1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
8 #include <linux/loop.h>
10 #include <selinux/selinux.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.h>
23 #include "sd-daemon.h"
26 #include "alloc-util.h"
27 #include "ether-addr-util.h"
29 #include "base-filesystem.h"
30 #include "blkid-util.h"
31 #include "btrfs-util.h"
33 #include "bus-error.h"
34 #include "bus-locator.h"
37 #include "capability-util.h"
38 #include "cgroup-util.h"
40 #include "common-signal.h"
42 #include "cpu-set-util.h"
43 #include "creds-util.h"
44 #include "dev-setup.h"
45 #include "discover-image.h"
46 #include "dissect-image.h"
52 #include "format-util.h"
55 #include "hexdecoct.h"
56 #include "hostname-setup.h"
57 #include "hostname-util.h"
58 #include "id128-util.h"
61 #include "loop-util.h"
62 #include "loopback-setup.h"
63 #include "machine-credential.h"
65 #include "main-func.h"
66 #include "missing_sched.h"
68 #include "mount-util.h"
69 #include "mountpoint-util.h"
70 #include "namespace-util.h"
71 #include "netlink-util.h"
72 #include "nspawn-bind-user.h"
73 #include "nspawn-cgroup.h"
74 #include "nspawn-def.h"
75 #include "nspawn-expose-ports.h"
76 #include "nspawn-mount.h"
77 #include "nspawn-network.h"
78 #include "nspawn-oci.h"
79 #include "nspawn-patch-uid.h"
80 #include "nspawn-register.h"
81 #include "nspawn-seccomp.h"
82 #include "nspawn-settings.h"
83 #include "nspawn-setuid.h"
84 #include "nspawn-stub-pid1.h"
85 #include "nspawn-util.h"
87 #include "nulstr-util.h"
90 #include "parse-argument.h"
91 #include "parse-util.h"
92 #include "pretty-print.h"
93 #include "process-util.h"
95 #include "random-util.h"
96 #include "raw-clone.h"
97 #include "resolve-util.h"
98 #include "rlimit-util.h"
100 #include "seccomp-util.h"
101 #include "selinux-util.h"
102 #include "signal-util.h"
103 #include "socket-util.h"
104 #include "stat-util.h"
105 #include "stdio-util.h"
106 #include "string-table.h"
107 #include "string-util.h"
109 #include "sysctl-util.h"
110 #include "terminal-util.h"
111 #include "tmpfile-util.h"
112 #include "umask-util.h"
113 #include "unit-name.h"
114 #include "user-util.h"
116 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
117 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
118 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
120 #define EXIT_FORCE_RESTART 133
122 typedef enum ContainerStatus
{
123 CONTAINER_TERMINATED
,
127 static char *arg_directory
= NULL
;
128 static char *arg_template
= NULL
;
129 static char *arg_chdir
= NULL
;
130 static char *arg_pivot_root_new
= NULL
;
131 static char *arg_pivot_root_old
= NULL
;
132 static char *arg_user
= NULL
;
133 static uid_t arg_uid
= UID_INVALID
;
134 static gid_t arg_gid
= GID_INVALID
;
135 static gid_t
* arg_supplementary_gids
= NULL
;
136 static size_t arg_n_supplementary_gids
= 0;
137 static sd_id128_t arg_uuid
= {};
138 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
139 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
140 static const char *arg_selinux_context
= NULL
;
141 static const char *arg_selinux_apifs_context
= NULL
;
142 static char *arg_slice
= NULL
;
143 static bool arg_private_network
= false;
144 static bool arg_read_only
= false;
145 static StartMode arg_start_mode
= START_PID1
;
146 static bool arg_ephemeral
= false;
147 static LinkJournal arg_link_journal
= LINK_AUTO
;
148 static bool arg_link_journal_try
= false;
149 static uint64_t arg_caps_retain
=
150 (1ULL << CAP_AUDIT_CONTROL
) |
151 (1ULL << CAP_AUDIT_WRITE
) |
152 (1ULL << CAP_CHOWN
) |
153 (1ULL << CAP_DAC_OVERRIDE
) |
154 (1ULL << CAP_DAC_READ_SEARCH
) |
155 (1ULL << CAP_FOWNER
) |
156 (1ULL << CAP_FSETID
) |
157 (1ULL << CAP_IPC_OWNER
) |
159 (1ULL << CAP_LEASE
) |
160 (1ULL << CAP_LINUX_IMMUTABLE
) |
161 (1ULL << CAP_MKNOD
) |
162 (1ULL << CAP_NET_BIND_SERVICE
) |
163 (1ULL << CAP_NET_BROADCAST
) |
164 (1ULL << CAP_NET_RAW
) |
165 (1ULL << CAP_SETFCAP
) |
166 (1ULL << CAP_SETGID
) |
167 (1ULL << CAP_SETPCAP
) |
168 (1ULL << CAP_SETUID
) |
169 (1ULL << CAP_SYS_ADMIN
) |
170 (1ULL << CAP_SYS_BOOT
) |
171 (1ULL << CAP_SYS_CHROOT
) |
172 (1ULL << CAP_SYS_NICE
) |
173 (1ULL << CAP_SYS_PTRACE
) |
174 (1ULL << CAP_SYS_RESOURCE
) |
175 (1ULL << CAP_SYS_TTY_CONFIG
);
176 static uint64_t arg_caps_ambient
= 0;
177 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
178 static CustomMount
*arg_custom_mounts
= NULL
;
179 static size_t arg_n_custom_mounts
= 0;
180 static char **arg_setenv
= NULL
;
181 static bool arg_quiet
= false;
182 static bool arg_register
= true;
183 static bool arg_keep_unit
= false;
184 static char **arg_network_interfaces
= NULL
;
185 static char **arg_network_macvlan
= NULL
;
186 static char **arg_network_ipvlan
= NULL
;
187 static bool arg_network_veth
= false;
188 static char **arg_network_veth_extra
= NULL
;
189 static char *arg_network_bridge
= NULL
;
190 static char *arg_network_zone
= NULL
;
191 static char *arg_network_namespace_path
= NULL
;
192 struct ether_addr arg_network_provided_mac
= {};
193 static PagerFlags arg_pager_flags
= 0;
194 static unsigned long arg_personality
= PERSONALITY_INVALID
;
195 static char *arg_image
= NULL
;
196 static char *arg_oci_bundle
= NULL
;
197 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
198 static ExposePort
*arg_expose_ports
= NULL
;
199 static char **arg_property
= NULL
;
200 static sd_bus_message
*arg_property_message
= NULL
;
201 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
202 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
203 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
204 static int arg_kill_signal
= 0;
205 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
206 static SettingsMask arg_settings_mask
= 0;
207 static int arg_settings_trusted
= -1;
208 static char **arg_parameters
= NULL
;
209 static const char *arg_container_service_name
= "systemd-nspawn";
210 static bool arg_notify_ready
= false;
211 static bool arg_use_cgns
= true;
212 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
213 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
214 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
215 static char **arg_syscall_allow_list
= NULL
;
216 static char **arg_syscall_deny_list
= NULL
;
218 static scmp_filter_ctx arg_seccomp
= NULL
;
220 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
221 static bool arg_no_new_privileges
= false;
222 static int arg_oom_score_adjust
= 0;
223 static bool arg_oom_score_adjust_set
= false;
224 static CPUSet arg_cpu_set
= {};
225 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
226 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
227 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
228 static DeviceNode
* arg_extra_nodes
= NULL
;
229 static size_t arg_n_extra_nodes
= 0;
230 static char **arg_sysctl
= NULL
;
231 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
232 static MachineCredential
*arg_credentials
= NULL
;
233 static size_t arg_n_credentials
= 0;
234 static char **arg_bind_user
= NULL
;
235 static bool arg_suppress_sync
= false;
236 static char *arg_settings_filename
= NULL
;
237 static Architecture arg_architecture
= _ARCHITECTURE_INVALID
;
238 static ImagePolicy
*arg_image_policy
= NULL
;
240 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
260 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
262 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
264 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
265 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
267 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
269 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
270 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
271 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
272 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
273 STATIC_DESTRUCTOR_REGISTER(arg_image_policy
, image_policy_freep
);
275 static int handle_arg_console(const char *arg
) {
276 if (streq(arg
, "help")) {
285 if (streq(arg
, "interactive"))
286 arg_console_mode
= CONSOLE_INTERACTIVE
;
287 else if (streq(arg
, "read-only"))
288 arg_console_mode
= CONSOLE_READ_ONLY
;
289 else if (streq(arg
, "passive"))
290 arg_console_mode
= CONSOLE_PASSIVE
;
291 else if (streq(arg
, "pipe")) {
292 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
293 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
294 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
295 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
296 "Proceeding anyway.");
298 arg_console_mode
= CONSOLE_PIPE
;
299 } else if (streq(arg
, "autopipe")) {
300 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
301 arg_console_mode
= CONSOLE_INTERACTIVE
;
303 arg_console_mode
= CONSOLE_PIPE
;
305 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
307 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
311 static int help(void) {
312 _cleanup_free_
char *link
= NULL
;
315 pager_open(arg_pager_flags
);
317 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
321 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
322 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
323 " -h --help Show this help\n"
324 " --version Print version string\n"
325 " -q --quiet Do not show status information\n"
326 " --no-pager Do not pipe output into a pager\n"
327 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
329 " -D --directory=PATH Root directory for the container\n"
330 " --template=PATH Initialize root directory from template directory,\n"
332 " -x --ephemeral Run container with snapshot of root directory, and\n"
333 " remove it after exit\n"
334 " -i --image=PATH Root file system disk image (or device node) for\n"
336 " --image-policy=POLICY Specify disk image dissection policy\n"
337 " --oci-bundle=PATH OCI bundle directory\n"
338 " --read-only Mount the root directory read-only\n"
339 " --volatile[=MODE] Run the system in volatile mode\n"
340 " --root-hash=HASH Specify verity root hash for root disk image\n"
341 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
342 " as a DER encoded PKCS7, either as a path to a file\n"
343 " or as an ASCII base64 encoded string prefixed by\n"
345 " --verity-data=PATH Specify hash device for verity\n"
346 " --pivot-root=PATH[:PATH]\n"
347 " Pivot root to given directory in the container\n\n"
348 "%3$sExecution:%4$s\n"
349 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
350 " -b --boot Boot up full system (i.e. invoke init)\n"
351 " --chdir=PATH Set working directory in the container\n"
352 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
353 " -u --user=USER Run the command under specified user or UID\n"
354 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
355 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
356 " --suppress-sync=BOOLEAN\n"
357 " Suppress any form of disk data synchronization\n\n"
358 "%3$sSystem Identity:%4$s\n"
359 " -M --machine=NAME Set the machine name for the container\n"
360 " --hostname=NAME Override the hostname for the container\n"
361 " --uuid=UUID Set a specific machine UUID for the container\n\n"
362 "%3$sProperties:%4$s\n"
363 " -S --slice=SLICE Place the container in the specified slice\n"
364 " --property=NAME=VALUE Set scope unit property\n"
365 " --register=BOOLEAN Register container as machine\n"
366 " --keep-unit Do not register a scope for the machine, reuse\n"
367 " the service unit nspawn is running in\n\n"
368 "%3$sUser Namespacing:%4$s\n"
369 " --private-users=no Run without user namespacing\n"
370 " --private-users=yes|pick|identity\n"
371 " Run within user namespace, autoselect UID/GID range\n"
372 " --private-users=UIDBASE[:NUIDS]\n"
373 " Similar, but with user configured UID/GID range\n"
374 " --private-users-ownership=MODE\n"
375 " Adjust ('chown') or map ('map') OS tree ownership\n"
376 " to private UID/GID range\n"
377 " -U Equivalent to --private-users=pick and\n"
378 " --private-users-ownership=auto\n\n"
379 "%3$sNetworking:%4$s\n"
380 " --private-network Disable network in container\n"
381 " --network-interface=HOSTIF[:CONTAINERIF]\n"
382 " Assign an existing network interface to the\n"
384 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
385 " Create a macvlan network interface based on an\n"
386 " existing network interface to the container\n"
387 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
388 " Create an ipvlan network interface based on an\n"
389 " existing network interface to the container\n"
390 " -n --network-veth Add a virtual Ethernet connection between host\n"
392 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
393 " Add an additional virtual Ethernet link between\n"
394 " host and container\n"
395 " --network-bridge=INTERFACE\n"
396 " Add a virtual Ethernet connection to the container\n"
397 " and attach it to an existing bridge on the host\n"
398 " --network-zone=NAME Similar, but attach the new interface to an\n"
399 " an automatically managed bridge interface\n"
400 " --network-namespace-path=PATH\n"
401 " Set network namespace to the one represented by\n"
402 " the specified kernel namespace file node\n"
403 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
404 " Expose a container IP port on the host\n\n"
405 "%3$sSecurity:%4$s\n"
406 " --capability=CAP In addition to the default, retain specified\n"
408 " --drop-capability=CAP Drop the specified capability from the default set\n"
409 " --ambient-capability=CAP\n"
410 " Sets the specified capability for the started\n"
411 " process. Not useful if booting a machine.\n"
412 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
413 " --system-call-filter=LIST|~LIST\n"
414 " Permit/prohibit specific system calls\n"
415 " -Z --selinux-context=SECLABEL\n"
416 " Set the SELinux security context to be used by\n"
417 " processes in the container\n"
418 " -L --selinux-apifs-context=SECLABEL\n"
419 " Set the SELinux security context to be used by\n"
420 " API/tmpfs file systems in the container\n\n"
421 "%3$sResources:%4$s\n"
422 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
423 " --oom-score-adjust=VALUE\n"
424 " Adjust the OOM score value for the payload\n"
425 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
426 " --personality=ARCH Pick personality for this container\n\n"
427 "%3$sIntegration:%4$s\n"
428 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
429 " --timezone=MODE Select mode of /etc/localtime initialization\n"
430 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
431 " host, try-guest, try-host\n"
432 " -j Equivalent to --link-journal=try-guest\n\n"
434 " --bind=PATH[:PATH[:OPTIONS]]\n"
435 " Bind mount a file or directory from the host into\n"
437 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
438 " Similar, but creates a read-only bind mount\n"
439 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
441 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
442 " --overlay=PATH[:PATH...]:PATH\n"
443 " Create an overlay mount from the host to \n"
445 " --overlay-ro=PATH[:PATH...]:PATH\n"
446 " Similar, but creates a read-only overlay mount\n"
447 " --bind-user=NAME Bind user from host to container\n\n"
448 "%3$sInput/Output:%4$s\n"
449 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
450 " set up for the container.\n"
451 " -P --pipe Equivalent to --console=pipe\n\n"
452 "%3$sCredentials:%4$s\n"
453 " --set-credential=ID:VALUE\n"
454 " Pass a credential with literal value to container.\n"
455 " --load-credential=ID:PATH\n"
456 " Load credential to pass to container from file or\n"
457 " AF_UNIX stream socket.\n"
458 "\nSee the %2$s for details.\n",
459 program_invocation_short_name
,
469 static int custom_mount_check_all(void) {
472 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
473 CustomMount
*m
= &arg_custom_mounts
[i
];
475 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
476 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
477 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
478 "--private-users-ownership=own may not be combined with custom root mounts.");
479 if (arg_uid_shift
== UID_INVALID
)
480 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
481 "--private-users with automatic UID shift may not be combined with custom root mounts.");
488 static int detect_unified_cgroup_hierarchy_from_environment(void) {
489 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
492 /* Allow the user to control whether the unified hierarchy is used */
496 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
497 var
= "UNIFIED_CGROUP_HIERARCHY";
502 r
= parse_boolean(e
);
504 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
506 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
508 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
514 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
517 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
518 * in the image actually supports. */
519 r
= cg_all_unified();
521 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
523 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
524 * routine only detects 231, so we'll have a false negative here for 230. */
525 r
= systemd_installation_has_version(directory
, "230");
527 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
529 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
531 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
532 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
533 /* Mixed cgroup hierarchy support was added in 233 */
534 r
= systemd_installation_has_version(directory
, "233");
536 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
538 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
540 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
542 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
544 log_debug("Using %s hierarchy for container.",
545 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
546 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
551 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
556 _cleanup_free_
char *t
= NULL
;
558 r
= extract_first_word(&spec
, &t
, ",", 0);
560 return log_error_errno(r
, "Failed to parse capability %s.", t
);
564 if (streq(t
, "help")) {
565 for (int i
= 0; i
< capability_list_length(); i
++) {
568 name
= capability_to_name(i
);
579 r
= capability_from_name(t
);
581 return log_error_errno(r
, "Failed to parse capability %s.", t
);
588 return 1; /* continue */
591 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
594 r
= getenv_bool(name
);
598 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
600 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
601 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
605 static int parse_mount_settings_env(void) {
609 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
610 if (r
< 0 && r
!= -ENXIO
)
611 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
613 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
615 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
616 if (streq_ptr(e
, "network"))
617 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
620 r
= parse_boolean(e
);
622 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
624 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
625 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
631 static int parse_environment(void) {
635 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
638 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
641 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
644 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
648 r
= parse_mount_settings_env();
652 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
653 * even if it is supported. If not supported, it has no effect. */
654 if (!cg_ns_supported())
655 arg_use_cgns
= false;
657 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
660 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
664 arg_use_cgns
= r
> 0;
665 arg_settings_mask
|= SETTING_USE_CGNS
;
669 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
671 arg_container_service_name
= e
;
673 e
= getenv("SYSTEMD_NSPAWN_NETWORK_MAC");
675 r
= parse_ether_addr(e
, &arg_network_provided_mac
);
677 return log_error_errno(r
, "Failed to parse provided MAC address via environment variable");
680 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
682 arg_suppress_sync
= r
;
683 else if (r
!= -ENXIO
)
684 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
686 return detect_unified_cgroup_hierarchy_from_environment();
689 static int parse_argv(int argc
, char *argv
[]) {
696 ARG_AMBIENT_CAPABILITY
,
708 ARG_NETWORK_INTERFACE
,
713 ARG_NETWORK_VETH_EXTRA
,
714 ARG_NETWORK_NAMESPACE_PATH
,
724 ARG_PRIVATE_USERS_CHOWN
,
725 ARG_PRIVATE_USERS_OWNERSHIP
,
730 ARG_SYSTEM_CALL_FILTER
,
733 ARG_NO_NEW_PRIVILEGES
,
734 ARG_OOM_SCORE_ADJUST
,
749 static const struct option options
[] = {
750 { "help", no_argument
, NULL
, 'h' },
751 { "version", no_argument
, NULL
, ARG_VERSION
},
752 { "directory", required_argument
, NULL
, 'D' },
753 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
754 { "ephemeral", no_argument
, NULL
, 'x' },
755 { "user", required_argument
, NULL
, 'u' },
756 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
757 { "as-pid2", no_argument
, NULL
, 'a' },
758 { "boot", no_argument
, NULL
, 'b' },
759 { "uuid", required_argument
, NULL
, ARG_UUID
},
760 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
761 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
762 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
763 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
764 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
765 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
766 { "bind", required_argument
, NULL
, ARG_BIND
},
767 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
768 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
769 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
770 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
771 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
772 { "machine", required_argument
, NULL
, 'M' },
773 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
774 { "slice", required_argument
, NULL
, 'S' },
775 { "setenv", required_argument
, NULL
, 'E' },
776 { "selinux-context", required_argument
, NULL
, 'Z' },
777 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
778 { "quiet", no_argument
, NULL
, 'q' },
779 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
780 { "register", required_argument
, NULL
, ARG_REGISTER
},
781 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
782 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
783 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
784 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
785 { "network-veth", no_argument
, NULL
, 'n' },
786 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
787 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
788 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
789 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
790 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
791 { "image", required_argument
, NULL
, 'i' },
792 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
793 { "port", required_argument
, NULL
, 'p' },
794 { "property", required_argument
, NULL
, ARG_PROPERTY
},
795 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
796 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
797 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
798 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
799 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
800 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
801 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
802 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
803 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
804 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
805 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
806 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
807 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
808 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
809 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
810 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
811 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
812 { "console", required_argument
, NULL
, ARG_CONSOLE
},
813 { "pipe", no_argument
, NULL
, ARG_PIPE
},
814 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
815 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
816 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
817 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
818 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
819 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
820 { "image-policy", required_argument
, NULL
, ARG_IMAGE_POLICY
},
825 uint64_t plus
= 0, minus
= 0;
826 bool mask_all_settings
= false, mask_no_settings
= false;
831 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
832 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
834 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
844 r
= parse_path_argument(optarg
, false, &arg_directory
);
848 arg_settings_mask
|= SETTING_DIRECTORY
;
852 r
= parse_path_argument(optarg
, false, &arg_template
);
856 arg_settings_mask
|= SETTING_DIRECTORY
;
860 r
= parse_path_argument(optarg
, false, &arg_image
);
864 arg_settings_mask
|= SETTING_DIRECTORY
;
868 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
875 arg_ephemeral
= true;
876 arg_settings_mask
|= SETTING_EPHEMERAL
;
880 r
= free_and_strdup(&arg_user
, optarg
);
884 arg_settings_mask
|= SETTING_USER
;
887 case ARG_NETWORK_ZONE
: {
888 _cleanup_free_
char *j
= NULL
;
890 j
= strjoin("vz-", optarg
);
894 if (!ifname_valid(j
))
895 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
896 "Network zone name not valid: %s", j
);
898 free_and_replace(arg_network_zone
, j
);
900 arg_network_veth
= true;
901 arg_private_network
= true;
902 arg_settings_mask
|= SETTING_NETWORK
;
906 case ARG_NETWORK_BRIDGE
:
908 if (!ifname_valid(optarg
))
909 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
910 "Bridge interface name not valid: %s", optarg
);
912 r
= free_and_strdup(&arg_network_bridge
, optarg
);
918 arg_network_veth
= true;
919 arg_private_network
= true;
920 arg_settings_mask
|= SETTING_NETWORK
;
923 case ARG_NETWORK_VETH_EXTRA
:
924 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
926 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
928 arg_private_network
= true;
929 arg_settings_mask
|= SETTING_NETWORK
;
932 case ARG_NETWORK_INTERFACE
:
933 r
= interface_pair_parse(&arg_network_interfaces
, optarg
);
937 arg_private_network
= true;
938 arg_settings_mask
|= SETTING_NETWORK
;
941 case ARG_NETWORK_MACVLAN
:
942 r
= macvlan_pair_parse(&arg_network_macvlan
, optarg
);
946 arg_private_network
= true;
947 arg_settings_mask
|= SETTING_NETWORK
;
950 case ARG_NETWORK_IPVLAN
:
951 r
= ipvlan_pair_parse(&arg_network_ipvlan
, optarg
);
956 case ARG_PRIVATE_NETWORK
:
957 arg_private_network
= true;
958 arg_settings_mask
|= SETTING_NETWORK
;
961 case ARG_NETWORK_NAMESPACE_PATH
:
962 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
966 arg_settings_mask
|= SETTING_NETWORK
;
970 if (arg_start_mode
== START_PID2
)
971 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
972 "--boot and --as-pid2 may not be combined.");
974 arg_start_mode
= START_BOOT
;
975 arg_settings_mask
|= SETTING_START_MODE
;
979 if (arg_start_mode
== START_BOOT
)
980 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
981 "--boot and --as-pid2 may not be combined.");
983 arg_start_mode
= START_PID2
;
984 arg_settings_mask
|= SETTING_START_MODE
;
988 r
= id128_from_string_nonzero(optarg
, &arg_uuid
);
990 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
991 "Machine UUID may not be all zeroes.");
993 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
995 arg_settings_mask
|= SETTING_MACHINE_ID
;
999 _cleanup_free_
char *mangled
= NULL
;
1001 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1005 free_and_replace(arg_slice
, mangled
);
1006 arg_settings_mask
|= SETTING_SLICE
;
1011 if (isempty(optarg
))
1012 arg_machine
= mfree(arg_machine
);
1014 if (!hostname_is_valid(optarg
, 0))
1015 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1016 "Invalid machine name: %s", optarg
);
1018 r
= free_and_strdup(&arg_machine
, optarg
);
1025 if (isempty(optarg
))
1026 arg_hostname
= mfree(arg_hostname
);
1028 if (!hostname_is_valid(optarg
, 0))
1029 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1030 "Invalid hostname: %s", optarg
);
1032 r
= free_and_strdup(&arg_hostname
, optarg
);
1037 arg_settings_mask
|= SETTING_HOSTNAME
;
1041 arg_selinux_context
= optarg
;
1045 arg_selinux_apifs_context
= optarg
;
1049 arg_read_only
= true;
1050 arg_settings_mask
|= SETTING_READ_ONLY
;
1053 case ARG_AMBIENT_CAPABILITY
: {
1055 r
= parse_capability_spec(optarg
, &m
);
1058 arg_caps_ambient
|= m
;
1059 arg_settings_mask
|= SETTING_CAPABILITY
;
1062 case ARG_CAPABILITY
:
1063 case ARG_DROP_CAPABILITY
: {
1065 r
= parse_capability_spec(optarg
, &m
);
1069 if (c
== ARG_CAPABILITY
)
1073 arg_settings_mask
|= SETTING_CAPABILITY
;
1076 case ARG_NO_NEW_PRIVILEGES
:
1077 r
= parse_boolean(optarg
);
1079 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1081 arg_no_new_privileges
= r
;
1082 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1086 arg_link_journal
= LINK_GUEST
;
1087 arg_link_journal_try
= true;
1088 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1091 case ARG_LINK_JOURNAL
:
1092 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1094 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1096 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1101 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1103 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1105 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1109 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1111 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1113 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1117 case ARG_OVERLAY_RO
:
1118 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1119 if (r
== -EADDRNOTAVAIL
)
1120 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1122 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1124 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1127 case ARG_INACCESSIBLE
:
1128 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1130 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1132 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1136 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1138 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1140 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1147 case ARG_SHARE_SYSTEM
:
1148 /* We don't officially support this anymore, except for compat reasons. People should use the
1149 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1150 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1151 arg_clone_ns_flags
= 0;
1155 r
= parse_boolean(optarg
);
1157 log_error("Failed to parse --register= argument: %s", optarg
);
1165 arg_keep_unit
= true;
1168 case ARG_PERSONALITY
:
1170 arg_personality
= personality_from_string(optarg
);
1171 if (arg_personality
== PERSONALITY_INVALID
)
1172 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1173 "Unknown or unsupported personality '%s'.", optarg
);
1175 arg_settings_mask
|= SETTING_PERSONALITY
;
1181 arg_volatile_mode
= VOLATILE_YES
;
1182 else if (streq(optarg
, "help")) {
1183 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1188 m
= volatile_mode_from_string(optarg
);
1190 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1191 "Failed to parse --volatile= argument: %s", optarg
);
1193 arg_volatile_mode
= m
;
1196 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1200 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1202 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1204 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1206 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1210 if (strv_extend(&arg_property
, optarg
) < 0)
1215 case ARG_PRIVATE_USERS
: {
1220 else if (!in_charset(optarg
, DIGITS
))
1221 /* do *not* parse numbers as booleans */
1222 boolean
= parse_boolean(optarg
);
1227 /* no: User namespacing off */
1228 arg_userns_mode
= USER_NAMESPACE_NO
;
1229 arg_uid_shift
= UID_INVALID
;
1230 arg_uid_range
= UINT32_C(0x10000);
1231 } else if (boolean
> 0) {
1232 /* yes: User namespacing on, UID range is read from root dir */
1233 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1234 arg_uid_shift
= UID_INVALID
;
1235 arg_uid_range
= UINT32_C(0x10000);
1236 } else if (streq(optarg
, "pick")) {
1237 /* pick: User namespacing on, UID range is picked randomly */
1238 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1239 * implied by USER_NAMESPACE_PICK
1241 arg_uid_shift
= UID_INVALID
;
1242 arg_uid_range
= UINT32_C(0x10000);
1244 } else if (streq(optarg
, "identity")) {
1245 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1246 * itself, i.e. we don't actually map anything, but do take benefit of
1247 * isolation of capability sets. */
1248 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1250 arg_uid_range
= UINT32_C(0x10000);
1252 _cleanup_free_
char *buffer
= NULL
;
1253 const char *range
, *shift
;
1255 /* anything else: User namespacing on, UID range is explicitly configured */
1257 range
= strchr(optarg
, ':');
1259 buffer
= strndup(optarg
, range
- optarg
);
1265 r
= safe_atou32(range
, &arg_uid_range
);
1267 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1271 r
= parse_uid(shift
, &arg_uid_shift
);
1273 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1275 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1277 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1278 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1281 arg_settings_mask
|= SETTING_USERNS
;
1286 if (userns_supported()) {
1287 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1288 * implied by USER_NAMESPACE_PICK
1290 arg_uid_shift
= UID_INVALID
;
1291 arg_uid_range
= UINT32_C(0x10000);
1293 arg_settings_mask
|= SETTING_USERNS
;
1298 case ARG_PRIVATE_USERS_CHOWN
:
1299 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1301 arg_settings_mask
|= SETTING_USERNS
;
1304 case ARG_PRIVATE_USERS_OWNERSHIP
:
1305 if (streq(optarg
, "help")) {
1306 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1310 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1311 if (arg_userns_ownership
< 0)
1312 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1314 arg_settings_mask
|= SETTING_USERNS
;
1317 case ARG_KILL_SIGNAL
:
1318 if (streq(optarg
, "help")) {
1319 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1323 arg_kill_signal
= signal_from_string(optarg
);
1324 if (arg_kill_signal
< 0)
1325 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1327 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1332 /* no → do not read files
1333 * yes → read files, do not override cmdline, trust only subset
1334 * override → read files, override cmdline, trust only subset
1335 * trusted → read files, do not override cmdline, trust all
1338 r
= parse_boolean(optarg
);
1340 if (streq(optarg
, "trusted")) {
1341 mask_all_settings
= false;
1342 mask_no_settings
= false;
1343 arg_settings_trusted
= true;
1345 } else if (streq(optarg
, "override")) {
1346 mask_all_settings
= false;
1347 mask_no_settings
= true;
1348 arg_settings_trusted
= -1;
1350 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1353 mask_all_settings
= false;
1354 mask_no_settings
= false;
1355 arg_settings_trusted
= -1;
1358 mask_all_settings
= true;
1359 mask_no_settings
= false;
1360 arg_settings_trusted
= false;
1366 if (!path_is_absolute(optarg
))
1367 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1368 "Working directory %s is not an absolute path.", optarg
);
1370 r
= free_and_strdup(&arg_chdir
, optarg
);
1374 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1377 case ARG_PIVOT_ROOT
:
1378 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1380 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1382 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1385 case ARG_NOTIFY_READY
:
1386 r
= parse_boolean(optarg
);
1388 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1389 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1390 arg_notify_ready
= r
;
1391 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1394 case ARG_ROOT_HASH
: {
1395 _cleanup_free_
void *k
= NULL
;
1398 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1400 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1401 if (l
< sizeof(sd_id128_t
))
1402 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128-bit long: %s", optarg
);
1404 free_and_replace(arg_verity_settings
.root_hash
, k
);
1405 arg_verity_settings
.root_hash_size
= l
;
1409 case ARG_ROOT_HASH_SIG
: {
1414 if ((value
= startswith(optarg
, "base64:"))) {
1415 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1417 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1420 r
= read_full_file(optarg
, (char**) &p
, &l
);
1422 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1425 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1426 arg_verity_settings
.root_hash_sig_size
= l
;
1430 case ARG_VERITY_DATA
:
1431 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1436 case ARG_SYSTEM_CALL_FILTER
: {
1440 negative
= optarg
[0] == '~';
1441 items
= negative
? optarg
+ 1 : optarg
;
1444 _cleanup_free_
char *word
= NULL
;
1446 r
= extract_first_word(&items
, &word
, NULL
, 0);
1452 return log_error_errno(r
, "Failed to parse system call filter: %m");
1455 r
= strv_extend(&arg_syscall_deny_list
, word
);
1457 r
= strv_extend(&arg_syscall_allow_list
, word
);
1462 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1468 _cleanup_free_
char *name
= NULL
;
1471 if (streq(optarg
, "help")) {
1472 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1476 eq
= strchr(optarg
, '=');
1478 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1479 "--rlimit= expects an '=' assignment.");
1481 name
= strndup(optarg
, eq
- optarg
);
1485 rl
= rlimit_from_string_harder(name
);
1487 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1489 if (!arg_rlimit
[rl
]) {
1490 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1491 if (!arg_rlimit
[rl
])
1495 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1497 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1499 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1503 case ARG_OOM_SCORE_ADJUST
:
1504 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1506 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1508 arg_oom_score_adjust_set
= true;
1509 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1512 case ARG_CPU_AFFINITY
: {
1515 r
= parse_cpu_set(optarg
, &cpuset
);
1517 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1519 cpu_set_reset(&arg_cpu_set
);
1520 arg_cpu_set
= cpuset
;
1521 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1525 case ARG_RESOLV_CONF
:
1526 if (streq(optarg
, "help")) {
1527 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1531 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1532 if (arg_resolv_conf
< 0)
1533 return log_error_errno(arg_resolv_conf
,
1534 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1536 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1540 if (streq(optarg
, "help")) {
1541 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1545 arg_timezone
= timezone_mode_from_string(optarg
);
1546 if (arg_timezone
< 0)
1547 return log_error_errno(arg_timezone
,
1548 "Failed to parse /etc/localtime mode: %s", optarg
);
1550 arg_settings_mask
|= SETTING_TIMEZONE
;
1554 r
= handle_arg_console(optarg
);
1561 r
= handle_arg_console("pipe");
1567 arg_pager_flags
|= PAGER_DISABLE
;
1570 case ARG_SET_CREDENTIAL
:
1571 r
= machine_credential_set(&arg_credentials
, &arg_n_credentials
, optarg
);
1575 arg_settings_mask
|= SETTING_CREDENTIALS
;
1578 case ARG_LOAD_CREDENTIAL
:
1579 r
= machine_credential_load(&arg_credentials
, &arg_n_credentials
, optarg
);
1583 arg_settings_mask
|= SETTING_CREDENTIALS
;
1587 if (!valid_user_group_name(optarg
, 0))
1588 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1590 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1593 arg_settings_mask
|= SETTING_BIND_USER
;
1596 case ARG_SUPPRESS_SYNC
:
1597 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1601 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1604 case ARG_IMAGE_POLICY
:
1605 r
= parse_image_policy_argument(optarg
, &arg_image_policy
);
1614 assert_not_reached();
1617 if (argc
> optind
) {
1618 strv_free(arg_parameters
);
1619 arg_parameters
= strv_copy(argv
+ optind
);
1620 if (!arg_parameters
)
1623 arg_settings_mask
|= SETTING_START_MODE
;
1626 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1627 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1628 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1629 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1631 arg_directory
= TAKE_PTR(arg_template
);
1633 arg_caps_retain
|= plus
;
1634 arg_caps_retain
|= arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0;
1635 arg_caps_retain
&= ~minus
;
1637 /* Make sure to parse environment before we reset the settings mask below */
1638 r
= parse_environment();
1642 /* Load all settings from .nspawn files */
1643 if (mask_no_settings
)
1644 arg_settings_mask
= 0;
1646 /* Don't load any settings from .nspawn files */
1647 if (mask_all_settings
)
1648 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1653 static int verify_arguments(void) {
1656 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1657 /* If we are running the stub init in the container, we don't need to look at what the init
1658 * in the container supports, because we are not using it. Let's immediately pick the right
1659 * setting based on the host system configuration.
1661 * We only do this, if the user didn't use an environment variable to override the detection.
1664 r
= cg_all_unified();
1666 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1668 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1669 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1670 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1672 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1675 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1676 arg_mount_settings
|= MOUNT_USE_USERNS
;
1678 if (arg_private_network
)
1679 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1681 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1682 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1683 arg_register
= false;
1684 if (arg_start_mode
!= START_PID1
)
1685 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1688 if (arg_userns_ownership
< 0)
1689 arg_userns_ownership
=
1690 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1691 USER_NAMESPACE_OWNERSHIP_OFF
;
1693 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1694 arg_kill_signal
= SIGRTMIN
+3;
1696 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 */
1697 arg_read_only
= true;
1699 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1700 arg_read_only
= true;
1702 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1703 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1704 * The latter is not technically a user session, but we don't need to labour the point. */
1705 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1707 if (arg_directory
&& arg_image
)
1708 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1710 if (arg_template
&& arg_image
)
1711 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1713 if (arg_template
&& !(arg_directory
|| arg_machine
))
1714 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1716 if (arg_ephemeral
&& arg_template
)
1717 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1719 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1720 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1722 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1723 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1725 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1726 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1727 "--read-only and --private-users-ownership=chown may not be combined.");
1729 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1730 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1731 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1732 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1733 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1735 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1736 * we need to error out, to avoid conflicts between different network options. */
1737 if (arg_network_namespace_path
&&
1738 (arg_network_interfaces
|| arg_network_macvlan
||
1739 arg_network_ipvlan
|| arg_network_veth_extra
||
1740 arg_network_bridge
|| arg_network_zone
||
1742 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1744 if (arg_network_bridge
&& arg_network_zone
)
1745 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1746 "--network-bridge= and --network-zone= may not be combined.");
1748 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1749 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1751 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1752 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1754 if (arg_expose_ports
&& !arg_private_network
)
1755 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1757 if (arg_caps_ambient
) {
1758 if (arg_caps_ambient
== UINT64_MAX
)
1759 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1761 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1762 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1764 if (arg_start_mode
== START_BOOT
)
1765 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1768 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1769 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1771 /* Drop duplicate --bind-user= entries */
1772 strv_uniq(arg_bind_user
);
1774 r
= custom_mount_check_all();
1781 static int verify_network_interfaces_initialized(void) {
1783 r
= test_network_interfaces_initialized(arg_network_interfaces
);
1787 r
= test_network_interfaces_initialized(arg_network_macvlan
);
1791 r
= test_network_interfaces_initialized(arg_network_ipvlan
);
1798 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1801 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1804 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1807 if (uid
!= UID_INVALID
) {
1808 uid
+= arg_uid_shift
;
1810 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1814 if (gid
!= GID_INVALID
) {
1815 gid
+= (gid_t
) arg_uid_shift
;
1817 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1821 return RET_NERRNO(lchown(p
, uid
, gid
));
1824 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1828 q
= prefix_roota(root
, path
);
1829 r
= RET_NERRNO(mkdir(q
, mode
));
1835 return userns_lchown(q
, uid
, gid
);
1838 static const char *timezone_from_path(const char *path
) {
1839 return PATH_STARTSWITH_SET(
1841 "../usr/share/zoneinfo/",
1842 "/usr/share/zoneinfo/");
1845 static bool etc_writable(void) {
1846 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1849 static int setup_timezone(const char *dest
) {
1850 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1851 const char *where
, *check
;
1857 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1858 r
= readlink_malloc("/etc/localtime", &p
);
1859 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1860 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1861 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1862 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1864 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1865 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1869 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1872 } else if (arg_timezone
== TIMEZONE_AUTO
)
1873 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1879 if (m
== TIMEZONE_OFF
)
1882 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1884 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1888 where
= strjoina(etc
, "/localtime");
1892 case TIMEZONE_DELETE
:
1893 if (unlink(where
) < 0)
1894 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1898 case TIMEZONE_SYMLINK
: {
1899 _cleanup_free_
char *q
= NULL
;
1900 const char *z
, *what
;
1902 z
= timezone_from_path(p
);
1904 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1908 r
= readlink_malloc(where
, &q
);
1909 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1910 return 0; /* Already pointing to the right place? Then do nothing .. */
1912 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1913 r
= chase(check
, dest
, 0, NULL
, NULL
);
1915 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1917 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1918 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1919 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1923 what
= strjoina("../usr/share/zoneinfo/", z
);
1924 if (symlink(what
, where
) < 0) {
1925 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1926 errno
, "Failed to correct timezone of container, ignoring: %m");
1936 case TIMEZONE_BIND
: {
1937 _cleanup_free_
char *resolved
= NULL
;
1940 found
= chase(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1942 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1946 if (found
== 0) /* missing? */
1947 (void) touch(resolved
);
1949 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1951 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1957 /* If mounting failed, try to copy */
1958 r
= copy_file_atomic("/etc/localtime", where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
1960 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1961 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1968 assert_not_reached();
1971 /* Fix permissions of the symlink or file copy we just created */
1972 r
= userns_lchown(where
, 0, 0);
1974 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1979 static int have_resolv_conf(const char *path
) {
1982 if (access(path
, F_OK
) < 0) {
1983 if (errno
== ENOENT
)
1986 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1992 static int resolved_listening(void) {
1993 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1994 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1995 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1998 /* Check if resolved is listening */
2000 r
= sd_bus_open_system(&bus
);
2002 return log_debug_errno(r
, "Failed to open system bus: %m");
2004 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2006 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2010 r
= bus_get_property_string(bus
, bus_resolve_mgr
, "DNSStubListener", &error
, &dns_stub_listener_mode
);
2012 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2014 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2017 static int setup_resolv_conf(const char *dest
) {
2018 _cleanup_free_
char *etc
= NULL
;
2019 const char *where
, *what
;
2025 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2026 if (arg_private_network
)
2027 m
= RESOLV_CONF_OFF
;
2028 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2029 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2030 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2031 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2033 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2036 m
= arg_resolv_conf
;
2038 if (m
== RESOLV_CONF_OFF
)
2041 r
= chase("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2043 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2047 where
= strjoina(etc
, "/resolv.conf");
2049 if (m
== RESOLV_CONF_DELETE
) {
2050 if (unlink(where
) < 0)
2051 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2056 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2057 what
= PRIVATE_STATIC_RESOLV_CONF
;
2058 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2059 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2060 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2061 what
= PRIVATE_STUB_RESOLV_CONF
;
2063 what
= "/etc/resolv.conf";
2065 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2066 _cleanup_free_
char *resolved
= NULL
;
2069 found
= chase(where
, dest
, CHASE_NONEXISTENT
|CHASE_NOFOLLOW
, &resolved
, NULL
);
2071 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2075 if (found
== 0) /* missing? */
2076 (void) touch(resolved
);
2078 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2080 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2082 /* If that didn't work, let's copy the file */
2085 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2086 r
= copy_file_atomic(what
, where
, 0644, COPY_REFLINK
|COPY_REPLACE
);
2088 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, COPY_REFLINK
);
2090 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2091 * resolved or something similar runs inside and the symlink points there.
2093 * If the disk image is read-only, there's also no point in complaining.
2095 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2096 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2097 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2101 r
= userns_lchown(where
, 0, 0);
2103 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2108 static int setup_boot_id(void) {
2109 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2110 _cleanup_free_
char *path
= NULL
;
2111 sd_id128_t rnd
= SD_ID128_NULL
;
2115 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2117 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2119 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2121 r
= sd_id128_randomize(&rnd
);
2123 return log_error_errno(r
, "Failed to generate random boot id: %m");
2125 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2127 return log_error_errno(r
, "Failed to write boot id: %m");
2129 from
= TAKE_PTR(path
);
2130 to
= "/proc/sys/kernel/random/boot_id";
2132 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2136 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2139 static int copy_devnodes(const char *dest
) {
2140 static const char devnodes
[] =
2153 BLOCK_WITH_UMASK(0000);
2155 /* Create /dev/net, so that we can create /dev/net/tun in it */
2156 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2157 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2159 NULSTR_FOREACH(d
, devnodes
) {
2160 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2163 from
= path_join("/dev/", d
);
2167 to
= path_join(dest
, from
);
2171 if (stat(from
, &st
) < 0) {
2173 if (errno
!= ENOENT
)
2174 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2176 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2177 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2178 "%s is not a char or block device, cannot copy.", from
);
2180 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2182 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2183 /* Explicitly warn the user when /dev is already populated. */
2184 if (errno
== EEXIST
)
2185 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
);
2187 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2189 /* Some systems abusively restrict mknod but allow bind mounts. */
2192 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2193 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2195 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2198 r
= userns_lchown(to
, 0, 0);
2200 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2202 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2206 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2208 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2210 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2213 prefixed
= path_join(dest
, sl
);
2217 t
= path_join("..", d
);
2221 if (symlink(t
, prefixed
) < 0)
2222 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2229 static int make_extra_nodes(const char *dest
) {
2233 BLOCK_WITH_UMASK(0000);
2235 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2236 _cleanup_free_
char *path
= NULL
;
2237 DeviceNode
*n
= arg_extra_nodes
+ i
;
2239 path
= path_join(dest
, n
->path
);
2243 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2244 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2246 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2248 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2254 static int setup_pts(const char *dest
) {
2255 _cleanup_free_
char *options
= NULL
;
2260 if (arg_selinux_apifs_context
)
2261 (void) asprintf(&options
,
2262 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2263 arg_uid_shift
+ TTY_GID
,
2264 arg_selinux_apifs_context
);
2267 (void) asprintf(&options
,
2268 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2269 arg_uid_shift
+ TTY_GID
);
2274 /* Mount /dev/pts itself */
2275 p
= prefix_roota(dest
, "/dev/pts");
2276 r
= RET_NERRNO(mkdir(p
, 0755));
2278 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2280 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2283 r
= userns_lchown(p
, 0, 0);
2285 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2287 /* Create /dev/ptmx symlink */
2288 p
= prefix_roota(dest
, "/dev/ptmx");
2289 if (symlink("pts/ptmx", p
) < 0)
2290 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2291 r
= userns_lchown(p
, 0, 0);
2293 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2295 /* And fix /dev/pts/ptmx ownership */
2296 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2297 r
= userns_lchown(p
, 0, 0);
2299 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2304 static int setup_stdio_as_dev_console(void) {
2305 _cleanup_close_
int terminal
= -EBADF
;
2308 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2309 * explicitly, if we are configured to. */
2310 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2312 return log_error_errno(terminal
, "Failed to open console: %m");
2314 /* Make sure we can continue logging to the original stderr, even if
2315 * stderr points elsewhere now */
2316 r
= log_dup_console();
2318 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2320 /* invalidates 'terminal' on success and failure */
2321 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2324 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2329 static int setup_dev_console(const char *console
) {
2330 _cleanup_free_
char *p
= NULL
;
2333 /* Create /dev/console symlink */
2334 r
= path_make_relative("/dev", console
, &p
);
2336 return log_error_errno(r
, "Failed to create relative path: %m");
2338 if (symlink(p
, "/dev/console") < 0)
2339 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2344 static int setup_keyring(void) {
2345 key_serial_t keyring
;
2347 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2348 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2349 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2350 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2351 * into the container. */
2353 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2354 if (keyring
== -1) {
2355 if (errno
== ENOSYS
)
2356 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2357 else if (ERRNO_IS_PRIVILEGE(errno
))
2358 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2360 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2366 static int setup_credentials(const char *root
) {
2370 if (arg_n_credentials
<= 0)
2373 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2375 return log_error_errno(r
, "Failed to create /run/host: %m");
2377 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2379 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2381 q
= prefix_roota(root
, "/run/host/credentials");
2382 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2386 for (size_t i
= 0; i
< arg_n_credentials
; i
++) {
2387 _cleanup_free_
char *j
= NULL
;
2388 _cleanup_close_
int fd
= -EBADF
;
2390 j
= path_join(q
, arg_credentials
[i
].id
);
2394 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2396 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2398 r
= loop_write(fd
, arg_credentials
[i
].data
, arg_credentials
[i
].size
);
2400 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2402 if (fchmod(fd
, 0400) < 0)
2403 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2405 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2406 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2407 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2411 if (chmod(q
, 0500) < 0)
2412 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2414 r
= userns_lchown(q
, 0, 0);
2418 /* Make both mount and superblock read-only now */
2419 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2423 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2426 static int setup_kmsg(int fd_inner_socket
) {
2427 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2428 _cleanup_free_
char *fifo
= NULL
;
2429 _cleanup_close_
int fd
= -EBADF
;
2432 assert(fd_inner_socket
>= 0);
2434 BLOCK_WITH_UMASK(0000);
2436 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2437 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2438 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2439 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2441 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2443 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2445 if (mkfifo(fifo
, 0600) < 0)
2446 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2448 from
= TAKE_PTR(fifo
);
2450 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2454 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2456 return log_error_errno(errno
, "Failed to open fifo: %m");
2458 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2459 r
= send_one_fd(fd_inner_socket
, fd
, 0);
2461 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2467 union in_addr_union address4
;
2468 union in_addr_union address6
;
2469 struct FirewallContext
*fw_ctx
;
2472 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2473 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2478 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2479 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2483 static int setup_hostname(void) {
2486 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2489 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2491 return log_error_errno(r
, "Failed to set hostname: %m");
2496 static int setup_journal(const char *directory
) {
2497 _cleanup_free_
char *d
= NULL
;
2503 /* Don't link journals in ephemeral mode */
2507 if (arg_link_journal
== LINK_NO
)
2510 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2512 r
= sd_id128_get_machine(&this_id
);
2514 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2516 if (sd_id128_equal(arg_uuid
, this_id
)) {
2517 log_full(try ? LOG_WARNING
: LOG_ERR
,
2518 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2524 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2525 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2527 bool ignore
= r
== -EROFS
&& try;
2528 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2529 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2530 return ignore
? 0 : r
;
2534 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2535 q
= prefix_roota(directory
, p
);
2537 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2541 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2542 "%s: already a mount point, refusing to use for journal", p
);
2545 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2549 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2550 "%s: already a mount point, refusing to use for journal", q
);
2553 r
= readlink_and_make_absolute(p
, &d
);
2555 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2558 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2560 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2565 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2566 } else if (r
== -EINVAL
) {
2568 if (arg_link_journal
== LINK_GUEST
&&
2571 if (errno
== ENOTDIR
) {
2572 log_error("%s already exists and is neither a symlink nor a directory", p
);
2575 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2577 } else if (r
!= -ENOENT
)
2578 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2580 if (arg_link_journal
== LINK_GUEST
) {
2582 if (symlink(q
, p
) < 0) {
2584 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2587 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2590 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2592 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2596 if (arg_link_journal
== LINK_HOST
) {
2597 /* don't create parents here — if the host doesn't have
2598 * permanent journal set up, don't force it here */
2600 r
= RET_NERRNO(mkdir(p
, 0755));
2601 if (r
< 0 && r
!= -EEXIST
) {
2603 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2606 return log_error_errno(r
, "Failed to create %s: %m", p
);
2609 } else if (access(p
, F_OK
) < 0)
2612 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2613 log_warning("%s is not empty, proceeding anyway.", q
);
2615 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2617 return log_error_errno(r
, "Failed to create %s: %m", q
);
2619 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2621 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2626 static int drop_capabilities(uid_t uid
) {
2627 CapabilityQuintet q
;
2629 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2630 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2631 * arg_caps_retain. */
2633 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2634 q
= arg_full_capabilities
;
2636 if (q
.bounding
== UINT64_MAX
)
2637 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2639 if (q
.effective
== UINT64_MAX
)
2640 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2642 if (q
.inheritable
== UINT64_MAX
)
2643 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2645 if (q
.permitted
== UINT64_MAX
)
2646 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2648 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2649 q
.ambient
= arg_caps_ambient
;
2651 if (capability_quintet_mangle(&q
))
2652 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2655 q
= (CapabilityQuintet
) {
2656 .bounding
= arg_caps_retain
,
2657 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2658 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2659 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2660 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2663 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2664 * in order to maintain the same behavior as systemd < 242. */
2665 if (capability_quintet_mangle(&q
))
2666 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2667 "Some capabilities will not be set because they are not in the current bounding set.");
2671 return capability_quintet_enforce(&q
);
2674 static int reset_audit_loginuid(void) {
2675 _cleanup_free_
char *p
= NULL
;
2678 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2681 r
= read_one_line_file("/proc/self/loginuid", &p
);
2685 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2687 /* Already reset? */
2688 if (streq(p
, "4294967295"))
2691 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2694 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2695 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2696 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2697 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2698 "using systemd-nspawn. Sleeping for 5s... (%m)");
2706 static int mount_tunnel_dig(const char *root
) {
2710 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2711 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2712 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2713 (void) mkdir_p(p
, 0600);
2715 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2717 return log_error_errno(r
, "Failed to create /run/host: %m");
2719 r
= userns_mkdir(root
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2721 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2723 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
2724 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2728 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2735 static int mount_tunnel_open(void) {
2738 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2745 static int setup_machine_id(const char *directory
) {
2748 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2749 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2750 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2751 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2752 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2753 * container behaves nicely). */
2755 r
= id128_get_machine(directory
, &arg_uuid
);
2756 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r
)) {
2757 /* If the file is missing, empty, or uninitialized, we don't mind */
2758 if (sd_id128_is_null(arg_uuid
)) {
2759 r
= sd_id128_randomize(&arg_uuid
);
2761 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2764 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2769 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2774 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2777 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2778 if (r
== -EOPNOTSUPP
)
2779 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.");
2781 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2783 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2785 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2787 log_debug("Patched directory tree to match UID/GID range.");
2794 * < 0 : wait_for_terminate() failed to get the state of the
2795 * container, the container was terminated by a signal, or
2796 * failed for an unknown reason. No change is made to the
2797 * container argument.
2798 * > 0 : The program executed in the container terminated with an
2799 * error. The exit code of the program executed in the
2800 * container is returned. The container argument has been set
2801 * to CONTAINER_TERMINATED.
2802 * 0 : The container is being rebooted, has been shut down or exited
2803 * successfully. The container argument has been set to either
2804 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2806 * That is, success is indicated by a return value of zero, and an
2807 * error is indicated by a non-zero value.
2809 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2813 r
= wait_for_terminate(pid
, &status
);
2815 return log_warning_errno(r
, "Failed to wait for container: %m");
2817 switch (status
.si_code
) {
2820 if (status
.si_status
== 0)
2821 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2823 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2825 *container
= CONTAINER_TERMINATED
;
2826 return status
.si_status
;
2829 if (status
.si_status
== SIGINT
) {
2830 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2831 *container
= CONTAINER_TERMINATED
;
2834 } else if (status
.si_status
== SIGHUP
) {
2835 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2836 *container
= CONTAINER_REBOOTED
;
2842 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2843 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2846 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2847 "Container %s failed due to unknown reason.", arg_machine
);
2851 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2854 pid
= PTR_TO_PID(userdata
);
2856 if (kill(pid
, arg_kill_signal
) >= 0) {
2857 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2858 sd_event_source_set_userdata(s
, NULL
);
2863 sd_event_exit(sd_event_source_get_event(s
), 0);
2867 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2873 pid
= PTR_TO_PID(userdata
);
2878 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2879 return log_error_errno(errno
, "Failed to waitid(): %m");
2880 if (si
.si_pid
== 0) /* No pending children. */
2882 if (si
.si_pid
== pid
) {
2883 /* The main process we care for has exited. Return from
2884 * signal handler but leave the zombie. */
2885 sd_event_exit(sd_event_source_get_event(s
), 0);
2889 /* Reap all other children. */
2890 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2896 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2901 pid
= PTR_TO_PID(userdata
);
2903 if (arg_kill_signal
> 0) {
2904 log_info("Container termination requested. Attempting to halt container.");
2905 (void) kill(pid
, arg_kill_signal
);
2907 log_info("Container termination requested. Exiting.");
2908 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2914 static int determine_names(void) {
2917 if (arg_template
&& !arg_directory
&& arg_machine
) {
2919 /* If --template= was specified then we should not
2920 * search for a machine, but instead create a new one
2921 * in /var/lib/machine. */
2923 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2928 if (!arg_image
&& !arg_directory
) {
2930 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2932 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
2934 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2936 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2938 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2939 r
= free_and_strdup(&arg_image
, i
->path
);
2941 r
= free_and_strdup(&arg_directory
, i
->path
);
2946 arg_read_only
= arg_read_only
|| i
->read_only
;
2948 r
= safe_getcwd(&arg_directory
);
2950 return log_error_errno(r
, "Failed to determine current directory: %m");
2953 if (!arg_directory
&& !arg_image
)
2954 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2958 if (arg_directory
&& path_equal(arg_directory
, "/"))
2959 arg_machine
= gethostname_malloc();
2960 else if (arg_image
) {
2963 r
= path_extract_filename(arg_image
, &arg_machine
);
2965 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_image
);
2967 /* Truncate suffix if there is one */
2968 e
= endswith(arg_machine
, ".raw");
2972 r
= path_extract_filename(arg_directory
, &arg_machine
);
2974 return log_error_errno(r
, "Failed to extract file name from '%s': %m", arg_directory
);
2977 hostname_cleanup(arg_machine
);
2978 if (!hostname_is_valid(arg_machine
, 0))
2979 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2981 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
2982 * to match fixed config file names. */
2983 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
2984 if (!arg_settings_filename
)
2987 /* Add a random suffix when this is an ephemeral machine, so that we can run many
2988 * instances at once without manually having to specify -M each time. */
2990 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
2993 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
2994 if (!arg_settings_filename
)
3001 static int chase_and_update(char **p
, unsigned flags
) {
3010 r
= chase(*p
, NULL
, flags
, &chased
, NULL
);
3012 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3014 return free_and_replace(*p
, chased
);
3017 static int determine_uid_shift(const char *directory
) {
3019 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3024 if (arg_uid_shift
== UID_INVALID
) {
3027 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3029 if (stat(directory
, &st
) < 0)
3030 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3032 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3034 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3035 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3036 "UID and GID base of %s don't match.", directory
);
3038 arg_uid_range
= UINT32_C(0x10000);
3040 if (arg_uid_shift
!= 0) {
3041 /* If the image is shifted already, then we'll fall back to classic chowning, for
3042 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3044 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3045 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3046 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3047 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3048 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3049 "UID base of %s is not zero, UID mapping not supported.", directory
);
3053 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3054 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3059 static unsigned long effective_clone_ns_flags(void) {
3060 unsigned long flags
= arg_clone_ns_flags
;
3062 if (arg_private_network
)
3063 flags
|= CLONE_NEWNET
;
3065 flags
|= CLONE_NEWCGROUP
;
3066 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3067 flags
|= CLONE_NEWUSER
;
3072 static int patch_sysctl(void) {
3074 /* This table is inspired by runc's sysctl() function */
3075 static const struct {
3078 unsigned long clone_flags
;
3080 { "kernel.hostname", false, CLONE_NEWUTS
},
3081 { "kernel.domainname", false, CLONE_NEWUTS
},
3082 { "kernel.msgmax", false, CLONE_NEWIPC
},
3083 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3084 { "kernel.msgmni", false, CLONE_NEWIPC
},
3085 { "kernel.sem", false, CLONE_NEWIPC
},
3086 { "kernel.shmall", false, CLONE_NEWIPC
},
3087 { "kernel.shmmax", false, CLONE_NEWIPC
},
3088 { "kernel.shmmni", false, CLONE_NEWIPC
},
3089 { "fs.mqueue.", true, CLONE_NEWIPC
},
3090 { "net.", true, CLONE_NEWNET
},
3093 unsigned long flags
;
3096 flags
= effective_clone_ns_flags();
3098 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3102 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3104 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3107 if (safe_sysctl
[i
].prefix
)
3108 good
= startswith(*k
, safe_sysctl
[i
].key
);
3110 good
= streq(*k
, safe_sysctl
[i
].key
);
3117 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3119 r
= sysctl_write(*k
, *v
);
3121 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3127 static int inner_child(
3129 int fd_inner_socket
,
3131 char **os_release_pairs
) {
3133 _cleanup_free_
char *home
= NULL
;
3136 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3137 NULL
, /* container */
3142 NULL
, /* container_uuid */
3143 NULL
, /* LISTEN_FDS */
3144 NULL
, /* LISTEN_PID */
3145 NULL
, /* NOTIFY_SOCKET */
3146 NULL
, /* CREDENTIALS_DIRECTORY */
3150 const char *exec_target
;
3151 _cleanup_strv_free_
char **env_use
= NULL
;
3152 int r
, which_failed
;
3154 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3155 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3156 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3157 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3158 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3159 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3162 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3163 * unshare(). See below. */
3166 assert(fd_inner_socket
>= 0);
3168 log_debug("Inner child is initializing.");
3170 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3171 /* Tell the parent, that it now can write the UID map. */
3172 (void) barrier_place(barrier
); /* #1 */
3174 /* Wait until the parent wrote the UID map */
3175 if (!barrier_place_and_sync(barrier
)) /* #2 */
3176 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3178 /* Become the new root user inside our namespace */
3179 r
= reset_uid_gid();
3181 return log_error_errno(r
, "Couldn't become new root: %m");
3183 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3184 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3185 * propagation, but simply create new peer groups for all our mounts). */
3186 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3192 arg_mount_settings
| MOUNT_IN_USERNS
,
3194 arg_selinux_apifs_context
);
3198 if (!arg_network_namespace_path
&& arg_private_network
) {
3199 r
= unshare(CLONE_NEWNET
);
3201 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3203 /* Tell the parent that it can setup network interfaces. */
3204 (void) barrier_place(barrier
); /* #3 */
3207 r
= mount_sysfs(NULL
, arg_mount_settings
);
3211 /* Wait until we are cgroup-ified, so that we
3212 * can mount the right cgroup path writable */
3213 if (!barrier_place_and_sync(barrier
)) /* #4 */
3214 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3215 "Parent died too early");
3218 r
= unshare(CLONE_NEWCGROUP
);
3220 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3223 arg_unified_cgroup_hierarchy
,
3224 arg_userns_mode
!= USER_NAMESPACE_NO
,
3227 arg_selinux_apifs_context
,
3230 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3234 r
= setup_boot_id();
3238 r
= setup_kmsg(fd_inner_socket
);
3245 arg_n_custom_mounts
,
3248 arg_selinux_apifs_context
,
3249 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3254 return log_error_errno(errno
, "setsid() failed: %m");
3256 if (arg_private_network
)
3257 (void) loopback_setup();
3259 if (arg_expose_ports
) {
3260 r
= expose_port_send_rtnl(fd_inner_socket
);
3265 if (arg_console_mode
!= CONSOLE_PIPE
) {
3266 _cleanup_close_
int master
= -EBADF
;
3267 _cleanup_free_
char *console
= NULL
;
3269 /* Allocate a pty and make it available as /dev/console. */
3270 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3272 return log_error_errno(master
, "Failed to allocate a pty: %m");
3274 r
= setup_dev_console(console
);
3276 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3278 r
= send_one_fd(fd_inner_socket
, master
, 0);
3280 return log_error_errno(r
, "Failed to send master fd: %m");
3282 r
= setup_stdio_as_dev_console();
3291 if (arg_oom_score_adjust_set
) {
3292 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3294 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3297 if (arg_cpu_set
.set
)
3298 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3299 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3301 (void) setup_hostname();
3303 if (arg_personality
!= PERSONALITY_INVALID
) {
3304 r
= safe_personality(arg_personality
);
3306 return log_error_errno(r
, "personality() failed: %m");
3307 #ifdef ARCHITECTURE_SECONDARY
3308 } else if (arg_architecture
== ARCHITECTURE_SECONDARY
) {
3309 r
= safe_personality(PER_LINUX32
);
3311 return log_error_errno(r
, "personality() failed: %m");
3313 } else if (!arg_quiet
&& arg_architecture
>= 0 && arg_architecture
!= native_architecture())
3314 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3315 "invocation with qemu userspace emulator (or equivalent) in effect.",
3316 architecture_to_string(arg_architecture
));
3318 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3320 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3325 if (is_seccomp_available()) {
3326 r
= seccomp_load(arg_seccomp
);
3327 if (ERRNO_IS_NEG_SECCOMP_FATAL(r
))
3328 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3330 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3335 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3340 if (arg_suppress_sync
) {
3342 r
= seccomp_suppress_sync();
3344 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3346 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3351 if (arg_selinux_context
)
3352 if (setexeccon(arg_selinux_context
) < 0)
3353 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3356 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3357 * if we need to later on. */
3358 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3359 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3361 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3362 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3364 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3368 r
= drop_capabilities(getuid());
3370 return log_error_errno(r
, "Dropping capabilities failed: %m");
3372 if (arg_no_new_privileges
)
3373 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3374 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3376 /* LXC sets container=lxc, so follow the scheme here */
3377 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3379 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3383 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3384 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3387 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3388 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3389 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
?: "root") < 0)
3392 assert(!sd_id128_is_null(arg_uuid
));
3394 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3397 if (fdset_size(fds
) > 0) {
3398 r
= fdset_cloexec(fds
, false);
3400 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3402 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3403 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3406 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3409 if (arg_n_credentials
> 0) {
3410 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3416 if (arg_start_mode
!= START_BOOT
) {
3417 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3423 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3427 /* Let the parent know that we are ready and wait until the parent is ready with the setup, too... */
3428 if (!barrier_place_and_sync(barrier
)) /* #5 */
3429 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3432 if (chdir(arg_chdir
) < 0)
3433 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3435 if (arg_start_mode
== START_PID2
) {
3436 r
= stub_pid1(arg_uuid
);
3441 if (arg_console_mode
!= CONSOLE_PIPE
) {
3442 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3443 * are configured for that. Acquire it as controlling tty. */
3444 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3445 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3448 log_debug("Inner child completed, invoking payload.");
3450 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3451 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3452 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3454 log_set_open_when_needed(true);
3455 log_settle_target();
3457 (void) fdset_close_others(fds
);
3459 if (arg_start_mode
== START_BOOT
) {
3463 /* Automatically search for the init system */
3465 m
= strv_length(arg_parameters
);
3466 a
= newa(char*, m
+ 2);
3467 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3470 FOREACH_STRING(init
,
3471 "/usr/lib/systemd/systemd",
3472 "/lib/systemd/systemd",
3474 a
[0] = (char*) init
;
3475 execve(a
[0], a
, env_use
);
3478 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3479 } else if (!strv_isempty(arg_parameters
)) {
3480 const char *dollar_path
;
3482 exec_target
= arg_parameters
[0];
3484 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3486 dollar_path
= strv_env_get(env_use
, "PATH");
3488 if (setenv("PATH", dollar_path
, 1) < 0)
3489 return log_error_errno(errno
, "Failed to update $PATH: %m");
3492 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3495 /* If we cannot change the directory, we'll end up in /, that is expected. */
3496 (void) chdir(home
?: "/root");
3498 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3499 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3500 execle("/bin/bash", "-bash", NULL
, env_use
);
3501 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3502 execle("/bin/sh", "-sh", NULL
, env_use
);
3504 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3507 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3510 static int setup_notify_child(void) {
3511 _cleanup_close_
int fd
= -EBADF
;
3512 static const union sockaddr_union sa
= {
3513 .un
.sun_family
= AF_UNIX
,
3514 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3518 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3520 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3522 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3523 (void) sockaddr_un_unlink(&sa
.un
);
3525 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3527 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3529 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3531 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3533 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3535 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3540 static int outer_child(
3542 const char *directory
,
3543 DissectedImage
*dissected_image
,
3544 int fd_outer_socket
,
3545 int fd_inner_socket
,
3549 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3550 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3551 _cleanup_close_
int fd
= -EBADF
, mntns_fd
= -EBADF
;
3558 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3559 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3560 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3561 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3562 * forked off it, and it exits. */
3566 assert(fd_outer_socket
>= 0);
3567 assert(fd_inner_socket
>= 0);
3569 log_debug("Outer child is initializing.");
3571 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3573 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3575 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3576 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3578 r
= reset_audit_loginuid();
3582 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3583 * mounts to the real root. */
3584 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3588 if (dissected_image
) {
3589 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3590 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3591 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3592 * right place right away. This makes sure ESP partitions and userns are compatible. */
3594 r
= dissected_image_mount_and_warn(
3599 /* userns_fd= */ -EBADF
,
3600 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3601 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3602 DISSECT_IMAGE_USR_NO_ROOT
|
3603 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3604 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3609 r
= determine_uid_shift(directory
);
3613 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3614 r
= namespace_open(0, NULL
, &mntns_fd
, NULL
, NULL
, NULL
);
3616 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3618 l
= send_one_fd(fd_outer_socket
, mntns_fd
, 0);
3620 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3621 mntns_fd
= safe_close(mntns_fd
);
3623 /* Let the parent know which UID shift we read from the image */
3624 l
= send(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3626 return log_error_errno(errno
, "Failed to send UID shift: %m");
3627 if (l
!= sizeof(arg_uid_shift
))
3628 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3629 "Short write while sending UID shift.");
3631 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3632 /* When we are supposed to pick the UID shift, the parent will check now whether the
3633 * UID shift we just read from the image is available. If yes, it will send the UID
3634 * shift back to us, if not it will pick a different one, and send it back to us. */
3636 l
= recv(fd_outer_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3638 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3639 if (l
!= sizeof(arg_uid_shift
))
3640 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3641 "Short read while receiving UID shift.");
3644 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3645 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3648 if (path_equal(directory
, "/")) {
3649 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3650 * place, so that we can make changes to its mount structure (for example, to implement
3651 * --volatile=) without this interfering with our ability to access files such as
3652 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3653 * (instead of a temporary directory, since we are living in our own mount namespace here
3654 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3655 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3657 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3661 directory
= "/run/systemd/nspawn-root";
3664 /* Make sure we always have a mount that we can move to root later on. */
3665 r
= make_mount_point(directory
);
3669 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3670 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3671 * we'll live in our own little world from now on, and propagation from the host may only happen via
3672 * the mount tunnel dir, or not at all. */
3673 r
= mount_follow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_PRIVATE
|MS_REC
, NULL
);
3677 r
= setup_pivot_root(
3680 arg_pivot_root_old
);
3684 r
= setup_volatile_mode(
3688 arg_selinux_apifs_context
);
3692 r
= bind_user_prepare(
3697 &arg_custom_mounts
, &arg_n_custom_mounts
,
3698 &bind_user_context
);
3702 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3703 /* Send the user maps we determined to the parent, so that it installs it in our user
3704 * namespace UID map table */
3706 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3708 bind_user_context
->data
[i
].payload_user
->uid
,
3709 bind_user_context
->data
[i
].host_user
->uid
,
3710 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3711 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3714 l
= send(fd_outer_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3716 return log_error_errno(errno
, "Failed to send user UID map: %m");
3717 if (l
!= sizeof(map
))
3718 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3719 "Short write while sending user UID map.");
3726 arg_n_custom_mounts
,
3729 arg_selinux_apifs_context
,
3734 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3735 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3736 arg_uid_shift
!= 0) {
3737 _cleanup_free_
char *usr_subtree
= NULL
;
3741 dirs
[i
++] = (char*) directory
;
3743 if (dissected_image
&& dissected_image
->partitions
[PARTITION_USR
].found
) {
3744 usr_subtree
= path_join(directory
, "/usr");
3748 dirs
[i
++] = usr_subtree
;
3753 r
= remount_idmap(dirs
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3754 if (r
== -EINVAL
|| ERRNO_IS_NEG_NOT_SUPPORTED(r
)) {
3755 /* This might fail because the kernel or file system doesn't support idmapping. We
3756 * can't really distinguish this nicely, nor do we have any guarantees about the
3757 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3758 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3759 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3760 "ID mapped mounts are apparently not available, sorry.");
3762 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3763 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3765 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3767 log_debug("ID mapped mounts available, making use of them.");
3772 if (dissected_image
) {
3773 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3774 r
= dissected_image_mount(
3779 /* userns_fd= */ -EBADF
,
3780 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3781 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3782 DISSECT_IMAGE_USR_NO_ROOT
|
3783 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3784 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3786 return log_error_errno(r
, "File system check for image failed: %m");
3788 return log_error_errno(r
, "Failed to mount image file system: %m");
3791 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3792 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3794 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3798 l
= send(fd_outer_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3800 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3801 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3802 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3803 "Short write while sending cgroup mode.");
3806 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3810 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3814 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3815 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3816 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3818 return log_error_errno(r
, "Failed to make tree read-only: %m");
3821 r
= mount_all(directory
,
3824 arg_selinux_apifs_context
);
3828 r
= copy_devnodes(directory
);
3832 r
= make_extra_nodes(directory
);
3836 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3838 p
= prefix_roota(directory
, "/run/host");
3839 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3841 r
= setup_pts(directory
);
3845 r
= mount_tunnel_dig(directory
);
3849 r
= setup_keyring();
3853 r
= setup_credentials(directory
);
3857 r
= bind_user_setup(bind_user_context
, directory
);
3864 arg_n_custom_mounts
,
3867 arg_selinux_apifs_context
,
3868 MOUNT_NON_ROOT_ONLY
);
3872 r
= setup_timezone(directory
);
3876 r
= setup_resolv_conf(directory
);
3880 r
= setup_machine_id(directory
);
3884 r
= setup_journal(directory
);
3888 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3889 p
= prefix_roota(directory
, "/run/host/container-manager");
3890 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
);
3892 /* The same stuff as the $container_uuid env var */
3893 p
= prefix_roota(directory
, "/run/host/container-uuid");
3894 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
3896 if (!arg_use_cgns
) {
3899 arg_unified_cgroup_hierarchy
,
3900 arg_userns_mode
!= USER_NAMESPACE_NO
,
3903 arg_selinux_apifs_context
,
3909 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3910 * mounts available in systemd services inside the container that create a new mount namespace. See
3911 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3912 * will inherit the shared propagation mode.
3914 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3915 * directory mount to root later on.
3916 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3918 r
= mount_switch_root(directory
, MS_SHARED
);
3920 return log_error_errno(r
, "Failed to move root directory: %m");
3922 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
3923 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
3925 r
= mount_tunnel_open();
3929 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3930 /* In order to mount procfs and sysfs in an unprivileged container the kernel
3931 * requires that a fully visible instance is already present in the target mount
3932 * namespace. Mount one here so the inner child can mount its own instances. Later
3933 * we umount the temporary instances created here before we actually exec the
3934 * payload. Since the rootfs is shared the umount will propagate into the container.
3935 * Note, the inner child wouldn't be able to unmount the instances on its own since
3936 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
3938 r
= pin_fully_visible_fs();
3943 fd
= setup_notify_child();
3947 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3948 arg_clone_ns_flags
|
3949 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3951 return log_error_errno(errno
, "Failed to fork inner child: %m");
3953 fd_outer_socket
= safe_close(fd_outer_socket
);
3955 /* The inner child has all namespaces that are requested, so that we all are owned by the
3956 * user if user namespaces are turned on. */
3958 if (arg_network_namespace_path
) {
3959 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3961 return log_error_errno(r
, "Failed to join network namespace: %m");
3964 r
= inner_child(barrier
, fd_inner_socket
, fds
, os_release_pairs
);
3966 _exit(EXIT_FAILURE
);
3968 _exit(EXIT_SUCCESS
);
3971 l
= send(fd_outer_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3973 return log_error_errno(errno
, "Failed to send PID: %m");
3974 if (l
!= sizeof(pid
))
3975 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3976 "Short write while sending PID.");
3978 l
= send(fd_outer_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3980 return log_error_errno(errno
, "Failed to send machine ID: %m");
3981 if (l
!= sizeof(arg_uuid
))
3982 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3983 "Short write while sending machine ID.");
3985 l
= send_one_fd(fd_outer_socket
, fd
, 0);
3987 return log_error_errno(l
, "Failed to send notify fd: %m");
3989 fd_outer_socket
= safe_close(fd_outer_socket
);
3990 fd_inner_socket
= safe_close(fd_inner_socket
);
3991 netns_fd
= safe_close(netns_fd
);
3996 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3997 bool tried_hashed
= false;
3998 unsigned n_tries
= 100;
4003 assert(ret_lock_file
);
4004 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4005 assert(arg_uid_range
== 0x10000U
);
4009 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4012 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4013 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4018 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4020 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4023 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4024 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4025 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4030 /* Make some superficial checks whether the range is currently known in the user database */
4031 if (getpwuid(candidate
))
4033 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4035 if (getgrgid(candidate
))
4037 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4040 *ret_lock_file
= lf
;
4041 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4046 if (arg_machine
&& !tried_hashed
) {
4047 /* Try to hash the base from the container name */
4049 static const uint8_t hash_key
[] = {
4050 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4051 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4054 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4056 tried_hashed
= true;
4058 random_bytes(&candidate
, sizeof(candidate
));
4060 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4061 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4065 static int add_one_uid_map(
4067 uid_t container_uid
,
4071 return strextendf(p
,
4072 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4073 container_uid
, host_uid
, range
);
4076 static int make_uid_map_string(
4077 const uid_t bind_user_uid
[],
4078 size_t n_bind_user_uid
,
4082 _cleanup_free_
char *s
= NULL
;
4083 uid_t previous_uid
= 0;
4086 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4087 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4090 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4091 * quadruplet, consisting of host and container UID + GID. */
4093 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4094 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4095 host_uid
= bind_user_uid
[i
*4+offset
+1];
4097 assert(previous_uid
<= payload_uid
);
4098 assert(payload_uid
< arg_uid_range
);
4100 /* Add a range to close the gap to previous entry */
4101 if (payload_uid
> previous_uid
) {
4102 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4107 /* Map this specific user */
4108 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4112 previous_uid
= payload_uid
+ 1;
4115 /* And add a range to close the gap to finish the range */
4116 if (arg_uid_range
> previous_uid
) {
4117 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4128 static int setup_uid_map(
4130 const uid_t bind_user_uid
[],
4131 size_t n_bind_user_uid
) {
4133 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4134 _cleanup_free_
char *s
= NULL
;
4139 /* Build the UID map string */
4140 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4143 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4144 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4146 return log_error_errno(r
, "Failed to write UID map: %m");
4148 /* And now build the GID map string */
4150 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4153 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4154 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4156 return log_error_errno(r
, "Failed to write GID map: %m");
4161 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4162 char buf
[NOTIFY_BUFFER_MAX
+1];
4164 struct iovec iovec
= {
4166 .iov_len
= sizeof(buf
)-1,
4168 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4169 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4170 struct msghdr msghdr
= {
4173 .msg_control
= &control
,
4174 .msg_controllen
= sizeof(control
),
4176 struct ucred
*ucred
;
4178 pid_t inner_child_pid
;
4179 _cleanup_strv_free_
char **tags
= NULL
;
4184 inner_child_pid
= PTR_TO_PID(userdata
);
4186 if (revents
!= EPOLLIN
) {
4187 log_warning("Got unexpected poll event for notify fd.");
4191 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4192 if (ERRNO_IS_NEG_TRANSIENT(n
))
4194 else if (n
== -EXFULL
) {
4195 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4198 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4200 cmsg_close_all(&msghdr
);
4202 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4203 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4204 log_debug("Received notify message without valid credentials. Ignoring.");
4208 if ((size_t) n
>= sizeof(buf
)) {
4209 log_warning("Received notify message exceeded maximum size. Ignoring.");
4214 tags
= strv_split(buf
, "\n\r");
4218 if (strv_contains(tags
, "READY=1")) {
4219 r
= sd_notify(false, "READY=1\n");
4221 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4224 p
= strv_find_startswith(tags
, "STATUS=");
4226 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4231 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4234 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4236 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4238 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4243 static int merge_settings(Settings
*settings
, const char *path
) {
4249 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4250 * that this steals the fields of the Settings* structure, and hence modifies it. */
4252 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4253 settings
->start_mode
>= 0) {
4254 arg_start_mode
= settings
->start_mode
;
4255 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4258 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4259 settings
->ephemeral
>= 0)
4260 arg_ephemeral
= settings
->ephemeral
;
4262 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4265 if (!arg_settings_trusted
)
4266 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4268 free_and_replace(arg_directory
, settings
->root
);
4271 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4272 settings
->pivot_root_new
) {
4273 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4274 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4277 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4278 settings
->working_directory
)
4279 free_and_replace(arg_chdir
, settings
->working_directory
);
4281 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4282 settings
->environment
)
4283 strv_free_and_replace(arg_setenv
, settings
->environment
);
4285 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4288 free_and_replace(arg_user
, settings
->user
);
4290 if (uid_is_valid(settings
->uid
))
4291 arg_uid
= settings
->uid
;
4292 if (gid_is_valid(settings
->gid
))
4293 arg_gid
= settings
->gid
;
4294 if (settings
->n_supplementary_gids
> 0) {
4295 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4296 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4300 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4301 uint64_t plus
, minus
;
4302 uint64_t network_minus
= 0;
4305 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4306 * Settings structure */
4308 plus
= settings
->capability
;
4309 minus
= settings
->drop_capability
;
4311 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4312 settings_network_configured(settings
)) {
4313 if (settings_private_network(settings
))
4314 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4316 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4319 if (!arg_settings_trusted
&& plus
!= 0) {
4320 if (settings
->capability
!= 0)
4321 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4323 arg_caps_retain
&= ~network_minus
;
4324 arg_caps_retain
|= plus
;
4327 arg_caps_retain
&= ~minus
;
4329 /* Copy the full capabilities over too */
4330 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4331 if (!arg_settings_trusted
)
4332 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4334 arg_full_capabilities
= settings
->full_capabilities
;
4337 ambient
= settings
->ambient_capability
;
4338 if (!arg_settings_trusted
&& ambient
!= 0)
4339 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4341 arg_caps_ambient
|= ambient
;
4344 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4345 settings
->kill_signal
> 0)
4346 arg_kill_signal
= settings
->kill_signal
;
4348 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4349 settings
->personality
!= PERSONALITY_INVALID
)
4350 arg_personality
= settings
->personality
;
4352 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4353 !sd_id128_is_null(settings
->machine_id
)) {
4355 if (!arg_settings_trusted
)
4356 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4358 arg_uuid
= settings
->machine_id
;
4361 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4362 settings
->read_only
>= 0)
4363 arg_read_only
= settings
->read_only
;
4365 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4366 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4367 arg_volatile_mode
= settings
->volatile_mode
;
4369 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4370 settings
->n_custom_mounts
> 0) {
4372 if (!arg_settings_trusted
)
4373 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4375 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4376 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4377 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4378 settings
->n_custom_mounts
= 0;
4382 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4383 settings_network_configured(settings
)) {
4385 if (!arg_settings_trusted
)
4386 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4388 arg_network_veth
= settings_network_veth(settings
);
4389 arg_private_network
= settings_private_network(settings
);
4391 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4392 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4393 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4394 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4396 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4397 free_and_replace(arg_network_zone
, settings
->network_zone
);
4399 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4403 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4404 settings
->expose_ports
) {
4406 if (!arg_settings_trusted
)
4407 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4409 expose_port_free_all(arg_expose_ports
);
4410 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4414 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4415 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4417 if (!arg_settings_trusted
)
4418 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4420 arg_userns_mode
= settings
->userns_mode
;
4421 arg_uid_shift
= settings
->uid_shift
;
4422 arg_uid_range
= settings
->uid_range
;
4423 arg_userns_ownership
= settings
->userns_ownership
;
4427 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4428 !strv_isempty(settings
->bind_user
))
4429 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4431 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4432 settings
->notify_ready
>= 0)
4433 arg_notify_ready
= settings
->notify_ready
;
4435 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4437 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4438 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4439 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4441 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4442 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4447 if (settings
->seccomp
) {
4448 if (!arg_settings_trusted
)
4449 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4451 seccomp_release(arg_seccomp
);
4452 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4458 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4459 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4462 if (!settings
->rlimit
[rl
])
4465 if (!arg_settings_trusted
) {
4466 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4470 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4473 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4475 free_and_replace(arg_hostname
, settings
->hostname
);
4477 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4478 settings
->no_new_privileges
>= 0)
4479 arg_no_new_privileges
= settings
->no_new_privileges
;
4481 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4482 settings
->oom_score_adjust_set
) {
4484 if (!arg_settings_trusted
)
4485 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4487 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4488 arg_oom_score_adjust_set
= true;
4492 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4493 settings
->cpu_set
.set
) {
4495 if (!arg_settings_trusted
)
4496 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4498 cpu_set_reset(&arg_cpu_set
);
4499 arg_cpu_set
= TAKE_STRUCT(settings
->cpu_set
);
4503 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4504 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4505 arg_resolv_conf
= settings
->resolv_conf
;
4507 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4508 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4510 if (!arg_settings_trusted
)
4511 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4513 arg_link_journal
= settings
->link_journal
;
4514 arg_link_journal_try
= settings
->link_journal_try
;
4518 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4519 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4520 arg_timezone
= settings
->timezone
;
4522 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4525 if (!arg_settings_trusted
)
4526 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4528 free_and_replace(arg_slice
, settings
->slice
);
4531 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4532 settings
->use_cgns
>= 0) {
4534 if (!arg_settings_trusted
)
4535 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4537 arg_use_cgns
= settings
->use_cgns
;
4540 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4541 settings
->clone_ns_flags
!= ULONG_MAX
) {
4543 if (!arg_settings_trusted
)
4544 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4546 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4549 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4550 settings
->console_mode
>= 0) {
4552 if (!arg_settings_trusted
)
4553 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4555 arg_console_mode
= settings
->console_mode
;
4558 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4559 settings
->suppress_sync
>= 0)
4560 arg_suppress_sync
= settings
->suppress_sync
;
4562 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4563 * don't consult arg_settings_mask for them. */
4565 sd_bus_message_unref(arg_property_message
);
4566 arg_property_message
= TAKE_PTR(settings
->properties
);
4568 arg_console_width
= settings
->console_width
;
4569 arg_console_height
= settings
->console_height
;
4571 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4572 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4573 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4574 settings
->n_extra_nodes
= 0;
4579 static int load_settings(void) {
4580 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4581 _cleanup_fclose_
FILE *f
= NULL
;
4582 _cleanup_free_
char *p
= NULL
;
4588 /* If all settings are masked, there's no point in looking for
4589 * the settings file */
4590 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4593 /* We first look in the admin's directories in /etc and /run */
4594 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4595 _cleanup_free_
char *j
= NULL
;
4597 j
= path_join(i
, arg_settings_filename
);
4605 /* By default, we trust configuration from /etc and /run */
4606 if (arg_settings_trusted
< 0)
4607 arg_settings_trusted
= true;
4612 if (errno
!= ENOENT
)
4613 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4617 /* After that, let's look for a file next to the
4618 * actual image we shall boot. */
4621 r
= file_in_same_dir(arg_image
, arg_settings_filename
, &p
);
4623 return log_error_errno(r
, "Failed to generate settings path from image path: %m");
4624 } else if (arg_directory
) {
4625 r
= file_in_same_dir(arg_directory
, arg_settings_filename
, &p
);
4626 if (r
< 0 && r
!= -EADDRNOTAVAIL
) /* if directory is root fs, don't complain */
4627 return log_error_errno(r
, "Failed to generate settings path from directory path: %m");
4632 if (!f
&& errno
!= ENOENT
)
4633 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4635 /* By default, we do not trust configuration from /var/lib/machines */
4636 if (arg_settings_trusted
< 0)
4637 arg_settings_trusted
= false;
4644 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4646 r
= settings_load(f
, p
, &settings
);
4650 return merge_settings(settings
, p
);
4653 static int load_oci_bundle(void) {
4654 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4657 if (!arg_oci_bundle
)
4660 /* By default let's trust OCI bundles */
4661 if (arg_settings_trusted
< 0)
4662 arg_settings_trusted
= true;
4664 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4668 return merge_settings(settings
, arg_oci_bundle
);
4671 static int run_container(
4672 DissectedImage
*dissected_image
,
4674 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4675 struct ExposeArgs
*expose_args
,
4676 int *master
, pid_t
*pid
, int *ret
) {
4678 static const struct sigaction sa
= {
4679 .sa_handler
= nop_signal_handler
,
4680 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4683 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4684 _cleanup_close_
int etc_passwd_lock
= -EBADF
;
4685 _cleanup_close_pair_
int
4686 fd_inner_socket_pair
[2] = EBADF_PAIR
,
4687 fd_outer_socket_pair
[2] = EBADF_PAIR
;
4689 _cleanup_close_
int notify_socket
= -EBADF
, mntns_fd
= -EBADF
, fd_kmsg_fifo
= -EBADF
;
4690 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4691 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4692 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4693 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4694 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4695 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4696 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4697 size_t n_bind_user_uid
= 0;
4698 ContainerStatus container_status
= 0;
4702 _cleanup_close_
int child_netns_fd
= -EBADF
;
4704 assert_se(sigemptyset(&mask_chld
) == 0);
4705 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4707 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4708 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4709 * check with getpwuid() if the specific user already exists. Note that /etc might be
4710 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4711 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4712 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4715 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4716 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4717 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4720 r
= barrier_create(&barrier
);
4722 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4724 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_inner_socket_pair
) < 0)
4725 return log_error_errno(errno
, "Failed to create inner socket pair: %m");
4727 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_outer_socket_pair
) < 0)
4728 return log_error_errno(errno
, "Failed to create outer socket pair: %m");
4730 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4731 * parent's blocking calls and give it a chance to call wait() and terminate. */
4732 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4734 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4736 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4738 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4740 if (arg_network_namespace_path
) {
4741 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4742 if (child_netns_fd
< 0)
4743 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4745 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4747 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
);
4749 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4751 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4752 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4755 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4757 return log_error_errno(errno
, "clone() failed%s: %m",
4759 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4762 /* The outer child only has a file system namespace. */
4763 barrier_set_role(&barrier
, BARRIER_CHILD
);
4765 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
4766 fd_outer_socket_pair
[0] = safe_close(fd_outer_socket_pair
[0]);
4768 (void) reset_all_signal_handlers();
4769 (void) reset_signal_mask();
4771 r
= outer_child(&barrier
,
4774 fd_outer_socket_pair
[1],
4775 fd_inner_socket_pair
[1],
4779 _exit(EXIT_FAILURE
);
4781 _exit(EXIT_SUCCESS
);
4784 barrier_set_role(&barrier
, BARRIER_PARENT
);
4788 fd_inner_socket_pair
[1] = safe_close(fd_inner_socket_pair
[1]);
4789 fd_outer_socket_pair
[1] = safe_close(fd_outer_socket_pair
[1]);
4791 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4792 mntns_fd
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4794 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4796 /* The child just let us know the UID shift it might have read from the image. */
4797 l
= recv(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4799 return log_error_errno(errno
, "Failed to read UID shift: %m");
4800 if (l
!= sizeof arg_uid_shift
)
4801 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4803 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4804 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4805 * image, but if that's already in use, pick a new one, and report back to the child,
4806 * which one we now picked. */
4808 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4810 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4812 l
= send(fd_outer_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4814 return log_error_errno(errno
, "Failed to send UID shift: %m");
4815 if (l
!= sizeof arg_uid_shift
)
4816 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4819 n_bind_user_uid
= strv_length(arg_bind_user
);
4820 if (n_bind_user_uid
> 0) {
4821 /* Right after the UID shift, we'll receive the list of UID mappings for the
4822 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4824 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4828 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4829 l
= recv(fd_outer_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4831 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4832 if (l
!= sizeof(uid_t
)*4)
4833 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4834 SYNTHETIC_ERRNO(EIO
),
4835 "Short read while reading bind user UID pairs.");
4840 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4841 /* The child let us know the support cgroup mode it might have read from the image. */
4842 l
= recv(fd_outer_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4844 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4845 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4846 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
4847 l
, l
== 0 ? " The child is most likely dead." : "");
4850 /* Wait for the outer child. */
4851 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4854 if (r
!= EXIT_SUCCESS
)
4857 /* And now retrieve the PID of the inner child. */
4858 l
= recv(fd_outer_socket_pair
[0], pid
, sizeof *pid
, 0);
4860 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4861 if (l
!= sizeof *pid
)
4862 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4864 /* We also retrieve container UUID in case it was generated by outer child */
4865 l
= recv(fd_outer_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4867 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4868 if (l
!= sizeof(arg_uuid
))
4869 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4871 /* We also retrieve the socket used for notifications generated by outer child */
4872 notify_socket
= receive_one_fd(fd_outer_socket_pair
[0], 0);
4873 if (notify_socket
< 0)
4874 return log_error_errno(notify_socket
,
4875 "Failed to receive notification socket from the outer child: %m");
4877 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4879 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4880 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4881 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4883 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4887 (void) barrier_place(&barrier
); /* #2 */
4890 if (arg_private_network
) {
4891 if (!arg_network_namespace_path
) {
4892 /* Wait until the child has unshared its network namespace. */
4893 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4894 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4897 if (child_netns_fd
< 0) {
4898 /* Make sure we have an open file descriptor to the child's network
4899 * namespace so it stays alive even if the child exits. */
4900 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4902 return log_error_errno(r
, "Failed to open child network namespace: %m");
4905 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4909 if (arg_network_veth
) {
4910 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4911 arg_network_bridge
|| arg_network_zone
, &arg_network_provided_mac
);
4917 if (arg_network_bridge
) {
4918 /* Add the interface to a bridge */
4919 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4924 } else if (arg_network_zone
) {
4925 /* Add the interface to a bridge, possibly creating it */
4926 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4934 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4938 /* We created the primary and extra veth links now; let's remember this, so that we know to
4939 remove them later on. Note that we don't bother with removing veth links that were created
4940 here when their setup failed half-way, because in that case the kernel should be able to
4941 remove them on its own, since they cannot be referenced by anything yet. */
4942 *veth_created
= true;
4944 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4948 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4953 if (arg_register
|| !arg_keep_unit
) {
4954 r
= sd_bus_default_system(&bus
);
4956 return log_error_errno(r
, "Failed to open system bus: %m");
4958 r
= sd_bus_set_close_on_exit(bus
, false);
4960 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4963 if (!arg_keep_unit
) {
4964 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4965 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4966 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4968 r
= sd_bus_match_signal_async(
4971 "org.freedesktop.systemd1",
4973 "org.freedesktop.systemd1.Scope",
4975 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4977 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4981 r
= register_machine(
4989 arg_custom_mounts
, arg_n_custom_mounts
,
4992 arg_property_message
,
4994 arg_container_service_name
,
4999 } else if (!arg_keep_unit
) {
5005 arg_custom_mounts
, arg_n_custom_mounts
,
5008 arg_property_message
,
5009 /* allow_pidfds= */ true,
5014 } else if (arg_slice
|| arg_property
)
5015 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5017 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5021 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5025 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5029 /* Notify the child that the parent is ready with all
5030 * its setup (including cgroup-ification), and that
5031 * the child can now hand over control to the code to
5032 * run inside the container. */
5033 (void) barrier_place(&barrier
); /* #4 */
5035 /* Block SIGCHLD here, before notifying child.
5036 * process_pty() will handle it with the other signals. */
5037 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5039 /* Reset signal to default */
5040 r
= default_signals(SIGCHLD
);
5042 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5044 r
= sd_event_new(&event
);
5046 return log_error_errno(r
, "Failed to get default event source: %m");
5048 (void) sd_event_set_watchdog(event
, true);
5051 r
= sd_bus_attach_event(bus
, event
, 0);
5053 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5056 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5060 /* Wait that the child is completely ready now, and has mounted their own copies of procfs and so on,
5061 * before we take the fully visible instances away. */
5062 if (!barrier_sync(&barrier
)) /* #5.1 */
5063 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5065 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5066 r
= wipe_fully_visible_fs(mntns_fd
);
5069 mntns_fd
= safe_close(mntns_fd
);
5072 /* And now let the child know that we completed removing the procfs instances, and it can start the
5074 if (!barrier_place(&barrier
)) /* #5.2 */
5075 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5077 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5078 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5079 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5081 (void) sd_notifyf(false,
5082 "STATUS=Container running.\n"
5083 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5084 if (!arg_notify_ready
) {
5085 r
= sd_notify(false, "READY=1\n");
5087 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5090 if (arg_kill_signal
> 0) {
5091 /* Try to kill the init system on SIGINT or SIGTERM */
5092 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5093 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5095 /* Immediately exit */
5096 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5097 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5100 (void) sd_event_add_signal(event
, NULL
, SIGRTMIN
+18, sigrtmin18_handler
, NULL
);
5102 r
= sd_event_add_memory_pressure(event
, NULL
, NULL
, NULL
);
5104 log_debug_errno(r
, "Failed allocate memory pressure event source, ignoring: %m");
5106 /* Exit when the child exits */
5107 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5109 /* Retrieve the kmsg fifo allocated by inner child */
5110 fd_kmsg_fifo
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5111 if (fd_kmsg_fifo
< 0)
5112 return log_error_errno(fd_kmsg_fifo
, "Failed to receive kmsg fifo from inner child: %m");
5114 if (arg_expose_ports
) {
5115 r
= expose_port_watch_rtnl(event
, fd_inner_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5119 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5120 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5123 if (arg_console_mode
!= CONSOLE_PIPE
) {
5124 _cleanup_close_
int fd
= -EBADF
;
5125 PTYForwardFlags flags
= 0;
5127 /* Retrieve the master pty allocated by inner child */
5128 fd
= receive_one_fd(fd_inner_socket_pair
[0], 0);
5130 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5132 switch (arg_console_mode
) {
5134 case CONSOLE_READ_ONLY
:
5135 flags
|= PTY_FORWARD_READ_ONLY
;
5139 case CONSOLE_INTERACTIVE
:
5140 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5142 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5144 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5146 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5147 (void) pty_forward_set_width_height(forward
,
5149 arg_console_height
);
5153 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5156 *master
= TAKE_FD(fd
);
5159 fd_inner_socket_pair
[0] = safe_close(fd_inner_socket_pair
[0]);
5161 r
= sd_event_loop(event
);
5163 return log_error_errno(r
, "Failed to run event loop: %m");
5168 (void) pty_forward_get_last_char(forward
, &last_char
);
5169 forward
= pty_forward_free(forward
);
5171 if (!arg_quiet
&& last_char
!= '\n')
5175 /* Kill if it is not dead yet anyway */
5176 if (!arg_register
&& !arg_keep_unit
&& bus
)
5177 terminate_scope(bus
, arg_machine
);
5179 /* Normally redundant, but better safe than sorry */
5180 (void) kill(*pid
, SIGKILL
);
5182 fd_kmsg_fifo
= safe_close(fd_kmsg_fifo
);
5184 if (arg_private_network
) {
5185 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5186 * to avoid having to move the parent to the child network namespace. */
5187 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG_SIGTERM
|FORK_WAIT
|FORK_LOG
, NULL
);
5192 _cleanup_close_
int parent_netns_fd
= -EBADF
;
5194 r
= namespace_open(getpid_cached(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5196 log_error_errno(r
, "Failed to open parent network namespace: %m");
5197 _exit(EXIT_FAILURE
);
5200 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5202 log_error_errno(r
, "Failed to enter child network namespace: %m");
5203 _exit(EXIT_FAILURE
);
5206 /* Reverse network interfaces pair list so that interfaces get their initial name back.
5207 * This is about ensuring interfaces get their old name back when being moved back. */
5208 arg_network_interfaces
= strv_reverse(arg_network_interfaces
);
5210 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5212 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5214 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5218 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5220 /* Tell machined that we are gone. */
5222 (void) unregister_machine(bus
, arg_machine
);
5225 /* We failed to wait for the container, or the container exited abnormally. */
5227 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5228 /* r > 0 → The container exited with a non-zero status.
5229 * As a special case, we need to replace 133 with a different value,
5230 * because 133 is special-cased in the service file to reboot the container.
5231 * otherwise → The container exited with zero status and a reboot was not requested.
5233 if (r
== EXIT_FORCE_RESTART
)
5234 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5236 return 0; /* finito */
5239 /* CONTAINER_REBOOTED, loop again */
5241 if (arg_keep_unit
) {
5242 /* Special handling if we are running as a service: instead of simply
5243 * restarting the machine we want to restart the entire service, so let's
5244 * inform systemd about this with the special exit code 133. The service
5245 * file uses RestartForceExitStatus=133 so that this results in a full
5246 * nspawn restart. This is necessary since we might have cgroup parameters
5247 * set we want to have flushed out. */
5248 *ret
= EXIT_FORCE_RESTART
;
5249 return 0; /* finito */
5252 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5253 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5255 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5256 *veth_created
= false;
5257 return 1; /* loop again */
5260 static int initialize_rlimits(void) {
5261 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5262 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5263 * container execution environments. */
5265 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5266 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5267 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5268 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5269 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5270 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5271 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5272 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5273 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5274 [RLIMIT_NICE
] = { 0, 0 },
5275 [RLIMIT_NOFILE
] = { 1024, 4096 },
5276 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5277 [RLIMIT_RTPRIO
] = { 0, 0 },
5278 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5279 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5281 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5282 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5283 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5284 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5285 * that PID 1 changes a number of other resource limits during early initialization which is why we
5286 * don't read the other limits from PID 1 but prefer the static table above. */
5291 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5292 /* Let's only fill in what the user hasn't explicitly configured anyway */
5293 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5294 const struct rlimit
*v
;
5295 struct rlimit buffer
;
5297 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5298 /* For these two let's read the limits off PID 1. See above for an explanation. */
5300 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
5301 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5304 } else if (rl
== RLIMIT_NOFILE
) {
5305 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5306 * userspace. Given that nspawn containers are often run without our PID 1,
5307 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5308 * so that container userspace gets similar resources as host userspace
5310 buffer
= kernel_defaults
[rl
];
5311 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5314 v
= kernel_defaults
+ rl
;
5316 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5317 if (!arg_rlimit
[rl
])
5321 if (DEBUG_LOGGING
) {
5322 _cleanup_free_
char *k
= NULL
;
5324 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5325 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5332 static int cant_be_in_netns(void) {
5333 _cleanup_close_
int fd
= -EBADF
;
5337 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5338 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5341 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5344 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5346 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5348 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5349 if (r
== -ENOENT
|| ERRNO_IS_NEG_DISCONNECT(r
))
5350 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5351 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5353 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5355 r
= getpeercred(fd
, &ucred
);
5357 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5359 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5361 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5363 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5364 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5368 static int run(int argc
, char *argv
[]) {
5369 bool remove_directory
= false, remove_image
= false, veth_created
= false, remove_tmprootdir
= false;
5370 _cleanup_close_
int master
= -EBADF
;
5371 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5372 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5373 char veth_name
[IFNAMSIZ
] = "";
5374 struct ExposeArgs expose_args
= {};
5375 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5376 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5377 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5378 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5379 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5382 log_parse_environment();
5385 r
= parse_argv(argc
, argv
);
5389 if (geteuid() != 0) {
5390 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5391 argc
>= 2 ? "Need to be root." :
5392 "Need to be root (and some arguments are usually required).\nHint: try --help");
5396 r
= cant_be_in_netns();
5400 r
= initialize_rlimits();
5404 r
= load_oci_bundle();
5408 r
= determine_names();
5412 r
= load_settings();
5416 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
5417 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
5419 if (!arg_private_network
&& arg_userns_mode
!= USER_NAMESPACE_NO
&& arg_uid_shift
> 0)
5420 arg_caps_retain
&= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE
);
5424 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5428 r
= verify_arguments();
5432 r
= verify_network_interfaces_initialized();
5436 /* Reapply environment settings. */
5437 (void) detect_unified_cgroup_hierarchy_from_environment();
5439 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5440 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5441 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5442 (void) ignore_signals(SIGPIPE
);
5444 n_fd_passed
= sd_listen_fds(false);
5445 if (n_fd_passed
> 0) {
5446 r
= fdset_new_listen_fds(&fds
, false);
5448 log_error_errno(r
, "Failed to collect file descriptors: %m");
5453 /* The "default" umask. This is appropriate for most file and directory
5454 * operations performed by nspawn, and is the umask that will be used for
5455 * the child. Functions like copy_devnodes() change the umask temporarily. */
5458 if (arg_directory
) {
5461 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5462 * /var from the host will propagate into container dynamically (because bad things happen if
5463 * two systems write to the same /var). Let's allow it for the special cases where /var is
5464 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5465 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5466 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5467 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5471 if (arg_ephemeral
) {
5472 _cleanup_free_
char *np
= NULL
;
5474 r
= chase_and_update(&arg_directory
, 0);
5478 /* If the specified path is a mount point we generate the new snapshot immediately
5479 * inside it under a random name. However if the specified is not a mount point we
5480 * create the new snapshot in the parent directory, just next to it. */
5481 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5483 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5487 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5489 r
= tempfn_random(arg_directory
, "machine.", &np
);
5491 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5495 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5496 * only owned by us and no one else. */
5497 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5499 log_error_errno(r
, "Failed to lock %s: %m", np
);
5504 BLOCK_SIGNALS(SIGINT
);
5505 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_directory
, AT_FDCWD
, np
,
5506 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5507 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5508 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5509 BTRFS_SNAPSHOT_RECURSIVE
|
5510 BTRFS_SNAPSHOT_QUOTA
|
5511 BTRFS_SNAPSHOT_SIGINT
);
5514 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5518 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5522 free_and_replace(arg_directory
, np
);
5523 remove_directory
= true;
5525 r
= chase_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5529 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5531 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5535 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5540 r
= chase_and_update(&arg_template
, 0);
5545 BLOCK_SIGNALS(SIGINT
);
5546 r
= btrfs_subvol_snapshot_at(AT_FDCWD
, arg_template
, AT_FDCWD
, arg_directory
,
5547 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5548 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5549 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5550 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5551 BTRFS_SNAPSHOT_RECURSIVE
|
5552 BTRFS_SNAPSHOT_QUOTA
|
5553 BTRFS_SNAPSHOT_SIGINT
);
5556 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5557 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5558 else if (r
== -EINTR
) {
5559 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5562 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5565 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5566 "Populated %s from template %s.", arg_directory
, arg_template
);
5570 if (arg_start_mode
== START_BOOT
) {
5571 _cleanup_free_
char *b
= NULL
;
5573 int check_os_release
, is_os_tree
;
5575 if (arg_pivot_root_new
) {
5576 b
= path_join(arg_directory
, arg_pivot_root_new
);
5586 check_os_release
= getenv_bool("SYSTEMD_NSPAWN_CHECK_OS_RELEASE");
5587 if (check_os_release
< 0 && check_os_release
!= -ENXIO
) {
5588 r
= log_error_errno(check_os_release
, "Failed to parse $SYSTEMD_NSPAWN_CHECK_OS_RELEASE: %m");
5592 is_os_tree
= path_is_os_tree(p
);
5593 if (is_os_tree
== 0 && check_os_release
== 0)
5594 log_debug("Directory %s is missing an os-release file, continuing anyway.", p
);
5595 else if (is_os_tree
<= 0) {
5596 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5597 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5601 _cleanup_free_
char *p
= NULL
;
5603 if (arg_pivot_root_new
)
5604 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5606 p
= path_join(arg_directory
, "/usr/");
5612 if (laccess(p
, F_OK
) < 0) {
5613 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5614 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5620 DissectImageFlags dissect_image_flags
=
5621 DISSECT_IMAGE_GENERIC_ROOT
|
5622 DISSECT_IMAGE_REQUIRE_ROOT
|
5623 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5624 DISSECT_IMAGE_USR_NO_ROOT
|
5625 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5626 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5628 assert(!arg_template
);
5630 r
= chase_and_update(&arg_image
, 0);
5634 if (arg_ephemeral
) {
5635 _cleanup_free_
char *np
= NULL
;
5637 r
= tempfn_random(arg_image
, "machine.", &np
);
5639 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5643 /* Always take an exclusive lock on our own ephemeral copy. */
5644 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5646 r
= log_error_errno(r
, "Failed to create image lock: %m");
5651 BLOCK_SIGNALS(SIGINT
);
5652 r
= copy_file_full(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600,
5653 FS_NOCOW_FL
, FS_NOCOW_FL
,
5654 COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
,
5658 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5662 r
= log_error_errno(r
, "Failed to copy image file: %m");
5666 free_and_replace(arg_image
, np
);
5667 remove_image
= true;
5669 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5671 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5675 r
= log_error_errno(r
, "Failed to create image lock: %m");
5679 r
= verity_settings_load(
5680 &arg_verity_settings
,
5681 arg_image
, NULL
, NULL
);
5683 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5687 if (arg_verity_settings
.data_path
)
5688 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5691 if (!mkdtemp(tmprootdir
)) {
5692 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5696 remove_tmprootdir
= true;
5698 arg_directory
= strdup(tmprootdir
);
5699 if (!arg_directory
) {
5704 r
= loop_device_make_by_path(
5706 arg_read_only
? O_RDONLY
: O_RDWR
,
5707 /* sector_size= */ UINT32_MAX
,
5708 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5712 log_error_errno(r
, "Failed to set up loopback block device: %m");
5716 r
= dissect_loop_device_and_warn(
5718 &arg_verity_settings
,
5719 /* mount_options=*/ NULL
,
5720 arg_image_policy
?: &image_policy_container
,
5721 dissect_image_flags
,
5724 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5725 log_notice("Note that the disk image needs to\n"
5726 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5727 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5728 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5729 " d) or contain a file system without a partition table\n"
5730 "in order to be bootable with systemd-nspawn.");
5736 r
= dissected_image_load_verity_sig_partition(
5739 &arg_verity_settings
);
5743 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5744 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5745 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5747 r
= dissected_image_decrypt_interactively(
5750 &arg_verity_settings
,
5755 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5756 if (remove_image
&& unlink(arg_image
) >= 0)
5757 remove_image
= false;
5759 if (arg_architecture
< 0)
5760 arg_architecture
= dissected_image_architecture(dissected_image
);
5763 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5767 if (arg_console_mode
< 0)
5769 isatty(STDIN_FILENO
) > 0 &&
5770 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5772 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5776 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5777 arg_machine
, arg_image
?: arg_directory
);
5779 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, SIGRTMIN
+18, -1) >= 0);
5781 r
= make_reaper_process(true);
5783 log_error_errno(r
, "Failed to become subreaper: %m");
5787 if (arg_expose_ports
) {
5788 r
= fw_ctx_new(&fw_ctx
);
5790 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5793 expose_args
.fw_ctx
= fw_ctx
;
5796 r
= run_container(dissected_image
,
5798 veth_name
, &veth_created
,
5799 &expose_args
, &master
,
5806 (void) sd_notify(false,
5807 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5808 "STOPPING=1\nSTATUS=Terminating...");
5811 (void) kill(pid
, SIGKILL
);
5813 /* Try to flush whatever is still queued in the pty */
5815 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5816 master
= safe_close(master
);
5820 (void) wait_for_terminate(pid
, NULL
);
5824 if (remove_directory
&& arg_directory
) {
5827 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5829 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5832 if (remove_image
&& arg_image
) {
5833 if (unlink(arg_image
) < 0)
5834 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5837 if (remove_tmprootdir
) {
5838 if (rmdir(tmprootdir
) < 0)
5839 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5845 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5846 (void) rm_rf(p
, REMOVE_ROOT
);
5849 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5850 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5853 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5854 (void) remove_bridge(arg_network_zone
);
5856 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5857 expose_port_free_all(arg_expose_ports
);
5858 rlimit_free_all(arg_rlimit
);
5859 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5860 machine_credential_free_all(arg_credentials
, arg_n_credentials
);
5868 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);