1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <linux/loop.h>
8 #include <selinux/selinux.h>
12 #include <sys/ioctl.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
15 #include <sys/types.h>
21 #include "sd-daemon.h"
24 #include "alloc-util.h"
25 #include "ether-addr-util.h"
27 #include "base-filesystem.h"
28 #include "blkid-util.h"
29 #include "btrfs-util.h"
31 #include "bus-error.h"
32 #include "bus-locator.h"
35 #include "capability-util.h"
36 #include "cgroup-util.h"
38 #include "common-signal.h"
40 #include "cpu-set-util.h"
41 #include "creds-util.h"
42 #include "dev-setup.h"
43 #include "discover-image.h"
44 #include "dissect-image.h"
50 #include "format-util.h"
53 #include "hexdecoct.h"
54 #include "hostname-setup.h"
55 #include "hostname-util.h"
56 #include "id128-util.h"
59 #include "loop-util.h"
60 #include "loopback-setup.h"
61 #include "machine-credential.h"
63 #include "main-func.h"
64 #include "missing_sched.h"
66 #include "mount-util.h"
67 #include "mountpoint-util.h"
68 #include "namespace-util.h"
69 #include "netlink-util.h"
70 #include "nspawn-bind-user.h"
71 #include "nspawn-cgroup.h"
72 #include "nspawn-def.h"
73 #include "nspawn-expose-ports.h"
74 #include "nspawn-mount.h"
75 #include "nspawn-network.h"
76 #include "nspawn-oci.h"
77 #include "nspawn-patch-uid.h"
78 #include "nspawn-register.h"
79 #include "nspawn-seccomp.h"
80 #include "nspawn-settings.h"
81 #include "nspawn-setuid.h"
82 #include "nspawn-stub-pid1.h"
83 #include "nspawn-util.h"
85 #include "nulstr-util.h"
88 #include "parse-argument.h"
89 #include "parse-util.h"
90 #include "pretty-print.h"
91 #include "process-util.h"
93 #include "random-util.h"
94 #include "raw-clone.h"
95 #include "resolve-util.h"
96 #include "rlimit-util.h"
98 #include "seccomp-util.h"
99 #include "selinux-util.h"
100 #include "signal-util.h"
101 #include "socket-util.h"
102 #include "stat-util.h"
103 #include "stdio-util.h"
104 #include "string-table.h"
105 #include "string-util.h"
107 #include "sysctl-util.h"
108 #include "terminal-util.h"
109 #include "tmpfile-util.h"
110 #include "umask-util.h"
111 #include "unit-name.h"
112 #include "user-util.h"
115 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
116 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
117 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
119 #define EXIT_FORCE_RESTART 133
121 typedef enum ContainerStatus
{
122 CONTAINER_TERMINATED
,
126 static char *arg_directory
= NULL
;
127 static char *arg_template
= NULL
;
128 static char *arg_chdir
= NULL
;
129 static char *arg_pivot_root_new
= NULL
;
130 static char *arg_pivot_root_old
= NULL
;
131 static char *arg_user
= NULL
;
132 static uid_t arg_uid
= UID_INVALID
;
133 static gid_t arg_gid
= GID_INVALID
;
134 static gid_t
* arg_supplementary_gids
= NULL
;
135 static size_t arg_n_supplementary_gids
= 0;
136 static sd_id128_t arg_uuid
= {};
137 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
138 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
139 static const char *arg_selinux_context
= NULL
;
140 static const char *arg_selinux_apifs_context
= NULL
;
141 static char *arg_slice
= NULL
;
142 static bool arg_private_network
= false;
143 static bool arg_read_only
= false;
144 static StartMode arg_start_mode
= START_PID1
;
145 static bool arg_ephemeral
= false;
146 static LinkJournal arg_link_journal
= LINK_AUTO
;
147 static bool arg_link_journal_try
= false;
148 static uint64_t arg_caps_retain
=
149 (1ULL << CAP_AUDIT_CONTROL
) |
150 (1ULL << CAP_AUDIT_WRITE
) |
151 (1ULL << CAP_CHOWN
) |
152 (1ULL << CAP_DAC_OVERRIDE
) |
153 (1ULL << CAP_DAC_READ_SEARCH
) |
154 (1ULL << CAP_FOWNER
) |
155 (1ULL << CAP_FSETID
) |
156 (1ULL << CAP_IPC_OWNER
) |
158 (1ULL << CAP_LEASE
) |
159 (1ULL << CAP_LINUX_IMMUTABLE
) |
160 (1ULL << CAP_MKNOD
) |
161 (1ULL << CAP_NET_BIND_SERVICE
) |
162 (1ULL << CAP_NET_BROADCAST
) |
163 (1ULL << CAP_NET_RAW
) |
164 (1ULL << CAP_SETFCAP
) |
165 (1ULL << CAP_SETGID
) |
166 (1ULL << CAP_SETPCAP
) |
167 (1ULL << CAP_SETUID
) |
168 (1ULL << CAP_SYS_ADMIN
) |
169 (1ULL << CAP_SYS_BOOT
) |
170 (1ULL << CAP_SYS_CHROOT
) |
171 (1ULL << CAP_SYS_NICE
) |
172 (1ULL << CAP_SYS_PTRACE
) |
173 (1ULL << CAP_SYS_RESOURCE
) |
174 (1ULL << CAP_SYS_TTY_CONFIG
);
175 static uint64_t arg_caps_ambient
= 0;
176 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
177 static CustomMount
*arg_custom_mounts
= NULL
;
178 static size_t arg_n_custom_mounts
= 0;
179 static char **arg_setenv
= NULL
;
180 static bool arg_quiet
= false;
181 static bool arg_register
= true;
182 static bool arg_keep_unit
= false;
183 static char **arg_network_interfaces
= NULL
;
184 static char **arg_network_macvlan
= NULL
;
185 static char **arg_network_ipvlan
= NULL
;
186 static bool arg_network_veth
= false;
187 static char **arg_network_veth_extra
= NULL
;
188 static char *arg_network_bridge
= NULL
;
189 static char *arg_network_zone
= NULL
;
190 static char *arg_network_namespace_path
= NULL
;
191 struct ether_addr arg_network_provided_mac
= {};
192 static PagerFlags arg_pager_flags
= 0;
193 static unsigned long arg_personality
= PERSONALITY_INVALID
;
194 static char *arg_image
= NULL
;
195 static char *arg_oci_bundle
= NULL
;
196 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
197 static ExposePort
*arg_expose_ports
= NULL
;
198 static char **arg_property
= NULL
;
199 static sd_bus_message
*arg_property_message
= NULL
;
200 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
201 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
202 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
203 static int arg_kill_signal
= 0;
204 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
205 static SettingsMask arg_settings_mask
= 0;
206 static int arg_settings_trusted
= -1;
207 static char **arg_parameters
= NULL
;
208 static const char *arg_container_service_name
= "systemd-nspawn";
209 static bool arg_notify_ready
= false;
210 static bool arg_use_cgns
= true;
211 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
212 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
213 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
214 static char **arg_syscall_allow_list
= NULL
;
215 static char **arg_syscall_deny_list
= NULL
;
217 static scmp_filter_ctx arg_seccomp
= NULL
;
219 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
220 static bool arg_no_new_privileges
= false;
221 static int arg_oom_score_adjust
= 0;
222 static bool arg_oom_score_adjust_set
= false;
223 static CPUSet arg_cpu_set
= {};
224 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
225 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
226 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
227 static DeviceNode
* arg_extra_nodes
= NULL
;
228 static size_t arg_n_extra_nodes
= 0;
229 static char **arg_sysctl
= NULL
;
230 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
231 static MachineCredentialContext arg_credentials
= {};
232 static char **arg_bind_user
= NULL
;
233 static bool arg_suppress_sync
= false;
234 static char *arg_settings_filename
= NULL
;
235 static Architecture arg_architecture
= _ARCHITECTURE_INVALID
;
236 static ImagePolicy
*arg_image_policy
= NULL
;
238 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
260 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
262 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
265 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
267 STATIC_DESTRUCTOR_REGISTER(arg_credentials
, machine_credential_context_done
);
268 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
269 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
270 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
271 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
272 STATIC_DESTRUCTOR_REGISTER(arg_image_policy
, image_policy_freep
);
274 static int handle_arg_console(const char *arg
) {
275 if (streq(arg
, "help")) {
284 if (streq(arg
, "interactive"))
285 arg_console_mode
= CONSOLE_INTERACTIVE
;
286 else if (streq(arg
, "read-only"))
287 arg_console_mode
= CONSOLE_READ_ONLY
;
288 else if (streq(arg
, "passive"))
289 arg_console_mode
= CONSOLE_PASSIVE
;
290 else if (streq(arg
, "pipe")) {
291 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
292 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
293 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
294 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
295 "Proceeding anyway.");
297 arg_console_mode
= CONSOLE_PIPE
;
298 } else if (streq(arg
, "autopipe")) {
299 if (isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
))
300 arg_console_mode
= CONSOLE_INTERACTIVE
;
302 arg_console_mode
= CONSOLE_PIPE
;
304 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
306 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
310 static int help(void) {
311 _cleanup_free_
char *link
= NULL
;
314 pager_open(arg_pager_flags
);
316 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
320 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
321 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
322 " -h --help Show this help\n"
323 " --version Print version string\n"
324 " -q --quiet Do not show status information\n"
325 " --no-pager Do not pipe output into a pager\n"
326 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
328 " -D --directory=PATH Root directory for the container\n"
329 " --template=PATH Initialize root directory from template directory,\n"
331 " -x --ephemeral Run container with snapshot of root directory, and\n"
332 " remove it after exit\n"
333 " -i --image=PATH Root file system disk image (or device node) for\n"
335 " --image-policy=POLICY Specify disk image dissection policy\n"
336 " --oci-bundle=PATH OCI bundle directory\n"
337 " --read-only Mount the root directory read-only\n"
338 " --volatile[=MODE] Run the system in volatile mode\n"
339 " --root-hash=HASH Specify verity root hash for root disk image\n"
340 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
341 " as a DER encoded PKCS7, either as a path to a file\n"
342 " or as an ASCII base64 encoded string prefixed by\n"
344 " --verity-data=PATH Specify hash device for verity\n"
345 " --pivot-root=PATH[:PATH]\n"
346 " Pivot root to given directory in the container\n\n"
347 "%3$sExecution:%4$s\n"
348 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
349 " -b --boot Boot up full system (i.e. invoke init)\n"
350 " --chdir=PATH Set working directory in the container\n"
351 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
352 " -u --user=USER Run the command under specified user or UID\n"
353 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
354 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
355 " --suppress-sync=BOOLEAN\n"
356 " Suppress any form of disk data synchronization\n\n"
357 "%3$sSystem Identity:%4$s\n"
358 " -M --machine=NAME Set the machine name for the container\n"
359 " --hostname=NAME Override the hostname for the container\n"
360 " --uuid=UUID Set a specific machine UUID for the container\n\n"
361 "%3$sProperties:%4$s\n"
362 " -S --slice=SLICE Place the container in the specified slice\n"
363 " --property=NAME=VALUE Set scope unit property\n"
364 " --register=BOOLEAN Register container as machine\n"
365 " --keep-unit Do not register a scope for the machine, reuse\n"
366 " the service unit nspawn is running in\n\n"
367 "%3$sUser Namespacing:%4$s\n"
368 " --private-users=no Run without user namespacing\n"
369 " --private-users=yes|pick|identity\n"
370 " Run within user namespace, autoselect UID/GID range\n"
371 " --private-users=UIDBASE[:NUIDS]\n"
372 " Similar, but with user configured UID/GID range\n"
373 " --private-users-ownership=MODE\n"
374 " Adjust ('chown') or map ('map') OS tree ownership\n"
375 " to private UID/GID range\n"
376 " -U Equivalent to --private-users=pick and\n"
377 " --private-users-ownership=auto\n\n"
378 "%3$sNetworking:%4$s\n"
379 " --private-network Disable network in container\n"
380 " --network-interface=HOSTIF[:CONTAINERIF]\n"
381 " Assign an existing network interface to the\n"
383 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
384 " Create a macvlan network interface based on an\n"
385 " existing network interface to the container\n"
386 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
387 " Create an ipvlan network interface based on an\n"
388 " existing network interface to the container\n"
389 " -n --network-veth Add a virtual Ethernet connection between host\n"
391 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
392 " Add an additional virtual Ethernet link between\n"
393 " host and container\n"
394 " --network-bridge=INTERFACE\n"
395 " Add a virtual Ethernet connection to the container\n"
396 " and attach it to an existing bridge on the host\n"
397 " --network-zone=NAME Similar, but attach the new interface to an\n"
398 " an automatically managed bridge interface\n"
399 " --network-namespace-path=PATH\n"
400 " Set network namespace to the one represented by\n"
401 " the specified kernel namespace file node\n"
402 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
403 " Expose a container IP port on the host\n\n"
404 "%3$sSecurity:%4$s\n"
405 " --capability=CAP In addition to the default, retain specified\n"
407 " --drop-capability=CAP Drop the specified capability from the default set\n"
408 " --ambient-capability=CAP\n"
409 " Sets the specified capability for the started\n"
410 " process. Not useful if booting a machine.\n"
411 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
412 " --system-call-filter=LIST|~LIST\n"
413 " Permit/prohibit specific system calls\n"
414 " -Z --selinux-context=SECLABEL\n"
415 " Set the SELinux security context to be used by\n"
416 " processes in the container\n"
417 " -L --selinux-apifs-context=SECLABEL\n"
418 " Set the SELinux security context to be used by\n"
419 " API/tmpfs file systems in the container\n\n"
420 "%3$sResources:%4$s\n"
421 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
422 " --oom-score-adjust=VALUE\n"
423 " Adjust the OOM score value for the payload\n"
424 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
425 " --personality=ARCH Pick personality for this container\n\n"
426 "%3$sIntegration:%4$s\n"
427 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
428 " --timezone=MODE Select mode of /etc/localtime initialization\n"
429 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
430 " host, try-guest, try-host\n"
431 " -j Equivalent to --link-journal=try-guest\n\n"
433 " --bind=PATH[:PATH[:OPTIONS]]\n"
434 " Bind mount a file or directory from the host into\n"
436 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
437 " Similar, but creates a read-only bind mount\n"
438 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
440 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
441 " --overlay=PATH[:PATH...]:PATH\n"
442 " Create an overlay mount from the host to \n"
444 " --overlay-ro=PATH[:PATH...]:PATH\n"
445 " Similar, but creates a read-only overlay mount\n"
446 " --bind-user=NAME Bind user from host to container\n\n"
447 "%3$sInput/Output:%4$s\n"
448 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
449 " set up for the container.\n"
450 " -P --pipe Equivalent to --console=pipe\n\n"
451 "%3$sCredentials:%4$s\n"
452 " --set-credential=ID:VALUE\n"
453 " Pass a credential with literal value to container.\n"
454 " --load-credential=ID:PATH\n"
455 " Load credential to pass to container from file or\n"
456 " AF_UNIX stream socket.\n"
457 "\nSee the %2$s for details.\n",
458 program_invocation_short_name
,
468 static int custom_mount_check_all(void) {
471 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
472 CustomMount
*m
= &arg_custom_mounts
[i
];
474 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
475 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
476 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
477 "--private-users-ownership=own may not be combined with custom root mounts.");
478 if (arg_uid_shift
== UID_INVALID
)
479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
480 "--private-users with automatic UID shift may not be combined with custom root mounts.");
487 static int detect_unified_cgroup_hierarchy_from_environment(void) {
488 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
491 /* Allow the user to control whether the unified hierarchy is used */
495 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
496 var
= "UNIFIED_CGROUP_HIERARCHY";
501 r
= parse_boolean(e
);
503 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
505 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
507 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
513 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
516 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
517 * in the image actually supports. */
518 r
= cg_all_unified();
520 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
522 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
523 * routine only detects 231, so we'll have a false negative here for 230. */
524 r
= systemd_installation_has_version(directory
, "230");
526 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
528 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
530 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
531 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
532 /* Mixed cgroup hierarchy support was added in 233 */
533 r
= systemd_installation_has_version(directory
, "233");
535 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
537 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
539 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
541 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
543 log_debug("Using %s hierarchy for container.",
544 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
545 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
550 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
555 _cleanup_free_
char *t
= NULL
;
557 r
= extract_first_word(&spec
, &t
, ",", 0);
559 return log_error_errno(r
, "Failed to parse capability %s.", t
);
563 if (streq(t
, "help")) {
564 for (int i
= 0; i
< capability_list_length(); i
++) {
567 name
= capability_to_name(i
);
578 r
= capability_from_name(t
);
580 return log_error_errno(r
, "Failed to parse capability %s.", t
);
587 return 1; /* continue */
590 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
593 r
= getenv_bool(name
);
597 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
599 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
600 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
604 static int parse_mount_settings_env(void) {
608 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
609 if (r
< 0 && r
!= -ENXIO
)
610 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
612 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
614 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
615 if (streq_ptr(e
, "network"))
616 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
619 r
= parse_boolean(e
);
621 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
623 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
624 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
630 static int parse_environment(void) {
634 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
637 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
640 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
643 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
647 r
= parse_mount_settings_env();
651 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
652 * even if it is supported. If not supported, it has no effect. */
653 if (!cg_ns_supported())
654 arg_use_cgns
= false;
656 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
659 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
663 arg_use_cgns
= r
> 0;
664 arg_settings_mask
|= SETTING_USE_CGNS
;
668 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
670 arg_container_service_name
= e
;
672 e
= getenv("SYSTEMD_NSPAWN_NETWORK_MAC");
674 r
= parse_ether_addr(e
, &arg_network_provided_mac
);
676 return log_error_errno(r
, "Failed to parse provided MAC address via environment variable");
679 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
681 arg_suppress_sync
= r
;
682 else if (r
!= -ENXIO
)
683 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
685 return detect_unified_cgroup_hierarchy_from_environment();
688 static int parse_argv(int argc
, char *argv
[]) {
695 ARG_AMBIENT_CAPABILITY
,
707 ARG_NETWORK_INTERFACE
,
712 ARG_NETWORK_VETH_EXTRA
,
713 ARG_NETWORK_NAMESPACE_PATH
,
723 ARG_PRIVATE_USERS_CHOWN
,
724 ARG_PRIVATE_USERS_OWNERSHIP
,
729 ARG_SYSTEM_CALL_FILTER
,
732 ARG_NO_NEW_PRIVILEGES
,
733 ARG_OOM_SCORE_ADJUST
,
748 static const struct option options
[] = {
749 { "help", no_argument
, NULL
, 'h' },
750 { "version", no_argument
, NULL
, ARG_VERSION
},
751 { "directory", required_argument
, NULL
, 'D' },
752 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
753 { "ephemeral", no_argument
, NULL
, 'x' },
754 { "user", required_argument
, NULL
, 'u' },
755 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
756 { "as-pid2", no_argument
, NULL
, 'a' },
757 { "boot", no_argument
, NULL
, 'b' },
758 { "uuid", required_argument
, NULL
, ARG_UUID
},
759 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
760 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
761 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
762 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
763 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
764 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
765 { "bind", required_argument
, NULL
, ARG_BIND
},
766 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
767 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
768 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
769 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
770 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
771 { "machine", required_argument
, NULL
, 'M' },
772 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
773 { "slice", required_argument
, NULL
, 'S' },
774 { "setenv", required_argument
, NULL
, 'E' },
775 { "selinux-context", required_argument
, NULL
, 'Z' },
776 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
777 { "quiet", no_argument
, NULL
, 'q' },
778 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
779 { "register", required_argument
, NULL
, ARG_REGISTER
},
780 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
781 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
782 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
783 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
784 { "network-veth", no_argument
, NULL
, 'n' },
785 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
786 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
787 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
788 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
789 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
790 { "image", required_argument
, NULL
, 'i' },
791 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
792 { "port", required_argument
, NULL
, 'p' },
793 { "property", required_argument
, NULL
, ARG_PROPERTY
},
794 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
795 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
796 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
797 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
798 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
799 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
800 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
801 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
802 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
803 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
804 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
805 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
806 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
807 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
808 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
809 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
810 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
811 { "console", required_argument
, NULL
, ARG_CONSOLE
},
812 { "pipe", no_argument
, NULL
, ARG_PIPE
},
813 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
814 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
815 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
816 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
817 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
818 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
819 { "image-policy", required_argument
, NULL
, ARG_IMAGE_POLICY
},
824 uint64_t plus
= 0, minus
= 0;
825 bool mask_all_settings
= false, mask_no_settings
= false;
830 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
831 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
833 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
843 r
= parse_path_argument(optarg
, false, &arg_directory
);
847 arg_settings_mask
|= SETTING_DIRECTORY
;
851 r
= parse_path_argument(optarg
, false, &arg_template
);
855 arg_settings_mask
|= SETTING_DIRECTORY
;
859 r
= parse_path_argument(optarg
, false, &arg_image
);
863 arg_settings_mask
|= SETTING_DIRECTORY
;
867 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
874 arg_ephemeral
= true;
875 arg_settings_mask
|= SETTING_EPHEMERAL
;
879 r
= free_and_strdup(&arg_user
, optarg
);
883 arg_settings_mask
|= SETTING_USER
;
886 case ARG_NETWORK_ZONE
: {
887 _cleanup_free_
char *j
= NULL
;
889 j
= strjoin("vz-", optarg
);
893 if (!ifname_valid(j
))
894 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
895 "Network zone name not valid: %s", j
);
897 free_and_replace(arg_network_zone
, j
);
899 arg_network_veth
= true;
900 arg_private_network
= true;
901 arg_settings_mask
|= SETTING_NETWORK
;
905 case ARG_NETWORK_BRIDGE
:
907 if (!ifname_valid(optarg
))
908 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
909 "Bridge interface name not valid: %s", optarg
);
911 r
= free_and_strdup(&arg_network_bridge
, optarg
);
917 arg_network_veth
= true;
918 arg_private_network
= true;
919 arg_settings_mask
|= SETTING_NETWORK
;
922 case ARG_NETWORK_VETH_EXTRA
:
923 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
925 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
927 arg_private_network
= true;
928 arg_settings_mask
|= SETTING_NETWORK
;
931 case ARG_NETWORK_INTERFACE
:
932 r
= interface_pair_parse(&arg_network_interfaces
, optarg
);
936 arg_private_network
= true;
937 arg_settings_mask
|= SETTING_NETWORK
;
940 case ARG_NETWORK_MACVLAN
:
941 r
= macvlan_pair_parse(&arg_network_macvlan
, optarg
);
945 arg_private_network
= true;
946 arg_settings_mask
|= SETTING_NETWORK
;
949 case ARG_NETWORK_IPVLAN
:
950 r
= ipvlan_pair_parse(&arg_network_ipvlan
, optarg
);
955 case ARG_PRIVATE_NETWORK
:
956 arg_private_network
= true;
957 arg_settings_mask
|= SETTING_NETWORK
;
960 case ARG_NETWORK_NAMESPACE_PATH
:
961 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
965 arg_settings_mask
|= SETTING_NETWORK
;
969 if (arg_start_mode
== START_PID2
)
970 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
971 "--boot and --as-pid2 may not be combined.");
973 arg_start_mode
= START_BOOT
;
974 arg_settings_mask
|= SETTING_START_MODE
;
978 if (arg_start_mode
== START_BOOT
)
979 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
980 "--boot and --as-pid2 may not be combined.");
982 arg_start_mode
= START_PID2
;
983 arg_settings_mask
|= SETTING_START_MODE
;
987 r
= id128_from_string_nonzero(optarg
, &arg_uuid
);
989 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
990 "Machine UUID may not be all zeroes.");
992 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
994 arg_settings_mask
|= SETTING_MACHINE_ID
;
998 _cleanup_free_
char *mangled
= NULL
;
1000 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1004 free_and_replace(arg_slice
, mangled
);
1005 arg_settings_mask
|= SETTING_SLICE
;
1010 if (isempty(optarg
))
1011 arg_machine
= mfree(arg_machine
);
1013 if (!hostname_is_valid(optarg
, 0))
1014 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1015 "Invalid machine name: %s", optarg
);
1017 r
= free_and_strdup(&arg_machine
, optarg
);
1024 if (isempty(optarg
))
1025 arg_hostname
= mfree(arg_hostname
);
1027 if (!hostname_is_valid(optarg
, 0))
1028 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1029 "Invalid hostname: %s", optarg
);
1031 r
= free_and_strdup(&arg_hostname
, optarg
);
1036 arg_settings_mask
|= SETTING_HOSTNAME
;
1040 arg_selinux_context
= optarg
;
1044 arg_selinux_apifs_context
= optarg
;
1048 arg_read_only
= true;
1049 arg_settings_mask
|= SETTING_READ_ONLY
;
1052 case ARG_AMBIENT_CAPABILITY
: {
1054 r
= parse_capability_spec(optarg
, &m
);
1057 arg_caps_ambient
|= m
;
1058 arg_settings_mask
|= SETTING_CAPABILITY
;
1061 case ARG_CAPABILITY
:
1062 case ARG_DROP_CAPABILITY
: {
1064 r
= parse_capability_spec(optarg
, &m
);
1068 if (c
== ARG_CAPABILITY
)
1072 arg_settings_mask
|= SETTING_CAPABILITY
;
1075 case ARG_NO_NEW_PRIVILEGES
:
1076 r
= parse_boolean(optarg
);
1078 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1080 arg_no_new_privileges
= r
;
1081 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1085 arg_link_journal
= LINK_GUEST
;
1086 arg_link_journal_try
= true;
1087 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1090 case ARG_LINK_JOURNAL
:
1091 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1093 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1095 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1100 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1102 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1104 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1108 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1110 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1112 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1116 case ARG_OVERLAY_RO
:
1117 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1118 if (r
== -EADDRNOTAVAIL
)
1119 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1121 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1123 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1126 case ARG_INACCESSIBLE
:
1127 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1129 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1131 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1135 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1137 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1139 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1146 case ARG_SHARE_SYSTEM
:
1147 /* We don't officially support this anymore, except for compat reasons. People should use the
1148 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1149 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1150 arg_clone_ns_flags
= 0;
1154 r
= parse_boolean(optarg
);
1156 log_error("Failed to parse --register= argument: %s", optarg
);
1164 arg_keep_unit
= true;
1167 case ARG_PERSONALITY
:
1169 arg_personality
= personality_from_string(optarg
);
1170 if (arg_personality
== PERSONALITY_INVALID
)
1171 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1172 "Unknown or unsupported personality '%s'.", optarg
);
1174 arg_settings_mask
|= SETTING_PERSONALITY
;
1180 arg_volatile_mode
= VOLATILE_YES
;
1181 else if (streq(optarg
, "help")) {
1182 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1187 m
= volatile_mode_from_string(optarg
);
1189 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1190 "Failed to parse --volatile= argument: %s", optarg
);
1192 arg_volatile_mode
= m
;
1195 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1199 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1201 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1203 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1205 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1209 if (strv_extend(&arg_property
, optarg
) < 0)
1214 case ARG_PRIVATE_USERS
: {
1219 else if (!in_charset(optarg
, DIGITS
))
1220 /* do *not* parse numbers as booleans */
1221 boolean
= parse_boolean(optarg
);
1226 /* no: User namespacing off */
1227 arg_userns_mode
= USER_NAMESPACE_NO
;
1228 arg_uid_shift
= UID_INVALID
;
1229 arg_uid_range
= UINT32_C(0x10000);
1230 } else if (boolean
> 0) {
1231 /* yes: User namespacing on, UID range is read from root dir */
1232 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1233 arg_uid_shift
= UID_INVALID
;
1234 arg_uid_range
= UINT32_C(0x10000);
1235 } else if (streq(optarg
, "pick")) {
1236 /* pick: User namespacing on, UID range is picked randomly */
1237 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1238 * implied by USER_NAMESPACE_PICK
1240 arg_uid_shift
= UID_INVALID
;
1241 arg_uid_range
= UINT32_C(0x10000);
1243 } else if (streq(optarg
, "identity")) {
1244 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1245 * itself, i.e. we don't actually map anything, but do take benefit of
1246 * isolation of capability sets. */
1247 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1249 arg_uid_range
= UINT32_C(0x10000);
1251 _cleanup_free_
char *buffer
= NULL
;
1252 const char *range
, *shift
;
1254 /* anything else: User namespacing on, UID range is explicitly configured */
1256 range
= strchr(optarg
, ':');
1258 buffer
= strndup(optarg
, range
- optarg
);
1264 r
= safe_atou32(range
, &arg_uid_range
);
1266 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1270 r
= parse_uid(shift
, &arg_uid_shift
);
1272 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1274 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1276 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1277 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1280 arg_settings_mask
|= SETTING_USERNS
;
1285 if (userns_supported()) {
1286 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1287 * implied by USER_NAMESPACE_PICK
1289 arg_uid_shift
= UID_INVALID
;
1290 arg_uid_range
= UINT32_C(0x10000);
1292 arg_settings_mask
|= SETTING_USERNS
;
1297 case ARG_PRIVATE_USERS_CHOWN
:
1298 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1300 arg_settings_mask
|= SETTING_USERNS
;
1303 case ARG_PRIVATE_USERS_OWNERSHIP
:
1304 if (streq(optarg
, "help")) {
1305 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1309 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1310 if (arg_userns_ownership
< 0)
1311 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1313 arg_settings_mask
|= SETTING_USERNS
;
1316 case ARG_KILL_SIGNAL
:
1317 if (streq(optarg
, "help")) {
1318 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1322 arg_kill_signal
= signal_from_string(optarg
);
1323 if (arg_kill_signal
< 0)
1324 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1326 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1331 /* no → do not read files
1332 * yes → read files, do not override cmdline, trust only subset
1333 * override → read files, override cmdline, trust only subset
1334 * trusted → read files, do not override cmdline, trust all
1337 r
= parse_boolean(optarg
);
1339 if (streq(optarg
, "trusted")) {
1340 mask_all_settings
= false;
1341 mask_no_settings
= false;
1342 arg_settings_trusted
= true;
1344 } else if (streq(optarg
, "override")) {
1345 mask_all_settings
= false;
1346 mask_no_settings
= true;
1347 arg_settings_trusted
= -1;
1349 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1352 mask_all_settings
= false;
1353 mask_no_settings
= false;
1354 arg_settings_trusted
= -1;
1357 mask_all_settings
= true;
1358 mask_no_settings
= false;
1359 arg_settings_trusted
= false;
1365 if (!path_is_absolute(optarg
))
1366 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1367 "Working directory %s is not an absolute path.", optarg
);
1369 r
= free_and_strdup(&arg_chdir
, optarg
);
1373 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1376 case ARG_PIVOT_ROOT
:
1377 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1379 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1381 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1384 case ARG_NOTIFY_READY
:
1385 r
= parse_boolean(optarg
);
1387 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1388 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1389 arg_notify_ready
= r
;
1390 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1393 case ARG_ROOT_HASH
: {
1394 _cleanup_free_
void *k
= NULL
;
1397 r
= unhexmem(optarg
, &k
, &l
);
1399 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1400 if (l
< sizeof(sd_id128_t
))
1401 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128-bit long: %s", optarg
);
1403 free_and_replace(arg_verity_settings
.root_hash
, k
);
1404 arg_verity_settings
.root_hash_size
= l
;
1408 case ARG_ROOT_HASH_SIG
: {
1413 if ((value
= startswith(optarg
, "base64:"))) {
1414 r
= unbase64mem(value
, &p
, &l
);
1416 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1419 r
= read_full_file(optarg
, (char**) &p
, &l
);
1421 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1424 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1425 arg_verity_settings
.root_hash_sig_size
= l
;
1429 case ARG_VERITY_DATA
:
1430 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1435 case ARG_SYSTEM_CALL_FILTER
: {
1439 negative
= optarg
[0] == '~';
1440 items
= negative
? optarg
+ 1 : optarg
;
1443 _cleanup_free_
char *word
= NULL
;
1445 r
= extract_first_word(&items
, &word
, NULL
, 0);
1451 return log_error_errno(r
, "Failed to parse system call filter: %m");
1454 r
= strv_extend(&arg_syscall_deny_list
, word
);
1456 r
= strv_extend(&arg_syscall_allow_list
, word
);
1461 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1467 _cleanup_free_
char *name
= NULL
;
1470 if (streq(optarg
, "help")) {
1471 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1475 eq
= strchr(optarg
, '=');
1477 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1478 "--rlimit= expects an '=' assignment.");
1480 name
= strndup(optarg
, eq
- optarg
);
1484 rl
= rlimit_from_string_harder(name
);
1486 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1488 if (!arg_rlimit
[rl
]) {
1489 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1490 if (!arg_rlimit
[rl
])
1494 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1496 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1498 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1502 case ARG_OOM_SCORE_ADJUST
:
1503 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1505 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1507 arg_oom_score_adjust_set
= true;
1508 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1511 case ARG_CPU_AFFINITY
: {
1514 r
= parse_cpu_set(optarg
, &cpuset
);
1516 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1518 cpu_set_reset(&arg_cpu_set
);
1519 arg_cpu_set
= cpuset
;
1520 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1524 case ARG_RESOLV_CONF
:
1525 if (streq(optarg
, "help")) {
1526 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1530 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1531 if (arg_resolv_conf
< 0)
1532 return log_error_errno(arg_resolv_conf
,
1533 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1535 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1539 if (streq(optarg
, "help")) {
1540 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1544 arg_timezone
= timezone_mode_from_string(optarg
);
1545 if (arg_timezone
< 0)
1546 return log_error_errno(arg_timezone
,
1547 "Failed to parse /etc/localtime mode: %s", optarg
);
1549 arg_settings_mask
|= SETTING_TIMEZONE
;
1553 r
= handle_arg_console(optarg
);
1560 r
= handle_arg_console("pipe");
1566 arg_pager_flags
|= PAGER_DISABLE
;
1569 case ARG_SET_CREDENTIAL
:
1570 r
= machine_credential_set(&arg_credentials
, optarg
);
1574 arg_settings_mask
|= SETTING_CREDENTIALS
;
1577 case ARG_LOAD_CREDENTIAL
:
1578 r
= machine_credential_load(&arg_credentials
, optarg
);
1582 arg_settings_mask
|= SETTING_CREDENTIALS
;
1586 if (!valid_user_group_name(optarg
, 0))
1587 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1589 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1592 arg_settings_mask
|= SETTING_BIND_USER
;
1595 case ARG_SUPPRESS_SYNC
:
1596 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1600 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1603 case ARG_IMAGE_POLICY
:
1604 r
= parse_image_policy_argument(optarg
, &arg_image_policy
);
1613 assert_not_reached();
1616 if (argc
> optind
) {
1617 strv_free(arg_parameters
);
1618 arg_parameters
= strv_copy(argv
+ optind
);
1619 if (!arg_parameters
)
1622 arg_settings_mask
|= SETTING_START_MODE
;
1625 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1626 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1627 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1628 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1630 arg_directory
= TAKE_PTR(arg_template
);
1632 arg_caps_retain
|= plus
;
1633 arg_caps_retain
|= arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0;
1634 arg_caps_retain
&= ~minus
;
1636 /* Make sure to parse environment before we reset the settings mask below */
1637 r
= parse_environment();
1641 /* Load all settings from .nspawn files */
1642 if (mask_no_settings
)
1643 arg_settings_mask
= 0;
1645 /* Don't load any settings from .nspawn files */
1646 if (mask_all_settings
)
1647 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1652 static int verify_arguments(void) {
1655 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1656 /* If we are running the stub init in the container, we don't need to look at what the init
1657 * in the container supports, because we are not using it. Let's immediately pick the right
1658 * setting based on the host system configuration.
1660 * We only do this, if the user didn't use an environment variable to override the detection.
1663 r
= cg_all_unified();
1665 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1667 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1668 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1669 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1671 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1674 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1675 arg_mount_settings
|= MOUNT_USE_USERNS
;
1677 if (arg_private_network
)
1678 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1680 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1681 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1682 arg_register
= false;
1683 if (arg_start_mode
!= START_PID1
)
1684 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1687 if (arg_userns_ownership
< 0)
1688 arg_userns_ownership
=
1689 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1690 USER_NAMESPACE_OWNERSHIP_OFF
;
1692 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1693 arg_kill_signal
= SIGRTMIN
+3;
1695 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 */
1696 arg_read_only
= true;
1698 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1699 arg_read_only
= true;
1701 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1702 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1703 * The latter is not technically a user session, but we don't need to labour the point. */
1704 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1706 if (arg_directory
&& arg_image
)
1707 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1709 if (arg_template
&& arg_image
)
1710 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1712 if (arg_template
&& !(arg_directory
|| arg_machine
))
1713 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1715 if (arg_ephemeral
&& arg_template
)
1716 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1718 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1719 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1721 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1722 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1724 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1725 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1726 "--read-only and --private-users-ownership=chown may not be combined.");
1728 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1729 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1730 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1731 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1732 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1734 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1735 * we need to error out, to avoid conflicts between different network options. */
1736 if (arg_network_namespace_path
&&
1737 (arg_network_interfaces
|| arg_network_macvlan
||
1738 arg_network_ipvlan
|| arg_network_veth_extra
||
1739 arg_network_bridge
|| arg_network_zone
||
1741 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1743 if (arg_network_bridge
&& arg_network_zone
)
1744 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1745 "--network-bridge= and --network-zone= may not be combined.");
1747 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1748 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1750 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1751 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1753 if (arg_expose_ports
&& !arg_private_network
)
1754 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1756 if (arg_caps_ambient
) {
1757 if (arg_caps_ambient
== UINT64_MAX
)
1758 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1760 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1761 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1763 if (arg_start_mode
== START_BOOT
)
1764 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1767 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1768 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1770 /* Drop duplicate --bind-user= entries */
1771 strv_uniq(arg_bind_user
);
1773 r
= custom_mount_check_all();
1780 static int verify_network_interfaces_initialized(void) {
1782 r
= test_network_interfaces_initialized(arg_network_interfaces
);
1786 r
= test_network_interfaces_initialized(arg_network_macvlan
);
1790 r
= test_network_interfaces_initialized(arg_network_ipvlan
);
1797 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1800 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1803 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1806 if (uid
!= UID_INVALID
) {
1807 uid
+= arg_uid_shift
;
1809 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1813 if (gid
!= GID_INVALID
) {
1814 gid
+= (gid_t
) arg_uid_shift
;
1816 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1820 return RET_NERRNO(lchown(p
, uid
, gid
));
1823 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1827 q
= prefix_roota(root
, path
);
1828 r
= RET_NERRNO(mkdir(q
, mode
));
1834 return userns_lchown(q
, uid
, gid
);
1837 static const char *timezone_from_path(const char *path
) {
1838 return PATH_STARTSWITH_SET(
1840 "../usr/share/zoneinfo/",
1841 "/usr/share/zoneinfo/");
1844 static bool etc_writable(void) {
1845 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1848 static int setup_timezone(const char *dest
) {
1849 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1850 const char *where
, *check
;
1856 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1857 r
= readlink_malloc("/etc/localtime", &p
);
1858 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1859 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1860 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1861 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1863 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1864 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1868 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1871 } else if (arg_timezone
== TIMEZONE_AUTO
)
1872 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1878 if (m
== TIMEZONE_OFF
)
1881 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1883 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1887 where
= strjoina(etc
, "/localtime");
1891 case TIMEZONE_DELETE
:
1892 if (unlink(where
) < 0)
1893 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1897 case TIMEZONE_SYMLINK
: {
1898 _cleanup_free_
char *q
= NULL
;
1899 const char *z
, *what
;
1901 z
= timezone_from_path(p
);
1903 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1907 r
= readlink_malloc(where
, &q
);
1908 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1909 return 0; /* Already pointing to the right place? Then do nothing .. */
1911 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1912 r
= chase(check
, dest
, 0, NULL
, NULL
);
1914 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1916 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1917 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1918 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1922 what
= strjoina("../usr/share/zoneinfo/", z
);
1923 if (symlink(what
, where
) < 0) {
1924 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1925 errno
, "Failed to correct timezone of container, ignoring: %m");
1935 case TIMEZONE_BIND
: {
1936 _cleanup_free_
char *resolved
= NULL
;
1939 found
= chase(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1941 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1945 if (found
== 0) /* missing? */
1946 (void) touch(resolved
);
1948 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1950 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1956 /* If mounting failed, try to copy */
1957 r
= copy_file_atomic("/etc/localtime", where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
1959 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1960 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1967 assert_not_reached();
1970 /* Fix permissions of the symlink or file copy we just created */
1971 r
= userns_lchown(where
, 0, 0);
1973 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1978 static int have_resolv_conf(const char *path
) {
1981 if (access(path
, F_OK
) < 0) {
1982 if (errno
== ENOENT
)
1985 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1991 static int resolved_listening(void) {
1992 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1993 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1994 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1997 /* Check if resolved is listening */
1999 r
= sd_bus_open_system(&bus
);
2001 return log_debug_errno(r
, "Failed to open system bus: %m");
2003 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2005 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2009 r
= bus_get_property_string(bus
, bus_resolve_mgr
, "DNSStubListener", &error
, &dns_stub_listener_mode
);
2011 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2013 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2016 static int setup_resolv_conf(const char *dest
) {
2017 _cleanup_free_
char *etc
= NULL
;
2018 const char *where
, *what
;
2024 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2025 if (arg_private_network
)
2026 m
= RESOLV_CONF_OFF
;
2027 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2028 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2029 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2030 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2032 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2035 m
= arg_resolv_conf
;
2037 if (m
== RESOLV_CONF_OFF
)
2040 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2042 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2046 where
= strjoina(etc
, "/resolv.conf");
2048 if (m
== RESOLV_CONF_DELETE
) {
2049 if (unlink(where
) < 0)
2050 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2055 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2056 what
= PRIVATE_STATIC_RESOLV_CONF
;
2057 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2058 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2059 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2060 what
= PRIVATE_STUB_RESOLV_CONF
;
2062 what
= "/etc/resolv.conf";
2064 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2065 _cleanup_free_
char *resolved
= NULL
;
2068 found
= chase(where
, dest
, CHASE_NONEXISTENT
|CHASE_NOFOLLOW
, &resolved
, NULL
);
2070 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2074 if (found
== 0) /* missing? */
2075 (void) touch(resolved
);
2077 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2079 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2081 /* If that didn't work, let's copy the file */
2084 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2085 r
= copy_file_atomic(what
, where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2087 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, COPY_REFLINK
);
2089 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2090 * resolved or something similar runs inside and the symlink points there.
2092 * If the disk image is read-only, there's also no point in complaining.
2094 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2095 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2096 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2100 r
= userns_lchown(where
, 0, 0);
2102 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2107 static int setup_boot_id(void) {
2108 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2109 _cleanup_free_
char *path
= NULL
;
2110 sd_id128_t rnd
= SD_ID128_NULL
;
2114 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2116 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2118 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2120 r
= sd_id128_randomize(&rnd
);
2122 return log_error_errno(r
, "Failed to generate random boot id: %m");
2124 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2126 return log_error_errno(r
, "Failed to write boot id: %m");
2128 from
= TAKE_PTR(path
);
2129 to
= "/proc/sys/kernel/random/boot_id";
2131 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2135 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2138 static int copy_devnodes(const char *dest
) {
2139 static const char devnodes
[] =
2152 BLOCK_WITH_UMASK(0000);
2154 /* Create /dev/net, so that we can create /dev/net/tun in it */
2155 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2156 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2158 NULSTR_FOREACH(d
, devnodes
) {
2159 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2162 from
= path_join("/dev/", d
);
2166 to
= path_join(dest
, from
);
2170 if (stat(from
, &st
) < 0) {
2172 if (errno
!= ENOENT
)
2173 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2175 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2176 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2177 "%s is not a char or block device, cannot copy.", from
);
2179 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2181 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2182 /* Explicitly warn the user when /dev is already populated. */
2183 if (errno
== EEXIST
)
2184 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
);
2186 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2188 /* Some systems abusively restrict mknod but allow bind mounts. */
2191 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2192 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2194 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2197 r
= userns_lchown(to
, 0, 0);
2199 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2201 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2205 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2207 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2209 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2212 prefixed
= path_join(dest
, sl
);
2216 t
= path_join("..", d
);
2220 if (symlink(t
, prefixed
) < 0)
2221 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2228 static int make_extra_nodes(const char *dest
) {
2232 BLOCK_WITH_UMASK(0000);
2234 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2235 _cleanup_free_
char *path
= NULL
;
2236 DeviceNode
*n
= arg_extra_nodes
+ i
;
2238 path
= path_join(dest
, n
->path
);
2242 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2243 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2245 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2247 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2253 static int setup_pts(const char *dest
) {
2254 _cleanup_free_
char *options
= NULL
;
2259 if (arg_selinux_apifs_context
)
2260 (void) asprintf(&options
,
2261 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2262 arg_uid_shift
+ TTY_GID
,
2263 arg_selinux_apifs_context
);
2266 (void) asprintf(&options
,
2267 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2268 arg_uid_shift
+ TTY_GID
);
2273 /* Mount /dev/pts itself */
2274 p
= prefix_roota(dest
, "/dev/pts");
2275 r
= RET_NERRNO(mkdir(p
, 0755));
2277 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2279 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2282 r
= userns_lchown(p
, 0, 0);
2284 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2286 /* Create /dev/ptmx symlink */
2287 p
= prefix_roota(dest
, "/dev/ptmx");
2288 if (symlink("pts/ptmx", p
) < 0)
2289 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2290 r
= userns_lchown(p
, 0, 0);
2292 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2294 /* And fix /dev/pts/ptmx ownership */
2295 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2296 r
= userns_lchown(p
, 0, 0);
2298 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2303 static int setup_stdio_as_dev_console(void) {
2304 _cleanup_close_
int terminal
= -EBADF
;
2307 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2308 * explicitly, if we are configured to. */
2309 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2311 return log_error_errno(terminal
, "Failed to open console: %m");
2313 /* Make sure we can continue logging to the original stderr, even if
2314 * stderr points elsewhere now */
2315 r
= log_dup_console();
2317 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2319 /* invalidates 'terminal' on success and failure */
2320 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2323 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2328 static int setup_dev_console(const char *console
) {
2329 _cleanup_free_
char *p
= NULL
;
2332 /* Create /dev/console symlink */
2333 r
= path_make_relative("/dev", console
, &p
);
2335 return log_error_errno(r
, "Failed to create relative path: %m");
2337 if (symlink(p
, "/dev/console") < 0)
2338 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2343 static int setup_keyring(void) {
2344 key_serial_t keyring
;
2346 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2347 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2348 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2349 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2350 * into the container. */
2352 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2353 if (keyring
== -1) {
2354 if (errno
== ENOSYS
)
2355 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2356 else if (ERRNO_IS_PRIVILEGE(errno
))
2357 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2359 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2365 int make_run_host(const char *root
) {
2370 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2372 return log_error_errno(r
, "Failed to create /run/host/: %m");
2377 static int setup_credentials(const char *root
) {
2381 if (arg_credentials
.n_credentials
== 0)
2384 r
= make_run_host(root
);
2388 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2390 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2392 q
= prefix_roota(root
, "/run/host/credentials");
2393 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2397 FOREACH_ARRAY(cred
, arg_credentials
.credentials
, arg_credentials
.n_credentials
) {
2398 _cleanup_free_
char *j
= NULL
;
2399 _cleanup_close_
int fd
= -EBADF
;
2401 j
= path_join(q
, cred
->id
);
2405 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2407 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2409 r
= loop_write(fd
, cred
->data
, cred
->size
);
2411 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2413 if (fchmod(fd
, 0400) < 0)
2414 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2416 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2417 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2418 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2422 if (chmod(q
, 0500) < 0)
2423 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2425 r
= userns_lchown(q
, 0, 0);
2429 /* Make both mount and superblock read-only now */
2430 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2434 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2437 static int setup_kmsg(int fd_inner_socket
) {
2438 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2439 _cleanup_free_
char *fifo
= NULL
;
2440 _cleanup_close_
int fd
= -EBADF
;
2443 assert(fd_inner_socket
>= 0);
2445 BLOCK_WITH_UMASK(0000);
2447 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2448 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2449 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2450 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2452 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2454 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2456 if (mkfifo(fifo
, 0600) < 0)
2457 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2459 from
= TAKE_PTR(fifo
);
2461 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2465 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2467 return log_error_errno(errno
, "Failed to open fifo: %m");
2469 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2470 r
= send_one_fd(fd_inner_socket
, fd
, 0);
2472 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2478 union in_addr_union address4
;
2479 union in_addr_union address6
;
2480 struct FirewallContext
*fw_ctx
;
2483 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2484 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2489 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2490 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2494 static int setup_hostname(void) {
2497 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2500 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2502 return log_error_errno(r
, "Failed to set hostname: %m");
2507 static int setup_journal(const char *directory
) {
2508 _cleanup_free_
char *d
= NULL
;
2514 /* Don't link journals in ephemeral mode */
2518 if (arg_link_journal
== LINK_NO
)
2521 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2523 r
= sd_id128_get_machine(&this_id
);
2525 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2527 if (sd_id128_equal(arg_uuid
, this_id
)) {
2528 log_full(try ? LOG_WARNING
: LOG_ERR
,
2529 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2535 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2536 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2538 bool ignore
= r
== -EROFS
&& try;
2539 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2540 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2541 return ignore
? 0 : r
;
2545 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2546 q
= prefix_roota(directory
, p
);
2548 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2552 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2553 "%s: already a mount point, refusing to use for journal", p
);
2556 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2560 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2561 "%s: already a mount point, refusing to use for journal", q
);
2564 r
= readlink_and_make_absolute(p
, &d
);
2566 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2569 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2571 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2576 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2577 } else if (r
== -EINVAL
) {
2579 if (arg_link_journal
== LINK_GUEST
&&
2582 if (errno
== ENOTDIR
) {
2583 log_error("%s already exists and is neither a symlink nor a directory", p
);
2586 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2588 } else if (r
!= -ENOENT
)
2589 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2591 if (arg_link_journal
== LINK_GUEST
) {
2593 if (symlink(q
, p
) < 0) {
2595 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2598 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2601 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2603 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2607 if (arg_link_journal
== LINK_HOST
) {
2608 /* don't create parents here — if the host doesn't have
2609 * permanent journal set up, don't force it here */
2611 r
= RET_NERRNO(mkdir(p
, 0755));
2612 if (r
< 0 && r
!= -EEXIST
) {
2614 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2617 return log_error_errno(r
, "Failed to create %s: %m", p
);
2620 } else if (access(p
, F_OK
) < 0)
2623 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2624 log_warning("%s is not empty, proceeding anyway.", q
);
2626 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2628 return log_error_errno(r
, "Failed to create %s: %m", q
);
2630 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2632 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2637 static int drop_capabilities(uid_t uid
) {
2638 CapabilityQuintet q
;
2640 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2641 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2642 * arg_caps_retain. */
2644 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2645 q
= arg_full_capabilities
;
2647 if (q
.bounding
== UINT64_MAX
)
2648 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2650 if (q
.effective
== UINT64_MAX
)
2651 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2653 if (q
.inheritable
== UINT64_MAX
)
2654 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2656 if (q
.permitted
== UINT64_MAX
)
2657 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2659 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2660 q
.ambient
= arg_caps_ambient
;
2662 if (capability_quintet_mangle(&q
))
2663 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2666 q
= (CapabilityQuintet
) {
2667 .bounding
= arg_caps_retain
,
2668 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2669 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2670 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2671 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2674 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2675 * in order to maintain the same behavior as systemd < 242. */
2676 if (capability_quintet_mangle(&q
))
2677 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2678 "Some capabilities will not be set because they are not in the current bounding set.");
2682 return capability_quintet_enforce(&q
);
2685 static int reset_audit_loginuid(void) {
2686 _cleanup_free_
char *p
= NULL
;
2689 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2692 r
= read_one_line_file("/proc/self/loginuid", &p
);
2696 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2698 /* Already reset? */
2699 if (streq(p
, "4294967295"))
2702 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2705 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2706 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2707 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2708 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2709 "using systemd-nspawn. Sleeping for 5s... (%m)");
2717 static int mount_tunnel_dig(const char *root
) {
2721 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2722 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2723 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2724 (void) mkdir_p(p
, 0600);
2726 r
= make_run_host(root
);
2730 r
= userns_mkdir(root
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2732 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2734 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
2735 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2739 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2746 static int mount_tunnel_open(void) {
2749 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2756 static int setup_machine_id(const char *directory
) {
2759 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2760 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2761 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2762 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2763 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2764 * container behaves nicely). */
2766 r
= id128_get_machine(directory
, &arg_uuid
);
2767 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) {
2768 /* If the file is missing, empty, or uninitialized, we don't mind */
2769 if (sd_id128_is_null(arg_uuid
)) {
2770 r
= sd_id128_randomize(&arg_uuid
);
2772 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2775 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2780 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2785 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2788 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2789 if (r
== -EOPNOTSUPP
)
2790 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.");
2792 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2794 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2796 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2798 log_debug("Patched directory tree to match UID/GID range.");
2805 * < 0 : wait_for_terminate() failed to get the state of the
2806 * container, the container was terminated by a signal, or
2807 * failed for an unknown reason. No change is made to the
2808 * container argument.
2809 * > 0 : The program executed in the container terminated with an
2810 * error. The exit code of the program executed in the
2811 * container is returned. The container argument has been set
2812 * to CONTAINER_TERMINATED.
2813 * 0 : The container is being rebooted, has been shut down or exited
2814 * successfully. The container argument has been set to either
2815 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2817 * That is, success is indicated by a return value of zero, and an
2818 * error is indicated by a non-zero value.
2820 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2824 r
= wait_for_terminate(pid
, &status
);
2826 return log_warning_errno(r
, "Failed to wait for container: %m");
2828 switch (status
.si_code
) {
2831 if (status
.si_status
== 0)
2832 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2834 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2836 *container
= CONTAINER_TERMINATED
;
2837 return status
.si_status
;
2840 if (status
.si_status
== SIGINT
) {
2841 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2842 *container
= CONTAINER_TERMINATED
;
2845 } else if (status
.si_status
== SIGHUP
) {
2846 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2847 *container
= CONTAINER_REBOOTED
;
2853 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2854 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2857 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2858 "Container %s failed due to unknown reason.", arg_machine
);
2862 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2865 pid
= PTR_TO_PID(userdata
);
2867 if (kill(pid
, arg_kill_signal
) >= 0) {
2868 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2869 sd_event_source_set_userdata(s
, NULL
);
2874 sd_event_exit(sd_event_source_get_event(s
), 0);
2878 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2884 pid
= PTR_TO_PID(userdata
);
2889 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2890 return log_error_errno(errno
, "Failed to waitid(): %m");
2891 if (si
.si_pid
== 0) /* No pending children. */
2893 if (si
.si_pid
== pid
) {
2894 /* The main process we care for has exited. Return from
2895 * signal handler but leave the zombie. */
2896 sd_event_exit(sd_event_source_get_event(s
), 0);
2900 /* Reap all other children. */
2901 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2907 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2912 pid
= PTR_TO_PID(userdata
);
2914 if (arg_kill_signal
> 0) {
2915 log_info("Container termination requested. Attempting to halt container.");
2916 (void) kill(pid
, arg_kill_signal
);
2918 log_info("Container termination requested. Exiting.");
2919 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2925 static int pick_paths(void) {
2928 if (arg_directory
) {
2929 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2930 PickFilter filter
= pick_filter_image_dir
;
2932 filter
.architecture
= arg_architecture
;
2934 r
= path_pick_update_warn(
2937 PICK_ARCHITECTURE
|PICK_TRIES
,
2940 /* Accept ENOENT here so that the --template= logic can work */
2944 arg_architecture
= result
.architecture
;
2948 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2949 PickFilter filter
= pick_filter_image_raw
;
2951 filter
.architecture
= arg_architecture
;
2953 r
= path_pick_update_warn(
2956 PICK_ARCHITECTURE
|PICK_TRIES
,
2961 arg_architecture
= result
.architecture
;
2965 _cleanup_(pick_result_done
) PickResult result
= PICK_RESULT_NULL
;
2966 PickFilter filter
= pick_filter_image_dir
;
2968 filter
.architecture
= arg_architecture
;
2970 r
= path_pick_update_warn(
2978 arg_architecture
= result
.architecture
;
2984 static int determine_names(void) {
2987 if (arg_template
&& !arg_directory
&& arg_machine
) {
2989 /* If --template= was specified then we should not search for a machine, but instead create a
2990 * new one in /var/lib/machine. */
2992 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2997 if (!arg_image
&& !arg_directory
) {
2999 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3001 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3003 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3005 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3007 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3008 r
= free_and_strdup(&arg_image
, i
->path
);
3010 r
= free_and_strdup(&arg_directory
, i
->path
);
3015 arg_read_only
= arg_read_only
|| i
->read_only
;
3017 r
= safe_getcwd(&arg_directory
);
3019 return log_error_errno(r
, "Failed to determine current directory: %m");
3022 if (!arg_directory
&& !arg_image
)
3023 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3027 if (arg_directory
&& path_equal(arg_directory
, "/"))
3028 arg_machine
= gethostname_malloc();
3029 else if (arg_image
) {
3032 r
= path_extract_filename(arg_image
, &arg_machine
);
3034 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_image
);
3036 /* Truncate suffix if there is one */
3037 e
= endswith(arg_machine
, ".raw");
3041 r
= path_extract_filename(arg_directory
, &arg_machine
);
3043 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_directory
);
3046 hostname_cleanup(arg_machine
);
3047 if (!hostname_is_valid(arg_machine
, 0))
3048 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3050 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3051 * to match fixed config file names. */
3052 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3053 if (!arg_settings_filename
)
3056 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3057 * instances at once without manually having to specify -M each time. */
3059 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3062 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3063 if (!arg_settings_filename
)
3070 static int chase_and_update(char **p
, unsigned flags
) {
3079 r
= chase(*p
, NULL
, flags
, &chased
, NULL
);
3081 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3083 return free_and_replace(*p
, chased
);
3086 static int determine_uid_shift(const char *directory
) {
3088 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3093 if (arg_uid_shift
== UID_INVALID
) {
3096 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3098 if (stat(directory
, &st
) < 0)
3099 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3101 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3103 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3104 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3105 "UID and GID base of %s don't match.", directory
);
3107 arg_uid_range
= UINT32_C(0x10000);
3109 if (arg_uid_shift
!= 0) {
3110 /* If the image is shifted already, then we'll fall back to classic chowning, for
3111 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3113 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3114 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3115 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3116 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3117 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3118 "UID base of %s is not zero, UID mapping not supported.", directory
);
3122 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3123 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3128 static unsigned long effective_clone_ns_flags(void) {
3129 unsigned long flags
= arg_clone_ns_flags
;
3131 if (arg_private_network
)
3132 flags
|= CLONE_NEWNET
;
3134 flags
|= CLONE_NEWCGROUP
;
3135 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3136 flags
|= CLONE_NEWUSER
;
3141 static int patch_sysctl(void) {
3143 /* This table is inspired by runc's sysctl() function */
3144 static const struct {
3147 unsigned long clone_flags
;
3149 { "kernel.hostname", false, CLONE_NEWUTS
},
3150 { "kernel.domainname", false, CLONE_NEWUTS
},
3151 { "kernel.msgmax", false, CLONE_NEWIPC
},
3152 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3153 { "kernel.msgmni", false, CLONE_NEWIPC
},
3154 { "kernel.sem", false, CLONE_NEWIPC
},
3155 { "kernel.shmall", false, CLONE_NEWIPC
},
3156 { "kernel.shmmax", false, CLONE_NEWIPC
},
3157 { "kernel.shmmni", false, CLONE_NEWIPC
},
3158 { "fs.mqueue.", true, CLONE_NEWIPC
},
3159 { "net.", true, CLONE_NEWNET
},
3162 unsigned long flags
;
3165 flags
= effective_clone_ns_flags();
3167 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3171 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3173 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3176 if (safe_sysctl
[i
].prefix
)
3177 good
= startswith(*k
, safe_sysctl
[i
].key
);
3179 good
= streq(*k
, safe_sysctl
[i
].key
);
3186 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3188 r
= sysctl_write(*k
, *v
);
3190 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3196 static int inner_child(
3198 int fd_inner_socket
,
3200 char **os_release_pairs
) {
3202 _cleanup_free_
char *home
= NULL
;
3205 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3206 NULL
, /* container */
3211 NULL
, /* container_uuid */
3212 NULL
, /* LISTEN_FDS */
3213 NULL
, /* LISTEN_PID */
3214 NULL
, /* NOTIFY_SOCKET */
3215 NULL
, /* CREDENTIALS_DIRECTORY */
3219 const char *exec_target
;
3220 _cleanup_strv_free_
char **env_use
= NULL
;
3221 int r
, which_failed
;
3223 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3224 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3225 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3226 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3227 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3228 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3231 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3232 * unshare(). See below. */
3235 assert(fd_inner_socket
>= 0);
3237 log_debug("Inner child is initializing.");
3239 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3240 /* Tell the parent, that it now can write the UID map. */
3241 (void) barrier_place(barrier
); /* #1 */
3243 /* Wait until the parent wrote the UID map */
3244 if (!barrier_place_and_sync(barrier
)) /* #2 */
3245 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3247 /* Become the new root user inside our namespace */
3248 r
= reset_uid_gid();
3250 return log_error_errno(r
, "Couldn't become new root: %m");
3252 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3253 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3254 * propagation, but simply create new peer groups for all our mounts). */
3255 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3261 arg_mount_settings
| MOUNT_IN_USERNS
,
3263 arg_selinux_apifs_context
);
3267 if (!arg_network_namespace_path
&& arg_private_network
) {
3268 r
= unshare(CLONE_NEWNET
);
3270 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3272 /* Tell the parent that it can setup network interfaces. */
3273 (void) barrier_place(barrier
); /* #3 */
3276 r
= mount_sysfs(NULL
, arg_mount_settings
);
3280 /* Wait until we are cgroup-ified, so that we
3281 * can mount the right cgroup path writable */
3282 if (!barrier_place_and_sync(barrier
)) /* #4 */
3283 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3284 "Parent died too early");
3287 r
= unshare(CLONE_NEWCGROUP
);
3289 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3292 arg_unified_cgroup_hierarchy
,
3293 arg_userns_mode
!= USER_NAMESPACE_NO
,
3296 arg_selinux_apifs_context
,
3299 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3303 r
= setup_boot_id();
3307 r
= setup_kmsg(fd_inner_socket
);
3314 arg_n_custom_mounts
,
3317 arg_selinux_apifs_context
,
3318 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3323 return log_error_errno(errno
, "setsid() failed: %m");
3325 if (arg_private_network
)
3326 (void) loopback_setup();
3328 if (arg_expose_ports
) {
3329 r
= expose_port_send_rtnl(fd_inner_socket
);
3334 if (arg_console_mode
!= CONSOLE_PIPE
) {
3335 _cleanup_close_
int master
= -EBADF
;
3336 _cleanup_free_
char *console
= NULL
;
3338 /* Allocate a pty and make it available as /dev/console. */
3339 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3341 return log_error_errno(master
, "Failed to allocate a pty: %m");
3343 r
= setup_dev_console(console
);
3345 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3347 r
= send_one_fd(fd_inner_socket
, master
, 0);
3349 return log_error_errno(r
, "Failed to send master fd: %m");
3351 r
= setup_stdio_as_dev_console();
3360 if (arg_oom_score_adjust_set
) {
3361 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3363 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3366 if (arg_cpu_set
.set
)
3367 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3368 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3370 (void) setup_hostname();
3372 if (arg_personality
!= PERSONALITY_INVALID
) {
3373 r
= safe_personality(arg_personality
);
3375 return log_error_errno(r
, "personality() failed: %m");
3376 #ifdef ARCHITECTURE_SECONDARY
3377 } else if (arg_architecture
== ARCHITECTURE_SECONDARY
) {
3378 r
= safe_personality(PER_LINUX32
);
3380 return log_error_errno(r
, "personality() failed: %m");
3382 } else if (!arg_quiet
&& arg_architecture
>= 0 && arg_architecture
!= native_architecture())
3383 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3384 "invocation with qemu userspace emulator (or equivalent) in effect.",
3385 architecture_to_string(arg_architecture
));
3387 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3389 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3394 if (is_seccomp_available()) {
3395 r
= seccomp_load(arg_seccomp
);
3396 if (ERRNO_IS_NEG_SECCOMP_FATAL(r
))
3397 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3399 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3404 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3409 if (arg_suppress_sync
) {
3411 r
= seccomp_suppress_sync();
3413 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3415 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3420 if (arg_selinux_context
)
3421 if (setexeccon(arg_selinux_context
) < 0)
3422 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3425 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3426 * if we need to later on. */
3427 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3428 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3430 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3431 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3433 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3437 r
= drop_capabilities(getuid());
3439 return log_error_errno(r
, "Dropping capabilities failed: %m");
3441 if (arg_no_new_privileges
)
3442 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3443 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3445 /* LXC sets container=lxc, so follow the scheme here */
3446 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3448 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3452 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3453 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3456 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3457 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3458 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
?: "root") < 0)
3461 assert(!sd_id128_is_null(arg_uuid
));
3463 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3466 if (!fdset_isempty(fds
)) {
3467 r
= fdset_cloexec(fds
, false);
3469 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3471 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3472 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3475 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3478 if (arg_credentials
.n_credentials
> 0) {
3479 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3485 if (arg_start_mode
!= START_BOOT
) {
3486 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3492 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3496 /* Let the parent know that we are ready and wait until the parent is ready with the setup, too... */
3497 if (!barrier_place_and_sync(barrier
)) /* #5 */
3498 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3501 if (chdir(arg_chdir
) < 0)
3502 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3504 if (arg_start_mode
== START_PID2
) {
3505 r
= stub_pid1(arg_uuid
);
3510 if (arg_console_mode
!= CONSOLE_PIPE
) {
3511 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3512 * are configured for that. Acquire it as controlling tty. */
3513 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3514 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3517 log_debug("Inner child completed, invoking payload.");
3519 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3520 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3521 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3523 log_set_open_when_needed(true);
3524 log_settle_target();
3526 (void) fdset_close_others(fds
);
3528 if (arg_start_mode
== START_BOOT
) {
3532 /* Automatically search for the init system */
3534 m
= strv_length(arg_parameters
);
3535 a
= newa(char*, m
+ 2);
3536 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3539 FOREACH_STRING(init
,
3540 "/usr/lib/systemd/systemd",
3541 "/lib/systemd/systemd",
3543 a
[0] = (char*) init
;
3544 execve(a
[0], a
, env_use
);
3547 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3548 } else if (!strv_isempty(arg_parameters
)) {
3549 const char *dollar_path
;
3551 exec_target
= arg_parameters
[0];
3553 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3555 dollar_path
= strv_env_get(env_use
, "PATH");
3557 if (setenv("PATH", dollar_path
, 1) < 0)
3558 return log_error_errno(errno
, "Failed to update $PATH: %m");
3561 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3564 /* If we cannot change the directory, we'll end up in /, that is expected. */
3565 (void) chdir(home
?: "/root");
3567 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3568 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3569 execle("/bin/bash", "-bash", NULL
, env_use
);
3570 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3571 execle("/bin/sh", "-sh", NULL
, env_use
);
3573 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3576 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3579 static int setup_notify_child(void) {
3580 _cleanup_close_
int fd
= -EBADF
;
3581 static const union sockaddr_union sa
= {
3582 .un
.sun_family
= AF_UNIX
,
3583 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3587 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3589 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3591 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3592 (void) sockaddr_un_unlink(&sa
.un
);
3594 WITH_UMASK(0577) { /* only set "w" bit, which is all that's necessary for connecting from the container */
3595 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3597 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3600 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3602 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3604 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3606 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3611 static int outer_child(
3613 const char *directory
,
3614 DissectedImage
*dissected_image
,
3615 int fd_outer_socket
,
3616 int fd_inner_socket
,
3620 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3621 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3622 _cleanup_close_
int fd
= -EBADF
, mntns_fd
= -EBADF
;
3629 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3630 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3631 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3632 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3633 * forked off it, and it exits. */
3637 assert(fd_outer_socket
>= 0);
3638 assert(fd_inner_socket
>= 0);
3640 log_debug("Outer child is initializing.");
3642 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3644 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3646 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3647 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3649 r
= reset_audit_loginuid();
3653 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3654 * mounts to the real root. */
3655 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3659 if (dissected_image
) {
3660 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3661 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3662 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3663 * right place right away. This makes sure ESP partitions and userns are compatible. */
3665 r
= dissected_image_mount_and_warn(
3670 /* userns_fd= */ -EBADF
,
3671 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3672 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3673 DISSECT_IMAGE_USR_NO_ROOT
|
3674 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3675 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3680 r
= determine_uid_shift(directory
);
3684 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3685 r
= namespace_open(0, NULL
, &mntns_fd
, NULL
, NULL
, NULL
);
3687 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3689 l
= send_one_fd(fd_outer_socket
, mntns_fd
, 0);
3691 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3692 mntns_fd
= safe_close(mntns_fd
);
3694 /* Let the parent know which UID shift we read from the image */
3695 l
= send(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3697 return log_error_errno(errno
, "Failed to send UID shift: %m");
3698 if (l
!= sizeof(arg_uid_shift
))
3699 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3700 "Short write while sending UID shift.");
3702 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3703 /* When we are supposed to pick the UID shift, the parent will check now whether the
3704 * UID shift we just read from the image is available. If yes, it will send the UID
3705 * shift back to us, if not it will pick a different one, and send it back to us. */
3707 l
= recv(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3709 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3710 if (l
!= sizeof(arg_uid_shift
))
3711 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3712 "Short read while receiving UID shift.");
3715 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3716 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3719 if (path_equal(directory
, "/")) {
3720 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3721 * place, so that we can make changes to its mount structure (for example, to implement
3722 * --volatile=) without this interfering with our ability to access files such as
3723 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3724 * (instead of a temporary directory, since we are living in our own mount namespace here
3725 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3726 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3728 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3732 directory
= "/run/systemd/nspawn-root";
3735 /* Make sure we always have a mount that we can move to root later on. */
3736 r
= make_mount_point(directory
);
3740 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3741 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3742 * we'll live in our own little world from now on, and propagation from the host may only happen via
3743 * the mount tunnel dir, or not at all. */
3744 r
= mount_follow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_PRIVATE
|MS_REC
, NULL
);
3748 r
= setup_pivot_root(
3751 arg_pivot_root_old
);
3755 r
= setup_volatile_mode(
3759 arg_selinux_apifs_context
);
3763 r
= bind_user_prepare(
3768 &arg_custom_mounts
, &arg_n_custom_mounts
,
3769 &bind_user_context
);
3773 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3774 /* Send the user maps we determined to the parent, so that it installs it in our user
3775 * namespace UID map table */
3777 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3779 bind_user_context
->data
[i
].payload_user
->uid
,
3780 bind_user_context
->data
[i
].host_user
->uid
,
3781 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3782 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3785 l
= send(fd_outer_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3787 return log_error_errno(errno
, "Failed to send user UID map: %m");
3788 if (l
!= sizeof(map
))
3789 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3790 "Short write while sending user UID map.");
3797 arg_n_custom_mounts
,
3800 arg_selinux_apifs_context
,
3805 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3806 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3807 arg_uid_shift
!= 0) {
3808 _cleanup_free_
char *usr_subtree
= NULL
;
3812 dirs
[i
++] = (char*) directory
;
3814 if (dissected_image
&& dissected_image
->partitions
[PARTITION_USR
].found
) {
3815 usr_subtree
= path_join(directory
, "/usr");
3819 dirs
[i
++] = usr_subtree
;
3824 r
= remount_idmap(dirs
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3825 if (r
== -EINVAL
|| ERRNO_IS_NEG_NOT_SUPPORTED(r
)) {
3826 /* This might fail because the kernel or file system doesn't support idmapping. We
3827 * can't really distinguish this nicely, nor do we have any guarantees about the
3828 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3829 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3830 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3831 "ID mapped mounts are apparently not available, sorry.");
3833 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3834 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3836 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3838 log_debug("ID mapped mounts available, making use of them.");
3843 if (dissected_image
) {
3844 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3845 r
= dissected_image_mount(
3850 /* userns_fd= */ -EBADF
,
3851 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3852 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3853 DISSECT_IMAGE_USR_NO_ROOT
|
3854 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3855 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3857 return log_error_errno(r
, "File system check for image failed: %m");
3859 return log_error_errno(r
, "Failed to mount image file system: %m");
3862 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3863 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3865 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3869 l
= send(fd_outer_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3871 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3872 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3873 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3874 "Short write while sending cgroup mode.");
3877 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3881 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3885 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3886 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3887 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3889 return log_error_errno(r
, "Failed to make tree read-only: %m");
3892 r
= mount_all(directory
,
3895 arg_selinux_apifs_context
);
3899 r
= copy_devnodes(directory
);
3903 r
= make_extra_nodes(directory
);
3907 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3909 p
= prefix_roota(directory
, "/run/host");
3910 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3912 r
= setup_pts(directory
);
3916 r
= mount_tunnel_dig(directory
);
3920 r
= setup_keyring();
3924 r
= setup_credentials(directory
);
3928 r
= bind_user_setup(bind_user_context
, directory
);
3935 arg_n_custom_mounts
,
3938 arg_selinux_apifs_context
,
3939 MOUNT_NON_ROOT_ONLY
);
3943 r
= setup_timezone(directory
);
3947 r
= setup_resolv_conf(directory
);
3951 r
= setup_machine_id(directory
);
3955 r
= setup_journal(directory
);
3959 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3960 p
= prefix_roota(directory
, "/run/host/container-manager");
3961 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
|WRITE_STRING_FILE_MODE_0444
);
3963 /* The same stuff as the $container_uuid env var */
3964 p
= prefix_roota(directory
, "/run/host/container-uuid");
3965 (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
));
3967 if (!arg_use_cgns
) {
3970 arg_unified_cgroup_hierarchy
,
3971 arg_userns_mode
!= USER_NAMESPACE_NO
,
3974 arg_selinux_apifs_context
,
3980 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3981 * mounts available in systemd services inside the container that create a new mount namespace. See
3982 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3983 * will inherit the shared propagation mode.
3985 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3986 * directory mount to root later on.
3987 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3989 r
= mount_switch_root(directory
, MS_SHARED
);
3991 return log_error_errno(r
, "Failed to move root directory: %m");
3993 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
3994 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
3996 r
= mount_tunnel_open();
4000 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4001 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4002 * requires that a fully visible instance is already present in the target mount
4003 * namespace. Mount one here so the inner child can mount its own instances. Later
4004 * we umount the temporary instances created here before we actually exec the
4005 * payload. Since the rootfs is shared the umount will propagate into the container.
4006 * Note, the inner child wouldn't be able to unmount the instances on its own since
4007 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4009 r
= pin_fully_visible_fs();
4014 fd
= setup_notify_child();
4018 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
4019 arg_clone_ns_flags
|
4020 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
4022 return log_error_errno(errno
, "Failed to fork inner child: %m");
4024 fd_outer_socket
= safe_close(fd_outer_socket
);
4026 /* The inner child has all namespaces that are requested, so that we all are owned by the
4027 * user if user namespaces are turned on. */
4029 if (arg_network_namespace_path
) {
4030 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
4032 return log_error_errno(r
, "Failed to join network namespace: %m");
4035 r
= inner_child(barrier
, fd_inner_socket
, fds
, os_release_pairs
);
4037 _exit(EXIT_FAILURE
);
4039 _exit(EXIT_SUCCESS
);
4042 l
= send(fd_outer_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4044 return log_error_errno(errno
, "Failed to send PID: %m");
4045 if (l
!= sizeof(pid
))
4046 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4047 "Short write while sending PID.");
4049 l
= send(fd_outer_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4051 return log_error_errno(errno
, "Failed to send machine ID: %m");
4052 if (l
!= sizeof(arg_uuid
))
4053 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4054 "Short write while sending machine ID.");
4056 l
= send_one_fd(fd_outer_socket
, fd
, 0);
4058 return log_error_errno(l
, "Failed to send notify fd: %m");
4060 fd_outer_socket
= safe_close(fd_outer_socket
);
4061 fd_inner_socket
= safe_close(fd_inner_socket
);
4062 netns_fd
= safe_close(netns_fd
);
4067 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4068 bool tried_hashed
= false;
4069 unsigned n_tries
= 100;
4074 assert(ret_lock_file
);
4075 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4076 assert(arg_uid_range
== 0x10000U
);
4080 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4083 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4084 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4089 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4091 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4094 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4095 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4096 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4101 /* Make some superficial checks whether the range is currently known in the user database */
4102 if (getpwuid(candidate
))
4104 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4106 if (getgrgid(candidate
))
4108 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4111 *ret_lock_file
= lf
;
4112 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4117 if (arg_machine
&& !tried_hashed
) {
4118 /* Try to hash the base from the container name */
4120 static const uint8_t hash_key
[] = {
4121 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4122 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4125 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4127 tried_hashed
= true;
4129 random_bytes(&candidate
, sizeof(candidate
));
4131 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4132 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4136 static int add_one_uid_map(
4138 uid_t container_uid
,
4142 return strextendf(p
,
4143 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4144 container_uid
, host_uid
, range
);
4147 static int make_uid_map_string(
4148 const uid_t bind_user_uid
[],
4149 size_t n_bind_user_uid
,
4153 _cleanup_free_
char *s
= NULL
;
4154 uid_t previous_uid
= 0;
4157 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4158 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4161 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4162 * quadruplet, consisting of host and container UID + GID. */
4164 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4165 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4166 host_uid
= bind_user_uid
[i
*4+offset
+1];
4168 assert(previous_uid
<= payload_uid
);
4169 assert(payload_uid
< arg_uid_range
);
4171 /* Add a range to close the gap to previous entry */
4172 if (payload_uid
> previous_uid
) {
4173 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4178 /* Map this specific user */
4179 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4183 previous_uid
= payload_uid
+ 1;
4186 /* And add a range to close the gap to finish the range */
4187 if (arg_uid_range
> previous_uid
) {
4188 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4199 static int setup_uid_map(
4201 const uid_t bind_user_uid
[],
4202 size_t n_bind_user_uid
) {
4204 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4205 _cleanup_free_
char *s
= NULL
;
4210 /* Build the UID map string */
4211 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4214 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4215 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4217 return log_error_errno(r
, "Failed to write UID map: %m");
4219 /* And now build the GID map string */
4221 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4224 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4225 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4227 return log_error_errno(r
, "Failed to write GID map: %m");
4232 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4233 char buf
[NOTIFY_BUFFER_MAX
+1];
4235 struct iovec iovec
= {
4237 .iov_len
= sizeof(buf
)-1,
4239 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4240 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4241 struct msghdr msghdr
= {
4244 .msg_control
= &control
,
4245 .msg_controllen
= sizeof(control
),
4247 struct ucred
*ucred
;
4249 pid_t inner_child_pid
;
4250 _cleanup_strv_free_
char **tags
= NULL
;
4255 inner_child_pid
= PTR_TO_PID(userdata
);
4257 if (revents
!= EPOLLIN
) {
4258 log_warning("Got unexpected poll event for notify fd.");
4262 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4263 if (ERRNO_IS_NEG_TRANSIENT(n
))
4265 else if (n
== -EXFULL
) {
4266 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4269 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4271 cmsg_close_all(&msghdr
);
4273 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4274 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4275 log_debug("Received notify message without valid credentials. Ignoring.");
4279 if ((size_t) n
>= sizeof(buf
)) {
4280 log_warning("Received notify message exceeded maximum size. Ignoring.");
4285 tags
= strv_split(buf
, "\n\r");
4289 if (strv_contains(tags
, "READY=1")) {
4290 r
= sd_notify(false, "READY=1\n");
4292 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4295 p
= strv_find_startswith(tags
, "STATUS=");
4297 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4302 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4305 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4307 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4309 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4314 static int merge_settings(Settings
*settings
, const char *path
) {
4320 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4321 * that this steals the fields of the Settings* structure, and hence modifies it. */
4323 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4324 settings
->start_mode
>= 0) {
4325 arg_start_mode
= settings
->start_mode
;
4326 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4329 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4330 settings
->ephemeral
>= 0)
4331 arg_ephemeral
= settings
->ephemeral
;
4333 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4336 if (!arg_settings_trusted
)
4337 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4339 free_and_replace(arg_directory
, settings
->root
);
4342 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4343 settings
->pivot_root_new
) {
4344 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4345 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4348 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4349 settings
->working_directory
)
4350 free_and_replace(arg_chdir
, settings
->working_directory
);
4352 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4353 settings
->environment
)
4354 strv_free_and_replace(arg_setenv
, settings
->environment
);
4356 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4359 free_and_replace(arg_user
, settings
->user
);
4361 if (uid_is_valid(settings
->uid
))
4362 arg_uid
= settings
->uid
;
4363 if (gid_is_valid(settings
->gid
))
4364 arg_gid
= settings
->gid
;
4365 if (settings
->n_supplementary_gids
> 0) {
4366 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4367 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4371 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4372 uint64_t plus
, minus
;
4373 uint64_t network_minus
= 0;
4376 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4377 * Settings structure */
4379 plus
= settings
->capability
;
4380 minus
= settings
->drop_capability
;
4382 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4383 settings_network_configured(settings
)) {
4384 if (settings_private_network(settings
))
4385 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4387 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4390 if (!arg_settings_trusted
&& plus
!= 0) {
4391 if (settings
->capability
!= 0)
4392 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4394 arg_caps_retain
&= ~network_minus
;
4395 arg_caps_retain
|= plus
;
4398 arg_caps_retain
&= ~minus
;
4400 /* Copy the full capabilities over too */
4401 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4402 if (!arg_settings_trusted
)
4403 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4405 arg_full_capabilities
= settings
->full_capabilities
;
4408 ambient
= settings
->ambient_capability
;
4409 if (!arg_settings_trusted
&& ambient
!= 0)
4410 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4412 arg_caps_ambient
|= ambient
;
4415 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4416 settings
->kill_signal
> 0)
4417 arg_kill_signal
= settings
->kill_signal
;
4419 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4420 settings
->personality
!= PERSONALITY_INVALID
)
4421 arg_personality
= settings
->personality
;
4423 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4424 !sd_id128_is_null(settings
->machine_id
)) {
4426 if (!arg_settings_trusted
)
4427 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4429 arg_uuid
= settings
->machine_id
;
4432 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4433 settings
->read_only
>= 0)
4434 arg_read_only
= settings
->read_only
;
4436 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4437 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4438 arg_volatile_mode
= settings
->volatile_mode
;
4440 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4441 settings
->n_custom_mounts
> 0) {
4443 if (!arg_settings_trusted
)
4444 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4446 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4447 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4448 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4449 settings
->n_custom_mounts
= 0;
4453 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4454 settings_network_configured(settings
)) {
4456 if (!arg_settings_trusted
)
4457 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4459 arg_network_veth
= settings_network_veth(settings
);
4460 arg_private_network
= settings_private_network(settings
);
4462 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4463 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4464 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4465 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4467 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4468 free_and_replace(arg_network_zone
, settings
->network_zone
);
4470 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4474 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4475 settings
->expose_ports
) {
4477 if (!arg_settings_trusted
)
4478 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4480 expose_port_free_all(arg_expose_ports
);
4481 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4485 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4486 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4488 if (!arg_settings_trusted
)
4489 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4491 arg_userns_mode
= settings
->userns_mode
;
4492 arg_uid_shift
= settings
->uid_shift
;
4493 arg_uid_range
= settings
->uid_range
;
4494 arg_userns_ownership
= settings
->userns_ownership
;
4498 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4499 !strv_isempty(settings
->bind_user
))
4500 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4502 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4503 settings
->notify_ready
>= 0)
4504 arg_notify_ready
= settings
->notify_ready
;
4506 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4508 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4509 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4510 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4512 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4513 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4518 if (settings
->seccomp
) {
4519 if (!arg_settings_trusted
)
4520 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4522 seccomp_release(arg_seccomp
);
4523 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4529 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4530 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4533 if (!settings
->rlimit
[rl
])
4536 if (!arg_settings_trusted
) {
4537 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4541 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4544 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4546 free_and_replace(arg_hostname
, settings
->hostname
);
4548 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4549 settings
->no_new_privileges
>= 0)
4550 arg_no_new_privileges
= settings
->no_new_privileges
;
4552 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4553 settings
->oom_score_adjust_set
) {
4555 if (!arg_settings_trusted
)
4556 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4558 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4559 arg_oom_score_adjust_set
= true;
4563 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4564 settings
->cpu_set
.set
) {
4566 if (!arg_settings_trusted
)
4567 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4569 cpu_set_reset(&arg_cpu_set
);
4570 arg_cpu_set
= TAKE_STRUCT(settings
->cpu_set
);
4574 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4575 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4576 arg_resolv_conf
= settings
->resolv_conf
;
4578 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4579 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4581 if (!arg_settings_trusted
)
4582 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4584 arg_link_journal
= settings
->link_journal
;
4585 arg_link_journal_try
= settings
->link_journal_try
;
4589 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4590 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4591 arg_timezone
= settings
->timezone
;
4593 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4596 if (!arg_settings_trusted
)
4597 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4599 free_and_replace(arg_slice
, settings
->slice
);
4602 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4603 settings
->use_cgns
>= 0) {
4605 if (!arg_settings_trusted
)
4606 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4608 arg_use_cgns
= settings
->use_cgns
;
4611 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4612 settings
->clone_ns_flags
!= ULONG_MAX
) {
4614 if (!arg_settings_trusted
)
4615 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4617 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4620 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4621 settings
->console_mode
>= 0) {
4623 if (!arg_settings_trusted
)
4624 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4626 arg_console_mode
= settings
->console_mode
;
4629 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4630 settings
->suppress_sync
>= 0)
4631 arg_suppress_sync
= settings
->suppress_sync
;
4633 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4634 * don't consult arg_settings_mask for them. */
4636 sd_bus_message_unref(arg_property_message
);
4637 arg_property_message
= TAKE_PTR(settings
->properties
);
4639 arg_console_width
= settings
->console_width
;
4640 arg_console_height
= settings
->console_height
;
4642 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4643 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4644 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4645 settings
->n_extra_nodes
= 0;
4650 static int load_settings(void) {
4651 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4652 _cleanup_fclose_
FILE *f
= NULL
;
4653 _cleanup_free_
char *p
= NULL
;
4659 /* If all settings are masked, there's no point in looking for
4660 * the settings file */
4661 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4664 /* We first look in the admin's directories in /etc and /run */
4665 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4666 _cleanup_free_
char *j
= NULL
;
4668 j
= path_join(i
, arg_settings_filename
);
4676 /* By default, we trust configuration from /etc and /run */
4677 if (arg_settings_trusted
< 0)
4678 arg_settings_trusted
= true;
4683 if (errno
!= ENOENT
)
4684 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4688 /* After that, let's look for a file next to the
4689 * actual image we shall boot. */
4692 r
= file_in_same_dir(arg_image
, arg_settings_filename
, &p
);
4694 return log_error_errno(r
, "Failed to generate settings path from image path: %m");
4695 } else if (arg_directory
) {
4696 r
= file_in_same_dir(arg_directory
, arg_settings_filename
, &p
);
4697 if (r
< 0 && r
!= -EADDRNOTAVAIL
) /* if directory is root fs, don't complain */
4698 return log_error_errno(r
, "Failed to generate settings path from directory path: %m");
4703 if (!f
&& errno
!= ENOENT
)
4704 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4706 /* By default, we do not trust configuration from /var/lib/machines */
4707 if (arg_settings_trusted
< 0)
4708 arg_settings_trusted
= false;
4715 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4717 r
= settings_load(f
, p
, &settings
);
4721 return merge_settings(settings
, p
);
4724 static int load_oci_bundle(void) {
4725 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4728 if (!arg_oci_bundle
)
4731 /* By default let's trust OCI bundles */
4732 if (arg_settings_trusted
< 0)
4733 arg_settings_trusted
= true;
4735 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4739 return merge_settings(settings
, arg_oci_bundle
);
4742 static int run_container(
4743 DissectedImage
*dissected_image
,
4745 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4746 struct ExposeArgs
*expose_args
,
4747 int *master
, pid_t
*pid
, int *ret
) {
4749 static const struct sigaction sa
= {
4750 .sa_handler
= nop_signal_handler
,
4751 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4754 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4755 _cleanup_close_
int etc_passwd_lock
= -EBADF
;
4756 _cleanup_close_pair_
int
4757 fd_inner_socket_pair
[2] = EBADF_PAIR
,
4758 fd_outer_socket_pair
[2] = EBADF_PAIR
;
4760 _cleanup_close_
int notify_socket
= -EBADF
, mntns_fd
= -EBADF
, fd_kmsg_fifo
= -EBADF
;
4761 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4762 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4763 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4764 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4765 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4766 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4767 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4768 size_t n_bind_user_uid
= 0;
4769 ContainerStatus container_status
= 0;
4773 _cleanup_close_
int child_netns_fd
= -EBADF
;
4775 assert_se(sigemptyset(&mask_chld
) == 0);
4776 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4778 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4779 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4780 * check with getpwuid() if the specific user already exists. Note that /etc might be
4781 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4782 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4783 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4786 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4787 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4788 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4791 r
= barrier_create(&barrier
);
4793 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4795 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_inner_socket_pair
) < 0)
4796 return log_error_errno(errno
, "Failed to create inner socket pair: %m");
4798 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_outer_socket_pair
) < 0)
4799 return log_error_errno(errno
, "Failed to create outer socket pair: %m");
4801 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4802 * parent's blocking calls and give it a chance to call wait() and terminate. */
4803 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4805 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4807 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4809 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4811 if (arg_network_namespace_path
) {
4812 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4813 if (child_netns_fd
< 0)
4814 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4816 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4818 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
);
4820 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4822 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4823 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4826 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4828 return log_error_errno(errno
, "clone() failed%s: %m",
4830 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4833 /* The outer child only has a file system namespace. */
4834 barrier_set_role(&barrier
, BARRIER_CHILD
);
4836 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
4837 fd_outer_socket_pair
[0] = safe_close(fd_outer_socket_pair
[0]);
4839 (void) reset_all_signal_handlers();
4840 (void) reset_signal_mask();
4842 r
= outer_child(&barrier
,
4845 fd_outer_socket_pair
[1],
4846 fd_inner_socket_pair
[1],
4850 _exit(EXIT_FAILURE
);
4852 _exit(EXIT_SUCCESS
);
4855 barrier_set_role(&barrier
, BARRIER_PARENT
);
4859 fd_inner_socket_pair
[1] = safe_close(fd_inner_socket_pair
[1]);
4860 fd_outer_socket_pair
[1] = safe_close(fd_outer_socket_pair
[1]);
4862 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4863 mntns_fd
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4865 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4867 /* The child just let us know the UID shift it might have read from the image. */
4868 l
= recv(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4870 return log_error_errno(errno
, "Failed to read UID shift: %m");
4871 if (l
!= sizeof arg_uid_shift
)
4872 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4874 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4875 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4876 * image, but if that's already in use, pick a new one, and report back to the child,
4877 * which one we now picked. */
4879 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4881 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4883 l
= send(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4885 return log_error_errno(errno
, "Failed to send UID shift: %m");
4886 if (l
!= sizeof arg_uid_shift
)
4887 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4890 n_bind_user_uid
= strv_length(arg_bind_user
);
4891 if (n_bind_user_uid
> 0) {
4892 /* Right after the UID shift, we'll receive the list of UID mappings for the
4893 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4895 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4899 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4900 l
= recv(fd_outer_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4902 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4903 if (l
!= sizeof(uid_t
)*4)
4904 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4905 SYNTHETIC_ERRNO(EIO
),
4906 "Short read while reading bind user UID pairs.");
4911 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4912 /* The child let us know the support cgroup mode it might have read from the image. */
4913 l
= recv(fd_outer_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4915 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4916 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4917 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
4918 l
, l
== 0 ? " The child is most likely dead." : "");
4921 /* Wait for the outer child. */
4922 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4925 if (r
!= EXIT_SUCCESS
)
4928 /* And now retrieve the PID of the inner child. */
4929 l
= recv(fd_outer_socket_pair
[0], pid
, sizeof *pid
, 0);
4931 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4932 if (l
!= sizeof *pid
)
4933 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4935 /* We also retrieve container UUID in case it was generated by outer child */
4936 l
= recv(fd_outer_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4938 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4939 if (l
!= sizeof(arg_uuid
))
4940 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4942 /* We also retrieve the socket used for notifications generated by outer child */
4943 notify_socket
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4944 if (notify_socket
< 0)
4945 return log_error_errno(notify_socket
,
4946 "Failed to receive notification socket from the outer child: %m");
4948 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4950 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4951 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4952 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4954 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4958 (void) barrier_place(&barrier
); /* #2 */
4961 if (arg_private_network
) {
4962 if (!arg_network_namespace_path
) {
4963 /* Wait until the child has unshared its network namespace. */
4964 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4965 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4968 if (child_netns_fd
< 0) {
4969 /* Make sure we have an open file descriptor to the child's network
4970 * namespace so it stays alive even if the child exits. */
4971 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4973 return log_error_errno(r
, "Failed to open child network namespace: %m");
4976 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4980 if (arg_network_veth
) {
4981 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4982 arg_network_bridge
|| arg_network_zone
, &arg_network_provided_mac
);
4988 if (arg_network_bridge
) {
4989 /* Add the interface to a bridge */
4990 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4995 } else if (arg_network_zone
) {
4996 /* Add the interface to a bridge, possibly creating it */
4997 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5005 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5009 /* We created the primary and extra veth links now; let's remember this, so that we know to
5010 remove them later on. Note that we don't bother with removing veth links that were created
5011 here when their setup failed half-way, because in that case the kernel should be able to
5012 remove them on its own, since they cannot be referenced by anything yet. */
5013 *veth_created
= true;
5015 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5019 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5024 if (arg_register
|| !arg_keep_unit
) {
5025 r
= sd_bus_default_system(&bus
);
5027 return log_error_errno(r
, "Failed to open system bus: %m");
5029 r
= sd_bus_set_close_on_exit(bus
, false);
5031 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5034 if (!arg_keep_unit
) {
5035 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5036 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5037 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5039 r
= sd_bus_match_signal_async(
5042 "org.freedesktop.systemd1",
5044 "org.freedesktop.systemd1.Scope",
5046 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5048 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5052 r
= register_machine(
5060 arg_custom_mounts
, arg_n_custom_mounts
,
5063 arg_property_message
,
5065 arg_container_service_name
,
5070 } else if (!arg_keep_unit
) {
5076 arg_custom_mounts
, arg_n_custom_mounts
,
5079 arg_property_message
,
5080 /* allow_pidfds= */ true,
5085 } else if (arg_slice
|| arg_property
)
5086 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5088 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5092 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5096 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5100 /* Notify the child that the parent is ready with all
5101 * its setup (including cgroup-ification), and that
5102 * the child can now hand over control to the code to
5103 * run inside the container. */
5104 (void) barrier_place(&barrier
); /* #4 */
5106 /* Block SIGCHLD here, before notifying child.
5107 * process_pty() will handle it with the other signals. */
5108 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5110 /* Reset signal to default */
5111 r
= default_signals(SIGCHLD
);
5113 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5115 r
= sd_event_new(&event
);
5117 return log_error_errno(r
, "Failed to get default event source: %m");
5119 (void) sd_event_set_watchdog(event
, true);
5122 r
= sd_bus_attach_event(bus
, event
, 0);
5124 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5127 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5131 /* Wait that the child is completely ready now, and has mounted their own copies of procfs and so on,
5132 * before we take the fully visible instances away. */
5133 if (!barrier_sync(&barrier
)) /* #5.1 */
5134 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5136 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5137 r
= wipe_fully_visible_fs(mntns_fd
);
5140 mntns_fd
= safe_close(mntns_fd
);
5143 /* And now let the child know that we completed removing the procfs instances, and it can start the
5145 if (!barrier_place(&barrier
)) /* #5.2 */
5146 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5148 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5149 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5150 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5152 (void) sd_notifyf(false,
5153 "STATUS=Container running.\n"
5154 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5155 if (!arg_notify_ready
) {
5156 r
= sd_notify(false, "READY=1\n");
5158 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5161 if (arg_kill_signal
> 0) {
5162 /* Try to kill the init system on SIGINT or SIGTERM */
5163 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5164 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5166 /* Immediately exit */
5167 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5168 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5171 (void) sd_event_add_signal(event
, NULL
, SIGRTMIN
+18, sigrtmin18_handler
, NULL
);
5173 r
= sd_event_add_memory_pressure(event
, NULL
, NULL
, NULL
);
5175 log_debug_errno(r
, "Failed allocate memory pressure event source, ignoring: %m");
5177 /* Exit when the child exits */
5178 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5180 /* Retrieve the kmsg fifo allocated by inner child */
5181 fd_kmsg_fifo
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5182 if (fd_kmsg_fifo
< 0)
5183 return log_error_errno(fd_kmsg_fifo
, "Failed to receive kmsg fifo from inner child: %m");
5185 if (arg_expose_ports
) {
5186 r
= expose_port_watch_rtnl(event
, fd_inner_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5190 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5191 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5194 if (arg_console_mode
!= CONSOLE_PIPE
) {
5195 _cleanup_close_
int fd
= -EBADF
;
5196 PTYForwardFlags flags
= 0;
5198 /* Retrieve the master pty allocated by inner child */
5199 fd
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5201 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5203 switch (arg_console_mode
) {
5205 case CONSOLE_READ_ONLY
:
5206 flags
|= PTY_FORWARD_READ_ONLY
;
5210 case CONSOLE_INTERACTIVE
:
5211 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5213 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5215 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5217 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5218 (void) pty_forward_set_width_height(forward
,
5220 arg_console_height
);
5224 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5227 *master
= TAKE_FD(fd
);
5230 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
5232 r
= sd_event_loop(event
);
5234 return log_error_errno(r
, "Failed to run event loop: %m");
5239 (void) pty_forward_get_last_char(forward
, &last_char
);
5240 forward
= pty_forward_free(forward
);
5242 if (!arg_quiet
&& last_char
!= '\n')
5246 /* Kill if it is not dead yet anyway */
5247 if (!arg_register
&& !arg_keep_unit
&& bus
)
5248 terminate_scope(bus
, arg_machine
);
5250 /* Normally redundant, but better safe than sorry */
5251 (void) kill(*pid
, SIGKILL
);
5253 fd_kmsg_fifo
= safe_close(fd_kmsg_fifo
);
5255 if (arg_private_network
) {
5256 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5257 * to avoid having to move the parent to the child network namespace. */
5258 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG_SIGTERM
|FORK_WAIT
|FORK_LOG
, NULL
);
5263 _cleanup_close_
int parent_netns_fd
= -EBADF
;
5265 r
= namespace_open(getpid_cached(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5267 log_error_errno(r
, "Failed to open parent network namespace: %m");
5268 _exit(EXIT_FAILURE
);
5271 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5273 log_error_errno(r
, "Failed to enter child network namespace: %m");
5274 _exit(EXIT_FAILURE
);
5277 /* Reverse network interfaces pair list so that interfaces get their initial name back.
5278 * This is about ensuring interfaces get their old name back when being moved back. */
5279 arg_network_interfaces
= strv_reverse(arg_network_interfaces
);
5281 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5283 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5285 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5289 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5291 /* Tell machined that we are gone. */
5293 (void) unregister_machine(bus
, arg_machine
);
5296 /* We failed to wait for the container, or the container exited abnormally. */
5298 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5299 /* r > 0 → The container exited with a non-zero status.
5300 * As a special case, we need to replace 133 with a different value,
5301 * because 133 is special-cased in the service file to reboot the container.
5302 * otherwise → The container exited with zero status and a reboot was not requested.
5304 if (r
== EXIT_FORCE_RESTART
)
5305 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5307 return 0; /* finito */
5310 /* CONTAINER_REBOOTED, loop again */
5312 if (arg_keep_unit
) {
5313 /* Special handling if we are running as a service: instead of simply
5314 * restarting the machine we want to restart the entire service, so let's
5315 * inform systemd about this with the special exit code 133. The service
5316 * file uses RestartForceExitStatus=133 so that this results in a full
5317 * nspawn restart. This is necessary since we might have cgroup parameters
5318 * set we want to have flushed out. */
5319 *ret
= EXIT_FORCE_RESTART
;
5320 return 0; /* finito */
5323 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5324 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5326 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5327 *veth_created
= false;
5328 return 1; /* loop again */
5331 static int initialize_rlimits(void) {
5332 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5333 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5334 * container execution environments. */
5336 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5337 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5338 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5339 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5340 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5341 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5342 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5343 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5344 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5345 [RLIMIT_NICE
] = { 0, 0 },
5346 [RLIMIT_NOFILE
] = { 1024, 4096 },
5347 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5348 [RLIMIT_RTPRIO
] = { 0, 0 },
5349 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5350 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5352 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5353 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5354 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5355 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5356 * that PID 1 changes a number of other resource limits during early initialization which is why we
5357 * don't read the other limits from PID 1 but prefer the static table above. */
5362 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5363 /* Let's only fill in what the user hasn't explicitly configured anyway */
5364 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5365 const struct rlimit
*v
;
5366 struct rlimit buffer
;
5368 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5369 /* For these two let's read the limits off PID 1. See above for an explanation. */
5371 r
= pid_getrlimit(1, rl
, &buffer
);
5373 return log_error_errno(r
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5376 } else if (rl
== RLIMIT_NOFILE
) {
5377 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5378 * userspace. Given that nspawn containers are often run without our PID 1,
5379 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5380 * so that container userspace gets similar resources as host userspace
5382 buffer
= kernel_defaults
[rl
];
5383 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5386 v
= kernel_defaults
+ rl
;
5388 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5389 if (!arg_rlimit
[rl
])
5393 if (DEBUG_LOGGING
) {
5394 _cleanup_free_
char *k
= NULL
;
5396 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5397 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5404 static int cant_be_in_netns(void) {
5405 _cleanup_close_
int fd
= -EBADF
;
5409 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5410 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5413 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5416 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5418 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5420 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5421 if (r
== -ENOENT
|| ERRNO_IS_NEG_DISCONNECT(r
))
5422 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5423 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5425 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5427 r
= getpeercred(fd
, &ucred
);
5429 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5431 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5433 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5435 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5436 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5440 static int run(int argc
, char *argv
[]) {
5441 bool remove_directory
= false, remove_image
= false, veth_created
= false, remove_tmprootdir
= false;
5442 _cleanup_close_
int master
= -EBADF
;
5443 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5444 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5445 char veth_name
[IFNAMSIZ
] = "";
5446 struct ExposeArgs expose_args
= {};
5447 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5448 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5449 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5450 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5451 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5454 log_parse_environment();
5457 r
= parse_argv(argc
, argv
);
5461 if (geteuid() != 0) {
5462 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5463 argc
>= 2 ? "Need to be root." :
5464 "Need to be root (and some arguments are usually required).\nHint: try --help");
5468 r
= cant_be_in_netns();
5472 r
= initialize_rlimits();
5476 r
= load_oci_bundle();
5484 r
= determine_names();
5488 r
= load_settings();
5492 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
5493 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
5495 if (!arg_private_network
&& arg_userns_mode
!= USER_NAMESPACE_NO
&& arg_uid_shift
> 0)
5496 arg_caps_retain
&= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE
);
5500 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5504 r
= verify_arguments();
5508 r
= verify_network_interfaces_initialized();
5512 /* Reapply environment settings. */
5513 (void) detect_unified_cgroup_hierarchy_from_environment();
5515 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5516 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5517 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5518 (void) ignore_signals(SIGPIPE
);
5520 n_fd_passed
= sd_listen_fds(false);
5521 if (n_fd_passed
> 0) {
5522 r
= fdset_new_listen_fds(&fds
, false);
5524 log_error_errno(r
, "Failed to collect file descriptors: %m");
5529 /* The "default" umask. This is appropriate for most file and directory
5530 * operations performed by nspawn, and is the umask that will be used for
5531 * the child. Functions like copy_devnodes() change the umask temporarily. */
5534 if (arg_directory
) {
5537 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5538 * /var from the host will propagate into container dynamically (because bad things happen if
5539 * two systems write to the same /var). Let's allow it for the special cases where /var is
5540 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5541 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5542 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5543 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5547 if (arg_ephemeral
) {
5548 _cleanup_free_
char *np
= NULL
;
5550 r
= chase_and_update(&arg_directory
, 0);
5554 /* If the specified path is a mount point we generate the new snapshot immediately
5555 * inside it under a random name. However if the specified is not a mount point we
5556 * create the new snapshot in the parent directory, just next to it. */
5557 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5559 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5563 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5565 r
= tempfn_random(arg_directory
, "machine.", &np
);
5567 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5571 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5572 * only owned by us and no one else. */
5573 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5575 log_error_errno(r
, "Failed to lock %s: %m", np
);
5580 BLOCK_SIGNALS(SIGINT
);
5581 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_directory
, AT_FDCWD
, np
,
5582 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5583 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5584 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5585 BTRFS_SNAPSHOT_RECURSIVE
|
5586 BTRFS_SNAPSHOT_QUOTA
|
5587 BTRFS_SNAPSHOT_SIGINT
);
5590 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5594 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5598 free_and_replace(arg_directory
, np
);
5599 remove_directory
= true;
5601 r
= chase_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5605 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5607 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5611 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5616 r
= chase_and_update(&arg_template
, 0);
5621 BLOCK_SIGNALS(SIGINT
);
5622 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_template
, AT_FDCWD
, arg_directory
,
5623 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5624 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5625 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5626 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5627 BTRFS_SNAPSHOT_RECURSIVE
|
5628 BTRFS_SNAPSHOT_QUOTA
|
5629 BTRFS_SNAPSHOT_SIGINT
);
5632 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5633 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5634 else if (r
== -EINTR
) {
5635 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5638 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5641 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5642 "Populated %s from template %s.", arg_directory
, arg_template
);
5646 if (arg_start_mode
== START_BOOT
) {
5647 _cleanup_free_
char *b
= NULL
;
5649 int check_os_release
, is_os_tree
;
5651 if (arg_pivot_root_new
) {
5652 b
= path_join(arg_directory
, arg_pivot_root_new
);
5662 check_os_release
= getenv_bool("SYSTEMD_NSPAWN_CHECK_OS_RELEASE");
5663 if (check_os_release
< 0 && check_os_release
!= -ENXIO
) {
5664 r
= log_error_errno(check_os_release
, "Failed to parse $SYSTEMD_NSPAWN_CHECK_OS_RELEASE: %m");
5668 is_os_tree
= path_is_os_tree(p
);
5669 if (is_os_tree
== 0 && check_os_release
== 0)
5670 log_debug("Directory %s is missing an os-release file, continuing anyway.", p
);
5671 else if (is_os_tree
<= 0) {
5672 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5673 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5677 _cleanup_free_
char *p
= NULL
;
5679 if (arg_pivot_root_new
)
5680 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5682 p
= path_join(arg_directory
, "/usr/");
5688 if (laccess(p
, F_OK
) < 0) {
5689 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5690 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5696 DissectImageFlags dissect_image_flags
=
5697 DISSECT_IMAGE_GENERIC_ROOT
|
5698 DISSECT_IMAGE_REQUIRE_ROOT
|
5699 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5700 DISSECT_IMAGE_USR_NO_ROOT
|
5701 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5702 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5704 assert(!arg_template
);
5706 r
= chase_and_update(&arg_image
, 0);
5710 if (arg_ephemeral
) {
5711 _cleanup_free_
char *np
= NULL
;
5713 r
= tempfn_random(arg_image
, "machine.", &np
);
5715 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5719 /* Always take an exclusive lock on our own ephemeral copy. */
5720 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5722 log_error_errno(r
, "Failed to create image lock: %m");
5727 BLOCK_SIGNALS(SIGINT
);
5728 r
= copy_file_full(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600,
5729 FS_NOCOW_FL
, FS_NOCOW_FL
,
5730 COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
,
5734 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5738 r
= log_error_errno(r
, "Failed to copy image file: %m");
5742 free_and_replace(arg_image
, np
);
5743 remove_image
= true;
5745 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5747 log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5751 log_error_errno(r
, "Failed to create image lock: %m");
5755 r
= verity_settings_load(
5756 &arg_verity_settings
,
5757 arg_image
, NULL
, NULL
);
5759 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5763 if (arg_verity_settings
.data_path
)
5764 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5767 if (!mkdtemp(tmprootdir
)) {
5768 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5772 remove_tmprootdir
= true;
5774 arg_directory
= strdup(tmprootdir
);
5775 if (!arg_directory
) {
5780 r
= loop_device_make_by_path(
5782 arg_read_only
? O_RDONLY
: O_RDWR
,
5783 /* sector_size= */ UINT32_MAX
,
5784 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5788 log_error_errno(r
, "Failed to set up loopback block device: %m");
5792 r
= dissect_loop_device_and_warn(
5794 &arg_verity_settings
,
5795 /* mount_options=*/ NULL
,
5796 arg_image_policy
?: &image_policy_container
,
5797 dissect_image_flags
,
5800 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5801 log_notice("Note that the disk image needs to\n"
5802 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5803 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5804 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5805 " d) or contain a file system without a partition table\n"
5806 "in order to be bootable with systemd-nspawn.");
5812 r
= dissected_image_load_verity_sig_partition(
5815 &arg_verity_settings
);
5819 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5820 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5821 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5823 r
= dissected_image_decrypt_interactively(
5826 &arg_verity_settings
,
5831 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5832 if (remove_image
&& unlink(arg_image
) >= 0)
5833 remove_image
= false;
5835 if (arg_architecture
< 0)
5836 arg_architecture
= dissected_image_architecture(dissected_image
);
5839 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5843 if (arg_console_mode
< 0)
5844 arg_console_mode
= isatty(STDIN_FILENO
) && isatty(STDOUT_FILENO
) ?
5845 CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5847 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5851 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5852 arg_machine
, arg_image
?: arg_directory
);
5854 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, SIGRTMIN
+18, -1) >= 0);
5856 r
= make_reaper_process(true);
5858 log_error_errno(r
, "Failed to become subreaper: %m");
5862 if (arg_expose_ports
) {
5863 r
= fw_ctx_new(&fw_ctx
);
5865 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5868 expose_args
.fw_ctx
= fw_ctx
;
5871 r
= run_container(dissected_image
,
5873 veth_name
, &veth_created
,
5874 &expose_args
, &master
,
5881 (void) sd_notify(false,
5882 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5883 "STOPPING=1\nSTATUS=Terminating...");
5886 (void) kill(pid
, SIGKILL
);
5888 /* Try to flush whatever is still queued in the pty */
5890 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5891 master
= safe_close(master
);
5895 (void) wait_for_terminate(pid
, NULL
);
5899 if (remove_directory
&& arg_directory
) {
5902 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5904 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5907 if (remove_image
&& arg_image
) {
5908 if (unlink(arg_image
) < 0)
5909 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5912 if (remove_tmprootdir
) {
5913 if (rmdir(tmprootdir
) < 0)
5914 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5920 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5921 (void) rm_rf(p
, REMOVE_ROOT
);
5924 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5925 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5928 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5929 (void) remove_bridge(arg_network_zone
);
5931 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5932 expose_port_free_all(arg_expose_ports
);
5933 rlimit_free_all(arg_rlimit
);
5934 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5942 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);