1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include <linux/loop.h>
7 #include <selinux/selinux.h>
11 #include <sys/ioctl.h>
12 #include <sys/mount.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
15 #include <sys/types.h>
20 #include <linux/fs.h> /* Must be included after <sys/mount.h> */
23 #include "sd-daemon.h"
26 #include "alloc-util.h"
27 #include "ether-addr-util.h"
29 #include "base-filesystem.h"
30 #include "blkid-util.h"
31 #include "btrfs-util.h"
33 #include "bus-error.h"
34 #include "bus-locator.h"
37 #include "capability-util.h"
38 #include "cgroup-util.h"
40 #include "common-signal.h"
42 #include "cpu-set-util.h"
43 #include "creds-util.h"
44 #include "dev-setup.h"
45 #include "discover-image.h"
46 #include "dissect-image.h"
52 #include "format-util.h"
55 #include "hexdecoct.h"
56 #include "hostname-setup.h"
57 #include "hostname-util.h"
58 #include "id128-util.h"
61 #include "loop-util.h"
62 #include "loopback-setup.h"
63 #include "machine-credential.h"
65 #include "main-func.h"
66 #include "missing_sched.h"
68 #include "mount-util.h"
69 #include "mountpoint-util.h"
70 #include "namespace-util.h"
71 #include "netlink-util.h"
72 #include "nspawn-bind-user.h"
73 #include "nspawn-cgroup.h"
74 #include "nspawn-def.h"
75 #include "nspawn-expose-ports.h"
76 #include "nspawn-mount.h"
77 #include "nspawn-network.h"
78 #include "nspawn-oci.h"
79 #include "nspawn-patch-uid.h"
80 #include "nspawn-register.h"
81 #include "nspawn-seccomp.h"
82 #include "nspawn-settings.h"
83 #include "nspawn-setuid.h"
84 #include "nspawn-stub-pid1.h"
85 #include "nspawn-util.h"
87 #include "nulstr-util.h"
90 #include "parse-argument.h"
91 #include "parse-util.h"
92 #include "pretty-print.h"
93 #include "process-util.h"
95 #include "random-util.h"
96 #include "raw-clone.h"
97 #include "resolve-util.h"
98 #include "rlimit-util.h"
100 #include "seccomp-util.h"
101 #include "selinux-util.h"
102 #include "signal-util.h"
103 #include "socket-util.h"
104 #include "stat-util.h"
105 #include "stdio-util.h"
106 #include "string-table.h"
107 #include "string-util.h"
109 #include "sysctl-util.h"
110 #include "terminal-util.h"
111 #include "tmpfile-util.h"
112 #include "umask-util.h"
113 #include "unit-name.h"
114 #include "user-util.h"
117 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
119 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
121 #define EXIT_FORCE_RESTART 133
123 typedef enum ContainerStatus
{
124 CONTAINER_TERMINATED
,
128 static char *arg_directory
= NULL
;
129 static char *arg_template
= NULL
;
130 static char *arg_chdir
= NULL
;
131 static char *arg_pivot_root_new
= NULL
;
132 static char *arg_pivot_root_old
= NULL
;
133 static char *arg_user
= NULL
;
134 static uid_t arg_uid
= UID_INVALID
;
135 static gid_t arg_gid
= GID_INVALID
;
136 static gid_t
* arg_supplementary_gids
= NULL
;
137 static size_t arg_n_supplementary_gids
= 0;
138 static sd_id128_t arg_uuid
= {};
139 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
140 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
141 static const char *arg_selinux_context
= NULL
;
142 static const char *arg_selinux_apifs_context
= NULL
;
143 static char *arg_slice
= NULL
;
144 static bool arg_private_network
= false;
145 static bool arg_read_only
= false;
146 static StartMode arg_start_mode
= START_PID1
;
147 static bool arg_ephemeral
= false;
148 static LinkJournal arg_link_journal
= LINK_AUTO
;
149 static bool arg_link_journal_try
= false;
150 static uint64_t arg_caps_retain
=
151 (1ULL << CAP_AUDIT_CONTROL
) |
152 (1ULL << CAP_AUDIT_WRITE
) |
153 (1ULL << CAP_CHOWN
) |
154 (1ULL << CAP_DAC_OVERRIDE
) |
155 (1ULL << CAP_DAC_READ_SEARCH
) |
156 (1ULL << CAP_FOWNER
) |
157 (1ULL << CAP_FSETID
) |
158 (1ULL << CAP_IPC_OWNER
) |
160 (1ULL << CAP_LEASE
) |
161 (1ULL << CAP_LINUX_IMMUTABLE
) |
162 (1ULL << CAP_MKNOD
) |
163 (1ULL << CAP_NET_BIND_SERVICE
) |
164 (1ULL << CAP_NET_BROADCAST
) |
165 (1ULL << CAP_NET_RAW
) |
166 (1ULL << CAP_SETFCAP
) |
167 (1ULL << CAP_SETGID
) |
168 (1ULL << CAP_SETPCAP
) |
169 (1ULL << CAP_SETUID
) |
170 (1ULL << CAP_SYS_ADMIN
) |
171 (1ULL << CAP_SYS_BOOT
) |
172 (1ULL << CAP_SYS_CHROOT
) |
173 (1ULL << CAP_SYS_NICE
) |
174 (1ULL << CAP_SYS_PTRACE
) |
175 (1ULL << CAP_SYS_RESOURCE
) |
176 (1ULL << CAP_SYS_TTY_CONFIG
);
177 static uint64_t arg_caps_ambient
= 0;
178 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
179 static CustomMount
*arg_custom_mounts
= NULL
;
180 static size_t arg_n_custom_mounts
= 0;
181 static char **arg_setenv
= NULL
;
182 static bool arg_quiet
= false;
183 static bool arg_register
= true;
184 static bool arg_keep_unit
= false;
185 static char **arg_network_interfaces
= NULL
;
186 static char **arg_network_macvlan
= NULL
;
187 static char **arg_network_ipvlan
= NULL
;
188 static bool arg_network_veth
= false;
189 static char **arg_network_veth_extra
= NULL
;
190 static char *arg_network_bridge
= NULL
;
191 static char *arg_network_zone
= NULL
;
192 static char *arg_network_namespace_path
= NULL
;
193 struct ether_addr arg_network_provided_mac
= {};
194 static PagerFlags arg_pager_flags
= 0;
195 static unsigned long arg_personality
= PERSONALITY_INVALID
;
196 static char *arg_image
= NULL
;
197 static char *arg_oci_bundle
= NULL
;
198 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
199 static ExposePort
*arg_expose_ports
= NULL
;
200 static char **arg_property
= NULL
;
201 static sd_bus_message
*arg_property_message
= NULL
;
202 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
203 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
204 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
205 static int arg_kill_signal
= 0;
206 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
207 static SettingsMask arg_settings_mask
= 0;
208 static int arg_settings_trusted
= -1;
209 static char **arg_parameters
= NULL
;
210 static const char *arg_container_service_name
= "systemd-nspawn";
211 static bool arg_notify_ready
= false;
212 static bool arg_use_cgns
= true;
213 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
214 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
215 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
216 static char **arg_syscall_allow_list
= NULL
;
217 static char **arg_syscall_deny_list
= NULL
;
219 static scmp_filter_ctx arg_seccomp
= NULL
;
221 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
222 static bool arg_no_new_privileges
= false;
223 static int arg_oom_score_adjust
= 0;
224 static bool arg_oom_score_adjust_set
= false;
225 static CPUSet arg_cpu_set
= {};
226 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
227 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
228 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
229 static DeviceNode
* arg_extra_nodes
= NULL
;
230 static size_t arg_n_extra_nodes
= 0;
231 static char **arg_sysctl
= NULL
;
232 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
233 static MachineCredentialContext arg_credentials
= {};
234 static char **arg_bind_user
= NULL
;
235 static bool arg_suppress_sync
= false;
236 static char *arg_settings_filename
= NULL
;
237 static Architecture arg_architecture
= _ARCHITECTURE_INVALID
;
238 static ImagePolicy
*arg_image_policy
= NULL
;
240 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
260 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
262 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
264 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
265 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
267 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
269 STATIC_DESTRUCTOR_REGISTER(arg_credentials
, machine_credential_context_done
);
270 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
271 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
272 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
273 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
274 STATIC_DESTRUCTOR_REGISTER(arg_image_policy
, image_policy_freep
);
276 static int handle_arg_console(const char *arg
) {
277 if (streq(arg
, "help")) {
286 if (streq(arg
, "interactive"))
287 arg_console_mode
= CONSOLE_INTERACTIVE
;
288 else if (streq(arg
, "read-only"))
289 arg_console_mode
= CONSOLE_READ_ONLY
;
290 else if (streq(arg
, "passive"))
291 arg_console_mode
= CONSOLE_PASSIVE
;
292 else if (streq(arg
, "pipe")) {
293 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
294 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
295 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
296 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
297 "Proceeding anyway.");
299 arg_console_mode
= CONSOLE_PIPE
;
300 } else if (streq(arg
, "autopipe")) {
301 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
302 arg_console_mode
= CONSOLE_INTERACTIVE
;
304 arg_console_mode
= CONSOLE_PIPE
;
306 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
308 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
312 static int help(void) {
313 _cleanup_free_
char *link
= NULL
;
316 pager_open(arg_pager_flags
);
318 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
322 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
323 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
324 " -h --help Show this help\n"
325 " --version Print version string\n"
326 " -q --quiet Do not show status information\n"
327 " --no-pager Do not pipe output into a pager\n"
328 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
330 " -D --directory=PATH Root directory for the container\n"
331 " --template=PATH Initialize root directory from template directory,\n"
333 " -x --ephemeral Run container with snapshot of root directory, and\n"
334 " remove it after exit\n"
335 " -i --image=PATH Root file system disk image (or device node) for\n"
337 " --image-policy=POLICY Specify disk image dissection policy\n"
338 " --oci-bundle=PATH OCI bundle directory\n"
339 " --read-only Mount the root directory read-only\n"
340 " --volatile[=MODE] Run the system in volatile mode\n"
341 " --root-hash=HASH Specify verity root hash for root disk image\n"
342 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
343 " as a DER encoded PKCS7, either as a path to a file\n"
344 " or as an ASCII base64 encoded string prefixed by\n"
346 " --verity-data=PATH Specify hash device for verity\n"
347 " --pivot-root=PATH[:PATH]\n"
348 " Pivot root to given directory in the container\n\n"
349 "%3$sExecution:%4$s\n"
350 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
351 " -b --boot Boot up full system (i.e. invoke init)\n"
352 " --chdir=PATH Set working directory in the container\n"
353 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
354 " -u --user=USER Run the command under specified user or UID\n"
355 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
356 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
357 " --suppress-sync=BOOLEAN\n"
358 " Suppress any form of disk data synchronization\n\n"
359 "%3$sSystem Identity:%4$s\n"
360 " -M --machine=NAME Set the machine name for the container\n"
361 " --hostname=NAME Override the hostname for the container\n"
362 " --uuid=UUID Set a specific machine UUID for the container\n\n"
363 "%3$sProperties:%4$s\n"
364 " -S --slice=SLICE Place the container in the specified slice\n"
365 " --property=NAME=VALUE Set scope unit property\n"
366 " --register=BOOLEAN Register container as machine\n"
367 " --keep-unit Do not register a scope for the machine, reuse\n"
368 " the service unit nspawn is running in\n\n"
369 "%3$sUser Namespacing:%4$s\n"
370 " --private-users=no Run without user namespacing\n"
371 " --private-users=yes|pick|identity\n"
372 " Run within user namespace, autoselect UID/GID range\n"
373 " --private-users=UIDBASE[:NUIDS]\n"
374 " Similar, but with user configured UID/GID range\n"
375 " --private-users-ownership=MODE\n"
376 " Adjust ('chown') or map ('map') OS tree ownership\n"
377 " to private UID/GID range\n"
378 " -U Equivalent to --private-users=pick and\n"
379 " --private-users-ownership=auto\n\n"
380 "%3$sNetworking:%4$s\n"
381 " --private-network Disable network in container\n"
382 " --network-interface=HOSTIF[:CONTAINERIF]\n"
383 " Assign an existing network interface to the\n"
385 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
386 " Create a macvlan network interface based on an\n"
387 " existing network interface to the container\n"
388 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
389 " Create an ipvlan network interface based on an\n"
390 " existing network interface to the container\n"
391 " -n --network-veth Add a virtual Ethernet connection between host\n"
393 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
394 " Add an additional virtual Ethernet link between\n"
395 " host and container\n"
396 " --network-bridge=INTERFACE\n"
397 " Add a virtual Ethernet connection to the container\n"
398 " and attach it to an existing bridge on the host\n"
399 " --network-zone=NAME Similar, but attach the new interface to an\n"
400 " an automatically managed bridge interface\n"
401 " --network-namespace-path=PATH\n"
402 " Set network namespace to the one represented by\n"
403 " the specified kernel namespace file node\n"
404 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
405 " Expose a container IP port on the host\n\n"
406 "%3$sSecurity:%4$s\n"
407 " --capability=CAP In addition to the default, retain specified\n"
409 " --drop-capability=CAP Drop the specified capability from the default set\n"
410 " --ambient-capability=CAP\n"
411 " Sets the specified capability for the started\n"
412 " process. Not useful if booting a machine.\n"
413 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
414 " --system-call-filter=LIST|~LIST\n"
415 " Permit/prohibit specific system calls\n"
416 " -Z --selinux-context=SECLABEL\n"
417 " Set the SELinux security context to be used by\n"
418 " processes in the container\n"
419 " -L --selinux-apifs-context=SECLABEL\n"
420 " Set the SELinux security context to be used by\n"
421 " API/tmpfs file systems in the container\n\n"
422 "%3$sResources:%4$s\n"
423 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
424 " --oom-score-adjust=VALUE\n"
425 " Adjust the OOM score value for the payload\n"
426 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
427 " --personality=ARCH Pick personality for this container\n\n"
428 "%3$sIntegration:%4$s\n"
429 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
430 " --timezone=MODE Select mode of /etc/localtime initialization\n"
431 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
432 " host, try-guest, try-host\n"
433 " -j Equivalent to --link-journal=try-guest\n\n"
435 " --bind=PATH[:PATH[:OPTIONS]]\n"
436 " Bind mount a file or directory from the host into\n"
438 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
439 " Similar, but creates a read-only bind mount\n"
440 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
442 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
443 " --overlay=PATH[:PATH...]:PATH\n"
444 " Create an overlay mount from the host to \n"
446 " --overlay-ro=PATH[:PATH...]:PATH\n"
447 " Similar, but creates a read-only overlay mount\n"
448 " --bind-user=NAME Bind user from host to container\n\n"
449 "%3$sInput/Output:%4$s\n"
450 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
451 " set up for the container.\n"
452 " -P --pipe Equivalent to --console=pipe\n\n"
453 "%3$sCredentials:%4$s\n"
454 " --set-credential=ID:VALUE\n"
455 " Pass a credential with literal value to container.\n"
456 " --load-credential=ID:PATH\n"
457 " Load credential to pass to container from file or\n"
458 " AF_UNIX stream socket.\n"
459 "\nSee the %2$s for details.\n",
460 program_invocation_short_name
,
470 static int custom_mount_check_all(void) {
473 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
474 CustomMount
*m
= &arg_custom_mounts
[i
];
476 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
477 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
478 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
479 "--private-users-ownership=own may not be combined with custom root mounts.");
480 if (arg_uid_shift
== UID_INVALID
)
481 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
482 "--private-users with automatic UID shift may not be combined with custom root mounts.");
489 static int detect_unified_cgroup_hierarchy_from_environment(void) {
490 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
493 /* Allow the user to control whether the unified hierarchy is used */
497 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
498 var
= "UNIFIED_CGROUP_HIERARCHY";
503 r
= parse_boolean(e
);
505 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
507 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
509 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
515 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
518 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
519 * in the image actually supports. */
520 r
= cg_all_unified();
522 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
524 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
525 * routine only detects 231, so we'll have a false negative here for 230. */
526 r
= systemd_installation_has_version(directory
, "230");
528 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
530 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
532 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
533 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
534 /* Mixed cgroup hierarchy support was added in 233 */
535 r
= systemd_installation_has_version(directory
, "233");
537 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
539 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
541 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
543 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
545 log_debug("Using %s hierarchy for container.",
546 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
547 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
552 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
557 _cleanup_free_
char *t
= NULL
;
559 r
= extract_first_word(&spec
, &t
, ",", 0);
561 return log_error_errno(r
, "Failed to parse capability %s.", t
);
565 if (streq(t
, "help")) {
566 for (int i
= 0; i
< capability_list_length(); i
++) {
569 name
= capability_to_name(i
);
580 r
= capability_from_name(t
);
582 return log_error_errno(r
, "Failed to parse capability %s.", t
);
589 return 1; /* continue */
592 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
595 r
= getenv_bool(name
);
599 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
601 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
602 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
606 static int parse_mount_settings_env(void) {
610 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
611 if (r
< 0 && r
!= -ENXIO
)
612 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
614 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
616 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
617 if (streq_ptr(e
, "network"))
618 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
621 r
= parse_boolean(e
);
623 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
625 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
626 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
632 static int parse_environment(void) {
636 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
639 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
642 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
645 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
649 r
= parse_mount_settings_env();
653 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
654 * even if it is supported. If not supported, it has no effect. */
655 if (!cg_ns_supported())
656 arg_use_cgns
= false;
658 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
661 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
665 arg_use_cgns
= r
> 0;
666 arg_settings_mask
|= SETTING_USE_CGNS
;
670 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
672 arg_container_service_name
= e
;
674 e
= getenv("SYSTEMD_NSPAWN_NETWORK_MAC");
676 r
= parse_ether_addr(e
, &arg_network_provided_mac
);
678 return log_error_errno(r
, "Failed to parse provided MAC address via environment variable");
681 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
683 arg_suppress_sync
= r
;
684 else if (r
!= -ENXIO
)
685 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
687 return detect_unified_cgroup_hierarchy_from_environment();
690 static int parse_argv(int argc
, char *argv
[]) {
697 ARG_AMBIENT_CAPABILITY
,
709 ARG_NETWORK_INTERFACE
,
714 ARG_NETWORK_VETH_EXTRA
,
715 ARG_NETWORK_NAMESPACE_PATH
,
725 ARG_PRIVATE_USERS_CHOWN
,
726 ARG_PRIVATE_USERS_OWNERSHIP
,
731 ARG_SYSTEM_CALL_FILTER
,
734 ARG_NO_NEW_PRIVILEGES
,
735 ARG_OOM_SCORE_ADJUST
,
750 static const struct option options
[] = {
751 { "help", no_argument
, NULL
, 'h' },
752 { "version", no_argument
, NULL
, ARG_VERSION
},
753 { "directory", required_argument
, NULL
, 'D' },
754 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
755 { "ephemeral", no_argument
, NULL
, 'x' },
756 { "user", required_argument
, NULL
, 'u' },
757 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
758 { "as-pid2", no_argument
, NULL
, 'a' },
759 { "boot", no_argument
, NULL
, 'b' },
760 { "uuid", required_argument
, NULL
, ARG_UUID
},
761 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
762 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
763 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
764 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
765 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
766 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
767 { "bind", required_argument
, NULL
, ARG_BIND
},
768 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
769 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
770 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
771 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
772 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
773 { "machine", required_argument
, NULL
, 'M' },
774 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
775 { "slice", required_argument
, NULL
, 'S' },
776 { "setenv", required_argument
, NULL
, 'E' },
777 { "selinux-context", required_argument
, NULL
, 'Z' },
778 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
779 { "quiet", no_argument
, NULL
, 'q' },
780 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
781 { "register", required_argument
, NULL
, ARG_REGISTER
},
782 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
783 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
784 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
785 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
786 { "network-veth", no_argument
, NULL
, 'n' },
787 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
788 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
789 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
790 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
791 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
792 { "image", required_argument
, NULL
, 'i' },
793 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
794 { "port", required_argument
, NULL
, 'p' },
795 { "property", required_argument
, NULL
, ARG_PROPERTY
},
796 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
797 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
798 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
799 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
800 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
801 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
802 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
803 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
804 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
805 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
806 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
807 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
808 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
809 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
810 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
811 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
812 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
813 { "console", required_argument
, NULL
, ARG_CONSOLE
},
814 { "pipe", no_argument
, NULL
, ARG_PIPE
},
815 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
816 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
817 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
818 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
819 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
820 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
821 { "image-policy", required_argument
, NULL
, ARG_IMAGE_POLICY
},
826 uint64_t plus
= 0, minus
= 0;
827 bool mask_all_settings
= false, mask_no_settings
= false;
832 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
833 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
835 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
845 r
= parse_path_argument(optarg
, false, &arg_directory
);
849 arg_settings_mask
|= SETTING_DIRECTORY
;
853 r
= parse_path_argument(optarg
, false, &arg_template
);
857 arg_settings_mask
|= SETTING_DIRECTORY
;
861 r
= parse_path_argument(optarg
, false, &arg_image
);
865 arg_settings_mask
|= SETTING_DIRECTORY
;
869 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
876 arg_ephemeral
= true;
877 arg_settings_mask
|= SETTING_EPHEMERAL
;
881 r
= free_and_strdup(&arg_user
, optarg
);
885 arg_settings_mask
|= SETTING_USER
;
888 case ARG_NETWORK_ZONE
: {
889 _cleanup_free_
char *j
= NULL
;
891 j
= strjoin("vz-", optarg
);
895 if (!ifname_valid(j
))
896 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
897 "Network zone name not valid: %s", j
);
899 free_and_replace(arg_network_zone
, j
);
901 arg_network_veth
= true;
902 arg_private_network
= true;
903 arg_settings_mask
|= SETTING_NETWORK
;
907 case ARG_NETWORK_BRIDGE
:
909 if (!ifname_valid(optarg
))
910 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
911 "Bridge interface name not valid: %s", optarg
);
913 r
= free_and_strdup(&arg_network_bridge
, optarg
);
919 arg_network_veth
= true;
920 arg_private_network
= true;
921 arg_settings_mask
|= SETTING_NETWORK
;
924 case ARG_NETWORK_VETH_EXTRA
:
925 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
927 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
929 arg_private_network
= true;
930 arg_settings_mask
|= SETTING_NETWORK
;
933 case ARG_NETWORK_INTERFACE
:
934 r
= interface_pair_parse(&arg_network_interfaces
, optarg
);
938 arg_private_network
= true;
939 arg_settings_mask
|= SETTING_NETWORK
;
942 case ARG_NETWORK_MACVLAN
:
943 r
= macvlan_pair_parse(&arg_network_macvlan
, optarg
);
947 arg_private_network
= true;
948 arg_settings_mask
|= SETTING_NETWORK
;
951 case ARG_NETWORK_IPVLAN
:
952 r
= ipvlan_pair_parse(&arg_network_ipvlan
, optarg
);
957 case ARG_PRIVATE_NETWORK
:
958 arg_private_network
= true;
959 arg_settings_mask
|= SETTING_NETWORK
;
962 case ARG_NETWORK_NAMESPACE_PATH
:
963 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
967 arg_settings_mask
|= SETTING_NETWORK
;
971 if (arg_start_mode
== START_PID2
)
972 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
973 "--boot and --as-pid2 may not be combined.");
975 arg_start_mode
= START_BOOT
;
976 arg_settings_mask
|= SETTING_START_MODE
;
980 if (arg_start_mode
== START_BOOT
)
981 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
982 "--boot and --as-pid2 may not be combined.");
984 arg_start_mode
= START_PID2
;
985 arg_settings_mask
|= SETTING_START_MODE
;
989 r
= id128_from_string_nonzero(optarg
, &arg_uuid
);
991 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
992 "Machine UUID may not be all zeroes.");
994 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
996 arg_settings_mask
|= SETTING_MACHINE_ID
;
1000 _cleanup_free_
char *mangled
= NULL
;
1002 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1006 free_and_replace(arg_slice
, mangled
);
1007 arg_settings_mask
|= SETTING_SLICE
;
1012 if (isempty(optarg
))
1013 arg_machine
= mfree(arg_machine
);
1015 if (!hostname_is_valid(optarg
, 0))
1016 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1017 "Invalid machine name: %s", optarg
);
1019 r
= free_and_strdup(&arg_machine
, optarg
);
1026 if (isempty(optarg
))
1027 arg_hostname
= mfree(arg_hostname
);
1029 if (!hostname_is_valid(optarg
, 0))
1030 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1031 "Invalid hostname: %s", optarg
);
1033 r
= free_and_strdup(&arg_hostname
, optarg
);
1038 arg_settings_mask
|= SETTING_HOSTNAME
;
1042 arg_selinux_context
= optarg
;
1046 arg_selinux_apifs_context
= optarg
;
1050 arg_read_only
= true;
1051 arg_settings_mask
|= SETTING_READ_ONLY
;
1054 case ARG_AMBIENT_CAPABILITY
: {
1056 r
= parse_capability_spec(optarg
, &m
);
1059 arg_caps_ambient
|= m
;
1060 arg_settings_mask
|= SETTING_CAPABILITY
;
1063 case ARG_CAPABILITY
:
1064 case ARG_DROP_CAPABILITY
: {
1066 r
= parse_capability_spec(optarg
, &m
);
1070 if (c
== ARG_CAPABILITY
)
1074 arg_settings_mask
|= SETTING_CAPABILITY
;
1077 case ARG_NO_NEW_PRIVILEGES
:
1078 r
= parse_boolean(optarg
);
1080 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1082 arg_no_new_privileges
= r
;
1083 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1087 arg_link_journal
= LINK_GUEST
;
1088 arg_link_journal_try
= true;
1089 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1092 case ARG_LINK_JOURNAL
:
1093 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1095 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1097 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1102 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1104 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1106 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1110 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1112 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1114 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1118 case ARG_OVERLAY_RO
:
1119 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1120 if (r
== -EADDRNOTAVAIL
)
1121 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1123 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1125 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1128 case ARG_INACCESSIBLE
:
1129 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1131 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1133 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1137 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1139 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1141 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1148 case ARG_SHARE_SYSTEM
:
1149 /* We don't officially support this anymore, except for compat reasons. People should use the
1150 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1151 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1152 arg_clone_ns_flags
= 0;
1156 r
= parse_boolean(optarg
);
1158 log_error("Failed to parse --register= argument: %s", optarg
);
1166 arg_keep_unit
= true;
1169 case ARG_PERSONALITY
:
1171 arg_personality
= personality_from_string(optarg
);
1172 if (arg_personality
== PERSONALITY_INVALID
)
1173 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1174 "Unknown or unsupported personality '%s'.", optarg
);
1176 arg_settings_mask
|= SETTING_PERSONALITY
;
1182 arg_volatile_mode
= VOLATILE_YES
;
1183 else if (streq(optarg
, "help")) {
1184 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1189 m
= volatile_mode_from_string(optarg
);
1191 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1192 "Failed to parse --volatile= argument: %s", optarg
);
1194 arg_volatile_mode
= m
;
1197 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1201 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1203 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1205 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1207 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1211 if (strv_extend(&arg_property
, optarg
) < 0)
1216 case ARG_PRIVATE_USERS
: {
1221 else if (!in_charset(optarg
, DIGITS
))
1222 /* do *not* parse numbers as booleans */
1223 boolean
= parse_boolean(optarg
);
1228 /* no: User namespacing off */
1229 arg_userns_mode
= USER_NAMESPACE_NO
;
1230 arg_uid_shift
= UID_INVALID
;
1231 arg_uid_range
= UINT32_C(0x10000);
1232 } else if (boolean
> 0) {
1233 /* yes: User namespacing on, UID range is read from root dir */
1234 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1235 arg_uid_shift
= UID_INVALID
;
1236 arg_uid_range
= UINT32_C(0x10000);
1237 } else if (streq(optarg
, "pick")) {
1238 /* pick: User namespacing on, UID range is picked randomly */
1239 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1240 * implied by USER_NAMESPACE_PICK
1242 arg_uid_shift
= UID_INVALID
;
1243 arg_uid_range
= UINT32_C(0x10000);
1245 } else if (streq(optarg
, "identity")) {
1246 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1247 * itself, i.e. we don't actually map anything, but do take benefit of
1248 * isolation of capability sets. */
1249 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1251 arg_uid_range
= UINT32_C(0x10000);
1253 _cleanup_free_
char *buffer
= NULL
;
1254 const char *range
, *shift
;
1256 /* anything else: User namespacing on, UID range is explicitly configured */
1258 range
= strchr(optarg
, ':');
1260 buffer
= strndup(optarg
, range
- optarg
);
1266 r
= safe_atou32(range
, &arg_uid_range
);
1268 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1272 r
= parse_uid(shift
, &arg_uid_shift
);
1274 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1276 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1278 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1279 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1282 arg_settings_mask
|= SETTING_USERNS
;
1287 if (userns_supported()) {
1288 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1289 * implied by USER_NAMESPACE_PICK
1291 arg_uid_shift
= UID_INVALID
;
1292 arg_uid_range
= UINT32_C(0x10000);
1294 arg_settings_mask
|= SETTING_USERNS
;
1299 case ARG_PRIVATE_USERS_CHOWN
:
1300 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1302 arg_settings_mask
|= SETTING_USERNS
;
1305 case ARG_PRIVATE_USERS_OWNERSHIP
:
1306 if (streq(optarg
, "help")) {
1307 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1311 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1312 if (arg_userns_ownership
< 0)
1313 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1315 arg_settings_mask
|= SETTING_USERNS
;
1318 case ARG_KILL_SIGNAL
:
1319 if (streq(optarg
, "help")) {
1320 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1324 arg_kill_signal
= signal_from_string(optarg
);
1325 if (arg_kill_signal
< 0)
1326 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1328 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1333 /* no → do not read files
1334 * yes → read files, do not override cmdline, trust only subset
1335 * override → read files, override cmdline, trust only subset
1336 * trusted → read files, do not override cmdline, trust all
1339 r
= parse_boolean(optarg
);
1341 if (streq(optarg
, "trusted")) {
1342 mask_all_settings
= false;
1343 mask_no_settings
= false;
1344 arg_settings_trusted
= true;
1346 } else if (streq(optarg
, "override")) {
1347 mask_all_settings
= false;
1348 mask_no_settings
= true;
1349 arg_settings_trusted
= -1;
1351 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1354 mask_all_settings
= false;
1355 mask_no_settings
= false;
1356 arg_settings_trusted
= -1;
1359 mask_all_settings
= true;
1360 mask_no_settings
= false;
1361 arg_settings_trusted
= false;
1367 if (!path_is_absolute(optarg
))
1368 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1369 "Working directory %s is not an absolute path.", optarg
);
1371 r
= free_and_strdup(&arg_chdir
, optarg
);
1375 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1378 case ARG_PIVOT_ROOT
:
1379 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1381 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1383 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1386 case ARG_NOTIFY_READY
:
1387 r
= parse_boolean(optarg
);
1389 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1390 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1391 arg_notify_ready
= r
;
1392 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1395 case ARG_ROOT_HASH
: {
1396 _cleanup_free_
void *k
= NULL
;
1399 r
= unhexmem(optarg
, &k
, &l
);
1401 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1402 if (l
< sizeof(sd_id128_t
))
1403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128-bit long: %s", optarg
);
1405 free_and_replace(arg_verity_settings
.root_hash
, k
);
1406 arg_verity_settings
.root_hash_size
= l
;
1410 case ARG_ROOT_HASH_SIG
: {
1415 if ((value
= startswith(optarg
, "base64:"))) {
1416 r
= unbase64mem(value
, &p
, &l
);
1418 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1421 r
= read_full_file(optarg
, (char**) &p
, &l
);
1423 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1426 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1427 arg_verity_settings
.root_hash_sig_size
= l
;
1431 case ARG_VERITY_DATA
:
1432 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1437 case ARG_SYSTEM_CALL_FILTER
: {
1441 negative
= optarg
[0] == '~';
1442 items
= negative
? optarg
+ 1 : optarg
;
1445 _cleanup_free_
char *word
= NULL
;
1447 r
= extract_first_word(&items
, &word
, NULL
, 0);
1453 return log_error_errno(r
, "Failed to parse system call filter: %m");
1456 r
= strv_extend(&arg_syscall_deny_list
, word
);
1458 r
= strv_extend(&arg_syscall_allow_list
, word
);
1463 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1469 _cleanup_free_
char *name
= NULL
;
1472 if (streq(optarg
, "help")) {
1473 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1477 eq
= strchr(optarg
, '=');
1479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1480 "--rlimit= expects an '=' assignment.");
1482 name
= strndup(optarg
, eq
- optarg
);
1486 rl
= rlimit_from_string_harder(name
);
1488 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1490 if (!arg_rlimit
[rl
]) {
1491 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1492 if (!arg_rlimit
[rl
])
1496 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1498 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1500 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1504 case ARG_OOM_SCORE_ADJUST
:
1505 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1507 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1509 arg_oom_score_adjust_set
= true;
1510 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1513 case ARG_CPU_AFFINITY
: {
1516 r
= parse_cpu_set(optarg
, &cpuset
);
1518 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1520 cpu_set_reset(&arg_cpu_set
);
1521 arg_cpu_set
= cpuset
;
1522 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1526 case ARG_RESOLV_CONF
:
1527 if (streq(optarg
, "help")) {
1528 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1532 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1533 if (arg_resolv_conf
< 0)
1534 return log_error_errno(arg_resolv_conf
,
1535 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1537 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1541 if (streq(optarg
, "help")) {
1542 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1546 arg_timezone
= timezone_mode_from_string(optarg
);
1547 if (arg_timezone
< 0)
1548 return log_error_errno(arg_timezone
,
1549 "Failed to parse /etc/localtime mode: %s", optarg
);
1551 arg_settings_mask
|= SETTING_TIMEZONE
;
1555 r
= handle_arg_console(optarg
);
1562 r
= handle_arg_console("pipe");
1568 arg_pager_flags
|= PAGER_DISABLE
;
1571 case ARG_SET_CREDENTIAL
:
1572 r
= machine_credential_set(&arg_credentials
, optarg
);
1576 arg_settings_mask
|= SETTING_CREDENTIALS
;
1579 case ARG_LOAD_CREDENTIAL
:
1580 r
= machine_credential_load(&arg_credentials
, optarg
);
1584 arg_settings_mask
|= SETTING_CREDENTIALS
;
1588 if (!valid_user_group_name(optarg
, 0))
1589 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1591 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1594 arg_settings_mask
|= SETTING_BIND_USER
;
1597 case ARG_SUPPRESS_SYNC
:
1598 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1602 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1605 case ARG_IMAGE_POLICY
:
1606 r
= parse_image_policy_argument(optarg
, &arg_image_policy
);
1615 assert_not_reached();
1618 if (argc
> optind
) {
1619 strv_free(arg_parameters
);
1620 arg_parameters
= strv_copy(argv
+ optind
);
1621 if (!arg_parameters
)
1624 arg_settings_mask
|= SETTING_START_MODE
;
1627 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1628 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1629 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1630 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1632 arg_directory
= TAKE_PTR(arg_template
);
1634 arg_caps_retain
|= plus
;
1635 arg_caps_retain
|= arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0;
1636 arg_caps_retain
&= ~minus
;
1638 /* Make sure to parse environment before we reset the settings mask below */
1639 r
= parse_environment();
1643 /* Load all settings from .nspawn files */
1644 if (mask_no_settings
)
1645 arg_settings_mask
= 0;
1647 /* Don't load any settings from .nspawn files */
1648 if (mask_all_settings
)
1649 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1654 static int verify_arguments(void) {
1657 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1658 /* If we are running the stub init in the container, we don't need to look at what the init
1659 * in the container supports, because we are not using it. Let's immediately pick the right
1660 * setting based on the host system configuration.
1662 * We only do this, if the user didn't use an environment variable to override the detection.
1665 r
= cg_all_unified();
1667 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1669 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1670 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1671 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1673 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1676 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1677 arg_mount_settings
|= MOUNT_USE_USERNS
;
1679 if (arg_private_network
)
1680 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1682 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1683 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1684 arg_register
= false;
1685 if (arg_start_mode
!= START_PID1
)
1686 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1689 if (arg_userns_ownership
< 0)
1690 arg_userns_ownership
=
1691 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1692 USER_NAMESPACE_OWNERSHIP_OFF
;
1694 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1695 arg_kill_signal
= SIGRTMIN
+3;
1697 if (arg_volatile_mode
!= VOLATILE_NO
) /* Make sure all file systems contained in the image are mounted read-only if we are in volatile mode */
1698 arg_read_only
= true;
1700 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1701 arg_read_only
= true;
1703 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1704 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1705 * The latter is not technically a user session, but we don't need to labour the point. */
1706 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1708 if (arg_directory
&& arg_image
)
1709 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1711 if (arg_template
&& arg_image
)
1712 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1714 if (arg_template
&& !(arg_directory
|| arg_machine
))
1715 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1717 if (arg_ephemeral
&& arg_template
)
1718 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1720 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1721 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1723 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1724 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1726 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1727 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1728 "--read-only and --private-users-ownership=chown may not be combined.");
1730 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1731 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1732 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1733 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1734 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1736 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1737 * we need to error out, to avoid conflicts between different network options. */
1738 if (arg_network_namespace_path
&&
1739 (arg_network_interfaces
|| arg_network_macvlan
||
1740 arg_network_ipvlan
|| arg_network_veth_extra
||
1741 arg_network_bridge
|| arg_network_zone
||
1743 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1745 if (arg_network_bridge
&& arg_network_zone
)
1746 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1747 "--network-bridge= and --network-zone= may not be combined.");
1749 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1750 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1752 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1753 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1755 if (arg_expose_ports
&& !arg_private_network
)
1756 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1758 if (arg_caps_ambient
) {
1759 if (arg_caps_ambient
== UINT64_MAX
)
1760 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1762 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1763 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1765 if (arg_start_mode
== START_BOOT
)
1766 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1769 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1770 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1772 /* Drop duplicate --bind-user= entries */
1773 strv_uniq(arg_bind_user
);
1775 r
= custom_mount_check_all();
1782 static int verify_network_interfaces_initialized(void) {
1784 r
= test_network_interfaces_initialized(arg_network_interfaces
);
1788 r
= test_network_interfaces_initialized(arg_network_macvlan
);
1792 r
= test_network_interfaces_initialized(arg_network_ipvlan
);
1799 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1802 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1805 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1808 if (uid
!= UID_INVALID
) {
1809 uid
+= arg_uid_shift
;
1811 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1815 if (gid
!= GID_INVALID
) {
1816 gid
+= (gid_t
) arg_uid_shift
;
1818 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1822 return RET_NERRNO(lchown(p
, uid
, gid
));
1825 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1829 q
= prefix_roota(root
, path
);
1830 r
= RET_NERRNO(mkdir(q
, mode
));
1836 return userns_lchown(q
, uid
, gid
);
1839 static const char *timezone_from_path(const char *path
) {
1840 return PATH_STARTSWITH_SET(
1842 "../usr/share/zoneinfo/",
1843 "/usr/share/zoneinfo/");
1846 static bool etc_writable(void) {
1847 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1850 static int setup_timezone(const char *dest
) {
1851 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1852 const char *where
, *check
;
1858 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1859 r
= readlink_malloc("/etc/localtime", &p
);
1860 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1861 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1862 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1863 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1865 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1866 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1870 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1873 } else if (arg_timezone
== TIMEZONE_AUTO
)
1874 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1880 if (m
== TIMEZONE_OFF
)
1883 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1885 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1889 where
= strjoina(etc
, "/localtime");
1893 case TIMEZONE_DELETE
:
1894 if (unlink(where
) < 0)
1895 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1899 case TIMEZONE_SYMLINK
: {
1900 _cleanup_free_
char *q
= NULL
;
1901 const char *z
, *what
;
1903 z
= timezone_from_path(p
);
1905 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1909 r
= readlink_malloc(where
, &q
);
1910 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1911 return 0; /* Already pointing to the right place? Then do nothing .. */
1913 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1914 r
= chase(check
, dest
, 0, NULL
, NULL
);
1916 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1918 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1919 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1920 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1924 what
= strjoina("../usr/share/zoneinfo/", z
);
1925 if (symlink(what
, where
) < 0) {
1926 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1927 errno
, "Failed to correct timezone of container, ignoring: %m");
1937 case TIMEZONE_BIND
: {
1938 _cleanup_free_
char *resolved
= NULL
;
1941 found
= chase(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1943 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1947 if (found
== 0) /* missing? */
1948 (void) touch(resolved
);
1950 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1952 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1958 /* If mounting failed, try to copy */
1959 r
= copy_file_atomic("/etc/localtime", where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
1961 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1962 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1969 assert_not_reached();
1972 /* Fix permissions of the symlink or file copy we just created */
1973 r
= userns_lchown(where
, 0, 0);
1975 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1980 static int have_resolv_conf(const char *path
) {
1983 if (access(path
, F_OK
) < 0) {
1984 if (errno
== ENOENT
)
1987 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1993 static int resolved_listening(void) {
1994 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1995 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1996 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1999 /* Check if resolved is listening */
2001 r
= sd_bus_open_system(&bus
);
2003 return log_debug_errno(r
, "Failed to open system bus: %m");
2005 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2007 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2011 r
= bus_get_property_string(bus
, bus_resolve_mgr
, "DNSStubListener", &error
, &dns_stub_listener_mode
);
2013 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2015 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2018 static int setup_resolv_conf(const char *dest
) {
2019 _cleanup_free_
char *etc
= NULL
;
2020 const char *where
, *what
;
2026 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2027 if (arg_private_network
)
2028 m
= RESOLV_CONF_OFF
;
2029 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2030 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2031 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2032 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2034 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2037 m
= arg_resolv_conf
;
2039 if (m
== RESOLV_CONF_OFF
)
2042 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2044 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2048 where
= strjoina(etc
, "/resolv.conf");
2050 if (m
== RESOLV_CONF_DELETE
) {
2051 if (unlink(where
) < 0)
2052 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2057 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2058 what
= PRIVATE_STATIC_RESOLV_CONF
;
2059 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2060 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2061 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2062 what
= PRIVATE_STUB_RESOLV_CONF
;
2064 what
= "/etc/resolv.conf";
2066 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2067 _cleanup_free_
char *resolved
= NULL
;
2070 found
= chase(where
, dest
, CHASE_NONEXISTENT
|CHASE_NOFOLLOW
, &resolved
, NULL
);
2072 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2076 if (found
== 0) /* missing? */
2077 (void) touch(resolved
);
2079 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2081 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2083 /* If that didn't work, let's copy the file */
2086 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2087 r
= copy_file_atomic(what
, where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2089 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, COPY_REFLINK
);
2091 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2092 * resolved or something similar runs inside and the symlink points there.
2094 * If the disk image is read-only, there's also no point in complaining.
2096 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2097 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2098 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2102 r
= userns_lchown(where
, 0, 0);
2104 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2109 static int setup_boot_id(void) {
2110 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2111 _cleanup_free_
char *path
= NULL
;
2112 sd_id128_t rnd
= SD_ID128_NULL
;
2116 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2118 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2120 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2122 r
= sd_id128_randomize(&rnd
);
2124 return log_error_errno(r
, "Failed to generate random boot id: %m");
2126 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2128 return log_error_errno(r
, "Failed to write boot id: %m");
2130 from
= TAKE_PTR(path
);
2131 to
= "/proc/sys/kernel/random/boot_id";
2133 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2137 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2140 static int copy_devnodes(const char *dest
) {
2141 static const char devnodes
[] =
2154 BLOCK_WITH_UMASK(0000);
2156 /* Create /dev/net, so that we can create /dev/net/tun in it */
2157 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2158 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2160 NULSTR_FOREACH(d
, devnodes
) {
2161 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2164 from
= path_join("/dev/", d
);
2168 to
= path_join(dest
, from
);
2172 if (stat(from
, &st
) < 0) {
2174 if (errno
!= ENOENT
)
2175 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2177 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2178 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2179 "%s is not a char or block device, cannot copy.", from
);
2181 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2183 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2184 /* Explicitly warn the user when /dev is already populated. */
2185 if (errno
== EEXIST
)
2186 log_notice("%s/dev/ is pre-mounted and pre-populated. If a pre-mounted /dev/ is provided it needs to be an unpopulated file system.", dest
);
2188 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2190 /* Some systems abusively restrict mknod but allow bind mounts. */
2193 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2194 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2196 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2199 r
= userns_lchown(to
, 0, 0);
2201 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2203 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2207 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2209 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2211 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2214 prefixed
= path_join(dest
, sl
);
2218 t
= path_join("..", d
);
2222 if (symlink(t
, prefixed
) < 0)
2223 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2230 static int make_extra_nodes(const char *dest
) {
2234 BLOCK_WITH_UMASK(0000);
2236 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2237 _cleanup_free_
char *path
= NULL
;
2238 DeviceNode
*n
= arg_extra_nodes
+ i
;
2240 path
= path_join(dest
, n
->path
);
2244 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2245 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2247 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2249 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2255 static int setup_pts(const char *dest
) {
2256 _cleanup_free_
char *options
= NULL
;
2261 if (arg_selinux_apifs_context
)
2262 (void) asprintf(&options
,
2263 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2264 arg_uid_shift
+ TTY_GID
,
2265 arg_selinux_apifs_context
);
2268 (void) asprintf(&options
,
2269 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2270 arg_uid_shift
+ TTY_GID
);
2275 /* Mount /dev/pts itself */
2276 p
= prefix_roota(dest
, "/dev/pts");
2277 r
= RET_NERRNO(mkdir(p
, 0755));
2279 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2281 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2284 r
= userns_lchown(p
, 0, 0);
2286 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2288 /* Create /dev/ptmx symlink */
2289 p
= prefix_roota(dest
, "/dev/ptmx");
2290 if (symlink("pts/ptmx", p
) < 0)
2291 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2292 r
= userns_lchown(p
, 0, 0);
2294 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2296 /* And fix /dev/pts/ptmx ownership */
2297 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2298 r
= userns_lchown(p
, 0, 0);
2300 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2305 static int setup_stdio_as_dev_console(void) {
2306 _cleanup_close_
int terminal
= -EBADF
;
2309 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2310 * explicitly, if we are configured to. */
2311 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2313 return log_error_errno(terminal
, "Failed to open console: %m");
2315 /* Make sure we can continue logging to the original stderr, even if
2316 * stderr points elsewhere now */
2317 r
= log_dup_console();
2319 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2321 /* invalidates 'terminal' on success and failure */
2322 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2325 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2330 static int setup_dev_console(const char *console
) {
2331 _cleanup_free_
char *p
= NULL
;
2334 /* Create /dev/console symlink */
2335 r
= path_make_relative("/dev", console
, &p
);
2337 return log_error_errno(r
, "Failed to create relative path: %m");
2339 if (symlink(p
, "/dev/console") < 0)
2340 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2345 static int setup_keyring(void) {
2346 key_serial_t keyring
;
2348 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2349 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2350 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2351 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2352 * into the container. */
2354 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2355 if (keyring
== -1) {
2356 if (errno
== ENOSYS
)
2357 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2358 else if (ERRNO_IS_PRIVILEGE(errno
))
2359 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2361 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2367 int make_run_host(const char *root
) {
2372 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2374 return log_error_errno(r
, "Failed to create /run/host/: %m");
2379 static int setup_credentials(const char *root
) {
2383 if (arg_credentials
.n_credentials
== 0)
2386 r
= make_run_host(root
);
2390 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2392 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2394 q
= prefix_roota(root
, "/run/host/credentials");
2395 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2399 FOREACH_ARRAY(cred
, arg_credentials
.credentials
, arg_credentials
.n_credentials
) {
2400 _cleanup_free_
char *j
= NULL
;
2401 _cleanup_close_
int fd
= -EBADF
;
2403 j
= path_join(q
, cred
->id
);
2407 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2409 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2411 r
= loop_write(fd
, cred
->data
, cred
->size
);
2413 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2415 if (fchmod(fd
, 0400) < 0)
2416 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2418 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2419 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2420 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2424 if (chmod(q
, 0500) < 0)
2425 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2427 r
= userns_lchown(q
, 0, 0);
2431 /* Make both mount and superblock read-only now */
2432 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2436 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2439 static int setup_kmsg(int fd_inner_socket
) {
2440 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2441 _cleanup_free_
char *fifo
= NULL
;
2442 _cleanup_close_
int fd
= -EBADF
;
2445 assert(fd_inner_socket
>= 0);
2447 BLOCK_WITH_UMASK(0000);
2449 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2450 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2451 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2452 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2454 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2456 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2458 if (mkfifo(fifo
, 0600) < 0)
2459 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2461 from
= TAKE_PTR(fifo
);
2463 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2467 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2469 return log_error_errno(errno
, "Failed to open fifo: %m");
2471 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2472 r
= send_one_fd(fd_inner_socket
, fd
, 0);
2474 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2480 union in_addr_union address4
;
2481 union in_addr_union address6
;
2482 struct FirewallContext
*fw_ctx
;
2485 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2486 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2491 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2492 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2496 static int setup_hostname(void) {
2499 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2502 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2504 return log_error_errno(r
, "Failed to set hostname: %m");
2509 static int setup_journal(const char *directory
) {
2510 _cleanup_free_
char *d
= NULL
;
2516 /* Don't link journals in ephemeral mode */
2520 if (arg_link_journal
== LINK_NO
)
2523 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2525 r
= sd_id128_get_machine(&this_id
);
2527 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2529 if (sd_id128_equal(arg_uuid
, this_id
)) {
2530 log_full(try ? LOG_WARNING
: LOG_ERR
,
2531 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2537 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2538 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2540 bool ignore
= r
== -EROFS
&& try;
2541 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2542 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2543 return ignore
? 0 : r
;
2547 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2548 q
= prefix_roota(directory
, p
);
2550 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2554 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2555 "%s: already a mount point, refusing to use for journal", p
);
2558 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2562 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2563 "%s: already a mount point, refusing to use for journal", q
);
2566 r
= readlink_and_make_absolute(p
, &d
);
2568 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2571 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2573 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2578 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2579 } else if (r
== -EINVAL
) {
2581 if (arg_link_journal
== LINK_GUEST
&&
2584 if (errno
== ENOTDIR
) {
2585 log_error("%s already exists and is neither a symlink nor a directory", p
);
2588 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2590 } else if (r
!= -ENOENT
)
2591 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2593 if (arg_link_journal
== LINK_GUEST
) {
2595 if (symlink(q
, p
) < 0) {
2597 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2600 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2603 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2605 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2609 if (arg_link_journal
== LINK_HOST
) {
2610 /* don't create parents here — if the host doesn't have
2611 * permanent journal set up, don't force it here */
2613 r
= RET_NERRNO(mkdir(p
, 0755));
2614 if (r
< 0 && r
!= -EEXIST
) {
2616 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2619 return log_error_errno(r
, "Failed to create %s: %m", p
);
2622 } else if (access(p
, F_OK
) < 0)
2625 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2626 log_warning("%s is not empty, proceeding anyway.", q
);
2628 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2630 return log_error_errno(r
, "Failed to create %s: %m", q
);
2632 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2634 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2639 static int drop_capabilities(uid_t uid
) {
2640 CapabilityQuintet q
;
2642 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2643 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2644 * arg_caps_retain. */
2646 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2647 q
= arg_full_capabilities
;
2649 if (q
.bounding
== UINT64_MAX
)
2650 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2652 if (q
.effective
== UINT64_MAX
)
2653 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2655 if (q
.inheritable
== UINT64_MAX
)
2656 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2658 if (q
.permitted
== UINT64_MAX
)
2659 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2661 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2662 q
.ambient
= arg_caps_ambient
;
2664 if (capability_quintet_mangle(&q
))
2665 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2668 q
= (CapabilityQuintet
) {
2669 .bounding
= arg_caps_retain
,
2670 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2671 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2672 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2673 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2676 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2677 * in order to maintain the same behavior as systemd < 242. */
2678 if (capability_quintet_mangle(&q
))
2679 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2680 "Some capabilities will not be set because they are not in the current bounding set.");
2684 return capability_quintet_enforce(&q
);
2687 static int reset_audit_loginuid(void) {
2688 _cleanup_free_
char *p
= NULL
;
2691 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2694 r
= read_one_line_file("/proc/self/loginuid", &p
);
2698 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2700 /* Already reset? */
2701 if (streq(p
, "4294967295"))
2704 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2707 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2708 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2709 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2710 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2711 "using systemd-nspawn. Sleeping for 5s... (%m)");
2719 static int mount_tunnel_dig(const char *root
) {
2723 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2724 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2725 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2726 (void) mkdir_p(p
, 0600);
2728 r
= make_run_host(root
);
2732 r
= userns_mkdir(root
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2734 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2736 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
2737 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2741 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2748 static int mount_tunnel_open(void) {
2751 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2758 static int setup_machine_id(const char *directory
) {
2761 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2762 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2763 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2764 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2765 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2766 * container behaves nicely). */
2768 r
= id128_get_machine(directory
, &arg_uuid
);
2769 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) {
2770 /* If the file is missing, empty, or uninitialized, we don't mind */
2771 if (sd_id128_is_null(arg_uuid
)) {
2772 r
= sd_id128_randomize(&arg_uuid
);
2774 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2777 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2782 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2787 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2790 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2791 if (r
== -EOPNOTSUPP
)
2792 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.");
2794 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2796 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2798 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2800 log_debug("Patched directory tree to match UID/GID range.");
2807 * < 0 : wait_for_terminate() failed to get the state of the
2808 * container, the container was terminated by a signal, or
2809 * failed for an unknown reason. No change is made to the
2810 * container argument.
2811 * > 0 : The program executed in the container terminated with an
2812 * error. The exit code of the program executed in the
2813 * container is returned. The container argument has been set
2814 * to CONTAINER_TERMINATED.
2815 * 0 : The container is being rebooted, has been shut down or exited
2816 * successfully. The container argument has been set to either
2817 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2819 * That is, success is indicated by a return value of zero, and an
2820 * error is indicated by a non-zero value.
2822 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2826 r
= wait_for_terminate(pid
, &status
);
2828 return log_warning_errno(r
, "Failed to wait for container: %m");
2830 switch (status
.si_code
) {
2833 if (status
.si_status
== 0)
2834 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2836 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2838 *container
= CONTAINER_TERMINATED
;
2839 return status
.si_status
;
2842 if (status
.si_status
== SIGINT
) {
2843 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2844 *container
= CONTAINER_TERMINATED
;
2847 } else if (status
.si_status
== SIGHUP
) {
2848 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2849 *container
= CONTAINER_REBOOTED
;
2855 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2856 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2859 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2860 "Container %s failed due to unknown reason.", arg_machine
);
2864 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2867 pid
= PTR_TO_PID(userdata
);
2869 if (kill(pid
, arg_kill_signal
) >= 0) {
2870 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2871 sd_event_source_set_userdata(s
, NULL
);
2876 sd_event_exit(sd_event_source_get_event(s
), 0);
2880 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2886 pid
= PTR_TO_PID(userdata
);
2891 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2892 return log_error_errno(errno
, "Failed to waitid(): %m");
2893 if (si
.si_pid
== 0) /* No pending children. */
2895 if (si
.si_pid
== pid
) {
2896 /* The main process we care for has exited. Return from
2897 * signal handler but leave the zombie. */
2898 sd_event_exit(sd_event_source_get_event(s
), 0);
2902 /* Reap all other children. */
2903 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2909 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2914 pid
= PTR_TO_PID(userdata
);
2916 if (arg_kill_signal
> 0) {
2917 log_info("Container termination requested. Attempting to halt container.");
2918 (void) kill(pid
, arg_kill_signal
);
2920 log_info("Container termination requested. Exiting.");
2921 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2927 static int pick_paths(void) {
2930 if (arg_directory
) {
2931 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2932 PickFilter filter
= pick_filter_image_dir
;
2934 filter
.architecture
= arg_architecture
;
2936 r
= path_pick_update_warn(
2939 PICK_ARCHITECTURE
|PICK_TRIES
,
2942 /* Accept ENOENT here so that the --template= logic can work */
2946 arg_architecture
= result
.architecture
;
2950 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2951 PickFilter filter
= pick_filter_image_raw
;
2953 filter
.architecture
= arg_architecture
;
2955 r
= path_pick_update_warn(
2958 PICK_ARCHITECTURE
|PICK_TRIES
,
2963 arg_architecture
= result
.architecture
;
2967 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2968 PickFilter filter
= pick_filter_image_dir
;
2970 filter
.architecture
= arg_architecture
;
2972 r
= path_pick_update_warn(
2980 arg_architecture
= result
.architecture
;
2986 static int determine_names(void) {
2989 if (arg_template
&& !arg_directory
&& arg_machine
) {
2991 /* If --template= was specified then we should not search for a machine, but instead create a
2992 * new one in /var/lib/machine. */
2994 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2999 if (!arg_image
&& !arg_directory
) {
3001 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3003 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3005 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3007 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3009 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3010 r
= free_and_strdup(&arg_image
, i
->path
);
3012 r
= free_and_strdup(&arg_directory
, i
->path
);
3017 arg_read_only
= arg_read_only
|| i
->read_only
;
3019 r
= safe_getcwd(&arg_directory
);
3021 return log_error_errno(r
, "Failed to determine current directory: %m");
3024 if (!arg_directory
&& !arg_image
)
3025 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3029 if (arg_directory
&& path_equal(arg_directory
, "/"))
3030 arg_machine
= gethostname_malloc();
3031 else if (arg_image
) {
3034 r
= path_extract_filename(arg_image
, &arg_machine
);
3036 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_image
);
3038 /* Truncate suffix if there is one */
3039 e
= endswith(arg_machine
, ".raw");
3043 r
= path_extract_filename(arg_directory
, &arg_machine
);
3045 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_directory
);
3048 hostname_cleanup(arg_machine
);
3049 if (!hostname_is_valid(arg_machine
, 0))
3050 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3052 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3053 * to match fixed config file names. */
3054 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3055 if (!arg_settings_filename
)
3058 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3059 * instances at once without manually having to specify -M each time. */
3061 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3064 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3065 if (!arg_settings_filename
)
3072 static int chase_and_update(char **p
, unsigned flags
) {
3081 r
= chase(*p
, NULL
, flags
, &chased
, NULL
);
3083 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3085 return free_and_replace(*p
, chased
);
3088 static int determine_uid_shift(const char *directory
) {
3090 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3095 if (arg_uid_shift
== UID_INVALID
) {
3098 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3100 if (stat(directory
, &st
) < 0)
3101 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3103 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3105 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3106 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3107 "UID and GID base of %s don't match.", directory
);
3109 arg_uid_range
= UINT32_C(0x10000);
3111 if (arg_uid_shift
!= 0) {
3112 /* If the image is shifted already, then we'll fall back to classic chowning, for
3113 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3115 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3116 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3117 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3118 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3119 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3120 "UID base of %s is not zero, UID mapping not supported.", directory
);
3124 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3125 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3130 static unsigned long effective_clone_ns_flags(void) {
3131 unsigned long flags
= arg_clone_ns_flags
;
3133 if (arg_private_network
)
3134 flags
|= CLONE_NEWNET
;
3136 flags
|= CLONE_NEWCGROUP
;
3137 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3138 flags
|= CLONE_NEWUSER
;
3143 static int patch_sysctl(void) {
3145 /* This table is inspired by runc's sysctl() function */
3146 static const struct {
3149 unsigned long clone_flags
;
3151 { "kernel.hostname", false, CLONE_NEWUTS
},
3152 { "kernel.domainname", false, CLONE_NEWUTS
},
3153 { "kernel.msgmax", false, CLONE_NEWIPC
},
3154 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3155 { "kernel.msgmni", false, CLONE_NEWIPC
},
3156 { "kernel.sem", false, CLONE_NEWIPC
},
3157 { "kernel.shmall", false, CLONE_NEWIPC
},
3158 { "kernel.shmmax", false, CLONE_NEWIPC
},
3159 { "kernel.shmmni", false, CLONE_NEWIPC
},
3160 { "fs.mqueue.", true, CLONE_NEWIPC
},
3161 { "net.", true, CLONE_NEWNET
},
3164 unsigned long flags
;
3167 flags
= effective_clone_ns_flags();
3169 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3173 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3175 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3178 if (safe_sysctl
[i
].prefix
)
3179 good
= startswith(*k
, safe_sysctl
[i
].key
);
3181 good
= streq(*k
, safe_sysctl
[i
].key
);
3188 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3190 r
= sysctl_write(*k
, *v
);
3192 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3198 static int inner_child(
3200 int fd_inner_socket
,
3202 char **os_release_pairs
) {
3204 _cleanup_free_
char *home
= NULL
;
3207 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3208 NULL
, /* container */
3213 NULL
, /* container_uuid */
3214 NULL
, /* LISTEN_FDS */
3215 NULL
, /* LISTEN_PID */
3216 NULL
, /* NOTIFY_SOCKET */
3217 NULL
, /* CREDENTIALS_DIRECTORY */
3221 const char *exec_target
;
3222 _cleanup_strv_free_
char **env_use
= NULL
;
3223 int r
, which_failed
;
3225 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3226 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3227 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3228 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3229 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3230 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3233 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3234 * unshare(). See below. */
3237 assert(fd_inner_socket
>= 0);
3239 log_debug("Inner child is initializing.");
3241 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3242 /* Tell the parent, that it now can write the UID map. */
3243 (void) barrier_place(barrier
); /* #1 */
3245 /* Wait until the parent wrote the UID map */
3246 if (!barrier_place_and_sync(barrier
)) /* #2 */
3247 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3249 /* Become the new root user inside our namespace */
3250 r
= reset_uid_gid();
3252 return log_error_errno(r
, "Couldn't become new root: %m");
3254 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3255 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3256 * propagation, but simply create new peer groups for all our mounts). */
3257 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3263 arg_mount_settings
| MOUNT_IN_USERNS
,
3265 arg_selinux_apifs_context
);
3269 if (!arg_network_namespace_path
&& arg_private_network
) {
3270 r
= unshare(CLONE_NEWNET
);
3272 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3274 /* Tell the parent that it can setup network interfaces. */
3275 (void) barrier_place(barrier
); /* #3 */
3278 r
= mount_sysfs(NULL
, arg_mount_settings
);
3282 /* Wait until we are cgroup-ified, so that we
3283 * can mount the right cgroup path writable */
3284 if (!barrier_place_and_sync(barrier
)) /* #4 */
3285 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3286 "Parent died too early");
3289 r
= unshare(CLONE_NEWCGROUP
);
3291 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3294 arg_unified_cgroup_hierarchy
,
3295 arg_userns_mode
!= USER_NAMESPACE_NO
,
3298 arg_selinux_apifs_context
,
3301 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3305 r
= setup_boot_id();
3309 r
= setup_kmsg(fd_inner_socket
);
3316 arg_n_custom_mounts
,
3319 arg_selinux_apifs_context
,
3320 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3325 return log_error_errno(errno
, "setsid() failed: %m");
3327 if (arg_private_network
)
3328 (void) loopback_setup();
3330 if (arg_expose_ports
) {
3331 r
= expose_port_send_rtnl(fd_inner_socket
);
3336 if (arg_console_mode
!= CONSOLE_PIPE
) {
3337 _cleanup_close_
int master
= -EBADF
;
3338 _cleanup_free_
char *console
= NULL
;
3340 /* Allocate a pty and make it available as /dev/console. */
3341 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3343 return log_error_errno(master
, "Failed to allocate a pty: %m");
3345 r
= setup_dev_console(console
);
3347 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3349 r
= send_one_fd(fd_inner_socket
, master
, 0);
3351 return log_error_errno(r
, "Failed to send master fd: %m");
3353 r
= setup_stdio_as_dev_console();
3362 if (arg_oom_score_adjust_set
) {
3363 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3365 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3368 if (arg_cpu_set
.set
)
3369 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3370 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3372 (void) setup_hostname();
3374 if (arg_personality
!= PERSONALITY_INVALID
) {
3375 r
= safe_personality(arg_personality
);
3377 return log_error_errno(r
, "personality() failed: %m");
3378 #ifdef ARCHITECTURE_SECONDARY
3379 } else if (arg_architecture
== ARCHITECTURE_SECONDARY
) {
3380 r
= safe_personality(PER_LINUX32
);
3382 return log_error_errno(r
, "personality() failed: %m");
3384 } else if (!arg_quiet
&& arg_architecture
>= 0 && arg_architecture
!= native_architecture())
3385 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3386 "invocation with qemu userspace emulator (or equivalent) in effect.",
3387 architecture_to_string(arg_architecture
));
3389 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3391 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3396 if (is_seccomp_available()) {
3397 r
= seccomp_load(arg_seccomp
);
3398 if (ERRNO_IS_NEG_SECCOMP_FATAL(r
))
3399 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3401 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3406 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3411 if (arg_suppress_sync
) {
3413 r
= seccomp_suppress_sync();
3415 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3417 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3422 if (arg_selinux_context
)
3423 if (setexeccon(arg_selinux_context
) < 0)
3424 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3427 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3428 * if we need to later on. */
3429 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3430 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3432 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3433 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3435 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3439 r
= drop_capabilities(getuid());
3441 return log_error_errno(r
, "Dropping capabilities failed: %m");
3443 if (arg_no_new_privileges
)
3444 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3445 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3447 /* LXC sets container=lxc, so follow the scheme here */
3448 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3450 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3454 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3455 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3458 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3459 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3460 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
?: "root") < 0)
3463 assert(!sd_id128_is_null(arg_uuid
));
3465 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3468 if (!fdset_isempty(fds
)) {
3469 r
= fdset_cloexec(fds
, false);
3471 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3473 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3474 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3477 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3480 if (arg_credentials
.n_credentials
> 0) {
3481 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3487 if (arg_start_mode
!= START_BOOT
) {
3488 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3494 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3498 /* Let the parent know that we are ready and wait until the parent is ready with the setup, too... */
3499 if (!barrier_place_and_sync(barrier
)) /* #5 */
3500 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3503 if (chdir(arg_chdir
) < 0)
3504 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3506 if (arg_start_mode
== START_PID2
) {
3507 r
= stub_pid1(arg_uuid
);
3512 if (arg_console_mode
!= CONSOLE_PIPE
) {
3513 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3514 * are configured for that. Acquire it as controlling tty. */
3515 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3516 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3519 log_debug("Inner child completed, invoking payload.");
3521 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3522 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3523 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3525 log_set_open_when_needed(true);
3526 log_settle_target();
3528 (void) fdset_close_others(fds
);
3530 if (arg_start_mode
== START_BOOT
) {
3534 /* Automatically search for the init system */
3536 m
= strv_length(arg_parameters
);
3537 a
= newa(char*, m
+ 2);
3538 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3541 FOREACH_STRING(init
,
3542 "/usr/lib/systemd/systemd",
3543 "/lib/systemd/systemd",
3545 a
[0] = (char*) init
;
3546 execve(a
[0], a
, env_use
);
3549 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3550 } else if (!strv_isempty(arg_parameters
)) {
3551 const char *dollar_path
;
3553 exec_target
= arg_parameters
[0];
3555 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3557 dollar_path
= strv_env_get(env_use
, "PATH");
3559 if (setenv("PATH", dollar_path
, 1) < 0)
3560 return log_error_errno(errno
, "Failed to update $PATH: %m");
3563 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3566 /* If we cannot change the directory, we'll end up in /, that is expected. */
3567 (void) chdir(home
?: "/root");
3569 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3570 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3571 execle("/bin/bash", "-bash", NULL
, env_use
);
3572 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3573 execle("/bin/sh", "-sh", NULL
, env_use
);
3575 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3578 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3581 static int setup_notify_child(void) {
3582 _cleanup_close_
int fd
= -EBADF
;
3583 static const union sockaddr_union sa
= {
3584 .un
.sun_family
= AF_UNIX
,
3585 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3589 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3591 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3593 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3594 (void) sockaddr_un_unlink(&sa
.un
);
3596 WITH_UMASK(0577) { /* only set "w" bit, which is all that's necessary for connecting from the container */
3597 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3599 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3602 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3604 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3606 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3608 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3613 static int setup_unix_export_dir_outside(char **ret
) {
3618 _cleanup_free_
char *p
= NULL
;
3619 p
= path_join("/run/systemd/nspawn/unix-export", arg_machine
);
3623 r
= path_is_mount_point(p
, /* root= */ NULL
, 0);
3625 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Mount point '%s' exists already, refusing.", p
);
3626 if (r
< 0 && r
!= -ENOENT
)
3627 return log_error_errno(r
, "Failed to detect if '%s' is a mount point: %m", p
);
3629 r
= mkdir_p(p
, 0755);
3631 return log_error_errno(r
, "Failed to create '%s': %m", p
);
3633 _cleanup_(rmdir_and_freep
) char *q
= TAKE_PTR(p
);
3635 /* Mount the "unix export" directory really tiny, just 64 inodes. We mark the superblock writable
3636 * (since the container shall bind sockets into it). */
3637 r
= mount_nofollow_verbose(
3642 MS_NODEV
|MS_NOEXEC
|MS_NOSUID
|ms_nosymfollow_supported(),
3643 "size=4M,nr_inodes=64,mode=0755");
3647 _cleanup_(umount_and_rmdir_and_freep
) char *w
= TAKE_PTR(q
);
3649 /* After creating the superblock we change the bind mount to be read-only. This means that the fs
3650 * itself is writable, but not through the mount accessible from the host. */
3651 r
= mount_nofollow_verbose(
3656 MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
|ms_nosymfollow_supported(),
3657 /* options= */ NULL
);
3665 static int setup_unix_export_host_inside(const char *directory
, const char *unix_export_path
) {
3669 assert(unix_export_path
);
3671 r
= make_run_host(directory
);
3675 _cleanup_free_
char *p
= path_join(directory
, "run/host/unix-export");
3679 if (mkdir(p
, 0755) < 0)
3680 return log_error_errno(errno
, "Failed to create '%s': %m", p
);
3682 r
= mount_nofollow_verbose(
3688 /* options= */ NULL
);
3692 r
= mount_nofollow_verbose(
3697 MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
|ms_nosymfollow_supported(),
3698 /* options= */ NULL
);
3702 r
= userns_lchown(p
, 0, 0);
3704 return log_error_errno(r
, "Failed to chown '%s': %m", p
);
3709 static int outer_child(
3711 const char *directory
,
3712 DissectedImage
*dissected_image
,
3713 int fd_outer_socket
,
3714 int fd_inner_socket
,
3717 const char *unix_export_path
) {
3719 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3720 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3721 _cleanup_close_
int fd
= -EBADF
, mntns_fd
= -EBADF
;
3728 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3729 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3730 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3731 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3732 * forked off it, and it exits. */
3736 assert(fd_outer_socket
>= 0);
3737 assert(fd_inner_socket
>= 0);
3739 log_debug("Outer child is initializing.");
3741 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3743 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3745 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3746 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3748 r
= reset_audit_loginuid();
3752 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3753 * mounts to the real root. */
3754 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3758 if (dissected_image
) {
3759 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3760 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3761 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3762 * right place right away. This makes sure ESP partitions and userns are compatible. */
3764 r
= dissected_image_mount_and_warn(
3769 /* userns_fd= */ -EBADF
,
3770 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3771 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3772 DISSECT_IMAGE_USR_NO_ROOT
|
3773 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3774 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3779 r
= determine_uid_shift(directory
);
3783 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3784 r
= namespace_open(0,
3785 /* ret_pidns_fd = */ NULL
,
3787 /* ret_netns_fd = */ NULL
,
3788 /* ret_userns_fd = */ NULL
,
3789 /* ret_root_fd = */ NULL
);
3791 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3793 l
= send_one_fd(fd_outer_socket
, mntns_fd
, 0);
3795 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3796 mntns_fd
= safe_close(mntns_fd
);
3798 /* Let the parent know which UID shift we read from the image */
3799 l
= send(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3801 return log_error_errno(errno
, "Failed to send UID shift: %m");
3802 if (l
!= sizeof(arg_uid_shift
))
3803 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3804 "Short write while sending UID shift.");
3806 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3807 /* When we are supposed to pick the UID shift, the parent will check now whether the
3808 * UID shift we just read from the image is available. If yes, it will send the UID
3809 * shift back to us, if not it will pick a different one, and send it back to us. */
3811 l
= recv(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3813 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3814 if (l
!= sizeof(arg_uid_shift
))
3815 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3816 "Short read while receiving UID shift.");
3819 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3820 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3823 if (path_equal(directory
, "/")) {
3824 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3825 * place, so that we can make changes to its mount structure (for example, to implement
3826 * --volatile=) without this interfering with our ability to access files such as
3827 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3828 * (instead of a temporary directory, since we are living in our own mount namespace here
3829 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3830 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3832 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3836 directory
= "/run/systemd/nspawn-root";
3839 /* Make sure we always have a mount that we can move to root later on. */
3840 r
= make_mount_point(directory
);
3844 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3845 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3846 * we'll live in our own little world from now on, and propagation from the host may only happen via
3847 * the mount tunnel dir, or not at all. */
3848 r
= mount_follow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_PRIVATE
|MS_REC
, NULL
);
3852 r
= setup_pivot_root(
3855 arg_pivot_root_old
);
3859 r
= setup_volatile_mode(
3863 arg_selinux_apifs_context
);
3867 r
= bind_user_prepare(
3872 &arg_custom_mounts
, &arg_n_custom_mounts
,
3873 &bind_user_context
);
3877 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3878 /* Send the user maps we determined to the parent, so that it installs it in our user
3879 * namespace UID map table */
3881 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3883 bind_user_context
->data
[i
].payload_user
->uid
,
3884 bind_user_context
->data
[i
].host_user
->uid
,
3885 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3886 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3889 l
= send(fd_outer_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3891 return log_error_errno(errno
, "Failed to send user UID map: %m");
3892 if (l
!= sizeof(map
))
3893 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3894 "Short write while sending user UID map.");
3901 arg_n_custom_mounts
,
3904 arg_selinux_apifs_context
,
3909 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3910 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3911 arg_uid_shift
!= 0) {
3912 _cleanup_free_
char *usr_subtree
= NULL
;
3916 dirs
[i
++] = (char*) directory
;
3918 if (dissected_image
&& dissected_image
->partitions
[PARTITION_USR
].found
) {
3919 usr_subtree
= path_join(directory
, "/usr");
3923 dirs
[i
++] = usr_subtree
;
3928 r
= remount_idmap(dirs
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3929 if (r
== -EINVAL
|| ERRNO_IS_NEG_NOT_SUPPORTED(r
)) {
3930 /* This might fail because the kernel or file system doesn't support idmapping. We
3931 * can't really distinguish this nicely, nor do we have any guarantees about the
3932 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3933 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3934 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3935 "ID mapped mounts are apparently not available, sorry.");
3937 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3938 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3940 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3942 log_debug("ID mapped mounts available, making use of them.");
3947 if (dissected_image
) {
3948 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3949 r
= dissected_image_mount(
3954 /* userns_fd= */ -EBADF
,
3955 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3956 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3957 DISSECT_IMAGE_USR_NO_ROOT
|
3958 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3959 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3961 return log_error_errno(r
, "File system check for image failed: %m");
3963 return log_error_errno(r
, "Failed to mount image file system: %m");
3966 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3967 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3969 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3973 l
= send(fd_outer_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3975 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3976 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3977 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3978 "Short write while sending cgroup mode.");
3981 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3985 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3989 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3990 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3991 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3993 return log_error_errno(r
, "Failed to make tree read-only: %m");
3996 r
= mount_all(directory
,
3999 arg_selinux_apifs_context
);
4003 r
= copy_devnodes(directory
);
4007 r
= make_extra_nodes(directory
);
4011 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
4013 p
= prefix_roota(directory
, "/run/host");
4014 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
4016 r
= setup_unix_export_host_inside(directory
, unix_export_path
);
4020 r
= setup_pts(directory
);
4024 r
= mount_tunnel_dig(directory
);
4028 r
= setup_keyring();
4032 r
= setup_credentials(directory
);
4036 r
= bind_user_setup(bind_user_context
, directory
);
4043 arg_n_custom_mounts
,
4046 arg_selinux_apifs_context
,
4047 MOUNT_NON_ROOT_ONLY
);
4051 r
= setup_timezone(directory
);
4055 r
= setup_resolv_conf(directory
);
4059 r
= setup_machine_id(directory
);
4063 r
= setup_journal(directory
);
4067 /* The same stuff as the $container env var, but nicely readable for the entire payload */
4068 p
= prefix_roota(directory
, "/run/host/container-manager");
4069 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
|WRITE_STRING_FILE_MODE_0444
);
4071 /* The same stuff as the $container_uuid env var */
4072 p
= prefix_roota(directory
, "/run/host/container-uuid");
4073 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
|WRITE_STRING_FILE_MODE_0444
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
4075 if (!arg_use_cgns
) {
4078 arg_unified_cgroup_hierarchy
,
4079 arg_userns_mode
!= USER_NAMESPACE_NO
,
4082 arg_selinux_apifs_context
,
4088 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
4089 * mounts available in systemd services inside the container that create a new mount namespace. See
4090 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
4091 * will inherit the shared propagation mode.
4093 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
4094 * directory mount to root later on.
4095 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
4097 r
= mount_switch_root(directory
, MS_SHARED
);
4099 return log_error_errno(r
, "Failed to move root directory: %m");
4101 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
4102 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
4104 r
= mount_tunnel_open();
4108 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4109 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4110 * requires that a fully visible instance is already present in the target mount
4111 * namespace. Mount one here so the inner child can mount its own instances. Later
4112 * we umount the temporary instances created here before we actually exec the
4113 * payload. Since the rootfs is shared the umount will propagate into the container.
4114 * Note, the inner child wouldn't be able to unmount the instances on its own since
4115 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4117 r
= pin_fully_visible_fs();
4122 fd
= setup_notify_child();
4126 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
4127 arg_clone_ns_flags
|
4128 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
4130 return log_error_errno(errno
, "Failed to fork inner child: %m");
4132 fd_outer_socket
= safe_close(fd_outer_socket
);
4134 /* The inner child has all namespaces that are requested, so that we all are owned by the
4135 * user if user namespaces are turned on. */
4137 if (arg_network_namespace_path
) {
4138 r
= namespace_enter(/* pidns_fd = */ -EBADF
,
4139 /* mntns_fd = */ -EBADF
,
4141 /* userns_fd = */ -EBADF
,
4142 /* root_fd = */ -EBADF
);
4144 return log_error_errno(r
, "Failed to join network namespace: %m");
4147 r
= inner_child(barrier
, fd_inner_socket
, fds
, os_release_pairs
);
4149 _exit(EXIT_FAILURE
);
4151 _exit(EXIT_SUCCESS
);
4154 l
= send(fd_outer_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4156 return log_error_errno(errno
, "Failed to send PID: %m");
4157 if (l
!= sizeof(pid
))
4158 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4159 "Short write while sending PID.");
4161 l
= send(fd_outer_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4163 return log_error_errno(errno
, "Failed to send machine ID: %m");
4164 if (l
!= sizeof(arg_uuid
))
4165 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4166 "Short write while sending machine ID.");
4168 l
= send_one_fd(fd_outer_socket
, fd
, 0);
4170 return log_error_errno(l
, "Failed to send notify fd: %m");
4172 fd_outer_socket
= safe_close(fd_outer_socket
);
4173 fd_inner_socket
= safe_close(fd_inner_socket
);
4174 netns_fd
= safe_close(netns_fd
);
4179 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4180 bool tried_hashed
= false;
4181 unsigned n_tries
= 100;
4186 assert(ret_lock_file
);
4187 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4188 assert(arg_uid_range
== 0x10000U
);
4192 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4195 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4196 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4201 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4203 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4206 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4207 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4208 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4213 /* Make some superficial checks whether the range is currently known in the user database */
4214 if (getpwuid(candidate
))
4216 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4218 if (getgrgid(candidate
))
4220 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4223 *ret_lock_file
= lf
;
4224 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4229 if (arg_machine
&& !tried_hashed
) {
4230 /* Try to hash the base from the container name */
4232 static const uint8_t hash_key
[] = {
4233 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4234 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4237 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4239 tried_hashed
= true;
4241 random_bytes(&candidate
, sizeof(candidate
));
4243 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4244 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4248 static int add_one_uid_map(
4250 uid_t container_uid
,
4254 return strextendf(p
,
4255 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4256 container_uid
, host_uid
, range
);
4259 static int make_uid_map_string(
4260 const uid_t bind_user_uid
[],
4261 size_t n_bind_user_uid
,
4265 _cleanup_free_
char *s
= NULL
;
4266 uid_t previous_uid
= 0;
4269 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4270 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4273 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4274 * quadruplet, consisting of host and container UID + GID. */
4276 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4277 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4278 host_uid
= bind_user_uid
[i
*4+offset
+1];
4280 assert(previous_uid
<= payload_uid
);
4281 assert(payload_uid
< arg_uid_range
);
4283 /* Add a range to close the gap to previous entry */
4284 if (payload_uid
> previous_uid
) {
4285 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4290 /* Map this specific user */
4291 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4295 previous_uid
= payload_uid
+ 1;
4298 /* And add a range to close the gap to finish the range */
4299 if (arg_uid_range
> previous_uid
) {
4300 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4311 static int setup_uid_map(
4313 const uid_t bind_user_uid
[],
4314 size_t n_bind_user_uid
) {
4316 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4317 _cleanup_free_
char *s
= NULL
;
4322 /* Build the UID map string */
4323 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4326 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4327 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4329 return log_error_errno(r
, "Failed to write UID map: %m");
4331 /* And now build the GID map string */
4333 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4336 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4337 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4339 return log_error_errno(r
, "Failed to write GID map: %m");
4344 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4345 char buf
[NOTIFY_BUFFER_MAX
+1];
4347 struct iovec iovec
= {
4349 .iov_len
= sizeof(buf
)-1,
4351 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4352 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4353 struct msghdr msghdr
= {
4356 .msg_control
= &control
,
4357 .msg_controllen
= sizeof(control
),
4359 struct ucred
*ucred
;
4361 pid_t inner_child_pid
;
4362 _cleanup_strv_free_
char **tags
= NULL
;
4367 inner_child_pid
= PTR_TO_PID(userdata
);
4369 if (revents
!= EPOLLIN
) {
4370 log_warning("Got unexpected poll event for notify fd.");
4374 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4375 if (ERRNO_IS_NEG_TRANSIENT(n
))
4377 else if (n
== -EXFULL
) {
4378 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4381 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4383 cmsg_close_all(&msghdr
);
4385 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4386 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4387 log_debug("Received notify message without valid credentials. Ignoring.");
4391 if ((size_t) n
>= sizeof(buf
)) {
4392 log_warning("Received notify message exceeded maximum size. Ignoring.");
4397 tags
= strv_split(buf
, "\n\r");
4401 if (strv_contains(tags
, "READY=1")) {
4402 r
= sd_notify(false, "READY=1\n");
4404 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4407 p
= strv_find_startswith(tags
, "STATUS=");
4409 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4414 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4417 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4419 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4421 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4426 static int merge_settings(Settings
*settings
, const char *path
) {
4432 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4433 * that this steals the fields of the Settings* structure, and hence modifies it. */
4435 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4436 settings
->start_mode
>= 0) {
4437 arg_start_mode
= settings
->start_mode
;
4438 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4441 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4442 settings
->ephemeral
>= 0)
4443 arg_ephemeral
= settings
->ephemeral
;
4445 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4448 if (!arg_settings_trusted
)
4449 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4451 free_and_replace(arg_directory
, settings
->root
);
4454 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4455 settings
->pivot_root_new
) {
4456 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4457 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4460 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4461 settings
->working_directory
)
4462 free_and_replace(arg_chdir
, settings
->working_directory
);
4464 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4465 settings
->environment
)
4466 strv_free_and_replace(arg_setenv
, settings
->environment
);
4468 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4471 free_and_replace(arg_user
, settings
->user
);
4473 if (uid_is_valid(settings
->uid
))
4474 arg_uid
= settings
->uid
;
4475 if (gid_is_valid(settings
->gid
))
4476 arg_gid
= settings
->gid
;
4477 if (settings
->n_supplementary_gids
> 0) {
4478 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4479 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4483 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4484 uint64_t plus
, minus
;
4485 uint64_t network_minus
= 0;
4488 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4489 * Settings structure */
4491 plus
= settings
->capability
;
4492 minus
= settings
->drop_capability
;
4494 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4495 settings_network_configured(settings
)) {
4496 if (settings_private_network(settings
))
4497 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4499 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4502 if (!arg_settings_trusted
&& plus
!= 0) {
4503 if (settings
->capability
!= 0)
4504 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4506 arg_caps_retain
&= ~network_minus
;
4507 arg_caps_retain
|= plus
;
4510 arg_caps_retain
&= ~minus
;
4512 /* Copy the full capabilities over too */
4513 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4514 if (!arg_settings_trusted
)
4515 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4517 arg_full_capabilities
= settings
->full_capabilities
;
4520 ambient
= settings
->ambient_capability
;
4521 if (!arg_settings_trusted
&& ambient
!= 0)
4522 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4524 arg_caps_ambient
|= ambient
;
4527 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4528 settings
->kill_signal
> 0)
4529 arg_kill_signal
= settings
->kill_signal
;
4531 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4532 settings
->personality
!= PERSONALITY_INVALID
)
4533 arg_personality
= settings
->personality
;
4535 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4536 !sd_id128_is_null(settings
->machine_id
)) {
4538 if (!arg_settings_trusted
)
4539 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4541 arg_uuid
= settings
->machine_id
;
4544 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4545 settings
->read_only
>= 0)
4546 arg_read_only
= settings
->read_only
;
4548 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4549 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4550 arg_volatile_mode
= settings
->volatile_mode
;
4552 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4553 settings
->n_custom_mounts
> 0) {
4555 if (!arg_settings_trusted
)
4556 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4558 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4559 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4560 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4561 settings
->n_custom_mounts
= 0;
4565 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4566 settings_network_configured(settings
)) {
4568 if (!arg_settings_trusted
)
4569 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4571 arg_network_veth
= settings_network_veth(settings
);
4572 arg_private_network
= settings_private_network(settings
);
4574 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4575 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4576 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4577 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4579 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4580 free_and_replace(arg_network_zone
, settings
->network_zone
);
4582 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4586 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4587 settings
->expose_ports
) {
4589 if (!arg_settings_trusted
)
4590 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4592 expose_port_free_all(arg_expose_ports
);
4593 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4597 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4598 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4600 if (!arg_settings_trusted
)
4601 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4603 arg_userns_mode
= settings
->userns_mode
;
4604 arg_uid_shift
= settings
->uid_shift
;
4605 arg_uid_range
= settings
->uid_range
;
4606 arg_userns_ownership
= settings
->userns_ownership
;
4610 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4611 !strv_isempty(settings
->bind_user
))
4612 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4614 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4615 settings
->notify_ready
>= 0)
4616 arg_notify_ready
= settings
->notify_ready
;
4618 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4620 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4621 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4622 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4624 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4625 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4630 if (settings
->seccomp
) {
4631 if (!arg_settings_trusted
)
4632 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4634 seccomp_release(arg_seccomp
);
4635 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4641 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4642 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4645 if (!settings
->rlimit
[rl
])
4648 if (!arg_settings_trusted
) {
4649 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4653 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4656 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4658 free_and_replace(arg_hostname
, settings
->hostname
);
4660 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4661 settings
->no_new_privileges
>= 0)
4662 arg_no_new_privileges
= settings
->no_new_privileges
;
4664 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4665 settings
->oom_score_adjust_set
) {
4667 if (!arg_settings_trusted
)
4668 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4670 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4671 arg_oom_score_adjust_set
= true;
4675 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4676 settings
->cpu_set
.set
) {
4678 if (!arg_settings_trusted
)
4679 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4681 cpu_set_reset(&arg_cpu_set
);
4682 arg_cpu_set
= TAKE_STRUCT(settings
->cpu_set
);
4686 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4687 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4688 arg_resolv_conf
= settings
->resolv_conf
;
4690 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4691 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4693 if (!arg_settings_trusted
)
4694 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4696 arg_link_journal
= settings
->link_journal
;
4697 arg_link_journal_try
= settings
->link_journal_try
;
4701 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4702 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4703 arg_timezone
= settings
->timezone
;
4705 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4708 if (!arg_settings_trusted
)
4709 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4711 free_and_replace(arg_slice
, settings
->slice
);
4714 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4715 settings
->use_cgns
>= 0) {
4717 if (!arg_settings_trusted
)
4718 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4720 arg_use_cgns
= settings
->use_cgns
;
4723 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4724 settings
->clone_ns_flags
!= ULONG_MAX
) {
4726 if (!arg_settings_trusted
)
4727 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4729 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4732 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4733 settings
->console_mode
>= 0) {
4735 if (!arg_settings_trusted
)
4736 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4738 arg_console_mode
= settings
->console_mode
;
4741 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4742 settings
->suppress_sync
>= 0)
4743 arg_suppress_sync
= settings
->suppress_sync
;
4745 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4746 * don't consult arg_settings_mask for them. */
4748 sd_bus_message_unref(arg_property_message
);
4749 arg_property_message
= TAKE_PTR(settings
->properties
);
4751 arg_console_width
= settings
->console_width
;
4752 arg_console_height
= settings
->console_height
;
4754 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4755 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4756 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4757 settings
->n_extra_nodes
= 0;
4762 static int load_settings(void) {
4763 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4764 _cleanup_fclose_
FILE *f
= NULL
;
4765 _cleanup_free_
char *p
= NULL
;
4771 /* If all settings are masked, there's no point in looking for
4772 * the settings file */
4773 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4776 /* We first look in the admin's directories in /etc and /run */
4777 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4778 _cleanup_free_
char *j
= NULL
;
4780 j
= path_join(i
, arg_settings_filename
);
4788 /* By default, we trust configuration from /etc and /run */
4789 if (arg_settings_trusted
< 0)
4790 arg_settings_trusted
= true;
4795 if (errno
!= ENOENT
)
4796 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4800 /* After that, let's look for a file next to the
4801 * actual image we shall boot. */
4804 r
= file_in_same_dir(arg_image
, arg_settings_filename
, &p
);
4806 return log_error_errno(r
, "Failed to generate settings path from image path: %m");
4807 } else if (arg_directory
) {
4808 r
= file_in_same_dir(arg_directory
, arg_settings_filename
, &p
);
4809 if (r
< 0 && r
!= -EADDRNOTAVAIL
) /* if directory is root fs, don't complain */
4810 return log_error_errno(r
, "Failed to generate settings path from directory path: %m");
4815 if (!f
&& errno
!= ENOENT
)
4816 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4818 /* By default, we do not trust configuration from /var/lib/machines */
4819 if (arg_settings_trusted
< 0)
4820 arg_settings_trusted
= false;
4827 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4829 r
= settings_load(f
, p
, &settings
);
4833 return merge_settings(settings
, p
);
4836 static int load_oci_bundle(void) {
4837 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4840 if (!arg_oci_bundle
)
4843 /* By default let's trust OCI bundles */
4844 if (arg_settings_trusted
< 0)
4845 arg_settings_trusted
= true;
4847 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4851 return merge_settings(settings
, arg_oci_bundle
);
4854 static int run_container(
4855 DissectedImage
*dissected_image
,
4857 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4858 struct ExposeArgs
*expose_args
,
4859 int *master
, pid_t
*pid
, int *ret
) {
4861 static const struct sigaction sa
= {
4862 .sa_handler
= nop_signal_handler
,
4863 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4866 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4867 _cleanup_close_
int etc_passwd_lock
= -EBADF
;
4868 _cleanup_close_pair_
int
4869 fd_inner_socket_pair
[2] = EBADF_PAIR
,
4870 fd_outer_socket_pair
[2] = EBADF_PAIR
;
4872 _cleanup_close_
int notify_socket
= -EBADF
, mntns_fd
= -EBADF
, fd_kmsg_fifo
= -EBADF
;
4873 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4874 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4875 _cleanup_(umount_and_rmdir_and_freep
) char *unix_export_host_dir
= NULL
;
4876 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4877 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4878 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4879 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4880 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4881 size_t n_bind_user_uid
= 0;
4882 ContainerStatus container_status
= 0;
4886 _cleanup_close_
int child_netns_fd
= -EBADF
;
4888 assert_se(sigemptyset(&mask_chld
) == 0);
4889 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4891 /* Set up the unix export host directory on the host first */
4892 r
= setup_unix_export_dir_outside(&unix_export_host_dir
);
4896 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4897 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4898 * check with getpwuid() if the specific user already exists. Note that /etc might be
4899 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4900 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4901 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4904 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4905 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4906 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4909 r
= barrier_create(&barrier
);
4911 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4913 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_inner_socket_pair
) < 0)
4914 return log_error_errno(errno
, "Failed to create inner socket pair: %m");
4916 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_outer_socket_pair
) < 0)
4917 return log_error_errno(errno
, "Failed to create outer socket pair: %m");
4919 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4920 * parent's blocking calls and give it a chance to call wait() and terminate. */
4921 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4923 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4925 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4927 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4929 if (arg_network_namespace_path
) {
4930 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4931 if (child_netns_fd
< 0)
4932 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4934 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4936 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
);
4938 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4940 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4941 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4944 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4946 return log_error_errno(errno
, "clone() failed%s: %m",
4948 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4951 /* The outer child only has a file system namespace. */
4952 barrier_set_role(&barrier
, BARRIER_CHILD
);
4954 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
4955 fd_outer_socket_pair
[0] = safe_close(fd_outer_socket_pair
[0]);
4957 (void) reset_all_signal_handlers();
4958 (void) reset_signal_mask();
4960 r
= outer_child(&barrier
,
4963 fd_outer_socket_pair
[1],
4964 fd_inner_socket_pair
[1],
4967 unix_export_host_dir
);
4969 _exit(EXIT_FAILURE
);
4971 _exit(EXIT_SUCCESS
);
4974 barrier_set_role(&barrier
, BARRIER_PARENT
);
4978 fd_inner_socket_pair
[1] = safe_close(fd_inner_socket_pair
[1]);
4979 fd_outer_socket_pair
[1] = safe_close(fd_outer_socket_pair
[1]);
4981 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4982 mntns_fd
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4984 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4986 /* The child just let us know the UID shift it might have read from the image. */
4987 l
= recv(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4989 return log_error_errno(errno
, "Failed to read UID shift: %m");
4990 if (l
!= sizeof arg_uid_shift
)
4991 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4993 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4994 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4995 * image, but if that's already in use, pick a new one, and report back to the child,
4996 * which one we now picked. */
4998 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
5000 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
5002 l
= send(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
5004 return log_error_errno(errno
, "Failed to send UID shift: %m");
5005 if (l
!= sizeof arg_uid_shift
)
5006 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
5009 n_bind_user_uid
= strv_length(arg_bind_user
);
5010 if (n_bind_user_uid
> 0) {
5011 /* Right after the UID shift, we'll receive the list of UID mappings for the
5012 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
5014 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
5018 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
5019 l
= recv(fd_outer_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
5021 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
5022 if (l
!= sizeof(uid_t
)*4)
5023 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
5024 SYNTHETIC_ERRNO(EIO
),
5025 "Short read while reading bind user UID pairs.");
5030 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
5031 /* The child let us know the support cgroup mode it might have read from the image. */
5032 l
= recv(fd_outer_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
5034 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
5035 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
5036 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
5037 l
, l
== 0 ? " The child is most likely dead." : "");
5040 /* Wait for the outer child. */
5041 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
5044 if (r
!= EXIT_SUCCESS
)
5047 /* And now retrieve the PID of the inner child. */
5048 l
= recv(fd_outer_socket_pair
[0], pid
, sizeof *pid
, 0);
5050 return log_error_errno(errno
, "Failed to read inner child PID: %m");
5051 if (l
!= sizeof *pid
)
5052 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
5054 /* We also retrieve container UUID in case it was generated by outer child */
5055 l
= recv(fd_outer_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
5057 return log_error_errno(errno
, "Failed to read container machine ID: %m");
5058 if (l
!= sizeof(arg_uuid
))
5059 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
5061 /* We also retrieve the socket used for notifications generated by outer child */
5062 notify_socket
= receive_one_fd(fd_outer_socket_pair
[0], 0);
5063 if (notify_socket
< 0)
5064 return log_error_errno(notify_socket
,
5065 "Failed to receive notification socket from the outer child: %m");
5067 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
5069 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5070 if (!barrier_place_and_sync(&barrier
)) /* #1 */
5071 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5073 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
5077 (void) barrier_place(&barrier
); /* #2 */
5080 if (arg_private_network
) {
5081 if (!arg_network_namespace_path
) {
5082 /* Wait until the child has unshared its network namespace. */
5083 if (!barrier_place_and_sync(&barrier
)) /* #3 */
5084 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
5087 if (child_netns_fd
< 0) {
5088 /* Make sure we have an open file descriptor to the child's network
5089 * namespace so it stays alive even if the child exits. */
5090 r
= namespace_open(*pid
,
5091 /* ret_pidns_fd = */ NULL
,
5092 /* ret_mntns_fd = */ NULL
,
5094 /* ret_userns_fd = */ NULL
,
5095 /* ret_root_fd = */ NULL
);
5097 return log_error_errno(r
, "Failed to open child network namespace: %m");
5100 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
5104 if (arg_network_veth
) {
5105 r
= setup_veth(arg_machine
, *pid
, veth_name
,
5106 arg_network_bridge
|| arg_network_zone
, &arg_network_provided_mac
);
5112 if (arg_network_bridge
) {
5113 /* Add the interface to a bridge */
5114 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
5119 } else if (arg_network_zone
) {
5120 /* Add the interface to a bridge, possibly creating it */
5121 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5129 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5133 /* We created the primary and extra veth links now; let's remember this, so that we know to
5134 remove them later on. Note that we don't bother with removing veth links that were created
5135 here when their setup failed half-way, because in that case the kernel should be able to
5136 remove them on its own, since they cannot be referenced by anything yet. */
5137 *veth_created
= true;
5139 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5143 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5148 if (arg_register
|| !arg_keep_unit
) {
5149 r
= sd_bus_default_system(&bus
);
5151 return log_error_errno(r
, "Failed to open system bus: %m");
5153 r
= sd_bus_set_close_on_exit(bus
, false);
5155 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5158 if (!arg_keep_unit
) {
5159 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5160 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5161 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5163 r
= sd_bus_match_signal_async(
5166 "org.freedesktop.systemd1",
5168 "org.freedesktop.systemd1.Scope",
5170 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5172 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5176 r
= register_machine(
5184 arg_custom_mounts
, arg_n_custom_mounts
,
5187 arg_property_message
,
5189 arg_container_service_name
,
5194 } else if (!arg_keep_unit
) {
5200 arg_custom_mounts
, arg_n_custom_mounts
,
5203 arg_property_message
,
5204 /* allow_pidfds= */ true,
5209 } else if (arg_slice
|| arg_property
)
5210 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5212 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5216 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5220 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5224 /* Notify the child that the parent is ready with all
5225 * its setup (including cgroup-ification), and that
5226 * the child can now hand over control to the code to
5227 * run inside the container. */
5228 (void) barrier_place(&barrier
); /* #4 */
5230 /* Block SIGCHLD here, before notifying child.
5231 * process_pty() will handle it with the other signals. */
5232 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5234 /* Reset signal to default */
5235 r
= default_signals(SIGCHLD
);
5237 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5239 r
= sd_event_new(&event
);
5241 return log_error_errno(r
, "Failed to get default event source: %m");
5243 (void) sd_event_set_watchdog(event
, true);
5246 r
= sd_bus_attach_event(bus
, event
, 0);
5248 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5251 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5255 /* Wait that the child is completely ready now, and has mounted their own copies of procfs and so on,
5256 * before we take the fully visible instances away. */
5257 if (!barrier_sync(&barrier
)) /* #5.1 */
5258 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5260 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5261 r
= wipe_fully_visible_fs(mntns_fd
);
5264 mntns_fd
= safe_close(mntns_fd
);
5267 /* And now let the child know that we completed removing the procfs instances, and it can start the
5269 if (!barrier_place(&barrier
)) /* #5.2 */
5270 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5272 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5273 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5274 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5276 (void) sd_notifyf(false,
5277 "STATUS=Container running.\n"
5278 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5279 if (!arg_notify_ready
) {
5280 r
= sd_notify(false, "READY=1\n");
5282 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5285 if (arg_kill_signal
> 0) {
5286 /* Try to kill the init system on SIGINT or SIGTERM */
5287 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5288 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5290 /* Immediately exit */
5291 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5292 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5295 (void) sd_event_add_signal(event
, NULL
, SIGRTMIN
+18, sigrtmin18_handler
, NULL
);
5297 r
= sd_event_add_memory_pressure(event
, NULL
, NULL
, NULL
);
5299 log_debug_errno(r
, "Failed allocate memory pressure event source, ignoring: %m");
5301 /* Exit when the child exits */
5302 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5304 /* Retrieve the kmsg fifo allocated by inner child */
5305 fd_kmsg_fifo
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5306 if (fd_kmsg_fifo
< 0)
5307 return log_error_errno(fd_kmsg_fifo
, "Failed to receive kmsg fifo from inner child: %m");
5309 if (arg_expose_ports
) {
5310 r
= expose_port_watch_rtnl(event
, fd_inner_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5314 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5315 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5318 if (arg_console_mode
!= CONSOLE_PIPE
) {
5319 _cleanup_close_
int fd
= -EBADF
;
5320 PTYForwardFlags flags
= 0;
5322 /* Retrieve the master pty allocated by inner child */
5323 fd
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5325 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5327 switch (arg_console_mode
) {
5329 case CONSOLE_READ_ONLY
:
5330 flags
|= PTY_FORWARD_READ_ONLY
;
5334 case CONSOLE_INTERACTIVE
:
5335 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5337 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5339 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5341 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5342 (void) pty_forward_set_width_height(forward
,
5344 arg_console_height
);
5348 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5351 *master
= TAKE_FD(fd
);
5354 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
5356 r
= sd_event_loop(event
);
5358 return log_error_errno(r
, "Failed to run event loop: %m");
5363 (void) pty_forward_get_last_char(forward
, &last_char
);
5364 forward
= pty_forward_free(forward
);
5366 if (!arg_quiet
&& last_char
!= '\n')
5370 /* Kill if it is not dead yet anyway */
5371 if (!arg_register
&& !arg_keep_unit
&& bus
)
5372 terminate_scope(bus
, arg_machine
);
5374 /* Normally redundant, but better safe than sorry */
5375 (void) kill(*pid
, SIGKILL
);
5377 fd_kmsg_fifo
= safe_close(fd_kmsg_fifo
);
5379 if (arg_private_network
) {
5380 r
= move_back_network_interfaces(child_netns_fd
, arg_network_interfaces
);
5385 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5387 /* Tell machined that we are gone. */
5389 (void) unregister_machine(bus
, arg_machine
);
5392 /* We failed to wait for the container, or the container exited abnormally. */
5394 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5395 /* r > 0 → The container exited with a non-zero status.
5396 * As a special case, we need to replace 133 with a different value,
5397 * because 133 is special-cased in the service file to reboot the container.
5398 * otherwise → The container exited with zero status and a reboot was not requested.
5400 if (r
== EXIT_FORCE_RESTART
)
5401 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5403 return 0; /* finito */
5406 /* CONTAINER_REBOOTED, loop again */
5408 if (arg_keep_unit
) {
5409 /* Special handling if we are running as a service: instead of simply
5410 * restarting the machine we want to restart the entire service, so let's
5411 * inform systemd about this with the special exit code 133. The service
5412 * file uses RestartForceExitStatus=133 so that this results in a full
5413 * nspawn restart. This is necessary since we might have cgroup parameters
5414 * set we want to have flushed out. */
5415 *ret
= EXIT_FORCE_RESTART
;
5416 return 0; /* finito */
5419 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5420 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5422 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5423 *veth_created
= false;
5424 return 1; /* loop again */
5427 static int initialize_rlimits(void) {
5428 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5429 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5430 * container execution environments. */
5432 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5433 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5434 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5435 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5436 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5437 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5438 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5439 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5440 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5441 [RLIMIT_NICE
] = { 0, 0 },
5442 [RLIMIT_NOFILE
] = { 1024, 4096 },
5443 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5444 [RLIMIT_RTPRIO
] = { 0, 0 },
5445 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5446 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5448 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5449 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5450 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5451 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5452 * that PID 1 changes a number of other resource limits during early initialization which is why we
5453 * don't read the other limits from PID 1 but prefer the static table above. */
5458 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5459 /* Let's only fill in what the user hasn't explicitly configured anyway */
5460 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5461 const struct rlimit
*v
;
5462 struct rlimit buffer
;
5464 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5465 /* For these two let's read the limits off PID 1. See above for an explanation. */
5467 r
= pid_getrlimit(1, rl
, &buffer
);
5469 return log_error_errno(r
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5472 } else if (rl
== RLIMIT_NOFILE
) {
5473 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5474 * userspace. Given that nspawn containers are often run without our PID 1,
5475 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5476 * so that container userspace gets similar resources as host userspace
5478 buffer
= kernel_defaults
[rl
];
5479 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5482 v
= kernel_defaults
+ rl
;
5484 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5485 if (!arg_rlimit
[rl
])
5489 if (DEBUG_LOGGING
) {
5490 _cleanup_free_
char *k
= NULL
;
5492 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5493 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5500 static int cant_be_in_netns(void) {
5501 _cleanup_close_
int fd
= -EBADF
;
5505 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5506 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5509 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5512 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5514 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5516 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5517 if (r
== -ENOENT
|| ERRNO_IS_NEG_DISCONNECT(r
))
5518 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5519 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5521 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5523 r
= getpeercred(fd
, &ucred
);
5525 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5527 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5529 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5531 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5532 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5536 static int run(int argc
, char *argv
[]) {
5537 bool remove_directory
= false, remove_image
= false, veth_created
= false, remove_tmprootdir
= false;
5538 _cleanup_close_
int master
= -EBADF
;
5539 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5540 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5541 char veth_name
[IFNAMSIZ
] = "";
5542 struct ExposeArgs expose_args
= {};
5543 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5544 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5545 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5546 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5547 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5550 log_parse_environment();
5553 r
= parse_argv(argc
, argv
);
5557 if (geteuid() != 0) {
5558 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5559 argc
>= 2 ? "Need to be root." :
5560 "Need to be root (and some arguments are usually required).\nHint: try --help");
5564 r
= cant_be_in_netns();
5568 r
= initialize_rlimits();
5572 r
= load_oci_bundle();
5580 r
= determine_names();
5584 r
= load_settings();
5588 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
5589 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
5591 if (!arg_private_network
&& arg_userns_mode
!= USER_NAMESPACE_NO
&& arg_uid_shift
> 0)
5592 arg_caps_retain
&= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE
);
5596 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5600 r
= verify_arguments();
5604 r
= verify_network_interfaces_initialized();
5608 /* Reapply environment settings. */
5609 (void) detect_unified_cgroup_hierarchy_from_environment();
5611 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5612 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5613 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5614 (void) ignore_signals(SIGPIPE
);
5616 n_fd_passed
= sd_listen_fds(false);
5617 if (n_fd_passed
> 0) {
5618 r
= fdset_new_listen_fds(&fds
, false);
5620 log_error_errno(r
, "Failed to collect file descriptors: %m");
5625 /* The "default" umask. This is appropriate for most file and directory
5626 * operations performed by nspawn, and is the umask that will be used for
5627 * the child. Functions like copy_devnodes() change the umask temporarily. */
5630 if (arg_directory
) {
5633 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5634 * /var from the host will propagate into container dynamically (because bad things happen if
5635 * two systems write to the same /var). Let's allow it for the special cases where /var is
5636 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5637 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5638 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5639 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5643 if (arg_ephemeral
) {
5644 _cleanup_free_
char *np
= NULL
;
5646 r
= chase_and_update(&arg_directory
, 0);
5650 /* If the specified path is a mount point we generate the new snapshot immediately
5651 * inside it under a random name. However if the specified is not a mount point we
5652 * create the new snapshot in the parent directory, just next to it. */
5653 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5655 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5659 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5661 r
= tempfn_random(arg_directory
, "machine.", &np
);
5663 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5667 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5668 * only owned by us and no one else. */
5669 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5671 log_error_errno(r
, "Failed to lock %s: %m", np
);
5676 BLOCK_SIGNALS(SIGINT
);
5677 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_directory
, AT_FDCWD
, np
,
5678 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5679 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5680 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5681 BTRFS_SNAPSHOT_RECURSIVE
|
5682 BTRFS_SNAPSHOT_QUOTA
|
5683 BTRFS_SNAPSHOT_SIGINT
);
5686 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5690 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5694 free_and_replace(arg_directory
, np
);
5695 remove_directory
= true;
5697 r
= chase_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5701 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5703 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5707 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5712 r
= chase_and_update(&arg_template
, 0);
5717 BLOCK_SIGNALS(SIGINT
);
5718 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_template
, AT_FDCWD
, arg_directory
,
5719 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5720 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5721 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5722 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5723 BTRFS_SNAPSHOT_RECURSIVE
|
5724 BTRFS_SNAPSHOT_QUOTA
|
5725 BTRFS_SNAPSHOT_SIGINT
);
5728 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5729 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5730 else if (r
== -EINTR
) {
5731 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5734 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5737 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5738 "Populated %s from template %s.", arg_directory
, arg_template
);
5742 if (arg_start_mode
== START_BOOT
) {
5743 _cleanup_free_
char *b
= NULL
;
5745 int check_os_release
, is_os_tree
;
5747 if (arg_pivot_root_new
) {
5748 b
= path_join(arg_directory
, arg_pivot_root_new
);
5758 check_os_release
= getenv_bool("SYSTEMD_NSPAWN_CHECK_OS_RELEASE");
5759 if (check_os_release
< 0 && check_os_release
!= -ENXIO
) {
5760 r
= log_error_errno(check_os_release
, "Failed to parse $SYSTEMD_NSPAWN_CHECK_OS_RELEASE: %m");
5764 is_os_tree
= path_is_os_tree(p
);
5765 if (is_os_tree
== 0 && check_os_release
== 0)
5766 log_debug("Directory %s is missing an os-release file, continuing anyway.", p
);
5767 else if (is_os_tree
<= 0) {
5768 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5769 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5773 _cleanup_free_
char *p
= NULL
;
5775 if (arg_pivot_root_new
)
5776 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5778 p
= path_join(arg_directory
, "/usr/");
5784 if (laccess(p
, F_OK
) < 0) {
5785 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5786 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5792 DissectImageFlags dissect_image_flags
=
5793 DISSECT_IMAGE_GENERIC_ROOT
|
5794 DISSECT_IMAGE_REQUIRE_ROOT
|
5795 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5796 DISSECT_IMAGE_USR_NO_ROOT
|
5797 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5798 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5800 assert(!arg_template
);
5802 r
= chase_and_update(&arg_image
, 0);
5806 if (arg_ephemeral
) {
5807 _cleanup_free_
char *np
= NULL
;
5809 r
= tempfn_random(arg_image
, "machine.", &np
);
5811 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5815 /* Always take an exclusive lock on our own ephemeral copy. */
5816 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5818 log_error_errno(r
, "Failed to create image lock: %m");
5823 BLOCK_SIGNALS(SIGINT
);
5824 r
= copy_file_full(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600,
5825 FS_NOCOW_FL
, FS_NOCOW_FL
,
5826 COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
,
5830 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5834 r
= log_error_errno(r
, "Failed to copy image file: %m");
5838 free_and_replace(arg_image
, np
);
5839 remove_image
= true;
5841 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5843 log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5847 log_error_errno(r
, "Failed to create image lock: %m");
5851 r
= verity_settings_load(
5852 &arg_verity_settings
,
5853 arg_image
, NULL
, NULL
);
5855 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5859 if (arg_verity_settings
.data_path
)
5860 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5863 if (!mkdtemp(tmprootdir
)) {
5864 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5868 remove_tmprootdir
= true;
5870 arg_directory
= strdup(tmprootdir
);
5871 if (!arg_directory
) {
5876 r
= loop_device_make_by_path(
5878 arg_read_only
? O_RDONLY
: O_RDWR
,
5879 /* sector_size= */ UINT32_MAX
,
5880 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5884 log_error_errno(r
, "Failed to set up loopback block device: %m");
5888 r
= dissect_loop_device_and_warn(
5890 &arg_verity_settings
,
5891 /* mount_options=*/ NULL
,
5892 arg_image_policy
?: &image_policy_container
,
5893 dissect_image_flags
,
5896 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5897 log_notice("Note that the disk image needs to\n"
5898 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5899 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5900 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5901 " d) or contain a file system without a partition table\n"
5902 "in order to be bootable with systemd-nspawn.");
5908 r
= dissected_image_load_verity_sig_partition(
5911 &arg_verity_settings
);
5915 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5916 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5917 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5919 r
= dissected_image_decrypt_interactively(
5922 &arg_verity_settings
,
5927 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5928 if (remove_image
&& unlink(arg_image
) >= 0)
5929 remove_image
= false;
5931 if (arg_architecture
< 0)
5932 arg_architecture
= dissected_image_architecture(dissected_image
);
5935 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5939 if (arg_console_mode
< 0)
5940 arg_console_mode
= isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
) ?
5941 CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5943 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5947 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5948 arg_machine
, arg_image
?: arg_directory
);
5950 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, SIGRTMIN
+18, -1) >= 0);
5952 r
= make_reaper_process(true);
5954 log_error_errno(r
, "Failed to become subreaper: %m");
5958 if (arg_expose_ports
) {
5959 r
= fw_ctx_new(&fw_ctx
);
5961 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5964 expose_args
.fw_ctx
= fw_ctx
;
5967 r
= run_container(dissected_image
,
5969 veth_name
, &veth_created
,
5970 &expose_args
, &master
,
5977 (void) sd_notify(false,
5978 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5979 "STOPPING=1\nSTATUS=Terminating...");
5982 (void) kill(pid
, SIGKILL
);
5984 /* Try to flush whatever is still queued in the pty */
5986 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5987 master
= safe_close(master
);
5991 (void) wait_for_terminate(pid
, NULL
);
5995 if (remove_directory
&& arg_directory
) {
5998 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
6000 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
6003 if (remove_image
&& arg_image
) {
6004 if (unlink(arg_image
) < 0)
6005 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
6008 if (remove_tmprootdir
) {
6009 if (rmdir(tmprootdir
) < 0)
6010 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
6016 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
6017 (void) rm_rf(p
, REMOVE_ROOT
);
6019 p
= strjoina("/run/systemd/nspawn/unix-export/", arg_machine
);
6020 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
6024 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
6025 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
6028 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
6029 (void) remove_bridge(arg_network_zone
);
6031 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
6032 expose_port_free_all(arg_expose_ports
);
6033 rlimit_free_all(arg_rlimit
);
6034 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
6042 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);