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"
28 #include "base-filesystem.h"
29 #include "blkid-util.h"
30 #include "btrfs-util.h"
32 #include "bus-error.h"
33 #include "bus-locator.h"
36 #include "capability-util.h"
37 #include "cgroup-util.h"
39 #include "common-signal.h"
41 #include "cpu-set-util.h"
42 #include "creds-util.h"
43 #include "dev-setup.h"
44 #include "discover-image.h"
45 #include "dissect-image.h"
51 #include "format-util.h"
54 #include "hexdecoct.h"
55 #include "hostname-setup.h"
56 #include "hostname-util.h"
57 #include "id128-util.h"
60 #include "loop-util.h"
61 #include "loopback-setup.h"
63 #include "main-func.h"
64 #include "missing_sched.h"
66 #include "mount-util.h"
67 #include "mountpoint-util.h"
68 #include "namespace-util.h"
69 #include "netlink-util.h"
70 #include "nspawn-bind-user.h"
71 #include "nspawn-cgroup.h"
72 #include "nspawn-creds.h"
73 #include "nspawn-def.h"
74 #include "nspawn-expose-ports.h"
75 #include "nspawn-mount.h"
76 #include "nspawn-network.h"
77 #include "nspawn-oci.h"
78 #include "nspawn-patch-uid.h"
79 #include "nspawn-register.h"
80 #include "nspawn-seccomp.h"
81 #include "nspawn-settings.h"
82 #include "nspawn-setuid.h"
83 #include "nspawn-stub-pid1.h"
84 #include "nspawn-util.h"
86 #include "nulstr-util.h"
89 #include "parse-argument.h"
90 #include "parse-util.h"
91 #include "pretty-print.h"
92 #include "process-util.h"
94 #include "random-util.h"
95 #include "raw-clone.h"
96 #include "resolve-util.h"
97 #include "rlimit-util.h"
100 #include "seccomp-util.h"
102 #include "selinux-util.h"
103 #include "signal-util.h"
104 #include "socket-util.h"
105 #include "stat-util.h"
106 #include "stdio-util.h"
107 #include "string-table.h"
108 #include "string-util.h"
110 #include "sysctl-util.h"
111 #include "terminal-util.h"
112 #include "tmpfile-util.h"
113 #include "umask-util.h"
114 #include "unit-name.h"
115 #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 static PagerFlags arg_pager_flags
= 0;
194 static unsigned long arg_personality
= PERSONALITY_INVALID
;
195 static char *arg_image
= NULL
;
196 static char *arg_oci_bundle
= NULL
;
197 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
198 static ExposePort
*arg_expose_ports
= NULL
;
199 static char **arg_property
= NULL
;
200 static sd_bus_message
*arg_property_message
= NULL
;
201 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
202 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
203 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
204 static int arg_kill_signal
= 0;
205 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
206 static SettingsMask arg_settings_mask
= 0;
207 static int arg_settings_trusted
= -1;
208 static char **arg_parameters
= NULL
;
209 static const char *arg_container_service_name
= "systemd-nspawn";
210 static bool arg_notify_ready
= false;
211 static bool arg_use_cgns
= true;
212 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
213 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
214 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
215 static char **arg_syscall_allow_list
= NULL
;
216 static char **arg_syscall_deny_list
= NULL
;
218 static scmp_filter_ctx arg_seccomp
= NULL
;
220 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
221 static bool arg_no_new_privileges
= false;
222 static int arg_oom_score_adjust
= 0;
223 static bool arg_oom_score_adjust_set
= false;
224 static CPUSet arg_cpu_set
= {};
225 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
226 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
227 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
228 static DeviceNode
* arg_extra_nodes
= NULL
;
229 static size_t arg_n_extra_nodes
= 0;
230 static char **arg_sysctl
= NULL
;
231 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
232 static Credential
*arg_credentials
= NULL
;
233 static size_t arg_n_credentials
= 0;
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_cpu_set
, cpu_set_reset
);
270 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
271 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
272 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
273 STATIC_DESTRUCTOR_REGISTER(arg_image_policy
, image_policy_freep
);
275 static int handle_arg_console(const char *arg
) {
276 if (streq(arg
, "help")) {
285 if (streq(arg
, "interactive"))
286 arg_console_mode
= CONSOLE_INTERACTIVE
;
287 else if (streq(arg
, "read-only"))
288 arg_console_mode
= CONSOLE_READ_ONLY
;
289 else if (streq(arg
, "passive"))
290 arg_console_mode
= CONSOLE_PASSIVE
;
291 else if (streq(arg
, "pipe")) {
292 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
293 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
294 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
295 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
296 "Proceeding anyway.");
298 arg_console_mode
= CONSOLE_PIPE
;
299 } else if (streq(arg
, "autopipe")) {
300 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
301 arg_console_mode
= CONSOLE_INTERACTIVE
;
303 arg_console_mode
= CONSOLE_PIPE
;
305 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
307 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
311 static int help(void) {
312 _cleanup_free_
char *link
= NULL
;
315 pager_open(arg_pager_flags
);
317 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
321 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
322 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
323 " -h --help Show this help\n"
324 " --version Print version string\n"
325 " -q --quiet Do not show status information\n"
326 " --no-pager Do not pipe output into a pager\n"
327 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
329 " -D --directory=PATH Root directory for the container\n"
330 " --template=PATH Initialize root directory from template directory,\n"
332 " -x --ephemeral Run container with snapshot of root directory, and\n"
333 " remove it after exit\n"
334 " -i --image=PATH Root file system disk image (or device node) for\n"
336 " --image-policy=POLICY Specify disk image dissection policy\n"
337 " --oci-bundle=PATH OCI bundle directory\n"
338 " --read-only Mount the root directory read-only\n"
339 " --volatile[=MODE] Run the system in volatile mode\n"
340 " --root-hash=HASH Specify verity root hash for root disk image\n"
341 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
342 " as a DER encoded PKCS7, either as a path to a file\n"
343 " or as an ASCII base64 encoded string prefixed by\n"
345 " --verity-data=PATH Specify hash device for verity\n"
346 " --pivot-root=PATH[:PATH]\n"
347 " Pivot root to given directory in the container\n\n"
348 "%3$sExecution:%4$s\n"
349 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
350 " -b --boot Boot up full system (i.e. invoke init)\n"
351 " --chdir=PATH Set working directory in the container\n"
352 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
353 " -u --user=USER Run the command under specified user or UID\n"
354 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
355 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
356 " --suppress-sync=BOOLEAN\n"
357 " Suppress any form of disk data synchronization\n\n"
358 "%3$sSystem Identity:%4$s\n"
359 " -M --machine=NAME Set the machine name for the container\n"
360 " --hostname=NAME Override the hostname for the container\n"
361 " --uuid=UUID Set a specific machine UUID for the container\n\n"
362 "%3$sProperties:%4$s\n"
363 " -S --slice=SLICE Place the container in the specified slice\n"
364 " --property=NAME=VALUE Set scope unit property\n"
365 " --register=BOOLEAN Register container as machine\n"
366 " --keep-unit Do not register a scope for the machine, reuse\n"
367 " the service unit nspawn is running in\n\n"
368 "%3$sUser Namespacing:%4$s\n"
369 " --private-users=no Run without user namespacing\n"
370 " --private-users=yes|pick|identity\n"
371 " Run within user namespace, autoselect UID/GID range\n"
372 " --private-users=UIDBASE[:NUIDS]\n"
373 " Similar, but with user configured UID/GID range\n"
374 " --private-users-ownership=MODE\n"
375 " Adjust ('chown') or map ('map') OS tree ownership\n"
376 " to private UID/GID range\n"
377 " -U Equivalent to --private-users=pick and\n"
378 " --private-users-ownership=auto\n\n"
379 "%3$sNetworking:%4$s\n"
380 " --private-network Disable network in container\n"
381 " --network-interface=HOSTIF[:CONTAINERIF]\n"
382 " Assign an existing network interface to the\n"
384 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
385 " Create a macvlan network interface based on an\n"
386 " existing network interface to the container\n"
387 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
388 " Create an ipvlan network interface based on an\n"
389 " existing network interface to the container\n"
390 " -n --network-veth Add a virtual Ethernet connection between host\n"
392 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
393 " Add an additional virtual Ethernet link between\n"
394 " host and container\n"
395 " --network-bridge=INTERFACE\n"
396 " Add a virtual Ethernet connection to the container\n"
397 " and attach it to an existing bridge on the host\n"
398 " --network-zone=NAME Similar, but attach the new interface to an\n"
399 " an automatically managed bridge interface\n"
400 " --network-namespace-path=PATH\n"
401 " Set network namespace to the one represented by\n"
402 " the specified kernel namespace file node\n"
403 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
404 " Expose a container IP port on the host\n\n"
405 "%3$sSecurity:%4$s\n"
406 " --capability=CAP In addition to the default, retain specified\n"
408 " --drop-capability=CAP Drop the specified capability from the default set\n"
409 " --ambient-capability=CAP\n"
410 " Sets the specified capability for the started\n"
411 " process. Not useful if booting a machine.\n"
412 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
413 " --system-call-filter=LIST|~LIST\n"
414 " Permit/prohibit specific system calls\n"
415 " -Z --selinux-context=SECLABEL\n"
416 " Set the SELinux security context to be used by\n"
417 " processes in the container\n"
418 " -L --selinux-apifs-context=SECLABEL\n"
419 " Set the SELinux security context to be used by\n"
420 " API/tmpfs file systems in the container\n\n"
421 "%3$sResources:%4$s\n"
422 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
423 " --oom-score-adjust=VALUE\n"
424 " Adjust the OOM score value for the payload\n"
425 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
426 " --personality=ARCH Pick personality for this container\n\n"
427 "%3$sIntegration:%4$s\n"
428 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
429 " --timezone=MODE Select mode of /etc/localtime initialization\n"
430 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
431 " host, try-guest, try-host\n"
432 " -j Equivalent to --link-journal=try-guest\n\n"
434 " --bind=PATH[:PATH[:OPTIONS]]\n"
435 " Bind mount a file or directory from the host into\n"
437 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
438 " Similar, but creates a read-only bind mount\n"
439 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
441 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
442 " --overlay=PATH[:PATH...]:PATH\n"
443 " Create an overlay mount from the host to \n"
445 " --overlay-ro=PATH[:PATH...]:PATH\n"
446 " Similar, but creates a read-only overlay mount\n"
447 " --bind-user=NAME Bind user from host to container\n\n"
448 "%3$sInput/Output:%4$s\n"
449 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
450 " set up for the container.\n"
451 " -P --pipe Equivalent to --console=pipe\n\n"
452 "%3$sCredentials:%4$s\n"
453 " --set-credential=ID:VALUE\n"
454 " Pass a credential with literal value to container.\n"
455 " --load-credential=ID:PATH\n"
456 " Load credential to pass to container from file or\n"
457 " AF_UNIX stream socket.\n"
458 "\nSee the %2$s for details.\n",
459 program_invocation_short_name
,
469 static int custom_mount_check_all(void) {
472 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
473 CustomMount
*m
= &arg_custom_mounts
[i
];
475 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
476 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
477 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
478 "--private-users-ownership=own may not be combined with custom root mounts.");
479 if (arg_uid_shift
== UID_INVALID
)
480 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
481 "--private-users with automatic UID shift may not be combined with custom root mounts.");
488 static int detect_unified_cgroup_hierarchy_from_environment(void) {
489 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
492 /* Allow the user to control whether the unified hierarchy is used */
496 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
497 var
= "UNIFIED_CGROUP_HIERARCHY";
502 r
= parse_boolean(e
);
504 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
506 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
508 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
514 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
517 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
518 * in the image actually supports. */
519 r
= cg_all_unified();
521 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
523 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
524 * routine only detects 231, so we'll have a false negative here for 230. */
525 r
= systemd_installation_has_version(directory
, "230");
527 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
529 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
531 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
532 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
533 /* Mixed cgroup hierarchy support was added in 233 */
534 r
= systemd_installation_has_version(directory
, "233");
536 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
538 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
540 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
542 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
544 log_debug("Using %s hierarchy for container.",
545 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
546 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
551 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
556 _cleanup_free_
char *t
= NULL
;
558 r
= extract_first_word(&spec
, &t
, ",", 0);
560 return log_error_errno(r
, "Failed to parse capability %s.", t
);
564 if (streq(t
, "help")) {
565 for (int i
= 0; i
< capability_list_length(); i
++) {
568 name
= capability_to_name(i
);
579 r
= capability_from_name(t
);
581 return log_error_errno(r
, "Failed to parse capability %s.", t
);
588 return 1; /* continue */
591 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
594 r
= getenv_bool(name
);
598 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
600 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
601 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
605 static int parse_mount_settings_env(void) {
609 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
610 if (r
< 0 && r
!= -ENXIO
)
611 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
613 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
615 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
616 if (streq_ptr(e
, "network"))
617 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
620 r
= parse_boolean(e
);
622 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
624 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
625 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
631 static int parse_environment(void) {
635 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
638 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
641 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
644 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
648 r
= parse_mount_settings_env();
652 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
653 * even if it is supported. If not supported, it has no effect. */
654 if (!cg_ns_supported())
655 arg_use_cgns
= false;
657 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
660 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
664 arg_use_cgns
= r
> 0;
665 arg_settings_mask
|= SETTING_USE_CGNS
;
669 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
671 arg_container_service_name
= e
;
673 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
675 arg_suppress_sync
= r
;
676 else if (r
!= -ENXIO
)
677 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
679 return detect_unified_cgroup_hierarchy_from_environment();
682 static int parse_argv(int argc
, char *argv
[]) {
689 ARG_AMBIENT_CAPABILITY
,
701 ARG_NETWORK_INTERFACE
,
706 ARG_NETWORK_VETH_EXTRA
,
707 ARG_NETWORK_NAMESPACE_PATH
,
717 ARG_PRIVATE_USERS_CHOWN
,
718 ARG_PRIVATE_USERS_OWNERSHIP
,
723 ARG_SYSTEM_CALL_FILTER
,
726 ARG_NO_NEW_PRIVILEGES
,
727 ARG_OOM_SCORE_ADJUST
,
742 static const struct option options
[] = {
743 { "help", no_argument
, NULL
, 'h' },
744 { "version", no_argument
, NULL
, ARG_VERSION
},
745 { "directory", required_argument
, NULL
, 'D' },
746 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
747 { "ephemeral", no_argument
, NULL
, 'x' },
748 { "user", required_argument
, NULL
, 'u' },
749 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
750 { "as-pid2", no_argument
, NULL
, 'a' },
751 { "boot", no_argument
, NULL
, 'b' },
752 { "uuid", required_argument
, NULL
, ARG_UUID
},
753 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
754 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
755 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
756 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
757 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
758 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
759 { "bind", required_argument
, NULL
, ARG_BIND
},
760 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
761 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
762 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
763 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
764 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
765 { "machine", required_argument
, NULL
, 'M' },
766 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
767 { "slice", required_argument
, NULL
, 'S' },
768 { "setenv", required_argument
, NULL
, 'E' },
769 { "selinux-context", required_argument
, NULL
, 'Z' },
770 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
771 { "quiet", no_argument
, NULL
, 'q' },
772 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
773 { "register", required_argument
, NULL
, ARG_REGISTER
},
774 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
775 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
776 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
777 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
778 { "network-veth", no_argument
, NULL
, 'n' },
779 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
780 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
781 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
782 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
783 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
784 { "image", required_argument
, NULL
, 'i' },
785 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
786 { "port", required_argument
, NULL
, 'p' },
787 { "property", required_argument
, NULL
, ARG_PROPERTY
},
788 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
789 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
790 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
791 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
792 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
793 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
794 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
795 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
796 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
797 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
798 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
799 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
800 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
801 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
802 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
803 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
804 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
805 { "console", required_argument
, NULL
, ARG_CONSOLE
},
806 { "pipe", no_argument
, NULL
, ARG_PIPE
},
807 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
808 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
809 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
810 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
811 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
812 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
813 { "image-policy", required_argument
, NULL
, ARG_IMAGE_POLICY
},
818 uint64_t plus
= 0, minus
= 0;
819 bool mask_all_settings
= false, mask_no_settings
= false;
824 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
825 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
827 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
837 r
= parse_path_argument(optarg
, false, &arg_directory
);
841 arg_settings_mask
|= SETTING_DIRECTORY
;
845 r
= parse_path_argument(optarg
, false, &arg_template
);
849 arg_settings_mask
|= SETTING_DIRECTORY
;
853 r
= parse_path_argument(optarg
, false, &arg_image
);
857 arg_settings_mask
|= SETTING_DIRECTORY
;
861 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
868 arg_ephemeral
= true;
869 arg_settings_mask
|= SETTING_EPHEMERAL
;
873 r
= free_and_strdup(&arg_user
, optarg
);
877 arg_settings_mask
|= SETTING_USER
;
880 case ARG_NETWORK_ZONE
: {
881 _cleanup_free_
char *j
= NULL
;
883 j
= strjoin("vz-", optarg
);
887 if (!ifname_valid(j
))
888 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
889 "Network zone name not valid: %s", j
);
891 free_and_replace(arg_network_zone
, j
);
893 arg_network_veth
= true;
894 arg_private_network
= true;
895 arg_settings_mask
|= SETTING_NETWORK
;
899 case ARG_NETWORK_BRIDGE
:
901 if (!ifname_valid(optarg
))
902 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
903 "Bridge interface name not valid: %s", optarg
);
905 r
= free_and_strdup(&arg_network_bridge
, optarg
);
911 arg_network_veth
= true;
912 arg_private_network
= true;
913 arg_settings_mask
|= SETTING_NETWORK
;
916 case ARG_NETWORK_VETH_EXTRA
:
917 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
919 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
921 arg_private_network
= true;
922 arg_settings_mask
|= SETTING_NETWORK
;
925 case ARG_NETWORK_INTERFACE
:
926 r
= interface_pair_parse(&arg_network_interfaces
, optarg
);
930 arg_private_network
= true;
931 arg_settings_mask
|= SETTING_NETWORK
;
934 case ARG_NETWORK_MACVLAN
:
935 r
= macvlan_pair_parse(&arg_network_macvlan
, optarg
);
939 arg_private_network
= true;
940 arg_settings_mask
|= SETTING_NETWORK
;
943 case ARG_NETWORK_IPVLAN
:
944 r
= ipvlan_pair_parse(&arg_network_ipvlan
, optarg
);
949 case ARG_PRIVATE_NETWORK
:
950 arg_private_network
= true;
951 arg_settings_mask
|= SETTING_NETWORK
;
954 case ARG_NETWORK_NAMESPACE_PATH
:
955 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
959 arg_settings_mask
|= SETTING_NETWORK
;
963 if (arg_start_mode
== START_PID2
)
964 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
965 "--boot and --as-pid2 may not be combined.");
967 arg_start_mode
= START_BOOT
;
968 arg_settings_mask
|= SETTING_START_MODE
;
972 if (arg_start_mode
== START_BOOT
)
973 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
974 "--boot and --as-pid2 may not be combined.");
976 arg_start_mode
= START_PID2
;
977 arg_settings_mask
|= SETTING_START_MODE
;
981 r
= sd_id128_from_string(optarg
, &arg_uuid
);
983 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
985 if (sd_id128_is_null(arg_uuid
))
986 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
987 "Machine UUID may not be all zeroes.");
989 arg_settings_mask
|= SETTING_MACHINE_ID
;
993 _cleanup_free_
char *mangled
= NULL
;
995 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
999 free_and_replace(arg_slice
, mangled
);
1000 arg_settings_mask
|= SETTING_SLICE
;
1005 if (isempty(optarg
))
1006 arg_machine
= mfree(arg_machine
);
1008 if (!hostname_is_valid(optarg
, 0))
1009 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1010 "Invalid machine name: %s", optarg
);
1012 r
= free_and_strdup(&arg_machine
, optarg
);
1019 if (isempty(optarg
))
1020 arg_hostname
= mfree(arg_hostname
);
1022 if (!hostname_is_valid(optarg
, 0))
1023 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1024 "Invalid hostname: %s", optarg
);
1026 r
= free_and_strdup(&arg_hostname
, optarg
);
1031 arg_settings_mask
|= SETTING_HOSTNAME
;
1035 arg_selinux_context
= optarg
;
1039 arg_selinux_apifs_context
= optarg
;
1043 arg_read_only
= true;
1044 arg_settings_mask
|= SETTING_READ_ONLY
;
1047 case ARG_AMBIENT_CAPABILITY
: {
1049 r
= parse_capability_spec(optarg
, &m
);
1052 arg_caps_ambient
|= m
;
1053 arg_settings_mask
|= SETTING_CAPABILITY
;
1056 case ARG_CAPABILITY
:
1057 case ARG_DROP_CAPABILITY
: {
1059 r
= parse_capability_spec(optarg
, &m
);
1063 if (c
== ARG_CAPABILITY
)
1067 arg_settings_mask
|= SETTING_CAPABILITY
;
1070 case ARG_NO_NEW_PRIVILEGES
:
1071 r
= parse_boolean(optarg
);
1073 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1075 arg_no_new_privileges
= r
;
1076 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1080 arg_link_journal
= LINK_GUEST
;
1081 arg_link_journal_try
= true;
1082 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1085 case ARG_LINK_JOURNAL
:
1086 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1088 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1090 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1095 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1097 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1099 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1103 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1105 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1107 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1111 case ARG_OVERLAY_RO
:
1112 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1113 if (r
== -EADDRNOTAVAIL
)
1114 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1116 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1118 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1121 case ARG_INACCESSIBLE
:
1122 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1124 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1126 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1130 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1132 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1134 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1141 case ARG_SHARE_SYSTEM
:
1142 /* We don't officially support this anymore, except for compat reasons. People should use the
1143 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1144 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1145 arg_clone_ns_flags
= 0;
1149 r
= parse_boolean(optarg
);
1151 log_error("Failed to parse --register= argument: %s", optarg
);
1159 arg_keep_unit
= true;
1162 case ARG_PERSONALITY
:
1164 arg_personality
= personality_from_string(optarg
);
1165 if (arg_personality
== PERSONALITY_INVALID
)
1166 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1167 "Unknown or unsupported personality '%s'.", optarg
);
1169 arg_settings_mask
|= SETTING_PERSONALITY
;
1175 arg_volatile_mode
= VOLATILE_YES
;
1176 else if (streq(optarg
, "help")) {
1177 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1182 m
= volatile_mode_from_string(optarg
);
1184 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1185 "Failed to parse --volatile= argument: %s", optarg
);
1187 arg_volatile_mode
= m
;
1190 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1194 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1196 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1198 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1200 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1204 if (strv_extend(&arg_property
, optarg
) < 0)
1209 case ARG_PRIVATE_USERS
: {
1214 else if (!in_charset(optarg
, DIGITS
))
1215 /* do *not* parse numbers as booleans */
1216 boolean
= parse_boolean(optarg
);
1221 /* no: User namespacing off */
1222 arg_userns_mode
= USER_NAMESPACE_NO
;
1223 arg_uid_shift
= UID_INVALID
;
1224 arg_uid_range
= UINT32_C(0x10000);
1225 } else if (boolean
> 0) {
1226 /* yes: User namespacing on, UID range is read from root dir */
1227 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1228 arg_uid_shift
= UID_INVALID
;
1229 arg_uid_range
= UINT32_C(0x10000);
1230 } else if (streq(optarg
, "pick")) {
1231 /* pick: User namespacing on, UID range is picked randomly */
1232 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1233 * implied by USER_NAMESPACE_PICK
1235 arg_uid_shift
= UID_INVALID
;
1236 arg_uid_range
= UINT32_C(0x10000);
1238 } else if (streq(optarg
, "identity")) {
1239 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1240 * itself, i.e. we don't actually map anything, but do take benefit of
1241 * isolation of capability sets. */
1242 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1244 arg_uid_range
= UINT32_C(0x10000);
1246 _cleanup_free_
char *buffer
= NULL
;
1247 const char *range
, *shift
;
1249 /* anything else: User namespacing on, UID range is explicitly configured */
1251 range
= strchr(optarg
, ':');
1253 buffer
= strndup(optarg
, range
- optarg
);
1259 r
= safe_atou32(range
, &arg_uid_range
);
1261 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1265 r
= parse_uid(shift
, &arg_uid_shift
);
1267 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1269 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1271 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1272 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1275 arg_settings_mask
|= SETTING_USERNS
;
1280 if (userns_supported()) {
1281 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1282 * implied by USER_NAMESPACE_PICK
1284 arg_uid_shift
= UID_INVALID
;
1285 arg_uid_range
= UINT32_C(0x10000);
1287 arg_settings_mask
|= SETTING_USERNS
;
1292 case ARG_PRIVATE_USERS_CHOWN
:
1293 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1295 arg_settings_mask
|= SETTING_USERNS
;
1298 case ARG_PRIVATE_USERS_OWNERSHIP
:
1299 if (streq(optarg
, "help")) {
1300 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1304 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1305 if (arg_userns_ownership
< 0)
1306 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1308 arg_settings_mask
|= SETTING_USERNS
;
1311 case ARG_KILL_SIGNAL
:
1312 if (streq(optarg
, "help")) {
1313 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1317 arg_kill_signal
= signal_from_string(optarg
);
1318 if (arg_kill_signal
< 0)
1319 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1321 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1326 /* no → do not read files
1327 * yes → read files, do not override cmdline, trust only subset
1328 * override → read files, override cmdline, trust only subset
1329 * trusted → read files, do not override cmdline, trust all
1332 r
= parse_boolean(optarg
);
1334 if (streq(optarg
, "trusted")) {
1335 mask_all_settings
= false;
1336 mask_no_settings
= false;
1337 arg_settings_trusted
= true;
1339 } else if (streq(optarg
, "override")) {
1340 mask_all_settings
= false;
1341 mask_no_settings
= true;
1342 arg_settings_trusted
= -1;
1344 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1347 mask_all_settings
= false;
1348 mask_no_settings
= false;
1349 arg_settings_trusted
= -1;
1352 mask_all_settings
= true;
1353 mask_no_settings
= false;
1354 arg_settings_trusted
= false;
1360 if (!path_is_absolute(optarg
))
1361 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1362 "Working directory %s is not an absolute path.", optarg
);
1364 r
= free_and_strdup(&arg_chdir
, optarg
);
1368 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1371 case ARG_PIVOT_ROOT
:
1372 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1374 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1376 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1379 case ARG_NOTIFY_READY
:
1380 r
= parse_boolean(optarg
);
1382 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1383 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1384 arg_notify_ready
= r
;
1385 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1388 case ARG_ROOT_HASH
: {
1389 _cleanup_free_
void *k
= NULL
;
1392 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1394 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1395 if (l
< sizeof(sd_id128_t
))
1396 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128-bit long: %s", optarg
);
1398 free_and_replace(arg_verity_settings
.root_hash
, k
);
1399 arg_verity_settings
.root_hash_size
= l
;
1403 case ARG_ROOT_HASH_SIG
: {
1408 if ((value
= startswith(optarg
, "base64:"))) {
1409 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1411 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1414 r
= read_full_file(optarg
, (char**) &p
, &l
);
1416 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1419 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1420 arg_verity_settings
.root_hash_sig_size
= l
;
1424 case ARG_VERITY_DATA
:
1425 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1430 case ARG_SYSTEM_CALL_FILTER
: {
1434 negative
= optarg
[0] == '~';
1435 items
= negative
? optarg
+ 1 : optarg
;
1438 _cleanup_free_
char *word
= NULL
;
1440 r
= extract_first_word(&items
, &word
, NULL
, 0);
1446 return log_error_errno(r
, "Failed to parse system call filter: %m");
1449 r
= strv_extend(&arg_syscall_deny_list
, word
);
1451 r
= strv_extend(&arg_syscall_allow_list
, word
);
1456 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1462 _cleanup_free_
char *name
= NULL
;
1465 if (streq(optarg
, "help")) {
1466 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1470 eq
= strchr(optarg
, '=');
1472 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1473 "--rlimit= expects an '=' assignment.");
1475 name
= strndup(optarg
, eq
- optarg
);
1479 rl
= rlimit_from_string_harder(name
);
1481 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1483 if (!arg_rlimit
[rl
]) {
1484 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1485 if (!arg_rlimit
[rl
])
1489 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1491 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1493 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1497 case ARG_OOM_SCORE_ADJUST
:
1498 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1500 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1502 arg_oom_score_adjust_set
= true;
1503 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1506 case ARG_CPU_AFFINITY
: {
1509 r
= parse_cpu_set(optarg
, &cpuset
);
1511 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1513 cpu_set_reset(&arg_cpu_set
);
1514 arg_cpu_set
= cpuset
;
1515 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1519 case ARG_RESOLV_CONF
:
1520 if (streq(optarg
, "help")) {
1521 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1525 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1526 if (arg_resolv_conf
< 0)
1527 return log_error_errno(arg_resolv_conf
,
1528 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1530 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1534 if (streq(optarg
, "help")) {
1535 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1539 arg_timezone
= timezone_mode_from_string(optarg
);
1540 if (arg_timezone
< 0)
1541 return log_error_errno(arg_timezone
,
1542 "Failed to parse /etc/localtime mode: %s", optarg
);
1544 arg_settings_mask
|= SETTING_TIMEZONE
;
1548 r
= handle_arg_console(optarg
);
1555 r
= handle_arg_console("pipe");
1561 arg_pager_flags
|= PAGER_DISABLE
;
1564 case ARG_SET_CREDENTIAL
: {
1565 _cleanup_free_
char *word
= NULL
, *data
= NULL
;
1566 const char *p
= optarg
;
1570 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1574 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1576 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1578 if (!credential_name_valid(word
))
1579 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1581 for (size_t i
= 0; i
< arg_n_credentials
; i
++)
1582 if (streq(arg_credentials
[i
].id
, word
))
1583 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1585 l
= cunescape(p
, UNESCAPE_ACCEPT_NUL
, &data
);
1587 return log_error_errno(l
, "Failed to unescape credential data: %s", p
);
1589 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1593 a
[arg_n_credentials
++] = (Credential
) {
1594 .id
= TAKE_PTR(word
),
1595 .data
= TAKE_PTR(data
),
1599 arg_credentials
= a
;
1601 arg_settings_mask
|= SETTING_CREDENTIALS
;
1605 case ARG_LOAD_CREDENTIAL
: {
1606 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
1607 _cleanup_(erase_and_freep
) char *data
= NULL
;
1608 _cleanup_free_
char *word
= NULL
, *j
= NULL
;
1609 const char *p
= optarg
;
1613 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1617 return log_error_errno(r
, "Failed to parse --load-credential= parameter: %m");
1619 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --load-credential=: %s", optarg
);
1621 if (!credential_name_valid(word
))
1622 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1624 for (i
= 0; i
< arg_n_credentials
; i
++)
1625 if (streq(arg_credentials
[i
].id
, word
))
1626 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1628 if (path_is_absolute(p
))
1629 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
1633 r
= get_credentials_dir(&e
);
1635 return log_error_errno(r
, "Credential not available (no credentials passed at all): %s", word
);
1637 j
= path_join(e
, p
);
1642 r
= read_full_file_full(AT_FDCWD
, j
?: p
, UINT64_MAX
, SIZE_MAX
,
1647 return log_error_errno(r
, "Failed to read credential '%s': %m", j
?: p
);
1649 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1653 a
[arg_n_credentials
++] = (Credential
) {
1654 .id
= TAKE_PTR(word
),
1655 .data
= TAKE_PTR(data
),
1659 arg_credentials
= a
;
1661 arg_settings_mask
|= SETTING_CREDENTIALS
;
1666 if (!valid_user_group_name(optarg
, 0))
1667 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1669 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1672 arg_settings_mask
|= SETTING_BIND_USER
;
1675 case ARG_SUPPRESS_SYNC
:
1676 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1680 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1683 case ARG_IMAGE_POLICY
:
1684 r
= parse_image_policy_argument(optarg
, &arg_image_policy
);
1693 assert_not_reached();
1696 if (argc
> optind
) {
1697 strv_free(arg_parameters
);
1698 arg_parameters
= strv_copy(argv
+ optind
);
1699 if (!arg_parameters
)
1702 arg_settings_mask
|= SETTING_START_MODE
;
1705 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1706 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1707 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1708 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1710 arg_directory
= TAKE_PTR(arg_template
);
1712 arg_caps_retain
|= plus
;
1713 arg_caps_retain
|= arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0;
1715 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
1716 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
1718 if (!arg_private_network
&& arg_userns_mode
!= USER_NAMESPACE_NO
&& arg_uid_shift
> 0)
1719 arg_caps_retain
&= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE
);
1721 arg_caps_retain
&= ~minus
;
1723 /* Make sure to parse environment before we reset the settings mask below */
1724 r
= parse_environment();
1728 /* Load all settings from .nspawn files */
1729 if (mask_no_settings
)
1730 arg_settings_mask
= 0;
1732 /* Don't load any settings from .nspawn files */
1733 if (mask_all_settings
)
1734 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1739 static int verify_arguments(void) {
1742 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1743 /* If we are running the stub init in the container, we don't need to look at what the init
1744 * in the container supports, because we are not using it. Let's immediately pick the right
1745 * setting based on the host system configuration.
1747 * We only do this, if the user didn't use an environment variable to override the detection.
1750 r
= cg_all_unified();
1752 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1754 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1755 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1756 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1758 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1761 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1762 arg_mount_settings
|= MOUNT_USE_USERNS
;
1764 if (arg_private_network
)
1765 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1767 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1768 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1769 arg_register
= false;
1770 if (arg_start_mode
!= START_PID1
)
1771 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1774 if (arg_userns_ownership
< 0)
1775 arg_userns_ownership
=
1776 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1777 USER_NAMESPACE_OWNERSHIP_OFF
;
1779 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1780 arg_kill_signal
= SIGRTMIN
+3;
1782 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 */
1783 arg_read_only
= true;
1785 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1786 arg_read_only
= true;
1788 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1789 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1790 * The latter is not technically a user session, but we don't need to labour the point. */
1791 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1793 if (arg_directory
&& arg_image
)
1794 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1796 if (arg_template
&& arg_image
)
1797 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1799 if (arg_template
&& !(arg_directory
|| arg_machine
))
1800 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1802 if (arg_ephemeral
&& arg_template
)
1803 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1805 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1806 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1808 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1809 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1811 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1812 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1813 "--read-only and --private-users-ownership=chown may not be combined.");
1815 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1816 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1817 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1818 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1819 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1821 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1822 * we need to error out, to avoid conflicts between different network options. */
1823 if (arg_network_namespace_path
&&
1824 (arg_network_interfaces
|| arg_network_macvlan
||
1825 arg_network_ipvlan
|| arg_network_veth_extra
||
1826 arg_network_bridge
|| arg_network_zone
||
1828 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1830 if (arg_network_bridge
&& arg_network_zone
)
1831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1832 "--network-bridge= and --network-zone= may not be combined.");
1834 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1835 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1837 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1838 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1840 if (arg_expose_ports
&& !arg_private_network
)
1841 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1843 if (arg_caps_ambient
) {
1844 if (arg_caps_ambient
== UINT64_MAX
)
1845 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1847 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1848 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1850 if (arg_start_mode
== START_BOOT
)
1851 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1854 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1855 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1857 /* Drop duplicate --bind-user= entries */
1858 strv_uniq(arg_bind_user
);
1860 r
= custom_mount_check_all();
1867 static int verify_network_interfaces_initialized(void) {
1869 r
= test_network_interfaces_initialized(arg_network_interfaces
);
1873 r
= test_network_interfaces_initialized(arg_network_macvlan
);
1877 r
= test_network_interfaces_initialized(arg_network_ipvlan
);
1884 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1887 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1890 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1893 if (uid
!= UID_INVALID
) {
1894 uid
+= arg_uid_shift
;
1896 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1900 if (gid
!= GID_INVALID
) {
1901 gid
+= (gid_t
) arg_uid_shift
;
1903 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1907 return RET_NERRNO(lchown(p
, uid
, gid
));
1910 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1914 q
= prefix_roota(root
, path
);
1915 r
= RET_NERRNO(mkdir(q
, mode
));
1921 return userns_lchown(q
, uid
, gid
);
1924 static const char *timezone_from_path(const char *path
) {
1925 return PATH_STARTSWITH_SET(
1927 "../usr/share/zoneinfo/",
1928 "/usr/share/zoneinfo/");
1931 static bool etc_writable(void) {
1932 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1935 static int setup_timezone(const char *dest
) {
1936 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1937 const char *where
, *check
;
1943 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1944 r
= readlink_malloc("/etc/localtime", &p
);
1945 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1946 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1947 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1948 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1950 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1951 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1955 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1958 } else if (arg_timezone
== TIMEZONE_AUTO
)
1959 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1965 if (m
== TIMEZONE_OFF
)
1968 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1970 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1974 where
= strjoina(etc
, "/localtime");
1978 case TIMEZONE_DELETE
:
1979 if (unlink(where
) < 0)
1980 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1984 case TIMEZONE_SYMLINK
: {
1985 _cleanup_free_
char *q
= NULL
;
1986 const char *z
, *what
;
1988 z
= timezone_from_path(p
);
1990 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1994 r
= readlink_malloc(where
, &q
);
1995 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1996 return 0; /* Already pointing to the right place? Then do nothing .. */
1998 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1999 r
= chase(check
, dest
, 0, NULL
, NULL
);
2001 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
2003 if (unlink(where
) < 0 && errno
!= ENOENT
) {
2004 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
2005 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
2009 what
= strjoina("../usr/share/zoneinfo/", z
);
2010 if (symlink(what
, where
) < 0) {
2011 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
2012 errno
, "Failed to correct timezone of container, ignoring: %m");
2022 case TIMEZONE_BIND
: {
2023 _cleanup_free_
char *resolved
= NULL
;
2026 found
= chase(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2028 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
2032 if (found
== 0) /* missing? */
2033 (void) touch(resolved
);
2035 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
2037 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2043 /* If mounting failed, try to copy */
2044 r
= copy_file_atomic("/etc/localtime", where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2046 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2047 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
2054 assert_not_reached();
2057 /* Fix permissions of the symlink or file copy we just created */
2058 r
= userns_lchown(where
, 0, 0);
2060 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
2065 static int have_resolv_conf(const char *path
) {
2068 if (access(path
, F_OK
) < 0) {
2069 if (errno
== ENOENT
)
2072 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
2078 static int resolved_listening(void) {
2079 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2080 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
2081 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
2084 /* Check if resolved is listening */
2086 r
= sd_bus_open_system(&bus
);
2088 return log_debug_errno(r
, "Failed to open system bus: %m");
2090 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2092 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2096 r
= bus_get_property_string(bus
, bus_resolve_mgr
, "DNSStubListener", &error
, &dns_stub_listener_mode
);
2098 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2100 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2103 static int setup_resolv_conf(const char *dest
) {
2104 _cleanup_free_
char *etc
= NULL
;
2105 const char *where
, *what
;
2111 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2112 if (arg_private_network
)
2113 m
= RESOLV_CONF_OFF
;
2114 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2115 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2116 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2117 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2119 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2122 m
= arg_resolv_conf
;
2124 if (m
== RESOLV_CONF_OFF
)
2127 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2129 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2133 where
= strjoina(etc
, "/resolv.conf");
2135 if (m
== RESOLV_CONF_DELETE
) {
2136 if (unlink(where
) < 0)
2137 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2142 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2143 what
= PRIVATE_STATIC_RESOLV_CONF
;
2144 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2145 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2146 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2147 what
= PRIVATE_STUB_RESOLV_CONF
;
2149 what
= "/etc/resolv.conf";
2151 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2152 _cleanup_free_
char *resolved
= NULL
;
2155 found
= chase(where
, dest
, CHASE_NONEXISTENT
|CHASE_NOFOLLOW
, &resolved
, NULL
);
2157 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2161 if (found
== 0) /* missing? */
2162 (void) touch(resolved
);
2164 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2166 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2168 /* If that didn't work, let's copy the file */
2171 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2172 r
= copy_file_atomic(what
, where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2174 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, COPY_REFLINK
);
2176 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2177 * resolved or something similar runs inside and the symlink points there.
2179 * If the disk image is read-only, there's also no point in complaining.
2181 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2182 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2183 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2187 r
= userns_lchown(where
, 0, 0);
2189 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2194 static int setup_boot_id(void) {
2195 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2196 _cleanup_free_
char *path
= NULL
;
2197 sd_id128_t rnd
= SD_ID128_NULL
;
2201 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2203 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2205 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2207 r
= sd_id128_randomize(&rnd
);
2209 return log_error_errno(r
, "Failed to generate random boot id: %m");
2211 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2213 return log_error_errno(r
, "Failed to write boot id: %m");
2215 from
= TAKE_PTR(path
);
2216 to
= "/proc/sys/kernel/random/boot_id";
2218 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2222 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2225 static int copy_devnodes(const char *dest
) {
2226 static const char devnodes
[] =
2239 BLOCK_WITH_UMASK(0000);
2241 /* Create /dev/net, so that we can create /dev/net/tun in it */
2242 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2243 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2245 NULSTR_FOREACH(d
, devnodes
) {
2246 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2249 from
= path_join("/dev/", d
);
2253 to
= path_join(dest
, from
);
2257 if (stat(from
, &st
) < 0) {
2259 if (errno
!= ENOENT
)
2260 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2262 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2263 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2264 "%s is not a char or block device, cannot copy.", from
);
2266 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2268 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2269 /* Explicitly warn the user when /dev is already populated. */
2270 if (errno
== EEXIST
)
2271 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
);
2273 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2275 /* Some systems abusively restrict mknod but allow bind mounts. */
2278 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2279 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2281 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2284 r
= userns_lchown(to
, 0, 0);
2286 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2288 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2292 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2294 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2296 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2299 prefixed
= path_join(dest
, sl
);
2303 t
= path_join("..", d
);
2307 if (symlink(t
, prefixed
) < 0)
2308 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2315 static int make_extra_nodes(const char *dest
) {
2319 BLOCK_WITH_UMASK(0000);
2321 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2322 _cleanup_free_
char *path
= NULL
;
2323 DeviceNode
*n
= arg_extra_nodes
+ i
;
2325 path
= path_join(dest
, n
->path
);
2329 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2330 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2332 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2334 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2340 static int setup_pts(const char *dest
) {
2341 _cleanup_free_
char *options
= NULL
;
2346 if (arg_selinux_apifs_context
)
2347 (void) asprintf(&options
,
2348 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2349 arg_uid_shift
+ TTY_GID
,
2350 arg_selinux_apifs_context
);
2353 (void) asprintf(&options
,
2354 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2355 arg_uid_shift
+ TTY_GID
);
2360 /* Mount /dev/pts itself */
2361 p
= prefix_roota(dest
, "/dev/pts");
2362 r
= RET_NERRNO(mkdir(p
, 0755));
2364 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2366 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2369 r
= userns_lchown(p
, 0, 0);
2371 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2373 /* Create /dev/ptmx symlink */
2374 p
= prefix_roota(dest
, "/dev/ptmx");
2375 if (symlink("pts/ptmx", p
) < 0)
2376 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2377 r
= userns_lchown(p
, 0, 0);
2379 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2381 /* And fix /dev/pts/ptmx ownership */
2382 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2383 r
= userns_lchown(p
, 0, 0);
2385 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2390 static int setup_stdio_as_dev_console(void) {
2391 _cleanup_close_
int terminal
= -EBADF
;
2394 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2395 * explicitly, if we are configured to. */
2396 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2398 return log_error_errno(terminal
, "Failed to open console: %m");
2400 /* Make sure we can continue logging to the original stderr, even if
2401 * stderr points elsewhere now */
2402 r
= log_dup_console();
2404 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2406 /* invalidates 'terminal' on success and failure */
2407 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2410 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2415 static int setup_dev_console(const char *console
) {
2416 _cleanup_free_
char *p
= NULL
;
2419 /* Create /dev/console symlink */
2420 r
= path_make_relative("/dev", console
, &p
);
2422 return log_error_errno(r
, "Failed to create relative path: %m");
2424 if (symlink(p
, "/dev/console") < 0)
2425 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2430 static int setup_keyring(void) {
2431 key_serial_t keyring
;
2433 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2434 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2435 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2436 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2437 * into the container. */
2439 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2440 if (keyring
== -1) {
2441 if (errno
== ENOSYS
)
2442 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2443 else if (ERRNO_IS_PRIVILEGE(errno
))
2444 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2446 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2452 static int setup_credentials(const char *root
) {
2456 if (arg_n_credentials
<= 0)
2459 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2461 return log_error_errno(r
, "Failed to create /run/host: %m");
2463 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2465 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2467 q
= prefix_roota(root
, "/run/host/credentials");
2468 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2472 for (size_t i
= 0; i
< arg_n_credentials
; i
++) {
2473 _cleanup_free_
char *j
= NULL
;
2474 _cleanup_close_
int fd
= -EBADF
;
2476 j
= path_join(q
, arg_credentials
[i
].id
);
2480 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2482 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2484 r
= loop_write(fd
, arg_credentials
[i
].data
, arg_credentials
[i
].size
, /* do_poll= */ false);
2486 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2488 if (fchmod(fd
, 0400) < 0)
2489 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2491 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2492 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2493 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2497 if (chmod(q
, 0500) < 0)
2498 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2500 r
= userns_lchown(q
, 0, 0);
2504 /* Make both mount and superblock read-only now */
2505 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2509 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2512 static int setup_kmsg(int fd_inner_socket
) {
2513 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2514 _cleanup_free_
char *fifo
= NULL
;
2515 _cleanup_close_
int fd
= -EBADF
;
2518 assert(fd_inner_socket
>= 0);
2520 BLOCK_WITH_UMASK(0000);
2522 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2523 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2524 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2525 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2527 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2529 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2531 if (mkfifo(fifo
, 0600) < 0)
2532 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2534 from
= TAKE_PTR(fifo
);
2536 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2540 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2542 return log_error_errno(errno
, "Failed to open fifo: %m");
2544 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2545 r
= send_one_fd(fd_inner_socket
, fd
, 0);
2547 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2553 union in_addr_union address4
;
2554 union in_addr_union address6
;
2555 struct FirewallContext
*fw_ctx
;
2558 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2559 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2564 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2565 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2569 static int setup_hostname(void) {
2572 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2575 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2577 return log_error_errno(r
, "Failed to set hostname: %m");
2582 static int setup_journal(const char *directory
) {
2583 _cleanup_free_
char *d
= NULL
;
2589 /* Don't link journals in ephemeral mode */
2593 if (arg_link_journal
== LINK_NO
)
2596 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2598 r
= sd_id128_get_machine(&this_id
);
2600 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2602 if (sd_id128_equal(arg_uuid
, this_id
)) {
2603 log_full(try ? LOG_WARNING
: LOG_ERR
,
2604 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2610 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2611 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2613 bool ignore
= r
== -EROFS
&& try;
2614 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2615 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2616 return ignore
? 0 : r
;
2620 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2621 q
= prefix_roota(directory
, p
);
2623 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2627 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2628 "%s: already a mount point, refusing to use for journal", p
);
2631 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2635 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2636 "%s: already a mount point, refusing to use for journal", q
);
2639 r
= readlink_and_make_absolute(p
, &d
);
2641 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2644 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2646 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2651 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2652 } else if (r
== -EINVAL
) {
2654 if (arg_link_journal
== LINK_GUEST
&&
2657 if (errno
== ENOTDIR
) {
2658 log_error("%s already exists and is neither a symlink nor a directory", p
);
2661 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2663 } else if (r
!= -ENOENT
)
2664 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2666 if (arg_link_journal
== LINK_GUEST
) {
2668 if (symlink(q
, p
) < 0) {
2670 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2673 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2676 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2678 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2682 if (arg_link_journal
== LINK_HOST
) {
2683 /* don't create parents here — if the host doesn't have
2684 * permanent journal set up, don't force it here */
2686 r
= RET_NERRNO(mkdir(p
, 0755));
2687 if (r
< 0 && r
!= -EEXIST
) {
2689 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2692 return log_error_errno(r
, "Failed to create %s: %m", p
);
2695 } else if (access(p
, F_OK
) < 0)
2698 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2699 log_warning("%s is not empty, proceeding anyway.", q
);
2701 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2703 return log_error_errno(r
, "Failed to create %s: %m", q
);
2705 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2707 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2712 static int drop_capabilities(uid_t uid
) {
2713 CapabilityQuintet q
;
2715 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2716 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2717 * arg_caps_retain. */
2719 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2720 q
= arg_full_capabilities
;
2722 if (q
.bounding
== UINT64_MAX
)
2723 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2725 if (q
.effective
== UINT64_MAX
)
2726 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2728 if (q
.inheritable
== UINT64_MAX
)
2729 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2731 if (q
.permitted
== UINT64_MAX
)
2732 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2734 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2735 q
.ambient
= arg_caps_ambient
;
2737 if (capability_quintet_mangle(&q
))
2738 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2741 q
= (CapabilityQuintet
) {
2742 .bounding
= arg_caps_retain
,
2743 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2744 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2745 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2746 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2749 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2750 * in order to maintain the same behavior as systemd < 242. */
2751 if (capability_quintet_mangle(&q
))
2752 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2753 "Some capabilities will not be set because they are not in the current bounding set.");
2757 return capability_quintet_enforce(&q
);
2760 static int reset_audit_loginuid(void) {
2761 _cleanup_free_
char *p
= NULL
;
2764 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2767 r
= read_one_line_file("/proc/self/loginuid", &p
);
2771 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2773 /* Already reset? */
2774 if (streq(p
, "4294967295"))
2777 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2780 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2781 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2782 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2783 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2784 "using systemd-nspawn. Sleeping for 5s... (%m)");
2792 static int mount_tunnel_dig(const char *root
) {
2796 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2797 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2798 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2799 (void) mkdir_p(p
, 0600);
2801 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2803 return log_error_errno(r
, "Failed to create /run/host: %m");
2805 r
= userns_mkdir(root
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2807 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2809 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
2810 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2814 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2821 static int mount_tunnel_open(void) {
2824 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2831 static int setup_machine_id(const char *directory
) {
2834 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2835 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2836 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2837 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2838 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2839 * container behaves nicely). */
2841 r
= id128_get_machine(directory
, &arg_uuid
);
2843 if (!ERRNO_IS_MACHINE_ID_UNSET(r
)) /* If the file is missing, empty, or uninitialized, we don't mind */
2844 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2846 if (sd_id128_is_null(arg_uuid
)) {
2847 r
= sd_id128_randomize(&arg_uuid
);
2849 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2856 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2861 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2864 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2865 if (r
== -EOPNOTSUPP
)
2866 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.");
2868 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2870 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2872 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2874 log_debug("Patched directory tree to match UID/GID range.");
2881 * < 0 : wait_for_terminate() failed to get the state of the
2882 * container, the container was terminated by a signal, or
2883 * failed for an unknown reason. No change is made to the
2884 * container argument.
2885 * > 0 : The program executed in the container terminated with an
2886 * error. The exit code of the program executed in the
2887 * container is returned. The container argument has been set
2888 * to CONTAINER_TERMINATED.
2889 * 0 : The container is being rebooted, has been shut down or exited
2890 * successfully. The container argument has been set to either
2891 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2893 * That is, success is indicated by a return value of zero, and an
2894 * error is indicated by a non-zero value.
2896 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2900 r
= wait_for_terminate(pid
, &status
);
2902 return log_warning_errno(r
, "Failed to wait for container: %m");
2904 switch (status
.si_code
) {
2907 if (status
.si_status
== 0)
2908 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2910 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2912 *container
= CONTAINER_TERMINATED
;
2913 return status
.si_status
;
2916 if (status
.si_status
== SIGINT
) {
2917 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2918 *container
= CONTAINER_TERMINATED
;
2921 } else if (status
.si_status
== SIGHUP
) {
2922 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2923 *container
= CONTAINER_REBOOTED
;
2929 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2930 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2933 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2934 "Container %s failed due to unknown reason.", arg_machine
);
2938 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2941 pid
= PTR_TO_PID(userdata
);
2943 if (kill(pid
, arg_kill_signal
) >= 0) {
2944 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2945 sd_event_source_set_userdata(s
, NULL
);
2950 sd_event_exit(sd_event_source_get_event(s
), 0);
2954 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2960 pid
= PTR_TO_PID(userdata
);
2965 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2966 return log_error_errno(errno
, "Failed to waitid(): %m");
2967 if (si
.si_pid
== 0) /* No pending children. */
2969 if (si
.si_pid
== pid
) {
2970 /* The main process we care for has exited. Return from
2971 * signal handler but leave the zombie. */
2972 sd_event_exit(sd_event_source_get_event(s
), 0);
2976 /* Reap all other children. */
2977 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2983 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2988 pid
= PTR_TO_PID(userdata
);
2990 if (arg_kill_signal
> 0) {
2991 log_info("Container termination requested. Attempting to halt container.");
2992 (void) kill(pid
, arg_kill_signal
);
2994 log_info("Container termination requested. Exiting.");
2995 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
3001 static int determine_names(void) {
3004 if (arg_template
&& !arg_directory
&& arg_machine
) {
3006 /* If --template= was specified then we should not
3007 * search for a machine, but instead create a new one
3008 * in /var/lib/machine. */
3010 arg_directory
= path_join("/var/lib/machines", arg_machine
);
3015 if (!arg_image
&& !arg_directory
) {
3017 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3019 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3021 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3023 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3025 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3026 r
= free_and_strdup(&arg_image
, i
->path
);
3028 r
= free_and_strdup(&arg_directory
, i
->path
);
3033 arg_read_only
= arg_read_only
|| i
->read_only
;
3035 r
= safe_getcwd(&arg_directory
);
3037 return log_error_errno(r
, "Failed to determine current directory: %m");
3040 if (!arg_directory
&& !arg_image
)
3041 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3045 if (arg_directory
&& path_equal(arg_directory
, "/"))
3046 arg_machine
= gethostname_malloc();
3047 else if (arg_image
) {
3050 r
= path_extract_filename(arg_image
, &arg_machine
);
3052 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_image
);
3054 /* Truncate suffix if there is one */
3055 e
= endswith(arg_machine
, ".raw");
3059 r
= path_extract_filename(arg_directory
, &arg_machine
);
3061 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_directory
);
3064 hostname_cleanup(arg_machine
);
3065 if (!hostname_is_valid(arg_machine
, 0))
3066 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3068 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3069 * to match fixed config file names. */
3070 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3071 if (!arg_settings_filename
)
3074 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3075 * instances at once without manually having to specify -M each time. */
3077 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3080 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3081 if (!arg_settings_filename
)
3088 static int chase_and_update(char **p
, unsigned flags
) {
3097 r
= chase(*p
, NULL
, flags
, &chased
, NULL
);
3099 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3101 return free_and_replace(*p
, chased
);
3104 static int determine_uid_shift(const char *directory
) {
3106 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3111 if (arg_uid_shift
== UID_INVALID
) {
3114 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3116 if (stat(directory
, &st
) < 0)
3117 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3119 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3121 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3122 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3123 "UID and GID base of %s don't match.", directory
);
3125 arg_uid_range
= UINT32_C(0x10000);
3127 if (arg_uid_shift
!= 0) {
3128 /* If the image is shifted already, then we'll fall back to classic chowning, for
3129 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3131 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3132 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3133 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3134 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3135 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3136 "UID base of %s is not zero, UID mapping not supported.", directory
);
3140 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3141 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3146 static unsigned long effective_clone_ns_flags(void) {
3147 unsigned long flags
= arg_clone_ns_flags
;
3149 if (arg_private_network
)
3150 flags
|= CLONE_NEWNET
;
3152 flags
|= CLONE_NEWCGROUP
;
3153 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3154 flags
|= CLONE_NEWUSER
;
3159 static int patch_sysctl(void) {
3161 /* This table is inspired by runc's sysctl() function */
3162 static const struct {
3165 unsigned long clone_flags
;
3167 { "kernel.hostname", false, CLONE_NEWUTS
},
3168 { "kernel.domainname", false, CLONE_NEWUTS
},
3169 { "kernel.msgmax", false, CLONE_NEWIPC
},
3170 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3171 { "kernel.msgmni", false, CLONE_NEWIPC
},
3172 { "kernel.sem", false, CLONE_NEWIPC
},
3173 { "kernel.shmall", false, CLONE_NEWIPC
},
3174 { "kernel.shmmax", false, CLONE_NEWIPC
},
3175 { "kernel.shmmni", false, CLONE_NEWIPC
},
3176 { "fs.mqueue.", true, CLONE_NEWIPC
},
3177 { "net.", true, CLONE_NEWNET
},
3180 unsigned long flags
;
3183 flags
= effective_clone_ns_flags();
3185 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3189 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3191 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3194 if (safe_sysctl
[i
].prefix
)
3195 good
= startswith(*k
, safe_sysctl
[i
].key
);
3197 good
= streq(*k
, safe_sysctl
[i
].key
);
3204 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3206 r
= sysctl_write(*k
, *v
);
3208 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3214 static int inner_child(
3216 int fd_inner_socket
,
3218 char **os_release_pairs
) {
3220 _cleanup_free_
char *home
= NULL
;
3223 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3224 NULL
, /* container */
3229 NULL
, /* container_uuid */
3230 NULL
, /* LISTEN_FDS */
3231 NULL
, /* LISTEN_PID */
3232 NULL
, /* NOTIFY_SOCKET */
3233 NULL
, /* CREDENTIALS_DIRECTORY */
3237 const char *exec_target
;
3238 _cleanup_strv_free_
char **env_use
= NULL
;
3239 int r
, which_failed
;
3241 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3242 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3243 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3244 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3245 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3246 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3249 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3250 * unshare(). See below. */
3253 assert(fd_inner_socket
>= 0);
3255 log_debug("Inner child is initializing.");
3257 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3258 /* Tell the parent, that it now can write the UID map. */
3259 (void) barrier_place(barrier
); /* #1 */
3261 /* Wait until the parent wrote the UID map */
3262 if (!barrier_place_and_sync(barrier
)) /* #2 */
3263 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3265 /* Become the new root user inside our namespace */
3266 r
= reset_uid_gid();
3268 return log_error_errno(r
, "Couldn't become new root: %m");
3270 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3271 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3272 * propagation, but simply create new peer groups for all our mounts). */
3273 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3279 arg_mount_settings
| MOUNT_IN_USERNS
,
3281 arg_selinux_apifs_context
);
3285 if (!arg_network_namespace_path
&& arg_private_network
) {
3286 r
= unshare(CLONE_NEWNET
);
3288 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3290 /* Tell the parent that it can setup network interfaces. */
3291 (void) barrier_place(barrier
); /* #3 */
3294 r
= mount_sysfs(NULL
, arg_mount_settings
);
3298 /* Wait until we are cgroup-ified, so that we
3299 * can mount the right cgroup path writable */
3300 if (!barrier_place_and_sync(barrier
)) /* #4 */
3301 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3302 "Parent died too early");
3305 r
= unshare(CLONE_NEWCGROUP
);
3307 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3310 arg_unified_cgroup_hierarchy
,
3311 arg_userns_mode
!= USER_NAMESPACE_NO
,
3314 arg_selinux_apifs_context
,
3317 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3321 r
= setup_boot_id();
3325 r
= setup_kmsg(fd_inner_socket
);
3332 arg_n_custom_mounts
,
3335 arg_selinux_apifs_context
,
3336 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3341 return log_error_errno(errno
, "setsid() failed: %m");
3343 if (arg_private_network
)
3344 (void) loopback_setup();
3346 if (arg_expose_ports
) {
3347 r
= expose_port_send_rtnl(fd_inner_socket
);
3352 if (arg_console_mode
!= CONSOLE_PIPE
) {
3353 _cleanup_close_
int master
= -EBADF
;
3354 _cleanup_free_
char *console
= NULL
;
3356 /* Allocate a pty and make it available as /dev/console. */
3357 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3359 return log_error_errno(master
, "Failed to allocate a pty: %m");
3361 r
= setup_dev_console(console
);
3363 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3365 r
= send_one_fd(fd_inner_socket
, master
, 0);
3367 return log_error_errno(r
, "Failed to send master fd: %m");
3369 r
= setup_stdio_as_dev_console();
3378 if (arg_oom_score_adjust_set
) {
3379 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3381 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3384 if (arg_cpu_set
.set
)
3385 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3386 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3388 (void) setup_hostname();
3390 if (arg_personality
!= PERSONALITY_INVALID
) {
3391 r
= safe_personality(arg_personality
);
3393 return log_error_errno(r
, "personality() failed: %m");
3394 #ifdef ARCHITECTURE_SECONDARY
3395 } else if (arg_architecture
== ARCHITECTURE_SECONDARY
) {
3396 r
= safe_personality(PER_LINUX32
);
3398 return log_error_errno(r
, "personality() failed: %m");
3400 } else if (!arg_quiet
&& arg_architecture
>= 0 && arg_architecture
!= native_architecture())
3401 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3402 "invocation with qemu userspace emulator (or equivalent) in effect.",
3403 architecture_to_string(arg_architecture
));
3405 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3407 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3412 if (is_seccomp_available()) {
3414 r
= seccomp_load(arg_seccomp
);
3415 if (ERRNO_IS_SECCOMP_FATAL(r
))
3416 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3418 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3423 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3428 if (arg_suppress_sync
) {
3430 r
= seccomp_suppress_sync();
3432 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3434 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3439 if (arg_selinux_context
)
3440 if (setexeccon(arg_selinux_context
) < 0)
3441 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3444 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3445 * if we need to later on. */
3446 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3447 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3449 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3450 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3452 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3456 r
= drop_capabilities(getuid());
3458 return log_error_errno(r
, "Dropping capabilities failed: %m");
3460 if (arg_no_new_privileges
)
3461 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3462 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3464 /* LXC sets container=lxc, so follow the scheme here */
3465 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3467 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3471 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3472 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3475 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3476 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3477 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
?: "root") < 0)
3480 assert(!sd_id128_is_null(arg_uuid
));
3482 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3485 if (fdset_size(fds
) > 0) {
3486 r
= fdset_cloexec(fds
, false);
3488 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3490 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3491 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3494 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3497 if (arg_n_credentials
> 0) {
3498 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3504 if (arg_start_mode
!= START_BOOT
) {
3505 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3511 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3515 /* Let the parent know that we are ready and
3516 * wait until the parent is ready with the
3518 if (!barrier_place_and_sync(barrier
)) /* #5 */
3519 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3522 if (chdir(arg_chdir
) < 0)
3523 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3525 if (arg_start_mode
== START_PID2
) {
3526 r
= stub_pid1(arg_uuid
);
3531 if (arg_console_mode
!= CONSOLE_PIPE
) {
3532 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3533 * are configured for that. Acquire it as controlling tty. */
3534 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3535 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3538 log_debug("Inner child completed, invoking payload.");
3540 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3541 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3542 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3544 log_set_open_when_needed(true);
3545 log_settle_target();
3547 (void) fdset_close_others(fds
);
3549 if (arg_start_mode
== START_BOOT
) {
3553 /* Automatically search for the init system */
3555 m
= strv_length(arg_parameters
);
3556 a
= newa(char*, m
+ 2);
3557 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3560 FOREACH_STRING(init
,
3561 "/usr/lib/systemd/systemd",
3562 "/lib/systemd/systemd",
3564 a
[0] = (char*) init
;
3565 execve(a
[0], a
, env_use
);
3568 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3569 } else if (!strv_isempty(arg_parameters
)) {
3570 const char *dollar_path
;
3572 exec_target
= arg_parameters
[0];
3574 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3576 dollar_path
= strv_env_get(env_use
, "PATH");
3578 if (setenv("PATH", dollar_path
, 1) < 0)
3579 return log_error_errno(errno
, "Failed to update $PATH: %m");
3582 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3585 /* If we cannot change the directory, we'll end up in /, that is expected. */
3586 (void) chdir(home
?: "/root");
3588 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3589 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3590 execle("/bin/bash", "-bash", NULL
, env_use
);
3591 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3592 execle("/bin/sh", "-sh", NULL
, env_use
);
3594 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3597 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3600 static int setup_notify_child(void) {
3601 _cleanup_close_
int fd
= -EBADF
;
3602 static const union sockaddr_union sa
= {
3603 .un
.sun_family
= AF_UNIX
,
3604 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3608 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3610 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3612 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3613 (void) sockaddr_un_unlink(&sa
.un
);
3615 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3617 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3619 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3621 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3623 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3625 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3630 static int outer_child(
3632 const char *directory
,
3633 DissectedImage
*dissected_image
,
3634 int fd_outer_socket
,
3635 int fd_inner_socket
,
3639 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3640 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3641 _cleanup_close_
int fd
= -EBADF
, mntns_fd
= -EBADF
;
3648 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3649 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3650 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3651 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3652 * forked off it, and it exits. */
3656 assert(fd_outer_socket
>= 0);
3657 assert(fd_inner_socket
>= 0);
3659 log_debug("Outer child is initializing.");
3661 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3663 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3665 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3666 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3668 r
= reset_audit_loginuid();
3672 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3673 * mounts to the real root. */
3674 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3678 if (dissected_image
) {
3679 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3680 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3681 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3682 * right place right away. This makes sure ESP partitions and userns are compatible. */
3684 r
= dissected_image_mount_and_warn(
3689 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3690 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3691 DISSECT_IMAGE_USR_NO_ROOT
|
3692 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3693 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3698 r
= determine_uid_shift(directory
);
3702 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3703 r
= namespace_open(0, NULL
, &mntns_fd
, NULL
, NULL
, NULL
);
3705 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3707 l
= send_one_fd(fd_outer_socket
, mntns_fd
, 0);
3709 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3710 mntns_fd
= safe_close(mntns_fd
);
3712 /* Let the parent know which UID shift we read from the image */
3713 l
= send(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3715 return log_error_errno(errno
, "Failed to send UID shift: %m");
3716 if (l
!= sizeof(arg_uid_shift
))
3717 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3718 "Short write while sending UID shift.");
3720 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3721 /* When we are supposed to pick the UID shift, the parent will check now whether the
3722 * UID shift we just read from the image is available. If yes, it will send the UID
3723 * shift back to us, if not it will pick a different one, and send it back to us. */
3725 l
= recv(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3727 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3728 if (l
!= sizeof(arg_uid_shift
))
3729 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3730 "Short read while receiving UID shift.");
3733 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3734 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3737 if (path_equal(directory
, "/")) {
3738 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3739 * place, so that we can make changes to its mount structure (for example, to implement
3740 * --volatile=) without this interfering with our ability to access files such as
3741 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3742 * (instead of a temporary directory, since we are living in our own mount namespace here
3743 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3744 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3746 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3750 directory
= "/run/systemd/nspawn-root";
3753 /* Make sure we always have a mount that we can move to root later on. */
3754 r
= make_mount_point(directory
);
3758 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3759 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3760 * we'll live in our own little world from now on, and propagation from the host may only happen via
3761 * the mount tunnel dir, or not at all. */
3762 r
= mount_follow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_PRIVATE
|MS_REC
, NULL
);
3766 r
= setup_pivot_root(
3769 arg_pivot_root_old
);
3773 r
= setup_volatile_mode(
3777 arg_selinux_apifs_context
);
3781 r
= bind_user_prepare(
3786 &arg_custom_mounts
, &arg_n_custom_mounts
,
3787 &bind_user_context
);
3791 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3792 /* Send the user maps we determined to the parent, so that it installs it in our user
3793 * namespace UID map table */
3795 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3797 bind_user_context
->data
[i
].payload_user
->uid
,
3798 bind_user_context
->data
[i
].host_user
->uid
,
3799 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3800 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3803 l
= send(fd_outer_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3805 return log_error_errno(errno
, "Failed to send user UID map: %m");
3806 if (l
!= sizeof(map
))
3807 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3808 "Short write while sending user UID map.");
3815 arg_n_custom_mounts
,
3818 arg_selinux_apifs_context
,
3823 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3824 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3825 arg_uid_shift
!= 0) {
3827 r
= remount_idmap(directory
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3829 if (r
== -EINVAL
|| ERRNO_IS_NOT_SUPPORTED(r
)) {
3830 /* This might fail because the kernel or file system doesn't support idmapping. We
3831 * can't really distinguish this nicely, nor do we have any guarantees about the
3832 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3833 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3834 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3835 "ID mapped mounts are apparently not available, sorry.");
3837 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3838 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3840 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3842 log_debug("ID mapped mounts available, making use of them.");
3847 if (dissected_image
) {
3848 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3849 r
= dissected_image_mount(
3854 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3855 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3856 DISSECT_IMAGE_USR_NO_ROOT
|
3857 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3858 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3860 return log_error_errno(r
, "File system check for image failed: %m");
3862 return log_error_errno(r
, "Failed to mount image file system: %m");
3865 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3866 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3868 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3872 l
= send(fd_outer_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3874 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3875 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3876 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3877 "Short write while sending cgroup mode.");
3880 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3884 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3888 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3889 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3890 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3892 return log_error_errno(r
, "Failed to make tree read-only: %m");
3895 r
= mount_all(directory
,
3898 arg_selinux_apifs_context
);
3902 r
= copy_devnodes(directory
);
3906 r
= make_extra_nodes(directory
);
3910 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3912 p
= prefix_roota(directory
, "/run/host");
3913 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3915 r
= setup_pts(directory
);
3919 r
= mount_tunnel_dig(directory
);
3923 r
= setup_keyring();
3927 r
= setup_credentials(directory
);
3931 r
= bind_user_setup(bind_user_context
, directory
);
3938 arg_n_custom_mounts
,
3941 arg_selinux_apifs_context
,
3942 MOUNT_NON_ROOT_ONLY
);
3946 r
= setup_timezone(directory
);
3950 r
= setup_resolv_conf(directory
);
3954 r
= setup_machine_id(directory
);
3958 r
= setup_journal(directory
);
3962 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3963 p
= prefix_roota(directory
, "/run/host/container-manager");
3964 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
);
3966 /* The same stuff as the $container_uuid env var */
3967 p
= prefix_roota(directory
, "/run/host/container-uuid");
3968 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
3970 if (!arg_use_cgns
) {
3973 arg_unified_cgroup_hierarchy
,
3974 arg_userns_mode
!= USER_NAMESPACE_NO
,
3977 arg_selinux_apifs_context
,
3983 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3984 * mounts available in systemd services inside the container that create a new mount namespace. See
3985 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3986 * will inherit the shared propagation mode.
3988 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3989 * directory mount to root later on.
3990 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3992 r
= mount_switch_root(directory
, MS_SHARED
);
3994 return log_error_errno(r
, "Failed to move root directory: %m");
3996 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
3997 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
3999 r
= mount_tunnel_open();
4003 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4004 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4005 * requires that a fully visible instance is already present in the target mount
4006 * namespace. Mount one here so the inner child can mount its own instances. Later
4007 * we umount the temporary instances created here before we actually exec the
4008 * payload. Since the rootfs is shared the umount will propagate into the container.
4009 * Note, the inner child wouldn't be able to unmount the instances on its own since
4010 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4012 r
= pin_fully_visible_fs();
4017 fd
= setup_notify_child();
4021 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
4022 arg_clone_ns_flags
|
4023 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
4025 return log_error_errno(errno
, "Failed to fork inner child: %m");
4027 fd_outer_socket
= safe_close(fd_outer_socket
);
4029 /* The inner child has all namespaces that are requested, so that we all are owned by the
4030 * user if user namespaces are turned on. */
4032 if (arg_network_namespace_path
) {
4033 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
4035 return log_error_errno(r
, "Failed to join network namespace: %m");
4038 r
= inner_child(barrier
, fd_inner_socket
, fds
, os_release_pairs
);
4040 _exit(EXIT_FAILURE
);
4042 _exit(EXIT_SUCCESS
);
4045 l
= send(fd_outer_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4047 return log_error_errno(errno
, "Failed to send PID: %m");
4048 if (l
!= sizeof(pid
))
4049 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4050 "Short write while sending PID.");
4052 l
= send(fd_outer_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4054 return log_error_errno(errno
, "Failed to send machine ID: %m");
4055 if (l
!= sizeof(arg_uuid
))
4056 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4057 "Short write while sending machine ID.");
4059 l
= send_one_fd(fd_outer_socket
, fd
, 0);
4061 return log_error_errno(l
, "Failed to send notify fd: %m");
4063 fd_outer_socket
= safe_close(fd_outer_socket
);
4064 fd_inner_socket
= safe_close(fd_inner_socket
);
4065 netns_fd
= safe_close(netns_fd
);
4070 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4071 bool tried_hashed
= false;
4072 unsigned n_tries
= 100;
4077 assert(ret_lock_file
);
4078 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4079 assert(arg_uid_range
== 0x10000U
);
4083 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4086 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4087 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4092 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4094 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4097 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4098 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4099 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4104 /* Make some superficial checks whether the range is currently known in the user database */
4105 if (getpwuid(candidate
))
4107 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4109 if (getgrgid(candidate
))
4111 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4114 *ret_lock_file
= lf
;
4115 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4120 if (arg_machine
&& !tried_hashed
) {
4121 /* Try to hash the base from the container name */
4123 static const uint8_t hash_key
[] = {
4124 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4125 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4128 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4130 tried_hashed
= true;
4132 random_bytes(&candidate
, sizeof(candidate
));
4134 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4135 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4139 static int add_one_uid_map(
4141 uid_t container_uid
,
4145 return strextendf(p
,
4146 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4147 container_uid
, host_uid
, range
);
4150 static int make_uid_map_string(
4151 const uid_t bind_user_uid
[],
4152 size_t n_bind_user_uid
,
4156 _cleanup_free_
char *s
= NULL
;
4157 uid_t previous_uid
= 0;
4160 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4161 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4164 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4165 * quadruplet, consisting of host and container UID + GID. */
4167 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4168 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4169 host_uid
= bind_user_uid
[i
*4+offset
+1];
4171 assert(previous_uid
<= payload_uid
);
4172 assert(payload_uid
< arg_uid_range
);
4174 /* Add a range to close the gap to previous entry */
4175 if (payload_uid
> previous_uid
) {
4176 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4181 /* Map this specific user */
4182 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4186 previous_uid
= payload_uid
+ 1;
4189 /* And add a range to close the gap to finish the range */
4190 if (arg_uid_range
> previous_uid
) {
4191 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4202 static int setup_uid_map(
4204 const uid_t bind_user_uid
[],
4205 size_t n_bind_user_uid
) {
4207 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4208 _cleanup_free_
char *s
= NULL
;
4213 /* Build the UID map string */
4214 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4217 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4218 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4220 return log_error_errno(r
, "Failed to write UID map: %m");
4222 /* And now build the GID map string */
4224 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4227 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4228 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4230 return log_error_errno(r
, "Failed to write GID map: %m");
4235 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4236 char buf
[NOTIFY_BUFFER_MAX
+1];
4238 struct iovec iovec
= {
4240 .iov_len
= sizeof(buf
)-1,
4242 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4243 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4244 struct msghdr msghdr
= {
4247 .msg_control
= &control
,
4248 .msg_controllen
= sizeof(control
),
4250 struct ucred
*ucred
;
4252 pid_t inner_child_pid
;
4253 _cleanup_strv_free_
char **tags
= NULL
;
4258 inner_child_pid
= PTR_TO_PID(userdata
);
4260 if (revents
!= EPOLLIN
) {
4261 log_warning("Got unexpected poll event for notify fd.");
4265 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4267 if (ERRNO_IS_TRANSIENT(n
))
4270 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4273 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4276 cmsg_close_all(&msghdr
);
4278 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4279 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4280 log_debug("Received notify message without valid credentials. Ignoring.");
4284 if ((size_t) n
>= sizeof(buf
)) {
4285 log_warning("Received notify message exceeded maximum size. Ignoring.");
4290 tags
= strv_split(buf
, "\n\r");
4294 if (strv_contains(tags
, "READY=1")) {
4295 r
= sd_notify(false, "READY=1\n");
4297 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4300 p
= strv_find_startswith(tags
, "STATUS=");
4302 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4307 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4310 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4312 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4314 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4319 static int merge_settings(Settings
*settings
, const char *path
) {
4325 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4326 * that this steals the fields of the Settings* structure, and hence modifies it. */
4328 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4329 settings
->start_mode
>= 0) {
4330 arg_start_mode
= settings
->start_mode
;
4331 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4334 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4335 settings
->ephemeral
>= 0)
4336 arg_ephemeral
= settings
->ephemeral
;
4338 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4341 if (!arg_settings_trusted
)
4342 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4344 free_and_replace(arg_directory
, settings
->root
);
4347 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4348 settings
->pivot_root_new
) {
4349 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4350 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4353 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4354 settings
->working_directory
)
4355 free_and_replace(arg_chdir
, settings
->working_directory
);
4357 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4358 settings
->environment
)
4359 strv_free_and_replace(arg_setenv
, settings
->environment
);
4361 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4364 free_and_replace(arg_user
, settings
->user
);
4366 if (uid_is_valid(settings
->uid
))
4367 arg_uid
= settings
->uid
;
4368 if (gid_is_valid(settings
->gid
))
4369 arg_gid
= settings
->gid
;
4370 if (settings
->n_supplementary_gids
> 0) {
4371 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4372 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4376 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4377 uint64_t plus
, minus
;
4378 uint64_t network_minus
= 0;
4381 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4382 * Settings structure */
4384 plus
= settings
->capability
;
4385 minus
= settings
->drop_capability
;
4387 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4388 settings_network_configured(settings
)) {
4389 if (settings_private_network(settings
))
4390 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4392 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4395 if (!arg_settings_trusted
&& plus
!= 0) {
4396 if (settings
->capability
!= 0)
4397 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4399 arg_caps_retain
&= ~network_minus
;
4400 arg_caps_retain
|= plus
;
4403 arg_caps_retain
&= ~minus
;
4405 /* Copy the full capabilities over too */
4406 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4407 if (!arg_settings_trusted
)
4408 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4410 arg_full_capabilities
= settings
->full_capabilities
;
4413 ambient
= settings
->ambient_capability
;
4414 if (!arg_settings_trusted
&& ambient
!= 0)
4415 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4417 arg_caps_ambient
|= ambient
;
4420 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4421 settings
->kill_signal
> 0)
4422 arg_kill_signal
= settings
->kill_signal
;
4424 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4425 settings
->personality
!= PERSONALITY_INVALID
)
4426 arg_personality
= settings
->personality
;
4428 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4429 !sd_id128_is_null(settings
->machine_id
)) {
4431 if (!arg_settings_trusted
)
4432 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4434 arg_uuid
= settings
->machine_id
;
4437 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4438 settings
->read_only
>= 0)
4439 arg_read_only
= settings
->read_only
;
4441 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4442 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4443 arg_volatile_mode
= settings
->volatile_mode
;
4445 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4446 settings
->n_custom_mounts
> 0) {
4448 if (!arg_settings_trusted
)
4449 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4451 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4452 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4453 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4454 settings
->n_custom_mounts
= 0;
4458 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4459 settings_network_configured(settings
)) {
4461 if (!arg_settings_trusted
)
4462 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4464 arg_network_veth
= settings_network_veth(settings
);
4465 arg_private_network
= settings_private_network(settings
);
4467 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4468 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4469 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4470 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4472 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4473 free_and_replace(arg_network_zone
, settings
->network_zone
);
4475 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4479 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4480 settings
->expose_ports
) {
4482 if (!arg_settings_trusted
)
4483 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4485 expose_port_free_all(arg_expose_ports
);
4486 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4490 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4491 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4493 if (!arg_settings_trusted
)
4494 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4496 arg_userns_mode
= settings
->userns_mode
;
4497 arg_uid_shift
= settings
->uid_shift
;
4498 arg_uid_range
= settings
->uid_range
;
4499 arg_userns_ownership
= settings
->userns_ownership
;
4503 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4504 !strv_isempty(settings
->bind_user
))
4505 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4507 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4508 settings
->notify_ready
>= 0)
4509 arg_notify_ready
= settings
->notify_ready
;
4511 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4513 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4514 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4515 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4517 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4518 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4523 if (settings
->seccomp
) {
4524 if (!arg_settings_trusted
)
4525 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4527 seccomp_release(arg_seccomp
);
4528 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4534 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4535 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4538 if (!settings
->rlimit
[rl
])
4541 if (!arg_settings_trusted
) {
4542 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4546 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4549 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4551 free_and_replace(arg_hostname
, settings
->hostname
);
4553 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4554 settings
->no_new_privileges
>= 0)
4555 arg_no_new_privileges
= settings
->no_new_privileges
;
4557 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4558 settings
->oom_score_adjust_set
) {
4560 if (!arg_settings_trusted
)
4561 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4563 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4564 arg_oom_score_adjust_set
= true;
4568 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4569 settings
->cpu_set
.set
) {
4571 if (!arg_settings_trusted
)
4572 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4574 cpu_set_reset(&arg_cpu_set
);
4575 arg_cpu_set
= TAKE_STRUCT(settings
->cpu_set
);
4579 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4580 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4581 arg_resolv_conf
= settings
->resolv_conf
;
4583 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4584 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4586 if (!arg_settings_trusted
)
4587 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4589 arg_link_journal
= settings
->link_journal
;
4590 arg_link_journal_try
= settings
->link_journal_try
;
4594 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4595 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4596 arg_timezone
= settings
->timezone
;
4598 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4601 if (!arg_settings_trusted
)
4602 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4604 free_and_replace(arg_slice
, settings
->slice
);
4607 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4608 settings
->use_cgns
>= 0) {
4610 if (!arg_settings_trusted
)
4611 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4613 arg_use_cgns
= settings
->use_cgns
;
4616 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4617 settings
->clone_ns_flags
!= ULONG_MAX
) {
4619 if (!arg_settings_trusted
)
4620 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4622 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4625 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4626 settings
->console_mode
>= 0) {
4628 if (!arg_settings_trusted
)
4629 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4631 arg_console_mode
= settings
->console_mode
;
4634 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4635 settings
->suppress_sync
>= 0)
4636 arg_suppress_sync
= settings
->suppress_sync
;
4638 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4639 * don't consult arg_settings_mask for them. */
4641 sd_bus_message_unref(arg_property_message
);
4642 arg_property_message
= TAKE_PTR(settings
->properties
);
4644 arg_console_width
= settings
->console_width
;
4645 arg_console_height
= settings
->console_height
;
4647 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4648 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4649 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4650 settings
->n_extra_nodes
= 0;
4655 static int load_settings(void) {
4656 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4657 _cleanup_fclose_
FILE *f
= NULL
;
4658 _cleanup_free_
char *p
= NULL
;
4664 /* If all settings are masked, there's no point in looking for
4665 * the settings file */
4666 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4669 /* We first look in the admin's directories in /etc and /run */
4670 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4671 _cleanup_free_
char *j
= NULL
;
4673 j
= path_join(i
, arg_settings_filename
);
4681 /* By default, we trust configuration from /etc and /run */
4682 if (arg_settings_trusted
< 0)
4683 arg_settings_trusted
= true;
4688 if (errno
!= ENOENT
)
4689 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4693 /* After that, let's look for a file next to the
4694 * actual image we shall boot. */
4697 r
= file_in_same_dir(arg_image
, arg_settings_filename
, &p
);
4699 return log_error_errno(r
, "Failed to generate settings path from image path: %m");
4700 } else if (arg_directory
) {
4701 r
= file_in_same_dir(arg_directory
, arg_settings_filename
, &p
);
4702 if (r
< 0 && r
!= -EADDRNOTAVAIL
) /* if directory is root fs, don't complain */
4703 return log_error_errno(r
, "Failed to generate settings path from directory path: %m");
4708 if (!f
&& errno
!= ENOENT
)
4709 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4711 /* By default, we do not trust configuration from /var/lib/machines */
4712 if (arg_settings_trusted
< 0)
4713 arg_settings_trusted
= false;
4720 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4722 r
= settings_load(f
, p
, &settings
);
4726 return merge_settings(settings
, p
);
4729 static int load_oci_bundle(void) {
4730 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4733 if (!arg_oci_bundle
)
4736 /* By default let's trust OCI bundles */
4737 if (arg_settings_trusted
< 0)
4738 arg_settings_trusted
= true;
4740 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4744 return merge_settings(settings
, arg_oci_bundle
);
4747 static int run_container(
4748 DissectedImage
*dissected_image
,
4750 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4751 struct ExposeArgs
*expose_args
,
4752 int *master
, pid_t
*pid
, int *ret
) {
4754 static const struct sigaction sa
= {
4755 .sa_handler
= nop_signal_handler
,
4756 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4759 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4760 _cleanup_close_
int etc_passwd_lock
= -EBADF
;
4761 _cleanup_close_pair_
int
4762 fd_inner_socket_pair
[2] = PIPE_EBADF
,
4763 fd_outer_socket_pair
[2] = PIPE_EBADF
;
4765 _cleanup_close_
int notify_socket
= -EBADF
, mntns_fd
= -EBADF
, fd_kmsg_fifo
= -EBADF
;
4766 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4767 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4768 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4769 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4770 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4771 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4772 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4773 size_t n_bind_user_uid
= 0;
4774 ContainerStatus container_status
= 0;
4778 _cleanup_close_
int child_netns_fd
= -EBADF
;
4780 assert_se(sigemptyset(&mask_chld
) == 0);
4781 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4783 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4784 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4785 * check with getpwuid() if the specific user already exists. Note that /etc might be
4786 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4787 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4788 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4791 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4792 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4793 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4796 r
= barrier_create(&barrier
);
4798 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4800 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_inner_socket_pair
) < 0)
4801 return log_error_errno(errno
, "Failed to create inner socket pair: %m");
4803 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_outer_socket_pair
) < 0)
4804 return log_error_errno(errno
, "Failed to create outer socket pair: %m");
4806 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4807 * parent's blocking calls and give it a chance to call wait() and terminate. */
4808 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4810 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4812 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4814 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4816 if (arg_network_namespace_path
) {
4817 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4818 if (child_netns_fd
< 0)
4819 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4821 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4823 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
);
4825 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4827 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4828 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4831 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4833 return log_error_errno(errno
, "clone() failed%s: %m",
4835 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4838 /* The outer child only has a file system namespace. */
4839 barrier_set_role(&barrier
, BARRIER_CHILD
);
4841 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
4842 fd_outer_socket_pair
[0] = safe_close(fd_outer_socket_pair
[0]);
4844 (void) reset_all_signal_handlers();
4845 (void) reset_signal_mask();
4847 r
= outer_child(&barrier
,
4850 fd_outer_socket_pair
[1],
4851 fd_inner_socket_pair
[1],
4855 _exit(EXIT_FAILURE
);
4857 _exit(EXIT_SUCCESS
);
4860 barrier_set_role(&barrier
, BARRIER_PARENT
);
4864 fd_inner_socket_pair
[1] = safe_close(fd_inner_socket_pair
[1]);
4865 fd_outer_socket_pair
[1] = safe_close(fd_outer_socket_pair
[1]);
4867 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4868 mntns_fd
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4870 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4872 /* The child just let us know the UID shift it might have read from the image. */
4873 l
= recv(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4875 return log_error_errno(errno
, "Failed to read UID shift: %m");
4876 if (l
!= sizeof arg_uid_shift
)
4877 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4879 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4880 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4881 * image, but if that's already in use, pick a new one, and report back to the child,
4882 * which one we now picked. */
4884 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4886 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4888 l
= send(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4890 return log_error_errno(errno
, "Failed to send UID shift: %m");
4891 if (l
!= sizeof arg_uid_shift
)
4892 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4895 n_bind_user_uid
= strv_length(arg_bind_user
);
4896 if (n_bind_user_uid
> 0) {
4897 /* Right after the UID shift, we'll receive the list of UID mappings for the
4898 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4900 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4904 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4905 l
= recv(fd_outer_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4907 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4908 if (l
!= sizeof(uid_t
)*4)
4909 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4910 SYNTHETIC_ERRNO(EIO
),
4911 "Short read while reading bind user UID pairs.");
4916 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4917 /* The child let us know the support cgroup mode it might have read from the image. */
4918 l
= recv(fd_outer_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4920 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4921 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4922 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
4923 l
, l
== 0 ? " The child is most likely dead." : "");
4926 /* Wait for the outer child. */
4927 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4930 if (r
!= EXIT_SUCCESS
)
4933 /* And now retrieve the PID of the inner child. */
4934 l
= recv(fd_outer_socket_pair
[0], pid
, sizeof *pid
, 0);
4936 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4937 if (l
!= sizeof *pid
)
4938 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4940 /* We also retrieve container UUID in case it was generated by outer child */
4941 l
= recv(fd_outer_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4943 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4944 if (l
!= sizeof(arg_uuid
))
4945 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4947 /* We also retrieve the socket used for notifications generated by outer child */
4948 notify_socket
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4949 if (notify_socket
< 0)
4950 return log_error_errno(notify_socket
,
4951 "Failed to receive notification socket from the outer child: %m");
4953 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4955 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4956 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4957 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4959 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4963 (void) barrier_place(&barrier
); /* #2 */
4966 if (arg_private_network
) {
4967 if (!arg_network_namespace_path
) {
4968 /* Wait until the child has unshared its network namespace. */
4969 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4970 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4973 if (child_netns_fd
< 0) {
4974 /* Make sure we have an open file descriptor to the child's network
4975 * namespace so it stays alive even if the child exits. */
4976 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4978 return log_error_errno(r
, "Failed to open child network namespace: %m");
4981 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4985 if (arg_network_veth
) {
4986 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4987 arg_network_bridge
|| arg_network_zone
);
4993 if (arg_network_bridge
) {
4994 /* Add the interface to a bridge */
4995 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
5000 } else if (arg_network_zone
) {
5001 /* Add the interface to a bridge, possibly creating it */
5002 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5010 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5014 /* We created the primary and extra veth links now; let's remember this, so that we know to
5015 remove them later on. Note that we don't bother with removing veth links that were created
5016 here when their setup failed half-way, because in that case the kernel should be able to
5017 remove them on its own, since they cannot be referenced by anything yet. */
5018 *veth_created
= true;
5020 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5024 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5029 if (arg_register
|| !arg_keep_unit
) {
5030 r
= sd_bus_default_system(&bus
);
5032 return log_error_errno(r
, "Failed to open system bus: %m");
5034 r
= sd_bus_set_close_on_exit(bus
, false);
5036 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5039 if (!arg_keep_unit
) {
5040 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5041 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5042 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5044 r
= sd_bus_match_signal_async(
5047 "org.freedesktop.systemd1",
5049 "org.freedesktop.systemd1.Scope",
5051 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5053 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5057 r
= register_machine(
5065 arg_custom_mounts
, arg_n_custom_mounts
,
5068 arg_property_message
,
5070 arg_container_service_name
);
5074 } else if (!arg_keep_unit
) {
5080 arg_custom_mounts
, arg_n_custom_mounts
,
5083 arg_property_message
);
5087 } else if (arg_slice
|| arg_property
)
5088 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5090 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5094 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5098 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5102 /* Notify the child that the parent is ready with all
5103 * its setup (including cgroup-ification), and that
5104 * the child can now hand over control to the code to
5105 * run inside the container. */
5106 (void) barrier_place(&barrier
); /* #4 */
5108 /* Block SIGCHLD here, before notifying child.
5109 * process_pty() will handle it with the other signals. */
5110 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5112 /* Reset signal to default */
5113 r
= default_signals(SIGCHLD
);
5115 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5117 r
= sd_event_new(&event
);
5119 return log_error_errno(r
, "Failed to get default event source: %m");
5121 (void) sd_event_set_watchdog(event
, true);
5124 r
= sd_bus_attach_event(bus
, event
, 0);
5126 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5129 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5133 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5134 r
= wipe_fully_visible_fs(mntns_fd
);
5137 mntns_fd
= safe_close(mntns_fd
);
5140 /* Let the child know that we are ready and wait that the child is completely ready now. */
5141 if (!barrier_place_and_sync(&barrier
)) /* #5 */
5142 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5144 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5145 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5146 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5148 (void) sd_notifyf(false,
5149 "STATUS=Container running.\n"
5150 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5151 if (!arg_notify_ready
) {
5152 r
= sd_notify(false, "READY=1\n");
5154 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5157 if (arg_kill_signal
> 0) {
5158 /* Try to kill the init system on SIGINT or SIGTERM */
5159 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5160 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5162 /* Immediately exit */
5163 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5164 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5167 (void) sd_event_add_signal(event
, NULL
, SIGRTMIN
+18, sigrtmin18_handler
, NULL
);
5169 r
= sd_event_add_memory_pressure(event
, NULL
, NULL
, NULL
);
5171 log_debug_errno(r
, "Failed allocate memory pressure event source, ignoring: %m");
5173 /* Exit when the child exits */
5174 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5176 /* Retrieve the kmsg fifo allocated by inner child */
5177 fd_kmsg_fifo
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5178 if (fd_kmsg_fifo
< 0)
5179 return log_error_errno(fd_kmsg_fifo
, "Failed to receive kmsg fifo from inner child: %m");
5181 if (arg_expose_ports
) {
5182 r
= expose_port_watch_rtnl(event
, fd_inner_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5186 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5187 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5190 if (arg_console_mode
!= CONSOLE_PIPE
) {
5191 _cleanup_close_
int fd
= -EBADF
;
5192 PTYForwardFlags flags
= 0;
5194 /* Retrieve the master pty allocated by inner child */
5195 fd
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5197 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5199 switch (arg_console_mode
) {
5201 case CONSOLE_READ_ONLY
:
5202 flags
|= PTY_FORWARD_READ_ONLY
;
5206 case CONSOLE_INTERACTIVE
:
5207 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5209 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5211 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5213 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5214 (void) pty_forward_set_width_height(forward
,
5216 arg_console_height
);
5220 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5223 *master
= TAKE_FD(fd
);
5226 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
5228 r
= sd_event_loop(event
);
5230 return log_error_errno(r
, "Failed to run event loop: %m");
5235 (void) pty_forward_get_last_char(forward
, &last_char
);
5236 forward
= pty_forward_free(forward
);
5238 if (!arg_quiet
&& last_char
!= '\n')
5242 /* Kill if it is not dead yet anyway */
5243 if (!arg_register
&& !arg_keep_unit
&& bus
)
5244 terminate_scope(bus
, arg_machine
);
5246 /* Normally redundant, but better safe than sorry */
5247 (void) kill(*pid
, SIGKILL
);
5249 fd_kmsg_fifo
= safe_close(fd_kmsg_fifo
);
5251 if (arg_private_network
) {
5252 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5253 * to avoid having to move the parent to the child network namespace. */
5254 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
5259 _cleanup_close_
int parent_netns_fd
= -EBADF
;
5261 r
= namespace_open(getpid_cached(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5263 log_error_errno(r
, "Failed to open parent network namespace: %m");
5264 _exit(EXIT_FAILURE
);
5267 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5269 log_error_errno(r
, "Failed to enter child network namespace: %m");
5270 _exit(EXIT_FAILURE
);
5273 /* Reverse network interfaces pair list so that interfaces get their initial name back.
5274 * This is about ensuring interfaces get their old name back when being moved back. */
5275 arg_network_interfaces
= strv_reverse(arg_network_interfaces
);
5277 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5279 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5281 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5285 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5287 /* Tell machined that we are gone. */
5289 (void) unregister_machine(bus
, arg_machine
);
5292 /* We failed to wait for the container, or the container exited abnormally. */
5294 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5295 /* r > 0 → The container exited with a non-zero status.
5296 * As a special case, we need to replace 133 with a different value,
5297 * because 133 is special-cased in the service file to reboot the container.
5298 * otherwise → The container exited with zero status and a reboot was not requested.
5300 if (r
== EXIT_FORCE_RESTART
)
5301 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5303 return 0; /* finito */
5306 /* CONTAINER_REBOOTED, loop again */
5308 if (arg_keep_unit
) {
5309 /* Special handling if we are running as a service: instead of simply
5310 * restarting the machine we want to restart the entire service, so let's
5311 * inform systemd about this with the special exit code 133. The service
5312 * file uses RestartForceExitStatus=133 so that this results in a full
5313 * nspawn restart. This is necessary since we might have cgroup parameters
5314 * set we want to have flushed out. */
5315 *ret
= EXIT_FORCE_RESTART
;
5316 return 0; /* finito */
5319 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5320 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5322 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5323 *veth_created
= false;
5324 return 1; /* loop again */
5327 static int initialize_rlimits(void) {
5328 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5329 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5330 * container execution environments. */
5332 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5333 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5334 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5335 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5336 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5337 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5338 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5339 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5340 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5341 [RLIMIT_NICE
] = { 0, 0 },
5342 [RLIMIT_NOFILE
] = { 1024, 4096 },
5343 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5344 [RLIMIT_RTPRIO
] = { 0, 0 },
5345 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5346 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5348 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5349 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5350 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5351 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5352 * that PID 1 changes a number of other resource limits during early initialization which is why we
5353 * don't read the other limits from PID 1 but prefer the static table above. */
5358 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5359 /* Let's only fill in what the user hasn't explicitly configured anyway */
5360 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5361 const struct rlimit
*v
;
5362 struct rlimit buffer
;
5364 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5365 /* For these two let's read the limits off PID 1. See above for an explanation. */
5367 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
5368 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5371 } else if (rl
== RLIMIT_NOFILE
) {
5372 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5373 * userspace. Given that nspawn containers are often run without our PID 1,
5374 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5375 * so that container userspace gets similar resources as host userspace
5377 buffer
= kernel_defaults
[rl
];
5378 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5381 v
= kernel_defaults
+ rl
;
5383 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5384 if (!arg_rlimit
[rl
])
5388 if (DEBUG_LOGGING
) {
5389 _cleanup_free_
char *k
= NULL
;
5391 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5392 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5399 static int cant_be_in_netns(void) {
5400 _cleanup_close_
int fd
= -EBADF
;
5404 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5405 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5408 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5411 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5413 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5415 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5417 if (r
== -ENOENT
|| ERRNO_IS_DISCONNECT(r
))
5418 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5419 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5421 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5424 r
= getpeercred(fd
, &ucred
);
5426 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5428 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5430 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5432 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5433 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5437 static int run(int argc
, char *argv
[]) {
5438 bool remove_directory
= false, remove_image
= false, veth_created
= false, remove_tmprootdir
= false;
5439 _cleanup_close_
int master
= -EBADF
;
5440 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5441 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5442 char veth_name
[IFNAMSIZ
] = "";
5443 struct ExposeArgs expose_args
= {};
5444 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5445 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5446 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5447 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5448 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5451 log_parse_environment();
5454 r
= parse_argv(argc
, argv
);
5458 if (geteuid() != 0) {
5459 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5460 argc
>= 2 ? "Need to be root." :
5461 "Need to be root (and some arguments are usually required).\nHint: try --help");
5465 r
= cant_be_in_netns();
5469 r
= initialize_rlimits();
5473 r
= load_oci_bundle();
5477 r
= determine_names();
5481 r
= load_settings();
5487 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5491 r
= verify_arguments();
5495 r
= verify_network_interfaces_initialized();
5499 /* Reapply environment settings. */
5500 (void) detect_unified_cgroup_hierarchy_from_environment();
5502 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5503 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5504 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5505 (void) ignore_signals(SIGPIPE
);
5507 n_fd_passed
= sd_listen_fds(false);
5508 if (n_fd_passed
> 0) {
5509 r
= fdset_new_listen_fds(&fds
, false);
5511 log_error_errno(r
, "Failed to collect file descriptors: %m");
5516 /* The "default" umask. This is appropriate for most file and directory
5517 * operations performed by nspawn, and is the umask that will be used for
5518 * the child. Functions like copy_devnodes() change the umask temporarily. */
5521 if (arg_directory
) {
5524 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5525 * /var from the host will propagate into container dynamically (because bad things happen if
5526 * two systems write to the same /var). Let's allow it for the special cases where /var is
5527 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5528 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5529 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5530 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5534 if (arg_ephemeral
) {
5535 _cleanup_free_
char *np
= NULL
;
5537 r
= chase_and_update(&arg_directory
, 0);
5541 /* If the specified path is a mount point we generate the new snapshot immediately
5542 * inside it under a random name. However if the specified is not a mount point we
5543 * create the new snapshot in the parent directory, just next to it. */
5544 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5546 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5550 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5552 r
= tempfn_random(arg_directory
, "machine.", &np
);
5554 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5558 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5559 * only owned by us and no one else. */
5560 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5562 log_error_errno(r
, "Failed to lock %s: %m", np
);
5567 BLOCK_SIGNALS(SIGINT
);
5568 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_directory
, AT_FDCWD
, np
,
5569 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5570 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5571 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5572 BTRFS_SNAPSHOT_RECURSIVE
|
5573 BTRFS_SNAPSHOT_QUOTA
|
5574 BTRFS_SNAPSHOT_SIGINT
);
5577 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5581 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5585 free_and_replace(arg_directory
, np
);
5586 remove_directory
= true;
5588 r
= chase_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5592 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5594 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5598 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5603 r
= chase_and_update(&arg_template
, 0);
5608 BLOCK_SIGNALS(SIGINT
);
5609 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_template
, AT_FDCWD
, arg_directory
,
5610 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5611 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5612 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5613 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5614 BTRFS_SNAPSHOT_RECURSIVE
|
5615 BTRFS_SNAPSHOT_QUOTA
|
5616 BTRFS_SNAPSHOT_SIGINT
);
5619 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5620 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5621 else if (r
== -EINTR
) {
5622 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5625 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5628 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5629 "Populated %s from template %s.", arg_directory
, arg_template
);
5633 if (arg_start_mode
== START_BOOT
) {
5634 _cleanup_free_
char *b
= NULL
;
5637 if (arg_pivot_root_new
) {
5638 b
= path_join(arg_directory
, arg_pivot_root_new
);
5646 if (path_is_os_tree(p
) <= 0) {
5647 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5648 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5652 _cleanup_free_
char *p
= NULL
;
5654 if (arg_pivot_root_new
)
5655 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5657 p
= path_join(arg_directory
, "/usr/");
5661 if (laccess(p
, F_OK
) < 0) {
5662 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5663 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5669 DissectImageFlags dissect_image_flags
=
5670 DISSECT_IMAGE_GENERIC_ROOT
|
5671 DISSECT_IMAGE_REQUIRE_ROOT
|
5672 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5673 DISSECT_IMAGE_USR_NO_ROOT
|
5674 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5675 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5677 assert(!arg_template
);
5679 r
= chase_and_update(&arg_image
, 0);
5683 if (arg_ephemeral
) {
5684 _cleanup_free_
char *np
= NULL
;
5686 r
= tempfn_random(arg_image
, "machine.", &np
);
5688 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5692 /* Always take an exclusive lock on our own ephemeral copy. */
5693 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5695 r
= log_error_errno(r
, "Failed to create image lock: %m");
5700 BLOCK_SIGNALS(SIGINT
);
5701 r
= copy_file_full(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600,
5702 FS_NOCOW_FL
, FS_NOCOW_FL
,
5703 COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
,
5707 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5711 r
= log_error_errno(r
, "Failed to copy image file: %m");
5715 free_and_replace(arg_image
, np
);
5716 remove_image
= true;
5718 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5720 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5724 r
= log_error_errno(r
, "Failed to create image lock: %m");
5728 r
= verity_settings_load(
5729 &arg_verity_settings
,
5730 arg_image
, NULL
, NULL
);
5732 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5736 if (arg_verity_settings
.data_path
)
5737 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5740 if (!mkdtemp(tmprootdir
)) {
5741 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5745 remove_tmprootdir
= true;
5747 arg_directory
= strdup(tmprootdir
);
5748 if (!arg_directory
) {
5753 r
= loop_device_make_by_path(
5755 arg_read_only
? O_RDONLY
: O_RDWR
,
5756 /* sector_size= */ UINT32_MAX
,
5757 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5761 log_error_errno(r
, "Failed to set up loopback block device: %m");
5765 r
= dissect_loop_device_and_warn(
5767 &arg_verity_settings
,
5768 /* mount_options=*/ NULL
,
5769 arg_image_policy
?: &image_policy_container
,
5770 dissect_image_flags
,
5773 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5774 log_notice("Note that the disk image needs to\n"
5775 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5776 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5777 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5778 " d) or contain a file system without a partition table\n"
5779 "in order to be bootable with systemd-nspawn.");
5785 r
= dissected_image_load_verity_sig_partition(
5788 &arg_verity_settings
);
5792 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5793 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5794 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5796 r
= dissected_image_decrypt_interactively(
5799 &arg_verity_settings
,
5804 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5805 if (remove_image
&& unlink(arg_image
) >= 0)
5806 remove_image
= false;
5808 if (arg_architecture
< 0)
5809 arg_architecture
= dissected_image_architecture(dissected_image
);
5812 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5816 if (arg_console_mode
< 0)
5818 isatty(STDIN_FILENO
) > 0 &&
5819 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5821 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5825 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5826 arg_machine
, arg_image
?: arg_directory
);
5828 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, SIGRTMIN
+18, -1) >= 0);
5830 r
= make_reaper_process(true);
5832 log_error_errno(r
, "Failed to become subreaper: %m");
5836 if (arg_expose_ports
) {
5837 r
= fw_ctx_new(&fw_ctx
);
5839 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5842 expose_args
.fw_ctx
= fw_ctx
;
5845 r
= run_container(dissected_image
,
5847 veth_name
, &veth_created
,
5848 &expose_args
, &master
,
5855 (void) sd_notify(false,
5856 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5857 "STOPPING=1\nSTATUS=Terminating...");
5860 (void) kill(pid
, SIGKILL
);
5862 /* Try to flush whatever is still queued in the pty */
5864 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5865 master
= safe_close(master
);
5869 (void) wait_for_terminate(pid
, NULL
);
5873 if (remove_directory
&& arg_directory
) {
5876 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5878 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5881 if (remove_image
&& arg_image
) {
5882 if (unlink(arg_image
) < 0)
5883 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5886 if (remove_tmprootdir
) {
5887 if (rmdir(tmprootdir
) < 0)
5888 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5894 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5895 (void) rm_rf(p
, REMOVE_ROOT
);
5898 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5899 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5902 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5903 (void) remove_bridge(arg_network_zone
);
5905 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5906 expose_port_free_all(arg_expose_ports
);
5907 rlimit_free_all(arg_rlimit
);
5908 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5909 credential_free_all(arg_credentials
, arg_n_credentials
);
5917 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);