required_argument
, NULL
, 'E' },
777 { "selinux-context", required_argument
, NULL
, 'Z' },
778 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
779 { "quiet", no_argument
, NULL
, 'q' },
780 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
781 { "register", required_argument
, NULL
, ARG_REGISTER
},
782 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
783 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
784 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
785 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
786 { "network-veth", no_argument
, NULL
, 'n' },
787 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
788 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
789 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
790 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
791 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
792 { "image", required_argument
, NULL
, 'i' },
793 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
794 { "port", required_argument
, NULL
, 'p' },
795 { "property", required_argument
, NULL
, ARG_PROPERTY
},
796 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
797 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
798 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
799 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
800 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
801 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
802 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
803 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
804 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
805 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
806 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
807 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
808 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
809 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
810 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
811 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
812 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
813 { "console", required_argument
, NULL
, ARG_CONSOLE
},
814 { "pipe", no_argument
, NULL
, ARG_PIPE
},
815 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
816 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
817 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
818 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
819 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
820 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
821 { "image-policy", required_argument
, NULL
, ARG_IMAGE_POLICY
},
826 uint64_t plus
= 0, minus
= 0;
827 bool mask_all_settings
= false, mask_no_settings
= false;
832 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
833 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
835 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
845 r
= parse_path_argument(optarg
, false, &arg_directory
);
849 arg_settings_mask
|= SETTING_DIRECTORY
;
853 r
= parse_path_argument(optarg
, false, &arg_template
);
857 arg_settings_mask
|= SETTING_DIRECTORY
;
861 r
= parse_path_argument(optarg
, false, &arg_image
);
865 arg_settings_mask
|= SETTING_DIRECTORY
;
869 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
876 arg_ephemeral
= true;
877 arg_settings_mask
|= SETTING_EPHEMERAL
;
881 r
= free_and_strdup(&arg_user
, optarg
);
885 arg_settings_mask
|= SETTING_USER
;
888 case ARG_NETWORK_ZONE
: {
889 _cleanup_free_
char *j
= NULL
;
891 j
= strjoin("vz-", optarg
);
895 if (!ifname_valid(j
))
896 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
897 "Network zone name not valid: %s", j
);
899 free_and_replace(arg_network_zone
, j
);
901 arg_network_veth
= true;
902 arg_private_network
= true;
903 arg_settings_mask
|= SETTING_NETWORK
;
907 case ARG_NETWORK_BRIDGE
:
909 if (!ifname_valid(optarg
))
910 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
911 "Bridge interface name not valid: %s", optarg
);
913 r
= free_and_strdup(&arg_network_bridge
, optarg
);
919 arg_network_veth
= true;
920 arg_private_network
= true;
921 arg_settings_mask
|= SETTING_NETWORK
;
924 case ARG_NETWORK_VETH_EXTRA
:
925 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
927 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
929 arg_private_network
= true;
930 arg_settings_mask
|= SETTING_NETWORK
;
933 case ARG_NETWORK_INTERFACE
:
934 r
= interface_pair_parse(&arg_network_interfaces
, optarg
);
938 arg_private_network
= true;
939 arg_settings_mask
|= SETTING_NETWORK
;
942 case ARG_NETWORK_MACVLAN
:
943 r
= macvlan_pair_parse(&arg_network_macvlan
, optarg
);
947 arg_private_network
= true;
948 arg_settings_mask
|= SETTING_NETWORK
;
951 case ARG_NETWORK_IPVLAN
:
952 r
= ipvlan_pair_parse(&arg_network_ipvlan
, optarg
);
957 case ARG_PRIVATE_NETWORK
:
958 arg_private_network
= true;
959 arg_settings_mask
|= SETTING_NETWORK
;
962 case ARG_NETWORK_NAMESPACE_PATH
:
963 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
967 arg_settings_mask
|= SETTING_NETWORK
;
971 if (arg_start_mode
== START_PID2
)
972 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
973 "--boot and --as-pid2 may not be combined.");
975 arg_start_mode
= START_BOOT
;
976 arg_settings_mask
|= SETTING_START_MODE
;
980 if (arg_start_mode
== START_BOOT
)
981 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
982 "--boot and --as-pid2 may not be combined.");
984 arg_start_mode
= START_PID2
;
985 arg_settings_mask
|= SETTING_START_MODE
;
989 r
= id128_from_string_nonzero(optarg
, &arg_uuid
);
991 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
992 "Machine UUID may not be all zeroes.");
994 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
996 arg_settings_mask
|= SETTING_MACHINE_ID
;
1000 _cleanup_free_
char *mangled
= NULL
;
1002 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1006 free_and_replace(arg_slice
, mangled
);
1007 arg_settings_mask
|= SETTING_SLICE
;
1012 if (isempty(optarg
))
1013 arg_machine
= mfree(arg_machine
);
1015 if (!hostname_is_valid(optarg
, 0))
1016 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1017 "Invalid machine name: %s", optarg
);
1019 r
= free_and_strdup(&arg_machine
, optarg
);
1026 if (isempty(optarg
))
1027 arg_hostname
= mfree(arg_hostname
);
1029 if (!hostname_is_valid(optarg
, 0))
1030 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1031 "Invalid hostname: %s", optarg
);
1033 r
= free_and_strdup(&arg_hostname
, optarg
);
1038 arg_settings_mask
|= SETTING_HOSTNAME
;
1042 arg_selinux_context
= optarg
;
1046 arg_selinux_apifs_context
= optarg
;
1050 arg_read_only
= true;
1051 arg_settings_mask
|= SETTING_READ_ONLY
;
1054 case ARG_AMBIENT_CAPABILITY
: {
1056 r
= parse_capability_spec(optarg
, &m
);
1059 arg_caps_ambient
|= m
;
1060 arg_settings_mask
|= SETTING_CAPABILITY
;
1063 case ARG_CAPABILITY
:
1064 case ARG_DROP_CAPABILITY
: {
1066 r
= parse_capability_spec(optarg
, &m
);
1070 if (c
== ARG_CAPABILITY
)
1074 arg_settings_mask
|= SETTING_CAPABILITY
;
1077 case ARG_NO_NEW_PRIVILEGES
:
1078 r
= parse_boolean(optarg
);
1080 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1082 arg_no_new_privileges
= r
;
1083 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1087 arg_link_journal
= LINK_GUEST
;
1088 arg_link_journal_try
= true;
1089 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1092 case ARG_LINK_JOURNAL
:
1093 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1095 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1097 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1102 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1104 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1106 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1110 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1112 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1114 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1118 case ARG_OVERLAY_RO
:
1119 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1120 if (r
== -EADDRNOTAVAIL
)
1121 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1123 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1125 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1128 case ARG_INACCESSIBLE
:
1129 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1131 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1133 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1137 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1139 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1141 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1148 case ARG_SHARE_SYSTEM
:
1149 /* We don't officially support this anymore, except for compat reasons. People should use the
1150 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1151 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1152 arg_clone_ns_flags
= 0;
1156 r
= parse_boolean(optarg
);
1158 log_error("Failed to parse --register= argument: %s", optarg
);
1166 arg_keep_unit
= true;
1169 case ARG_PERSONALITY
:
1171 arg_personality
= personality_from_string(optarg
);
1172 if (arg_personality
== PERSONALITY_INVALID
)
1173 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1174 "Unknown or unsupported personality '%s'.", optarg
);
1176 arg_settings_mask
|= SETTING_PERSONALITY
;
1182 arg_volatile_mode
= VOLATILE_YES
;
1183 else if (streq(optarg
, "help")) {
1184 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1189 m
= volatile_mode_from_string(optarg
);
1191 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1192 "Failed to parse --volatile= argument: %s", optarg
);
1194 arg_volatile_mode
= m
;
1197 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1201 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1203 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1205 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1207 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1211 if (strv_extend(&arg_property
, optarg
) < 0)
1216 case ARG_PRIVATE_USERS
: {
1221 else if (!in_charset(optarg
, DIGITS
))
1222 /* do *not* parse numbers as booleans */
1223 boolean
= parse_boolean(optarg
);
1228 /* no: User namespacing off */
1229 arg_userns_mode
= USER_NAMESPACE_NO
;
1230 arg_uid_shift
= UID_INVALID
;
1231 arg_uid_range
= UINT32_C(0x10000);
1232 } else if (boolean
> 0) {
1233 /* yes: User namespacing on, UID range is read from root dir */
1234 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1235 arg_uid_shift
= UID_INVALID
;
1236 arg_uid_range
= UINT32_C(0x10000);
1237 } else if (streq(optarg
, "pick")) {
1238 /* pick: User namespacing on, UID range is picked randomly */
1239 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1240 * implied by USER_NAMESPACE_PICK
1242 arg_uid_shift
= UID_INVALID
;
1243 arg_uid_range
= UINT32_C(0x10000);
1245 } else if (streq(optarg
, "identity")) {
1246 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1247 * itself, i.e. we don't actually map anything, but do take benefit of
1248 * isolation of capability sets. */
1249 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1251 arg_uid_range
= UINT32_C(0x10000);
1253 _cleanup_free_
char *buffer
= NULL
;
1254 const char *range
, *shift
;
1256 /* anything else: User namespacing on, UID range is explicitly configured */
1258 range
= strchr(optarg
, ':');
1260 buffer
= strndup(optarg
, range
- optarg
);
1266 r
= safe_atou32(range
, &arg_uid_range
);
1268 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1272 r
= parse_uid(shift
, &arg_uid_shift
);
1274 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1276 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1278 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1279 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1282 arg_settings_mask
|= SETTING_USERNS
;
1287 if (userns_supported()) {
1288 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1289 * implied by USER_NAMESPACE_PICK
1291 arg_uid_shift
= UID_INVALID
;
1292 arg_uid_range
= UINT32_C(0x10000);
1294 arg_settings_mask
|= SETTING_USERNS
;
1299 case ARG_PRIVATE_USERS_CHOWN
:
1300 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1302 arg_settings_mask
|= SETTING_USERNS
;
1305 case ARG_PRIVATE_USERS_OWNERSHIP
:
1306 if (streq(optarg
, "help")) {
1307 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1311 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1312 if (arg_userns_ownership
< 0)
1313 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1315 arg_settings_mask
|= SETTING_USERNS
;
1318 case ARG_KILL_SIGNAL
:
1319 if (streq(optarg
, "help")) {
1320 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1324 arg_kill_signal
= signal_from_string(optarg
);
1325 if (arg_kill_signal
< 0)
1326 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1328 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1333 /* no → do not read files
1334 * yes → read files, do not override cmdline, trust only subset
1335 * override → read files, override cmdline, trust only subset
1336 * trusted → read files, do not override cmdline, trust all
1339 r
= parse_boolean(optarg
);
1341 if (streq(optarg
, "trusted")) {
1342 mask_all_settings
= false;
1343 mask_no_settings
= false;
1344 arg_settings_trusted
= true;
1346 } else if (streq(optarg
, "override")) {
1347 mask_all_settings
= false;
1348 mask_no_settings
= true;
1349 arg_settings_trusted
= -1;
1351 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1354 mask_all_settings
= false;
1355 mask_no_settings
= false;
1356 arg_settings_trusted
= -1;
1359 mask_all_settings
= true;
1360 mask_no_settings
= false;
1361 arg_settings_trusted
= false;
1367 if (!path_is_absolute(optarg
))
1368 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1369 "Working directory %s is not an absolute path.", optarg
);
1371 r
= free_and_strdup(&arg_chdir
, optarg
);
1375 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1378 case ARG_PIVOT_ROOT
:
1379 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1381 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1383 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1386 case ARG_NOTIFY_READY
:
1387 r
= parse_boolean(optarg
);
1389 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1390 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1391 arg_notify_ready
= r
;
1392 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1395 case ARG_ROOT_HASH
: {
1396 _cleanup_free_
void *k
= NULL
;
1399 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1401 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1402 if (l
< sizeof(sd_id128_t
))
1403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128-bit long: %s", optarg
);
1405 free_and_replace(arg_verity_settings
.root_hash
, k
);
1406 arg_verity_settings
.root_hash_size
= l
;
1410 case ARG_ROOT_HASH_SIG
: {
1415 if ((value
= startswith(optarg
, "base64:"))) {
1416 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1418 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1421 r
= read_full_file(optarg
, (char**) &p
, &l
);
1423 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1426 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1427 arg_verity_settings
.root_hash_sig_size
= l
;
1431 case ARG_VERITY_DATA
:
1432 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1437 case ARG_SYSTEM_CALL_FILTER
: {
1441 negative
= optarg
[0] == '~';
1442 items
= negative
? optarg
+ 1 : optarg
;
1445 _cleanup_free_
char *word
= NULL
;
1447 r
= extract_first_word(&items
, &word
, NULL
, 0);
1453 return log_error_errno(r
, "Failed to parse system call filter: %m");
1456 r
= strv_extend(&arg_syscall_deny_list
, word
);
1458 r
= strv_extend(&arg_syscall_allow_list
, word
);
1463 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1469 _cleanup_free_
char *name
= NULL
;
1472 if (streq(optarg
, "help")) {
1473 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1477 eq
= strchr(optarg
, '=');
1479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1480 "--rlimit= expects an '=' assignment.");
1482 name
= strndup(optarg
, eq
- optarg
);
1486 rl
= rlimit_from_string_harder(name
);
1488 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1490 if (!arg_rlimit
[rl
]) {
1491 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1492 if (!arg_rlimit
[rl
])
1496 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1498 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1500 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1504 case ARG_OOM_SCORE_ADJUST
:
1505 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1507 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1509 arg_oom_score_adjust_set
= true;
1510 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1513 case ARG_CPU_AFFINITY
: {
1516 r
= parse_cpu_set(optarg
, &cpuset
);
1518 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1520 cpu_set_reset(&arg_cpu_set
);
1521 arg_cpu_set
= cpuset
;
1522 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1526 case ARG_RESOLV_CONF
:
1527 if (streq(optarg
, "help")) {
1528 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1532 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1533 if (arg_resolv_conf
< 0)
1534 return log_error_errno(arg_resolv_conf
,
1535 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1537 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1541 if (streq(optarg
, "help")) {
1542 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1546 arg_timezone
= timezone_mode_from_string(optarg
);
1547 if (arg_timezone
< 0)
1548 return log_error_errno(arg_timezone
,
1549 "Failed to parse /etc/localtime mode: %s", optarg
);
1551 arg_settings_mask
|= SETTING_TIMEZONE
;
1555 r
= handle_arg_console(optarg
);
1562 r
= handle_arg_console("pipe");
1568 arg_pager_flags
|= PAGER_DISABLE
;
1571 case ARG_SET_CREDENTIAL
:
1572 r
= machine_credential_set(&arg_credentials
, optarg
);
1576 arg_settings_mask
|= SETTING_CREDENTIALS
;
1579 case ARG_LOAD_CREDENTIAL
:
1580 r
= machine_credential_load(&arg_credentials
, optarg
);
1584 arg_settings_mask
|= SETTING_CREDENTIALS
;
1588 if (!valid_user_group_name(optarg
, 0))
1589 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1591 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1594 arg_settings_mask
|= SETTING_BIND_USER
;
1597 case ARG_SUPPRESS_SYNC
:
1598 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1602 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1605 case ARG_IMAGE_POLICY
:
1606 r
= parse_image_policy_argument(optarg
, &arg_image_policy
);
1615 assert_not_reached();
1618 if (argc
> optind
) {
1619 strv_free(arg_parameters
);
1620 arg_parameters
= strv_copy(argv
+ optind
);
1621 if (!arg_parameters
)
1624 arg_settings_mask
|= SETTING_START_MODE
;
1627 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1628 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1629 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1630 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1632 arg_directory
= TAKE_PTR(arg_template
);
1634 arg_caps_retain
|= plus
;
1635 arg_caps_retain
|= arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0;
1636 arg_caps_retain
&= ~minus
;
1638 /* Make sure to parse environment before we reset the settings mask below */
1639 r
= parse_environment();
1643 /* Load all settings from .nspawn files */
1644 if (mask_no_settings
)
1645 arg_settings_mask
= 0;
1647 /* Don't load any settings from .nspawn files */
1648 if (mask_all_settings
)
1649 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1654 static int verify_arguments(void) {
1657 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1658 /* If we are running the stub init in the container, we don't need to look at what the init
1659 * in the container supports, because we are not using it. Let's immediately pick the right
1660 * setting based on the host system configuration.
1662 * We only do this, if the user didn't use an environment variable to override the detection.
1665 r
= cg_all_unified();
1667 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1669 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1670 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1671 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1673 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1676 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1677 arg_mount_settings
|= MOUNT_USE_USERNS
;
1679 if (arg_private_network
)
1680 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1682 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1683 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1684 arg_register
= false;
1685 if (arg_start_mode
!= START_PID1
)
1686 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1689 if (arg_userns_ownership
< 0)
1690 arg_userns_ownership
=
1691 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1692 USER_NAMESPACE_OWNERSHIP_OFF
;
1694 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1695 arg_kill_signal
= SIGRTMIN
+3;
1697 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 */
1698 arg_read_only
= true;
1700 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1701 arg_read_only
= true;
1703 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1704 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1705 * The latter is not technically a user session, but we don't need to labour the point. */
1706 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1708 if (arg_directory
&& arg_image
)
1709 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1711 if (arg_template
&& arg_image
)
1712 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1714 if (arg_template
&& !(arg_directory
|| arg_machine
))
1715 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1717 if (arg_ephemeral
&& arg_template
)
1718 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1720 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1721 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1723 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1724 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1726 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1727 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1728 "--read-only and --private-users-ownership=chown may not be combined.");
1730 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1731 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1732 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1733 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1734 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1736 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1737 * we need to error out, to avoid conflicts between different network options. */
1738 if (arg_network_namespace_path
&&
1739 (arg_network_interfaces
|| arg_network_macvlan
||
1740 arg_network_ipvlan
|| arg_network_veth_extra
||
1741 arg_network_bridge
|| arg_network_zone
||
1743 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1745 if (arg_network_bridge
&& arg_network_zone
)
1746 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1747 "--network-bridge= and --network-zone= may not be combined.");
1749 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1750 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1752 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1753 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1755 if (arg_expose_ports
&& !arg_private_network
)
1756 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1758 if (arg_caps_ambient
) {
1759 if (arg_caps_ambient
== UINT64_MAX
)
1760 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1762 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1763 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1765 if (arg_start_mode
== START_BOOT
)
1766 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1769 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1770 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1772 /* Drop duplicate --bind-user= entries */
1773 strv_uniq(arg_bind_user
);
1775 r
= custom_mount_check_all();
1782 static int verify_network_interfaces_initialized(void) {
1784 r
= test_network_interfaces_initialized(arg_network_interfaces
);
1788 r
= test_network_interfaces_initialized(arg_network_macvlan
);
1792 r
= test_network_interfaces_initialized(arg_network_ipvlan
);
1799 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1802 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1805 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1808 if (uid
!= UID_INVALID
) {
1809 uid
+= arg_uid_shift
;
1811 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1815 if (gid
!= GID_INVALID
) {
1816 gid
+= (gid_t
) arg_uid_shift
;
1818 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1822 return RET_NERRNO(lchown(p
, uid
, gid
));
1825 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1829 q
= prefix_roota(root
, path
);
1830 r
= RET_NERRNO(mkdir(q
, mode
));
1836 return userns_lchown(q
, uid
, gid
);
1839 static const char *timezone_from_path(const char *path
) {
1840 return PATH_STARTSWITH_SET(
1842 "../usr/share/zoneinfo/",
1843 "/usr/share/zoneinfo/");
1846 static bool etc_writable(void) {
1847 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1850 static int setup_timezone(const char *dest
) {
1851 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1852 const char *where
, *check
;
1858 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1859 r
= readlink_malloc("/etc/localtime", &p
);
1860 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1861 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1862 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1863 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1865 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1866 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1870 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1873 } else if (arg_timezone
== TIMEZONE_AUTO
)
1874 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1880 if (m
== TIMEZONE_OFF
)
1883 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1885 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1889 where
= strjoina(etc
, "/localtime");
1893 case TIMEZONE_DELETE
:
1894 if (unlink(where
) < 0)
1895 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1899 case TIMEZONE_SYMLINK
: {
1900 _cleanup_free_
char *q
= NULL
;
1901 const char *z
, *what
;
1903 z
= timezone_from_path(p
);
1905 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1909 r
= readlink_malloc(where
, &q
);
1910 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1911 return 0; /* Already pointing to the right place? Then do nothing .. */
1913 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1914 r
= chase(check
, dest
, 0, NULL
, NULL
);
1916 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1918 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1919 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1920 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1924 what
= strjoina("../usr/share/zoneinfo/", z
);
1925 if (symlink(what
, where
) < 0) {
1926 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1927 errno
, "Failed to correct timezone of container, ignoring: %m");
1937 case TIMEZONE_BIND
: {
1938 _cleanup_free_
char *resolved
= NULL
;
1941 found
= chase(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1943 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1947 if (found
== 0) /* missing? */
1948 (void) touch(resolved
);
1950 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1952 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1958 /* If mounting failed, try to copy */
1959 r
= copy_file_atomic("/etc/localtime", where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
1961 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1962 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1969 assert_not_reached();
1972 /* Fix permissions of the symlink or file copy we just created */
1973 r
= userns_lchown(where
, 0, 0);
1975 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1980 static int have_resolv_conf(const char *path
) {
1983 if (access(path
, F_OK
) < 0) {
1984 if (errno
== ENOENT
)
1987 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1993 static int resolved_listening(void) {
1994 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1995 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1996 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1999 /* Check if resolved is listening */
2001 r
= sd_bus_open_system(&bus
);
2003 return log_debug_errno(r
, "Failed to open system bus: %m");
2005 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2007 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2011 r
= bus_get_property_string(bus
, bus_resolve_mgr
, "DNSStubListener", &error
, &dns_stub_listener_mode
);
2013 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2015 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2018 static int setup_resolv_conf(const char *dest
) {
2019 _cleanup_free_
char *etc
= NULL
;
2020 const char *where
, *what
;
2026 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2027 if (arg_private_network
)
2028 m
= RESOLV_CONF_OFF
;
2029 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2030 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2031 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2032 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2034 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2037 m
= arg_resolv_conf
;
2039 if (m
== RESOLV_CONF_OFF
)
2042 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2044 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2048 where
= strjoina(etc
, "/resolv.conf");
2050 if (m
== RESOLV_CONF_DELETE
) {
2051 if (unlink(where
) < 0)
2052 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2057 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2058 what
= PRIVATE_STATIC_RESOLV_CONF
;
2059 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2060 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2061 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2062 what
= PRIVATE_STUB_RESOLV_CONF
;
2064 what
= "/etc/resolv.conf";
2066 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2067 _cleanup_free_
char *resolved
= NULL
;
2070 found
= chase(where
, dest
, CHASE_NONEXISTENT
|CHASE_NOFOLLOW
, &resolved
, NULL
);
2072 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2076 if (found
== 0) /* missing? */
2077 (void) touch(resolved
);
2079 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2081 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2083 /* If that didn't work, let's copy the file */
2086 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2087 r
= copy_file_atomic(what
, where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2089 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, COPY_REFLINK
);
2091 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2092 * resolved or something similar runs inside and the symlink points there.
2094 * If the disk image is read-only, there's also no point in complaining.
2096 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2097 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2098 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2102 r
= userns_lchown(where
, 0, 0);
2104 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2109 static int setup_boot_id(void) {
2110 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2111 _cleanup_free_
char *path
= NULL
;
2112 sd_id128_t rnd
= SD_ID128_NULL
;
2116 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2118 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2120 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2122 r
= sd_id128_randomize(&rnd
);
2124 return log_error_errno(r
, "Failed to generate random boot id: %m");
2126 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2128 return log_error_errno(r
, "Failed to write boot id: %m");
2130 from
= TAKE_PTR(path
);
2131 to
= "/proc/sys/kernel/random/boot_id";
2133 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2137 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2140 static int copy_devnodes(const char *dest
) {
2141 static const char devnodes
[] =
2154 BLOCK_WITH_UMASK(0000);
2156 /* Create /dev/net, so that we can create /dev/net/tun in it */
2157 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2158 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2160 NULSTR_FOREACH(d
, devnodes
) {
2161 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2164 from
= path_join("/dev/", d
);
2168 to
= path_join(dest
, from
);
2172 if (stat(from
, &st
) < 0) {
2174 if (errno
!= ENOENT
)
2175 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2177 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2178 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2179 "%s is not a char or block device, cannot copy.", from
);
2181 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2183 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2184 /* Explicitly warn the user when /dev is already populated. */
2185 if (errno
== EEXIST
)
2186 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
);
2188 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2190 /* Some systems abusively restrict mknod but allow bind mounts. */
2193 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2194 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2196 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2199 r
= userns_lchown(to
, 0, 0);
2201 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2203 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2207 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2209 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2211 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2214 prefixed
= path_join(dest
, sl
);
2218 t
= path_join("..", d
);
2222 if (symlink(t
, prefixed
) < 0)
2223 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2230 static int make_extra_nodes(const char *dest
) {
2234 BLOCK_WITH_UMASK(0000);
2236 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2237 _cleanup_free_
char *path
= NULL
;
2238 DeviceNode
*n
= arg_extra_nodes
+ i
;
2240 path
= path_join(dest
, n
->path
);
2244 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2245 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2247 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2249 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2255 static int setup_pts(const char *dest
) {
2256 _cleanup_free_
char *options
= NULL
;
2261 if (arg_selinux_apifs_context
)
2262 (void) asprintf(&options
,
2263 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2264 arg_uid_shift
+ TTY_GID
,
2265 arg_selinux_apifs_context
);
2268 (void) asprintf(&options
,
2269 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2270 arg_uid_shift
+ TTY_GID
);
2275 /* Mount /dev/pts itself */
2276 p
= prefix_roota(dest
, "/dev/pts");
2277 r
= RET_NERRNO(mkdir(p
, 0755));
2279 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2281 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2284 r
= userns_lchown(p
, 0, 0);
2286 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2288 /* Create /dev/ptmx symlink */
2289 p
= prefix_roota(dest
, "/dev/ptmx");
2290 if (symlink("pts/ptmx", p
) < 0)
2291 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2292 r
= userns_lchown(p
, 0, 0);
2294 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2296 /* And fix /dev/pts/ptmx ownership */
2297 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2298 r
= userns_lchown(p
, 0, 0);
2300 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2305 static int setup_stdio_as_dev_console(void) {
2306 _cleanup_close_
int terminal
= -EBADF
;
2309 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2310 * explicitly, if we are configured to. */
2311 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2313 return log_error_errno(terminal
, "Failed to open console: %m");
2315 /* Make sure we can continue logging to the original stderr, even if
2316 * stderr points elsewhere now */
2317 r
= log_dup_console();
2319 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2321 /* invalidates 'terminal' on success and failure */
2322 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2325 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2330 static int setup_dev_console(const char *console
) {
2331 _cleanup_free_
char *p
= NULL
;
2334 /* Create /dev/console symlink */
2335 r
= path_make_relative("/dev", console
, &p
);
2337 return log_error_errno(r
, "Failed to create relative path: %m");
2339 if (symlink(p
, "/dev/console") < 0)
2340 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2345 static int setup_keyring(void) {
2346 key_serial_t keyring
;
2348 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2349 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2350 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2351 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2352 * into the container. */
2354 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2355 if (keyring
== -1) {
2356 if (errno
== ENOSYS
)
2357 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2358 else if (ERRNO_IS_PRIVILEGE(errno
))
2359 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2361 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2367 static int setup_credentials(const char *root
) {
2371 if (arg_credentials
.n_credentials
== 0)
2374 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2376 return log_error_errno(r
, "Failed to create /run/host: %m");
2378 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2380 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2382 q
= prefix_roota(root
, "/run/host/credentials");
2383 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2387 FOREACH_ARRAY(cred
, arg_credentials
.credentials
, arg_credentials
.n_credentials
) {
2388 _cleanup_free_
char *j
= NULL
;
2389 _cleanup_close_
int fd
= -EBADF
;
2391 j
= path_join(q
, cred
->id
);
2395 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2397 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2399 r
= loop_write(fd
, cred
->data
, cred
->size
);
2401 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2403 if (fchmod(fd
, 0400) < 0)
2404 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2406 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2407 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2408 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2412 if (chmod(q
, 0500) < 0)
2413 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2415 r
= userns_lchown(q
, 0, 0);
2419 /* Make both mount and superblock read-only now */
2420 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2424 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2427 static int setup_kmsg(int fd_inner_socket
) {
2428 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2429 _cleanup_free_
char *fifo
= NULL
;
2430 _cleanup_close_
int fd
= -EBADF
;
2433 assert(fd_inner_socket
>= 0);
2435 BLOCK_WITH_UMASK(0000);
2437 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2438 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2439 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2440 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2442 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2444 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2446 if (mkfifo(fifo
, 0600) < 0)
2447 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2449 from
= TAKE_PTR(fifo
);
2451 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2455 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2457 return log_error_errno(errno
, "Failed to open fifo: %m");
2459 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2460 r
= send_one_fd(fd_inner_socket
, fd
, 0);
2462 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2468 union in_addr_union address4
;
2469 union in_addr_union address6
;
2470 struct FirewallContext
*fw_ctx
;
2473 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2474 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2479 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2480 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2484 static int setup_hostname(void) {
2487 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2490 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2492 return log_error_errno(r
, "Failed to set hostname: %m");
2497 static int setup_journal(const char *directory
) {
2498 _cleanup_free_
char *d
= NULL
;
2504 /* Don't link journals in ephemeral mode */
2508 if (arg_link_journal
== LINK_NO
)
2511 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2513 r
= sd_id128_get_machine(&this_id
);
2515 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2517 if (sd_id128_equal(arg_uuid
, this_id
)) {
2518 log_full(try ? LOG_WARNING
: LOG_ERR
,
2519 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2525 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2526 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2528 bool ignore
= r
== -EROFS
&& try;
2529 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2530 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2531 return ignore
? 0 : r
;
2535 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2536 q
= prefix_roota(directory
, p
);
2538 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2542 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2543 "%s: already a mount point, refusing to use for journal", p
);
2546 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2550 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2551 "%s: already a mount point, refusing to use for journal", q
);
2554 r
= readlink_and_make_absolute(p
, &d
);
2556 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2559 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2561 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2566 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2567 } else if (r
== -EINVAL
) {
2569 if (arg_link_journal
== LINK_GUEST
&&
2572 if (errno
== ENOTDIR
) {
2573 log_error("%s already exists and is neither a symlink nor a directory", p
);
2576 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2578 } else if (r
!= -ENOENT
)
2579 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2581 if (arg_link_journal
== LINK_GUEST
) {
2583 if (symlink(q
, p
) < 0) {
2585 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2588 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2591 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2593 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2597 if (arg_link_journal
== LINK_HOST
) {
2598 /* don't create parents here — if the host doesn't have
2599 * permanent journal set up, don't force it here */
2601 r
= RET_NERRNO(mkdir(p
, 0755));
2602 if (r
< 0 && r
!= -EEXIST
) {
2604 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2607 return log_error_errno(r
, "Failed to create %s: %m", p
);
2610 } else if (access(p
, F_OK
) < 0)
2613 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2614 log_warning("%s is not empty, proceeding anyway.", q
);
2616 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2618 return log_error_errno(r
, "Failed to create %s: %m", q
);
2620 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2622 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2627 static int drop_capabilities(uid_t uid
) {
2628 CapabilityQuintet q
;
2630 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2631 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2632 * arg_caps_retain. */
2634 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2635 q
= arg_full_capabilities
;
2637 if (q
.bounding
== UINT64_MAX
)
2638 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2640 if (q
.effective
== UINT64_MAX
)
2641 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2643 if (q
.inheritable
== UINT64_MAX
)
2644 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2646 if (q
.permitted
== UINT64_MAX
)
2647 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2649 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2650 q
.ambient
= arg_caps_ambient
;
2652 if (capability_quintet_mangle(&q
))
2653 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2656 q
= (CapabilityQuintet
) {
2657 .bounding
= arg_caps_retain
,
2658 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2659 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2660 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2661 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2664 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2665 * in order to maintain the same behavior as systemd < 242. */
2666 if (capability_quintet_mangle(&q
))
2667 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2668 "Some capabilities will not be set because they are not in the current bounding set.");
2672 return capability_quintet_enforce(&q
);
2675 static int reset_audit_loginuid(void) {
2676 _cleanup_free_
char *p
= NULL
;
2679 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2682 r
= read_one_line_file("/proc/self/loginuid", &p
);
2686 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2688 /* Already reset? */
2689 if (streq(p
, "4294967295"))
2692 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2695 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2696 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2697 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2698 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2699 "using systemd-nspawn. Sleeping for 5s... (%m)");
2707 static int mount_tunnel_dig(const char *root
) {
2711 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2712 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2713 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2714 (void) mkdir_p(p
, 0600);
2716 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2718 return log_error_errno(r
, "Failed to create /run/host: %m");
2720 r
= userns_mkdir(root
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2722 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2724 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
2725 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2729 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2736 static int mount_tunnel_open(void) {
2739 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2746 static int setup_machine_id(const char *directory
) {
2749 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2750 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2751 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2752 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2753 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2754 * container behaves nicely). */
2756 r
= id128_get_machine(directory
, &arg_uuid
);
2757 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) {
2758 /* If the file is missing, empty, or uninitialized, we don't mind */
2759 if (sd_id128_is_null(arg_uuid
)) {
2760 r
= sd_id128_randomize(&arg_uuid
);
2762 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2765 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2770 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2775 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2778 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2779 if (r
== -EOPNOTSUPP
)
2780 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.");
2782 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2784 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2786 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2788 log_debug("Patched directory tree to match UID/GID range.");
2795 * < 0 : wait_for_terminate() failed to get the state of the
2796 * container, the container was terminated by a signal, or
2797 * failed for an unknown reason. No change is made to the
2798 * container argument.
2799 * > 0 : The program executed in the container terminated with an
2800 * error. The exit code of the program executed in the
2801 * container is returned. The container argument has been set
2802 * to CONTAINER_TERMINATED.
2803 * 0 : The container is being rebooted, has been shut down or exited
2804 * successfully. The container argument has been set to either
2805 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2807 * That is, success is indicated by a return value of zero, and an
2808 * error is indicated by a non-zero value.
2810 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2814 r
= wait_for_terminate(pid
, &status
);
2816 return log_warning_errno(r
, "Failed to wait for container: %m");
2818 switch (status
.si_code
) {
2821 if (status
.si_status
== 0)
2822 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2824 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2826 *container
= CONTAINER_TERMINATED
;
2827 return status
.si_status
;
2830 if (status
.si_status
== SIGINT
) {
2831 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2832 *container
= CONTAINER_TERMINATED
;
2835 } else if (status
.si_status
== SIGHUP
) {
2836 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2837 *container
= CONTAINER_REBOOTED
;
2843 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2844 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2847 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2848 "Container %s failed due to unknown reason.", arg_machine
);
2852 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2855 pid
= PTR_TO_PID(userdata
);
2857 if (kill(pid
, arg_kill_signal
) >= 0) {
2858 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2859 sd_event_source_set_userdata(s
, NULL
);
2864 sd_event_exit(sd_event_source_get_event(s
), 0);
2868 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2874 pid
= PTR_TO_PID(userdata
);
2879 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2880 return log_error_errno(errno
, "Failed to waitid(): %m");
2881 if (si
.si_pid
== 0) /* No pending children. */
2883 if (si
.si_pid
== pid
) {
2884 /* The main process we care for has exited. Return from
2885 * signal handler but leave the zombie. */
2886 sd_event_exit(sd_event_source_get_event(s
), 0);
2890 /* Reap all other children. */
2891 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2897 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2902 pid
= PTR_TO_PID(userdata
);
2904 if (arg_kill_signal
> 0) {
2905 log_info("Container termination requested. Attempting to halt container.");
2906 (void) kill(pid
, arg_kill_signal
);
2908 log_info("Container termination requested. Exiting.");
2909 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2915 static int pick_paths(void) {
2918 if (arg_directory
) {
2919 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2920 PickFilter filter
= pick_filter_image_dir
;
2922 filter
.architecture
= arg_architecture
;
2924 r
= path_pick_update_warn(
2927 PICK_ARCHITECTURE
|PICK_TRIES
,
2930 /* Accept ENOENT here so that the --template= logic can work */
2934 arg_architecture
= result
.architecture
;
2938 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2939 PickFilter filter
= pick_filter_image_raw
;
2941 filter
.architecture
= arg_architecture
;
2943 r
= path_pick_update_warn(
2946 PICK_ARCHITECTURE
|PICK_TRIES
,
2951 arg_architecture
= result
.architecture
;
2955 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2956 PickFilter filter
= pick_filter_image_dir
;
2958 filter
.architecture
= arg_architecture
;
2960 r
= path_pick_update_warn(
2968 arg_architecture
= result
.architecture
;
2974 static int determine_names(void) {
2977 if (arg_template
&& !arg_directory
&& arg_machine
) {
2979 /* If --template= was specified then we should not search for a machine, but instead create a
2980 * new one in /var/lib/machine. */
2982 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2987 if (!arg_image
&& !arg_directory
) {
2989 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2991 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
2993 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2995 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2997 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2998 r
= free_and_strdup(&arg_image
, i
->path
);
3000 r
= free_and_strdup(&arg_directory
, i
->path
);
3005 arg_read_only
= arg_read_only
|| i
->read_only
;
3007 r
= safe_getcwd(&arg_directory
);
3009 return log_error_errno(r
, "Failed to determine current directory: %m");
3012 if (!arg_directory
&& !arg_image
)
3013 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3017 if (arg_directory
&& path_equal(arg_directory
, "/"))
3018 arg_machine
= gethostname_malloc();
3019 else if (arg_image
) {
3022 r
= path_extract_filename(arg_image
, &arg_machine
);
3024 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_image
);
3026 /* Truncate suffix if there is one */
3027 e
= endswith(arg_machine
, ".raw");
3031 r
= path_extract_filename(arg_directory
, &arg_machine
);
3033 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_directory
);
3036 hostname_cleanup(arg_machine
);
3037 if (!hostname_is_valid(arg_machine
, 0))
3038 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3040 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3041 * to match fixed config file names. */
3042 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3043 if (!arg_settings_filename
)
3046 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3047 * instances at once without manually having to specify -M each time. */
3049 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3052 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3053 if (!arg_settings_filename
)
3060 static int chase_and_update(char **p
, unsigned flags
) {
3069 r
= chase(*p
, NULL
, flags
, &chased
, NULL
);
3071 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3073 return free_and_replace(*p
, chased
);
3076 static int determine_uid_shift(const char *directory
) {
3078 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3083 if (arg_uid_shift
== UID_INVALID
) {
3086 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3088 if (stat(directory
, &st
) < 0)
3089 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3091 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3093 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3094 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3095 "UID and GID base of %s don't match.", directory
);
3097 arg_uid_range
= UINT32_C(0x10000);
3099 if (arg_uid_shift
!= 0) {
3100 /* If the image is shifted already, then we'll fall back to classic chowning, for
3101 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3103 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3104 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3105 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3106 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3107 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3108 "UID base of %s is not zero, UID mapping not supported.", directory
);
3112 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3113 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3118 static unsigned long effective_clone_ns_flags(void) {
3119 unsigned long flags
= arg_clone_ns_flags
;
3121 if (arg_private_network
)
3122 flags
|= CLONE_NEWNET
;
3124 flags
|= CLONE_NEWCGROUP
;
3125 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3126 flags
|= CLONE_NEWUSER
;
3131 static int patch_sysctl(void) {
3133 /* This table is inspired by runc's sysctl() function */
3134 static const struct {
3137 unsigned long clone_flags
;
3139 { "kernel.hostname", false, CLONE_NEWUTS
},
3140 { "kernel.domainname", false, CLONE_NEWUTS
},
3141 { "kernel.msgmax", false, CLONE_NEWIPC
},
3142 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3143 { "kernel.msgmni", false, CLONE_NEWIPC
},
3144 { "kernel.sem", false, CLONE_NEWIPC
},
3145 { "kernel.shmall", false, CLONE_NEWIPC
},
3146 { "kernel.shmmax", false, CLONE_NEWIPC
},
3147 { "kernel.shmmni", false, CLONE_NEWIPC
},
3148 { "fs.mqueue.", true, CLONE_NEWIPC
},
3149 { "net.", true, CLONE_NEWNET
},
3152 unsigned long flags
;
3155 flags
= effective_clone_ns_flags();
3157 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3161 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3163 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3166 if (safe_sysctl
[i
].prefix
)
3167 good
= startswith(*k
, safe_sysctl
[i
].key
);
3169 good
= streq(*k
, safe_sysctl
[i
].key
);
3176 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3178 r
= sysctl_write(*k
, *v
);
3180 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3186 static int inner_child(
3188 int fd_inner_socket
,
3190 char **os_release_pairs
) {
3192 _cleanup_free_
char *home
= NULL
;
3195 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3196 NULL
, /* container */
3201 NULL
, /* container_uuid */
3202 NULL
, /* LISTEN_FDS */
3203 NULL
, /* LISTEN_PID */
3204 NULL
, /* NOTIFY_SOCKET */
3205 NULL
, /* CREDENTIALS_DIRECTORY */
3209 const char *exec_target
;
3210 _cleanup_strv_free_
char **env_use
= NULL
;
3211 int r
, which_failed
;
3213 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3214 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3215 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3216 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3217 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3218 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3221 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3222 * unshare(). See below. */
3225 assert(fd_inner_socket
>= 0);
3227 log_debug("Inner child is initializing.");
3229 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3230 /* Tell the parent, that it now can write the UID map. */
3231 (void) barrier_place(barrier
); /* #1 */
3233 /* Wait until the parent wrote the UID map */
3234 if (!barrier_place_and_sync(barrier
)) /* #2 */
3235 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3237 /* Become the new root user inside our namespace */
3238 r
= reset_uid_gid();
3240 return log_error_errno(r
, "Couldn't become new root: %m");
3242 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3243 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3244 * propagation, but simply create new peer groups for all our mounts). */
3245 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3251 arg_mount_settings
| MOUNT_IN_USERNS
,
3253 arg_selinux_apifs_context
);
3257 if (!arg_network_namespace_path
&& arg_private_network
) {
3258 r
= unshare(CLONE_NEWNET
);
3260 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3262 /* Tell the parent that it can setup network interfaces. */
3263 (void) barrier_place(barrier
); /* #3 */
3266 r
= mount_sysfs(NULL
, arg_mount_settings
);
3270 /* Wait until we are cgroup-ified, so that we
3271 * can mount the right cgroup path writable */
3272 if (!barrier_place_and_sync(barrier
)) /* #4 */
3273 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3274 "Parent died too early");
3277 r
= unshare(CLONE_NEWCGROUP
);
3279 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3282 arg_unified_cgroup_hierarchy
,
3283 arg_userns_mode
!= USER_NAMESPACE_NO
,
3286 arg_selinux_apifs_context
,
3289 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3293 r
= setup_boot_id();
3297 r
= setup_kmsg(fd_inner_socket
);
3304 arg_n_custom_mounts
,
3307 arg_selinux_apifs_context
,
3308 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3313 return log_error_errno(errno
, "setsid() failed: %m");
3315 if (arg_private_network
)
3316 (void) loopback_setup();
3318 if (arg_expose_ports
) {
3319 r
= expose_port_send_rtnl(fd_inner_socket
);
3324 if (arg_console_mode
!= CONSOLE_PIPE
) {
3325 _cleanup_close_
int master
= -EBADF
;
3326 _cleanup_free_
char *console
= NULL
;
3328 /* Allocate a pty and make it available as /dev/console. */
3329 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3331 return log_error_errno(master
, "Failed to allocate a pty: %m");
3333 r
= setup_dev_console(console
);
3335 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3337 r
= send_one_fd(fd_inner_socket
, master
, 0);
3339 return log_error_errno(r
, "Failed to send master fd: %m");
3341 r
= setup_stdio_as_dev_console();
3350 if (arg_oom_score_adjust_set
) {
3351 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3353 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3356 if (arg_cpu_set
.set
)
3357 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3358 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3360 (void) setup_hostname();
3362 if (arg_personality
!= PERSONALITY_INVALID
) {
3363 r
= safe_personality(arg_personality
);
3365 return log_error_errno(r
, "personality() failed: %m");
3366 #ifdef ARCHITECTURE_SECONDARY
3367 } else if (arg_architecture
== ARCHITECTURE_SECONDARY
) {
3368 r
= safe_personality(PER_LINUX32
);
3370 return log_error_errno(r
, "personality() failed: %m");
3372 } else if (!arg_quiet
&& arg_architecture
>= 0 && arg_architecture
!= native_architecture())
3373 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3374 "invocation with qemu userspace emulator (or equivalent) in effect.",
3375 architecture_to_string(arg_architecture
));
3377 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3379 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3384 if (is_seccomp_available()) {
3385 r
= seccomp_load(arg_seccomp
);
3386 if (ERRNO_IS_NEG_SECCOMP_FATAL(r
))
3387 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3389 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3394 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3399 if (arg_suppress_sync
) {
3401 r
= seccomp_suppress_sync();
3403 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3405 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3410 if (arg_selinux_context
)
3411 if (setexeccon(arg_selinux_context
) < 0)
3412 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3415 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3416 * if we need to later on. */
3417 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3418 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3420 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3421 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3423 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3427 r
= drop_capabilities(getuid());
3429 return log_error_errno(r
, "Dropping capabilities failed: %m");
3431 if (arg_no_new_privileges
)
3432 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3433 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3435 /* LXC sets container=lxc, so follow the scheme here */
3436 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3438 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3442 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3443 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3446 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3447 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3448 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
?: "root") < 0)
3451 assert(!sd_id128_is_null(arg_uuid
));
3453 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3456 if (!fdset_isempty(fds
)) {
3457 r
= fdset_cloexec(fds
, false);
3459 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3461 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3462 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3465 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3468 if (arg_credentials
.n_credentials
> 0) {
3469 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3475 if (arg_start_mode
!= START_BOOT
) {
3476 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3482 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3486 /* Let the parent know that we are ready and wait until the parent is ready with the setup, too... */
3487 if (!barrier_place_and_sync(barrier
)) /* #5 */
3488 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3491 if (chdir(arg_chdir
) < 0)
3492 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3494 if (arg_start_mode
== START_PID2
) {
3495 r
= stub_pid1(arg_uuid
);
3500 if (arg_console_mode
!= CONSOLE_PIPE
) {
3501 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3502 * are configured for that. Acquire it as controlling tty. */
3503 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3504 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3507 log_debug("Inner child completed, invoking payload.");
3509 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3510 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3511 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3513 log_set_open_when_needed(true);
3514 log_settle_target();
3516 (void) fdset_close_others(fds
);
3518 if (arg_start_mode
== START_BOOT
) {
3522 /* Automatically search for the init system */
3524 m
= strv_length(arg_parameters
);
3525 a
= newa(char*, m
+ 2);
3526 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3529 FOREACH_STRING(init
,
3530 "/usr/lib/systemd/systemd",
3531 "/lib/systemd/systemd",
3533 a
[0] = (char*) init
;
3534 execve(a
[0], a
, env_use
);
3537 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3538 } else if (!strv_isempty(arg_parameters
)) {
3539 const char *dollar_path
;
3541 exec_target
= arg_parameters
[0];
3543 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3545 dollar_path
= strv_env_get(env_use
, "PATH");
3547 if (setenv("PATH", dollar_path
, 1) < 0)
3548 return log_error_errno(errno
, "Failed to update $PATH: %m");
3551 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3554 /* If we cannot change the directory, we'll end up in /, that is expected. */
3555 (void) chdir(home
?: "/root");
3557 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3558 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3559 execle("/bin/bash", "-bash", NULL
, env_use
);
3560 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3561 execle("/bin/sh", "-sh", NULL
, env_use
);
3563 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3566 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3569 static int setup_notify_child(void) {
3570 _cleanup_close_
int fd
= -EBADF
;
3571 static const union sockaddr_union sa
= {
3572 .un
.sun_family
= AF_UNIX
,
3573 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3577 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3579 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3581 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3582 (void) sockaddr_un_unlink(&sa
.un
);
3584 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3586 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3588 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3590 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3592 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3594 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3599 static int outer_child(
3601 const char *directory
,
3602 DissectedImage
*dissected_image
,
3603 int fd_outer_socket
,
3604 int fd_inner_socket
,
3608 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3609 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3610 _cleanup_close_
int fd
= -EBADF
, mntns_fd
= -EBADF
;
3617 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3618 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3619 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3620 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3621 * forked off it, and it exits. */
3625 assert(fd_outer_socket
>= 0);
3626 assert(fd_inner_socket
>= 0);
3628 log_debug("Outer child is initializing.");
3630 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3632 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3634 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3635 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3637 r
= reset_audit_loginuid();
3641 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3642 * mounts to the real root. */
3643 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3647 if (dissected_image
) {
3648 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3649 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3650 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3651 * right place right away. This makes sure ESP partitions and userns are compatible. */
3653 r
= dissected_image_mount_and_warn(
3658 /* userns_fd= */ -EBADF
,
3659 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3660 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3661 DISSECT_IMAGE_USR_NO_ROOT
|
3662 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3663 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3668 r
= determine_uid_shift(directory
);
3672 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3673 r
= namespace_open(0, NULL
, &mntns_fd
, NULL
, NULL
, NULL
);
3675 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3677 l
= send_one_fd(fd_outer_socket
, mntns_fd
, 0);
3679 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3680 mntns_fd
= safe_close(mntns_fd
);
3682 /* Let the parent know which UID shift we read from the image */
3683 l
= send(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3685 return log_error_errno(errno
, "Failed to send UID shift: %m");
3686 if (l
!= sizeof(arg_uid_shift
))
3687 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3688 "Short write while sending UID shift.");
3690 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3691 /* When we are supposed to pick the UID shift, the parent will check now whether the
3692 * UID shift we just read from the image is available. If yes, it will send the UID
3693 * shift back to us, if not it will pick a different one, and send it back to us. */
3695 l
= recv(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3697 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3698 if (l
!= sizeof(arg_uid_shift
))
3699 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3700 "Short read while receiving UID shift.");
3703 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3704 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3707 if (path_equal(directory
, "/")) {
3708 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3709 * place, so that we can make changes to its mount structure (for example, to implement
3710 * --volatile=) without this interfering with our ability to access files such as
3711 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3712 * (instead of a temporary directory, since we are living in our own mount namespace here
3713 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3714 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3716 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3720 directory
= "/run/systemd/nspawn-root";
3723 /* Make sure we always have a mount that we can move to root later on. */
3724 r
= make_mount_point(directory
);
3728 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3729 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3730 * we'll live in our own little world from now on, and propagation from the host may only happen via
3731 * the mount tunnel dir, or not at all. */
3732 r
= mount_follow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_PRIVATE
|MS_REC
, NULL
);
3736 r
= setup_pivot_root(
3739 arg_pivot_root_old
);
3743 r
= setup_volatile_mode(
3747 arg_selinux_apifs_context
);
3751 r
= bind_user_prepare(
3756 &arg_custom_mounts
, &arg_n_custom_mounts
,
3757 &bind_user_context
);
3761 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3762 /* Send the user maps we determined to the parent, so that it installs it in our user
3763 * namespace UID map table */
3765 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3767 bind_user_context
->data
[i
].payload_user
->uid
,
3768 bind_user_context
->data
[i
].host_user
->uid
,
3769 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3770 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3773 l
= send(fd_outer_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3775 return log_error_errno(errno
, "Failed to send user UID map: %m");
3776 if (l
!= sizeof(map
))
3777 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3778 "Short write while sending user UID map.");
3785 arg_n_custom_mounts
,
3788 arg_selinux_apifs_context
,
3793 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3794 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3795 arg_uid_shift
!= 0) {
3796 _cleanup_free_
char *usr_subtree
= NULL
;
3800 dirs
[i
++] = (char*) directory
;
3802 if (dissected_image
&& dissected_image
->partitions
[PARTITION_USR
].found
) {
3803 usr_subtree
= path_join(directory
, "/usr");
3807 dirs
[i
++] = usr_subtree
;
3812 r
= remount_idmap(dirs
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3813 if (r
== -EINVAL
|| ERRNO_IS_NEG_NOT_SUPPORTED(r
)) {
3814 /* This might fail because the kernel or file system doesn't support idmapping. We
3815 * can't really distinguish this nicely, nor do we have any guarantees about the
3816 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3817 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3818 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3819 "ID mapped mounts are apparently not available, sorry.");
3821 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3822 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3824 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3826 log_debug("ID mapped mounts available, making use of them.");
3831 if (dissected_image
) {
3832 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3833 r
= dissected_image_mount(
3838 /* userns_fd= */ -EBADF
,
3839 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3840 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3841 DISSECT_IMAGE_USR_NO_ROOT
|
3842 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3843 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3845 return log_error_errno(r
, "File system check for image failed: %m");
3847 return log_error_errno(r
, "Failed to mount image file system: %m");
3850 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3851 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3853 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3857 l
= send(fd_outer_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3859 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3860 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3861 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3862 "Short write while sending cgroup mode.");
3865 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3869 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3873 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3874 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3875 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3877 return log_error_errno(r
, "Failed to make tree read-only: %m");
3880 r
= mount_all(directory
,
3883 arg_selinux_apifs_context
);
3887 r
= copy_devnodes(directory
);
3891 r
= make_extra_nodes(directory
);
3895 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3897 p
= prefix_roota(directory
, "/run/host");
3898 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3900 r
= setup_pts(directory
);
3904 r
= mount_tunnel_dig(directory
);
3908 r
= setup_keyring();
3912 r
= setup_credentials(directory
);
3916 r
= bind_user_setup(bind_user_context
, directory
);
3923 arg_n_custom_mounts
,
3926 arg_selinux_apifs_context
,
3927 MOUNT_NON_ROOT_ONLY
);
3931 r
= setup_timezone(directory
);
3935 r
= setup_resolv_conf(directory
);
3939 r
= setup_machine_id(directory
);
3943 r
= setup_journal(directory
);
3947 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3948 p
= prefix_roota(directory
, "/run/host/container-manager");
3949 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
);
3951 /* The same stuff as the $container_uuid env var */
3952 p
= prefix_roota(directory
, "/run/host/container-uuid");
3953 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
3955 if (!arg_use_cgns
) {
3958 arg_unified_cgroup_hierarchy
,
3959 arg_userns_mode
!= USER_NAMESPACE_NO
,
3962 arg_selinux_apifs_context
,
3968 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3969 * mounts available in systemd services inside the container that create a new mount namespace. See
3970 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3971 * will inherit the shared propagation mode.
3973 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3974 * directory mount to root later on.
3975 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3977 r
= mount_switch_root(directory
, MS_SHARED
);
3979 return log_error_errno(r
, "Failed to move root directory: %m");
3981 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
3982 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
3984 r
= mount_tunnel_open();
3988 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3989 /* In order to mount procfs and sysfs in an unprivileged container the kernel
3990 * requires that a fully visible instance is already present in the target mount
3991 * namespace. Mount one here so the inner child can mount its own instances. Later
3992 * we umount the temporary instances created here before we actually exec the
3993 * payload. Since the rootfs is shared the umount will propagate into the container.
3994 * Note, the inner child wouldn't be able to unmount the instances on its own since
3995 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
3997 r
= pin_fully_visible_fs();
4002 fd
= setup_notify_child();
4006 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
4007 arg_clone_ns_flags
|
4008 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
4010 return log_error_errno(errno
, "Failed to fork inner child: %m");
4012 fd_outer_socket
= safe_close(fd_outer_socket
);
4014 /* The inner child has all namespaces that are requested, so that we all are owned by the
4015 * user if user namespaces are turned on. */
4017 if (arg_network_namespace_path
) {
4018 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
4020 return log_error_errno(r
, "Failed to join network namespace: %m");
4023 r
= inner_child(barrier
, fd_inner_socket
, fds
, os_release_pairs
);
4025 _exit(EXIT_FAILURE
);
4027 _exit(EXIT_SUCCESS
);
4030 l
= send(fd_outer_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4032 return log_error_errno(errno
, "Failed to send PID: %m");
4033 if (l
!= sizeof(pid
))
4034 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4035 "Short write while sending PID.");
4037 l
= send(fd_outer_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4039 return log_error_errno(errno
, "Failed to send machine ID: %m");
4040 if (l
!= sizeof(arg_uuid
))
4041 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4042 "Short write while sending machine ID.");
4044 l
= send_one_fd(fd_outer_socket
, fd
, 0);
4046 return log_error_errno(l
, "Failed to send notify fd: %m");
4048 fd_outer_socket
= safe_close(fd_outer_socket
);
4049 fd_inner_socket
= safe_close(fd_inner_socket
);
4050 netns_fd
= safe_close(netns_fd
);
4055 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4056 bool tried_hashed
= false;
4057 unsigned n_tries
= 100;
4062 assert(ret_lock_file
);
4063 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4064 assert(arg_uid_range
== 0x10000U
);
4068 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4071 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4072 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4077 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4079 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4082 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4083 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4084 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4089 /* Make some superficial checks whether the range is currently known in the user database */
4090 if (getpwuid(candidate
))
4092 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4094 if (getgrgid(candidate
))
4096 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4099 *ret_lock_file
= lf
;
4100 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4105 if (arg_machine
&& !tried_hashed
) {
4106 /* Try to hash the base from the container name */
4108 static const uint8_t hash_key
[] = {
4109 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4110 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4113 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4115 tried_hashed
= true;
4117 random_bytes(&candidate
, sizeof(candidate
));
4119 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4120 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4124 static int add_one_uid_map(
4126 uid_t container_uid
,
4130 return strextendf(p
,
4131 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4132 container_uid
, host_uid
, range
);
4135 static int make_uid_map_string(
4136 const uid_t bind_user_uid
[],
4137 size_t n_bind_user_uid
,
4141 _cleanup_free_
char *s
= NULL
;
4142 uid_t previous_uid
= 0;
4145 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4146 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4149 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4150 * quadruplet, consisting of host and container UID + GID. */
4152 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4153 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4154 host_uid
= bind_user_uid
[i
*4+offset
+1];
4156 assert(previous_uid
<= payload_uid
);
4157 assert(payload_uid
< arg_uid_range
);
4159 /* Add a range to close the gap to previous entry */
4160 if (payload_uid
> previous_uid
) {
4161 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4166 /* Map this specific user */
4167 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4171 previous_uid
= payload_uid
+ 1;
4174 /* And add a range to close the gap to finish the range */
4175 if (arg_uid_range
> previous_uid
) {
4176 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4187 static int setup_uid_map(
4189 const uid_t bind_user_uid
[],
4190 size_t n_bind_user_uid
) {
4192 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4193 _cleanup_free_
char *s
= NULL
;
4198 /* Build the UID map string */
4199 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4202 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4203 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4205 return log_error_errno(r
, "Failed to write UID map: %m");
4207 /* And now build the GID map string */
4209 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4212 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4213 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4215 return log_error_errno(r
, "Failed to write GID map: %m");
4220 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4221 char buf
[NOTIFY_BUFFER_MAX
+1];
4223 struct iovec iovec
= {
4225 .iov_len
= sizeof(buf
)-1,
4227 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4228 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4229 struct msghdr msghdr
= {
4232 .msg_control
= &control
,
4233 .msg_controllen
= sizeof(control
),
4235 struct ucred
*ucred
;
4237 pid_t inner_child_pid
;
4238 _cleanup_strv_free_
char **tags
= NULL
;
4243 inner_child_pid
= PTR_TO_PID(userdata
);
4245 if (revents
!= EPOLLIN
) {
4246 log_warning("Got unexpected poll event for notify fd.");
4250 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4251 if (ERRNO_IS_NEG_TRANSIENT(n
))
4253 else if (n
== -EXFULL
) {
4254 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4257 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4259 cmsg_close_all(&msghdr
);
4261 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4262 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4263 log_debug("Received notify message without valid credentials. Ignoring.");
4267 if ((size_t) n
>= sizeof(buf
)) {
4268 log_warning("Received notify message exceeded maximum size. Ignoring.");
4273 tags
= strv_split(buf
, "\n\r");
4277 if (strv_contains(tags
, "READY=1")) {
4278 r
= sd_notify(false, "READY=1\n");
4280 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4283 p
= strv_find_startswith(tags
, "STATUS=");
4285 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4290 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4293 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4295 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4297 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4302 static int merge_settings(Settings
*settings
, const char *path
) {
4308 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4309 * that this steals the fields of the Settings* structure, and hence modifies it. */
4311 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4312 settings
->start_mode
>= 0) {
4313 arg_start_mode
= settings
->start_mode
;
4314 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4317 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4318 settings
->ephemeral
>= 0)
4319 arg_ephemeral
= settings
->ephemeral
;
4321 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4324 if (!arg_settings_trusted
)
4325 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4327 free_and_replace(arg_directory
, settings
->root
);
4330 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4331 settings
->pivot_root_new
) {
4332 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4333 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4336 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4337 settings
->working_directory
)
4338 free_and_replace(arg_chdir
, settings
->working_directory
);
4340 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4341 settings
->environment
)
4342 strv_free_and_replace(arg_setenv
, settings
->environment
);
4344 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4347 free_and_replace(arg_user
, settings
->user
);
4349 if (uid_is_valid(settings
->uid
))
4350 arg_uid
= settings
->uid
;
4351 if (gid_is_valid(settings
->gid
))
4352 arg_gid
= settings
->gid
;
4353 if (settings
->n_supplementary_gids
> 0) {
4354 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4355 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4359 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4360 uint64_t plus
, minus
;
4361 uint64_t network_minus
= 0;
4364 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4365 * Settings structure */
4367 plus
= settings
->capability
;
4368 minus
= settings
->drop_capability
;
4370 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4371 settings_network_configured(settings
)) {
4372 if (settings_private_network(settings
))
4373 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4375 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4378 if (!arg_settings_trusted
&& plus
!= 0) {
4379 if (settings
->capability
!= 0)
4380 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4382 arg_caps_retain
&= ~network_minus
;
4383 arg_caps_retain
|= plus
;
4386 arg_caps_retain
&= ~minus
;
4388 /* Copy the full capabilities over too */
4389 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4390 if (!arg_settings_trusted
)
4391 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4393 arg_full_capabilities
= settings
->full_capabilities
;
4396 ambient
= settings
->ambient_capability
;
4397 if (!arg_settings_trusted
&& ambient
!= 0)
4398 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4400 arg_caps_ambient
|= ambient
;
4403 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4404 settings
->kill_signal
> 0)
4405 arg_kill_signal
= settings
->kill_signal
;
4407 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4408 settings
->personality
!= PERSONALITY_INVALID
)
4409 arg_personality
= settings
->personality
;
4411 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4412 !sd_id128_is_null(settings
->machine_id
)) {
4414 if (!arg_settings_trusted
)
4415 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4417 arg_uuid
= settings
->machine_id
;
4420 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4421 settings
->read_only
>= 0)
4422 arg_read_only
= settings
->read_only
;
4424 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4425 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4426 arg_volatile_mode
= settings
->volatile_mode
;
4428 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4429 settings
->n_custom_mounts
> 0) {
4431 if (!arg_settings_trusted
)
4432 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4434 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4435 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4436 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4437 settings
->n_custom_mounts
= 0;
4441 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4442 settings_network_configured(settings
)) {
4444 if (!arg_settings_trusted
)
4445 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4447 arg_network_veth
= settings_network_veth(settings
);
4448 arg_private_network
= settings_private_network(settings
);
4450 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4451 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4452 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4453 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4455 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4456 free_and_replace(arg_network_zone
, settings
->network_zone
);
4458 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4462 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4463 settings
->expose_ports
) {
4465 if (!arg_settings_trusted
)
4466 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4468 expose_port_free_all(arg_expose_ports
);
4469 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4473 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4474 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4476 if (!arg_settings_trusted
)
4477 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4479 arg_userns_mode
= settings
->userns_mode
;
4480 arg_uid_shift
= settings
->uid_shift
;
4481 arg_uid_range
= settings
->uid_range
;
4482 arg_userns_ownership
= settings
->userns_ownership
;
4486 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4487 !strv_isempty(settings
->bind_user
))
4488 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4490 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4491 settings
->notify_ready
>= 0)
4492 arg_notify_ready
= settings
->notify_ready
;
4494 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4496 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4497 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4498 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4500 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4501 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4506 if (settings
->seccomp
) {
4507 if (!arg_settings_trusted
)
4508 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4510 seccomp_release(arg_seccomp
);
4511 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4517 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4518 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4521 if (!settings
->rlimit
[rl
])
4524 if (!arg_settings_trusted
) {
4525 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4529 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4532 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4534 free_and_replace(arg_hostname
, settings
->hostname
);
4536 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4537 settings
->no_new_privileges
>= 0)
4538 arg_no_new_privileges
= settings
->no_new_privileges
;
4540 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4541 settings
->oom_score_adjust_set
) {
4543 if (!arg_settings_trusted
)
4544 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4546 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4547 arg_oom_score_adjust_set
= true;
4551 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4552 settings
->cpu_set
.set
) {
4554 if (!arg_settings_trusted
)
4555 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4557 cpu_set_reset(&arg_cpu_set
);
4558 arg_cpu_set
= TAKE_STRUCT(settings
->cpu_set
);
4562 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4563 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4564 arg_resolv_conf
= settings
->resolv_conf
;
4566 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4567 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4569 if (!arg_settings_trusted
)
4570 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4572 arg_link_journal
= settings
->link_journal
;
4573 arg_link_journal_try
= settings
->link_journal_try
;
4577 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4578 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4579 arg_timezone
= settings
->timezone
;
4581 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4584 if (!arg_settings_trusted
)
4585 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4587 free_and_replace(arg_slice
, settings
->slice
);
4590 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4591 settings
->use_cgns
>= 0) {
4593 if (!arg_settings_trusted
)
4594 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4596 arg_use_cgns
= settings
->use_cgns
;
4599 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4600 settings
->clone_ns_flags
!= ULONG_MAX
) {
4602 if (!arg_settings_trusted
)
4603 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4605 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4608 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4609 settings
->console_mode
>= 0) {
4611 if (!arg_settings_trusted
)
4612 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4614 arg_console_mode
= settings
->console_mode
;
4617 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4618 settings
->suppress_sync
>= 0)
4619 arg_suppress_sync
= settings
->suppress_sync
;
4621 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4622 * don't consult arg_settings_mask for them. */
4624 sd_bus_message_unref(arg_property_message
);
4625 arg_property_message
= TAKE_PTR(settings
->properties
);
4627 arg_console_width
= settings
->console_width
;
4628 arg_console_height
= settings
->console_height
;
4630 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4631 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4632 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4633 settings
->n_extra_nodes
= 0;
4638 static int load_settings(void) {
4639 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4640 _cleanup_fclose_
FILE *f
= NULL
;
4641 _cleanup_free_
char *p
= NULL
;
4647 /* If all settings are masked, there's no point in looking for
4648 * the settings file */
4649 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4652 /* We first look in the admin's directories in /etc and /run */
4653 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4654 _cleanup_free_
char *j
= NULL
;
4656 j
= path_join(i
, arg_settings_filename
);
4664 /* By default, we trust configuration from /etc and /run */
4665 if (arg_settings_trusted
< 0)
4666 arg_settings_trusted
= true;
4671 if (errno
!= ENOENT
)
4672 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4676 /* After that, let's look for a file next to the
4677 * actual image we shall boot. */
4680 r
= file_in_same_dir(arg_image
, arg_settings_filename
, &p
);
4682 return log_error_errno(r
, "Failed to generate settings path from image path: %m");
4683 } else if (arg_directory
) {
4684 r
= file_in_same_dir(arg_directory
, arg_settings_filename
, &p
);
4685 if (r
< 0 && r
!= -EADDRNOTAVAIL
) /* if directory is root fs, don't complain */
4686 return log_error_errno(r
, "Failed to generate settings path from directory path: %m");
4691 if (!f
&& errno
!= ENOENT
)
4692 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4694 /* By default, we do not trust configuration from /var/lib/machines */
4695 if (arg_settings_trusted
< 0)
4696 arg_settings_trusted
= false;
4703 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4705 r
= settings_load(f
, p
, &settings
);
4709 return merge_settings(settings
, p
);
4712 static int load_oci_bundle(void) {
4713 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4716 if (!arg_oci_bundle
)
4719 /* By default let's trust OCI bundles */
4720 if (arg_settings_trusted
< 0)
4721 arg_settings_trusted
= true;
4723 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4727 return merge_settings(settings
, arg_oci_bundle
);
4730 static int run_container(
4731 DissectedImage
*dissected_image
,
4733 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4734 struct ExposeArgs
*expose_args
,
4735 int *master
, pid_t
*pid
, int *ret
) {
4737 static const struct sigaction sa
= {
4738 .sa_handler
= nop_signal_handler
,
4739 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4742 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4743 _cleanup_close_
int etc_passwd_lock
= -EBADF
;
4744 _cleanup_close_pair_
int
4745 fd_inner_socket_pair
[2] = EBADF_PAIR
,
4746 fd_outer_socket_pair
[2] = EBADF_PAIR
;
4748 _cleanup_close_
int notify_socket
= -EBADF
, mntns_fd
= -EBADF
, fd_kmsg_fifo
= -EBADF
;
4749 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4750 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4751 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4752 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4753 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4754 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4755 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4756 size_t n_bind_user_uid
= 0;
4757 ContainerStatus container_status
= 0;
4761 _cleanup_close_
int child_netns_fd
= -EBADF
;
4763 assert_se(sigemptyset(&mask_chld
) == 0);
4764 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4766 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4767 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4768 * check with getpwuid() if the specific user already exists. Note that /etc might be
4769 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4770 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4771 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4774 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4775 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4776 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4779 r
= barrier_create(&barrier
);
4781 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4783 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_inner_socket_pair
) < 0)
4784 return log_error_errno(errno
, "Failed to create inner socket pair: %m");
4786 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_outer_socket_pair
) < 0)
4787 return log_error_errno(errno
, "Failed to create outer socket pair: %m");
4789 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4790 * parent's blocking calls and give it a chance to call wait() and terminate. */
4791 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4793 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4795 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4797 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4799 if (arg_network_namespace_path
) {
4800 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4801 if (child_netns_fd
< 0)
4802 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4804 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4806 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
);
4808 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4810 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4811 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4814 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4816 return log_error_errno(errno
, "clone() failed%s: %m",
4818 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4821 /* The outer child only has a file system namespace. */
4822 barrier_set_role(&barrier
, BARRIER_CHILD
);
4824 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
4825 fd_outer_socket_pair
[0] = safe_close(fd_outer_socket_pair
[0]);
4827 (void) reset_all_signal_handlers();
4828 (void) reset_signal_mask();
4830 r
= outer_child(&barrier
,
4833 fd_outer_socket_pair
[1],
4834 fd_inner_socket_pair
[1],
4838 _exit(EXIT_FAILURE
);
4840 _exit(EXIT_SUCCESS
);
4843 barrier_set_role(&barrier
, BARRIER_PARENT
);
4847 fd_inner_socket_pair
[1] = safe_close(fd_inner_socket_pair
[1]);
4848 fd_outer_socket_pair
[1] = safe_close(fd_outer_socket_pair
[1]);
4850 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4851 mntns_fd
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4853 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4855 /* The child just let us know the UID shift it might have read from the image. */
4856 l
= recv(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4858 return log_error_errno(errno
, "Failed to read UID shift: %m");
4859 if (l
!= sizeof arg_uid_shift
)
4860 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4862 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4863 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4864 * image, but if that's already in use, pick a new one, and report back to the child,
4865 * which one we now picked. */
4867 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4869 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4871 l
= send(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4873 return log_error_errno(errno
, "Failed to send UID shift: %m");
4874 if (l
!= sizeof arg_uid_shift
)
4875 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4878 n_bind_user_uid
= strv_length(arg_bind_user
);
4879 if (n_bind_user_uid
> 0) {
4880 /* Right after the UID shift, we'll receive the list of UID mappings for the
4881 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4883 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4887 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4888 l
= recv(fd_outer_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4890 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4891 if (l
!= sizeof(uid_t
)*4)
4892 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4893 SYNTHETIC_ERRNO(EIO
),
4894 "Short read while reading bind user UID pairs.");
4899 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4900 /* The child let us know the support cgroup mode it might have read from the image. */
4901 l
= recv(fd_outer_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4903 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4904 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4905 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
4906 l
, l
== 0 ? " The child is most likely dead." : "");
4909 /* Wait for the outer child. */
4910 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4913 if (r
!= EXIT_SUCCESS
)
4916 /* And now retrieve the PID of the inner child. */
4917 l
= recv(fd_outer_socket_pair
[0], pid
, sizeof *pid
, 0);
4919 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4920 if (l
!= sizeof *pid
)
4921 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4923 /* We also retrieve container UUID in case it was generated by outer child */
4924 l
= recv(fd_outer_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4926 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4927 if (l
!= sizeof(arg_uuid
))
4928 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4930 /* We also retrieve the socket used for notifications generated by outer child */
4931 notify_socket
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4932 if (notify_socket
< 0)
4933 return log_error_errno(notify_socket
,
4934 "Failed to receive notification socket from the outer child: %m");
4936 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4938 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4939 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4940 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4942 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4946 (void) barrier_place(&barrier
); /* #2 */
4949 if (arg_private_network
) {
4950 if (!arg_network_namespace_path
) {
4951 /* Wait until the child has unshared its network namespace. */
4952 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4953 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4956 if (child_netns_fd
< 0) {
4957 /* Make sure we have an open file descriptor to the child's network
4958 * namespace so it stays alive even if the child exits. */
4959 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4961 return log_error_errno(r
, "Failed to open child network namespace: %m");
4964 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4968 if (arg_network_veth
) {
4969 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4970 arg_network_bridge
|| arg_network_zone
, &arg_network_provided_mac
);
4976 if (arg_network_bridge
) {
4977 /* Add the interface to a bridge */
4978 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4983 } else if (arg_network_zone
) {
4984 /* Add the interface to a bridge, possibly creating it */
4985 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4993 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4997 /* We created the primary and extra veth links now; let's remember this, so that we know to
4998 remove them later on. Note that we don't bother with removing veth links that were created
4999 here when their setup failed half-way, because in that case the kernel should be able to
5000 remove them on its own, since they cannot be referenced by anything yet. */
5001 *veth_created
= true;
5003 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5007 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5012 if (arg_register
|| !arg_keep_unit
) {
5013 r
= sd_bus_default_system(&bus
);
5015 return log_error_errno(r
, "Failed to open system bus: %m");
5017 r
= sd_bus_set_close_on_exit(bus
, false);
5019 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5022 if (!arg_keep_unit
) {
5023 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5024 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5025 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5027 r
= sd_bus_match_signal_async(
5030 "org.freedesktop.systemd1",
5032 "org.freedesktop.systemd1.Scope",
5034 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5036 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5040 r
= register_machine(
5048 arg_custom_mounts
, arg_n_custom_mounts
,
5051 arg_property_message
,
5053 arg_container_service_name
,
5058 } else if (!arg_keep_unit
) {
5064 arg_custom_mounts
, arg_n_custom_mounts
,
5067 arg_property_message
,
5068 /* allow_pidfds= */ true,
5073 } else if (arg_slice
|| arg_property
)
5074 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5076 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5080 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5084 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5088 /* Notify the child that the parent is ready with all
5089 * its setup (including cgroup-ification), and that
5090 * the child can now hand over control to the code to
5091 * run inside the container. */
5092 (void) barrier_place(&barrier
); /* #4 */
5094 /* Block SIGCHLD here, before notifying child.
5095 * process_pty() will handle it with the other signals. */
5096 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5098 /* Reset signal to default */
5099 r
= default_signals(SIGCHLD
);
5101 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5103 r
= sd_event_new(&event
);
5105 return log_error_errno(r
, "Failed to get default event source: %m");
5107 (void) sd_event_set_watchdog(event
, true);
5110 r
= sd_bus_attach_event(bus
, event
, 0);
5112 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5115 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5119 /* Wait that the child is completely ready now, and has mounted their own copies of procfs and so on,
5120 * before we take the fully visible instances away. */
5121 if (!barrier_sync(&barrier
)) /* #5.1 */
5122 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5124 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5125 r
= wipe_fully_visible_fs(mntns_fd
);
5128 mntns_fd
= safe_close(mntns_fd
);
5131 /* And now let the child know that we completed removing the procfs instances, and it can start the
5133 if (!barrier_place(&barrier
)) /* #5.2 */
5134 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5136 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5137 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5138 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5140 (void) sd_notifyf(false,
5141 "STATUS=Container running.\n"
5142 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5143 if (!arg_notify_ready
) {
5144 r
= sd_notify(false, "READY=1\n");
5146 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5149 if (arg_kill_signal
> 0) {
5150 /* Try to kill the init system on SIGINT or SIGTERM */
5151 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5152 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5154 /* Immediately exit */
5155 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5156 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5159 (void) sd_event_add_signal(event
, NULL
, SIGRTMIN
+18, sigrtmin18_handler
, NULL
);
5161 r
= sd_event_add_memory_pressure(event
, NULL
, NULL
, NULL
);
5163 log_debug_errno(r
, "Failed allocate memory pressure event source, ignoring: %m");
5165 /* Exit when the child exits */
5166 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5168 /* Retrieve the kmsg fifo allocated by inner child */
5169 fd_kmsg_fifo
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5170 if (fd_kmsg_fifo
< 0)
5171 return log_error_errno(fd_kmsg_fifo
, "Failed to receive kmsg fifo from inner child: %m");
5173 if (arg_expose_ports
) {
5174 r
= expose_port_watch_rtnl(event
, fd_inner_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5178 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5179 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5182 if (arg_console_mode
!= CONSOLE_PIPE
) {
5183 _cleanup_close_
int fd
= -EBADF
;
5184 PTYForwardFlags flags
= 0;
5186 /* Retrieve the master pty allocated by inner child */
5187 fd
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5189 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5191 switch (arg_console_mode
) {
5193 case CONSOLE_READ_ONLY
:
5194 flags
|= PTY_FORWARD_READ_ONLY
;
5198 case CONSOLE_INTERACTIVE
:
5199 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5201 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5203 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5205 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5206 (void) pty_forward_set_width_height(forward
,
5208 arg_console_height
);
5212 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5215 *master
= TAKE_FD(fd
);
5218 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
5220 r
= sd_event_loop(event
);
5222 return log_error_errno(r
, "Failed to run event loop: %m");
5227 (void) pty_forward_get_last_char(forward
, &last_char
);
5228 forward
= pty_forward_free(forward
);
5230 if (!arg_quiet
&& last_char
!= '\n')
5234 /* Kill if it is not dead yet anyway */
5235 if (!arg_register
&& !arg_keep_unit
&& bus
)
5236 terminate_scope(bus
, arg_machine
);
5238 /* Normally redundant, but better safe than sorry */
5239 (void) kill(*pid
, SIGKILL
);
5241 fd_kmsg_fifo
= safe_close(fd_kmsg_fifo
);
5243 if (arg_private_network
) {
5244 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5245 * to avoid having to move the parent to the child network namespace. */
5246 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG_SIGTERM
|FORK_WAIT
|FORK_LOG
, NULL
);
5251 _cleanup_close_
int parent_netns_fd
= -EBADF
;
5253 r
= namespace_open(getpid_cached(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5255 log_error_errno(r
, "Failed to open parent network namespace: %m");
5256 _exit(EXIT_FAILURE
);
5259 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5261 log_error_errno(r
, "Failed to enter child network namespace: %m");
5262 _exit(EXIT_FAILURE
);
5265 /* Reverse network interfaces pair list so that interfaces get their initial name back.
5266 * This is about ensuring interfaces get their old name back when being moved back. */
5267 arg_network_interfaces
= strv_reverse(arg_network_interfaces
);
5269 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5271 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5273 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5277 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5279 /* Tell machined that we are gone. */
5281 (void) unregister_machine(bus
, arg_machine
);
5284 /* We failed to wait for the container, or the container exited abnormally. */
5286 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5287 /* r > 0 → The container exited with a non-zero status.
5288 * As a special case, we need to replace 133 with a different value,
5289 * because 133 is special-cased in the service file to reboot the container.
5290 * otherwise → The container exited with zero status and a reboot was not requested.
5292 if (r
== EXIT_FORCE_RESTART
)
5293 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5295 return 0; /* finito */
5298 /* CONTAINER_REBOOTED, loop again */
5300 if (arg_keep_unit
) {
5301 /* Special handling if we are running as a service: instead of simply
5302 * restarting the machine we want to restart the entire service, so let's
5303 * inform systemd about this with the special exit code 133. The service
5304 * file uses RestartForceExitStatus=133 so that this results in a full
5305 * nspawn restart. This is necessary since we might have cgroup parameters
5306 * set we want to have flushed out. */
5307 *ret
= EXIT_FORCE_RESTART
;
5308 return 0; /* finito */
5311 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5312 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5314 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5315 *veth_created
= false;
5316 return 1; /* loop again */
5319 static int initialize_rlimits(void) {
5320 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5321 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5322 * container execution environments. */
5324 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5325 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5326 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5327 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5328 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5329 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5330 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5331 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5332 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5333 [RLIMIT_NICE
] = { 0, 0 },
5334 [RLIMIT_NOFILE
] = { 1024, 4096 },
5335 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5336 [RLIMIT_RTPRIO
] = { 0, 0 },
5337 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5338 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5340 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5341 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5342 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5343 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5344 * that PID 1 changes a number of other resource limits during early initialization which is why we
5345 * don't read the other limits from PID 1 but prefer the static table above. */
5350 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5351 /* Let's only fill in what the user hasn't explicitly configured anyway */
5352 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5353 const struct rlimit
*v
;
5354 struct rlimit buffer
;
5356 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5357 /* For these two let's read the limits off PID 1. See above for an explanation. */
5359 r
= pid_getrlimit(1, rl
, &buffer
);
5361 return log_error_errno(r
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5364 } else if (rl
== RLIMIT_NOFILE
) {
5365 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5366 * userspace. Given that nspawn containers are often run without our PID 1,
5367 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5368 * so that container userspace gets similar resources as host userspace
5370 buffer
= kernel_defaults
[rl
];
5371 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5374 v
= kernel_defaults
+ rl
;
5376 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5377 if (!arg_rlimit
[rl
])
5381 if (DEBUG_LOGGING
) {
5382 _cleanup_free_
char *k
= NULL
;
5384 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5385 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5392 static int cant_be_in_netns(void) {
5393 _cleanup_close_
int fd
= -EBADF
;
5397 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5398 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5401 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5404 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5406 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5408 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5409 if (r
== -ENOENT
|| ERRNO_IS_NEG_DISCONNECT(r
))
5410 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5411 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5413 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5415 r
= getpeercred(fd
, &ucred
);
5417 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5419 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5421 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5423 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5424 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5428 static int run(int argc
, char *argv
[]) {
5429 bool remove_directory
= false, remove_image
= false, veth_created
= false, remove_tmprootdir
= false;
5430 _cleanup_close_
int master
= -EBADF
;
5431 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5432 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5433 char veth_name
[IFNAMSIZ
] = "";
5434 struct ExposeArgs expose_args
= {};
5435 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5436 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5437 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5438 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5439 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5442 log_parse_environment();
5445 r
= parse_argv(argc
, argv
);
5449 if (geteuid() != 0) {
5450 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5451 argc
>= 2 ? "Need to be root." :
5452 "Need to be root (and some arguments are usually required).\nHint: try --help");
5456 r
= cant_be_in_netns();
5460 r
= initialize_rlimits();
5464 r
= load_oci_bundle();
5472 r
= determine_names();
5476 r
= load_settings();
5480 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
5481 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
5483 if (!arg_private_network
&& arg_userns_mode
!= USER_NAMESPACE_NO
&& arg_uid_shift
> 0)
5484 arg_caps_retain
&= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE
);
5488 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5492 r
= verify_arguments();
5496 r
= verify_network_interfaces_initialized();
5500 /* Reapply environment settings. */
5501 (void) detect_unified_cgroup_hierarchy_from_environment();
5503 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5504 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5505 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5506 (void) ignore_signals(SIGPIPE
);
5508 n_fd_passed
= sd_listen_fds(false);
5509 if (n_fd_passed
> 0) {
5510 r
= fdset_new_listen_fds(&fds
, false);
5512 log_error_errno(r
, "Failed to collect file descriptors: %m");
5517 /* The "default" umask. This is appropriate for most file and directory
5518 * operations performed by nspawn, and is the umask that will be used for
5519 * the child. Functions like copy_devnodes() change the umask temporarily. */
5522 if (arg_directory
) {
5525 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5526 * /var from the host will propagate into container dynamically (because bad things happen if
5527 * two systems write to the same /var). Let's allow it for the special cases where /var is
5528 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5529 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5530 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5531 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5535 if (arg_ephemeral
) {
5536 _cleanup_free_
char *np
= NULL
;
5538 r
= chase_and_update(&arg_directory
, 0);
5542 /* If the specified path is a mount point we generate the new snapshot immediately
5543 * inside it under a random name. However if the specified is not a mount point we
5544 * create the new snapshot in the parent directory, just next to it. */
5545 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5547 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5551 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5553 r
= tempfn_random(arg_directory
, "machine.", &np
);
5555 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5559 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5560 * only owned by us and no one else. */
5561 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5563 log_error_errno(r
, "Failed to lock %s: %m", np
);
5568 BLOCK_SIGNALS(SIGINT
);
5569 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_directory
, AT_FDCWD
, np
,
5570 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5571 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5572 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5573 BTRFS_SNAPSHOT_RECURSIVE
|
5574 BTRFS_SNAPSHOT_QUOTA
|
5575 BTRFS_SNAPSHOT_SIGINT
);
5578 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5582 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5586 free_and_replace(arg_directory
, np
);
5587 remove_directory
= true;
5589 r
= chase_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5593 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5595 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5599 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5604 r
= chase_and_update(&arg_template
, 0);
5609 BLOCK_SIGNALS(SIGINT
);
5610 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_template
, AT_FDCWD
, arg_directory
,
5611 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5612 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5613 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5614 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5615 BTRFS_SNAPSHOT_RECURSIVE
|
5616 BTRFS_SNAPSHOT_QUOTA
|
5617 BTRFS_SNAPSHOT_SIGINT
);
5620 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5621 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5622 else if (r
== -EINTR
) {
5623 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5626 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5629 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5630 "Populated %s from template %s.", arg_directory
, arg_template
);
5634 if (arg_start_mode
== START_BOOT
) {
5635 _cleanup_free_
char *b
= NULL
;
5637 int check_os_release
, is_os_tree
;
5639 if (arg_pivot_root_new
) {
5640 b
= path_join(arg_directory
, arg_pivot_root_new
);
5650 check_os_release
= getenv_bool("SYSTEMD_NSPAWN_CHECK_OS_RELEASE");
5651 if (check_os_release
< 0 && check_os_release
!= -ENXIO
) {
5652 r
= log_error_errno(check_os_release
, "Failed to parse $SYSTEMD_NSPAWN_CHECK_OS_RELEASE: %m");
5656 is_os_tree
= path_is_os_tree(p
);
5657 if (is_os_tree
== 0 && check_os_release
== 0)
5658 log_debug("Directory %s is missing an os-release file, continuing anyway.", p
);
5659 else if (is_os_tree
<= 0) {
5660 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5661 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5665 _cleanup_free_
char *p
= NULL
;
5667 if (arg_pivot_root_new
)
5668 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5670 p
= path_join(arg_directory
, "/usr/");
5676 if (laccess(p
, F_OK
) < 0) {
5677 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5678 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5684 DissectImageFlags dissect_image_flags
=
5685 DISSECT_IMAGE_GENERIC_ROOT
|
5686 DISSECT_IMAGE_REQUIRE_ROOT
|
5687 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5688 DISSECT_IMAGE_USR_NO_ROOT
|
5689 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5690 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5692 assert(!arg_template
);
5694 r
= chase_and_update(&arg_image
, 0);
5698 if (arg_ephemeral
) {
5699 _cleanup_free_
char *np
= NULL
;
5701 r
= tempfn_random(arg_image
, "machine.", &np
);
5703 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5707 /* Always take an exclusive lock on our own ephemeral copy. */
5708 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5710 log_error_errno(r
, "Failed to create image lock: %m");
5715 BLOCK_SIGNALS(SIGINT
);
5716 r
= copy_file_full(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600,
5717 FS_NOCOW_FL
, FS_NOCOW_FL
,
5718 COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
,
5722 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5726 r
= log_error_errno(r
, "Failed to copy image file: %m");
5730 free_and_replace(arg_image
, np
);
5731 remove_image
= true;
5733 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5735 log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5739 log_error_errno(r
, "Failed to create image lock: %m");
5743 r
= verity_settings_load(
5744 &arg_verity_settings
,
5745 arg_image
, NULL
, NULL
);
5747 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5751 if (arg_verity_settings
.data_path
)
5752 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5755 if (!mkdtemp(tmprootdir
)) {
5756 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5760 remove_tmprootdir
= true;
5762 arg_directory
= strdup(tmprootdir
);
5763 if (!arg_directory
) {
5768 r
= loop_device_make_by_path(
5770 arg_read_only
? O_RDONLY
: O_RDWR
,
5771 /* sector_size= */ UINT32_MAX
,
5772 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5776 log_error_errno(r
, "Failed to set up loopback block device: %m");
5780 r
= dissect_loop_device_and_warn(
5782 &arg_verity_settings
,
5783 /* mount_options=*/ NULL
,
5784 arg_image_policy
?: &image_policy_container
,
5785 dissect_image_flags
,
5788 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5789 log_notice("Note that the disk image needs to\n"
5790 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5791 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5792 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5793 " d) or contain a file system without a partition table\n"
5794 "in order to be bootable with systemd-nspawn.");
5800 r
= dissected_image_load_verity_sig_partition(
5803 &arg_verity_settings
);
5807 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5808 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5809 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5811 r
= dissected_image_decrypt_interactively(
5814 &arg_verity_settings
,
5819 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5820 if (remove_image
&& unlink(arg_image
) >= 0)
5821 remove_image
= false;
5823 if (arg_architecture
< 0)
5824 arg_architecture
= dissected_image_architecture(dissected_image
);
5827 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5831 if (arg_console_mode
< 0)
5832 arg_console_mode
= isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
) ?
5833 CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5835 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5839 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5840 arg_machine
, arg_image
?: arg_directory
);
5842 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, SIGRTMIN
+18, -1) >= 0);
5844 r
= make_reaper_process(true);
5846 log_error_errno(r
, "Failed to become subreaper: %m");
5850 if (arg_expose_ports
) {
5851 r
= fw_ctx_new(&fw_ctx
);
5853 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5856 expose_args
.fw_ctx
= fw_ctx
;
5859 r
= run_container(dissected_image
,
5861 veth_name
, &veth_created
,
5862 &expose_args
, &master
,
5869 (void) sd_notify(false,
5870 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5871 "STOPPING=1\nSTATUS=Terminating...");
5874 (void) kill(pid
, SIGKILL
);
5876 /* Try to flush whatever is still queued in the pty */
5878 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5879 master
= safe_close(master
);
5883 (void) wait_for_terminate(pid
, NULL
);
5887 if (remove_directory
&& arg_directory
) {
5890 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5892 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5895 if (remove_image
&& arg_image
) {
5896 if (unlink(arg_image
) < 0)
5897 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5900 if (remove_tmprootdir
) {
5901 if (rmdir(tmprootdir
) < 0)
5902 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5908 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5909 (void) rm_rf(p
, REMOVE_ROOT
);
5912 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5913 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5916 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5917 (void) remove_bridge(arg_network_zone
);
5919 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5920 expose_port_free_all(arg_expose_ports
);
5921 rlimit_free_all(arg_rlimit
);
5922 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5930 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);