1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <linux/loop.h>
14 #include <selinux/selinux.h>
21 #include <sys/personality.h>
22 #include <sys/prctl.h>
23 #include <sys/types.h>
28 #include "sd-daemon.h"
31 #include "alloc-util.h"
33 #include "base-filesystem.h"
34 #include "blkid-util.h"
35 #include "btrfs-util.h"
36 #include "bus-error.h"
39 #include "capability-util.h"
40 #include "cgroup-util.h"
42 #include "cpu-set-util.h"
43 #include "dev-setup.h"
44 #include "dissect-image.h"
49 #include "format-util.h"
52 #include "hexdecoct.h"
53 #include "hostname-util.h"
54 #include "id128-util.h"
56 #include "loop-util.h"
57 #include "loopback-setup.h"
58 #include "machine-image.h"
60 #include "main-func.h"
63 #include "mount-util.h"
64 #include "mountpoint-util.h"
65 #include "namespace-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-def.h"
69 #include "nspawn-expose-ports.h"
70 #include "nspawn-mount.h"
71 #include "nspawn-network.h"
72 #include "nspawn-oci.h"
73 #include "nspawn-patch-uid.h"
74 #include "nspawn-register.h"
75 #include "nspawn-seccomp.h"
76 #include "nspawn-settings.h"
77 #include "nspawn-setuid.h"
78 #include "nspawn-stub-pid1.h"
79 #include "nulstr-util.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "pretty-print.h"
85 #include "process-util.h"
87 #include "random-util.h"
88 #include "raw-clone.h"
89 #include "rlimit-util.h"
92 #include "seccomp-util.h"
94 #include "selinux-util.h"
95 #include "signal-util.h"
96 #include "socket-util.h"
97 #include "stat-util.h"
98 #include "stdio-util.h"
99 #include "string-table.h"
100 #include "string-util.h"
102 #include "sysctl-util.h"
103 #include "terminal-util.h"
104 #include "tmpfile-util.h"
105 #include "umask-util.h"
106 #include "user-util.h"
110 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
112 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
115 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
116 * nspawn_notify_socket_path is relative to the container
117 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
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 CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
177 static CustomMount
*arg_custom_mounts
= NULL
;
178 static size_t arg_n_custom_mounts
= 0;
179 static char **arg_setenv
= NULL
;
180 static bool arg_quiet
= false;
181 static bool arg_register
= true;
182 static bool arg_keep_unit
= false;
183 static char **arg_network_interfaces
= NULL
;
184 static char **arg_network_macvlan
= NULL
;
185 static char **arg_network_ipvlan
= NULL
;
186 static bool arg_network_veth
= false;
187 static char **arg_network_veth_extra
= NULL
;
188 static char *arg_network_bridge
= NULL
;
189 static char *arg_network_zone
= NULL
;
190 static char *arg_network_namespace_path
= NULL
;
191 static PagerFlags arg_pager_flags
= 0;
192 static unsigned long arg_personality
= PERSONALITY_INVALID
;
193 static char *arg_image
= NULL
;
194 static char *arg_oci_bundle
= NULL
;
195 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
196 static ExposePort
*arg_expose_ports
= NULL
;
197 static char **arg_property
= NULL
;
198 static sd_bus_message
*arg_property_message
= NULL
;
199 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
200 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
201 static bool arg_userns_chown
= false;
202 static int arg_kill_signal
= 0;
203 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
204 static SettingsMask arg_settings_mask
= 0;
205 static int arg_settings_trusted
= -1;
206 static char **arg_parameters
= NULL
;
207 static const char *arg_container_service_name
= "systemd-nspawn";
208 static bool arg_notify_ready
= false;
209 static bool arg_use_cgns
= true;
210 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
211 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
212 static void *arg_root_hash
= NULL
;
213 static size_t arg_root_hash_size
= 0;
214 static char **arg_syscall_whitelist
= NULL
;
215 static char **arg_syscall_blacklist
= NULL
;
217 static scmp_filter_ctx arg_seccomp
= NULL
;
219 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
220 static bool arg_no_new_privileges
= false;
221 static int arg_oom_score_adjust
= 0;
222 static bool arg_oom_score_adjust_set
= false;
223 static CPUSet arg_cpu_set
= {};
224 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
225 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
226 static unsigned arg_console_width
= (unsigned) -1, arg_console_height
= (unsigned) -1;
227 static DeviceNode
* arg_extra_nodes
= NULL
;
228 static size_t arg_n_extra_nodes
= 0;
229 static char **arg_sysctl
= NULL
;
230 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
232 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
233 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
254 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_root_hash
, freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist
, strv_freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist
, strv_freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
262 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
264 static int help(void) {
265 _cleanup_free_
char *link
= NULL
;
268 (void) pager_open(arg_pager_flags
);
270 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
274 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
275 "Spawn a command or OS in a light-weight container.\n\n"
276 " -h --help Show this help\n"
277 " --version Print version string\n"
278 " -q --quiet Do not show status information\n"
279 " --no-pager Do not pipe output into a pager\n"
280 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
282 " -D --directory=PATH Root directory for the container\n"
283 " --template=PATH Initialize root directory from template directory,\n"
285 " -x --ephemeral Run container with snapshot of root directory, and\n"
286 " remove it after exit\n"
287 " -i --image=PATH Root file system disk image (or device node) for\n"
289 " --oci-bundle=PATH OCI bundle directory\n"
290 " --read-only Mount the root directory read-only\n"
291 " --volatile[=MODE] Run the system in volatile mode\n"
292 " --root-hash=HASH Specify verity root hash for root disk image\n"
293 " --pivot-root=PATH[:PATH]\n"
294 " Pivot root to given directory in the container\n\n"
295 "%3$sExecution:%4$s\n"
296 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
297 " -b --boot Boot up full system (i.e. invoke init)\n"
298 " --chdir=PATH Set working directory in the container\n"
299 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
300 " -u --user=USER Run the command under specified user or UID\n"
301 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
302 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
303 "%3$sSystem Identity:%4$s\n"
304 " -M --machine=NAME Set the machine name for the container\n"
305 " --hostname=NAME Override the hostname for the container\n"
306 " --uuid=UUID Set a specific machine UUID for the container\n\n"
307 "%3$sProperties:%4$s\n"
308 " -S --slice=SLICE Place the container in the specified slice\n"
309 " --property=NAME=VALUE Set scope unit property\n"
310 " --register=BOOLEAN Register container as machine\n"
311 " --keep-unit Do not register a scope for the machine, reuse\n"
312 " the service unit nspawn is running in\n\n"
313 "%3$sUser Namespacing:%4$s\n"
314 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
315 " --private-users[=UIDBASE[:NUIDS]]\n"
316 " Similar, but with user configured UID/GID range\n"
317 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
318 "%3$sNetworking:%4$s\n"
319 " --private-network Disable network in container\n"
320 " --network-interface=INTERFACE\n"
321 " Assign an existing network interface to the\n"
323 " --network-macvlan=INTERFACE\n"
324 " Create a macvlan network interface based on an\n"
325 " existing network interface to the container\n"
326 " --network-ipvlan=INTERFACE\n"
327 " Create a ipvlan network interface based on an\n"
328 " existing network interface to the container\n"
329 " -n --network-veth Add a virtual Ethernet connection between host\n"
331 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
332 " Add an additional virtual Ethernet link between\n"
333 " host and container\n"
334 " --network-bridge=INTERFACE\n"
335 " Add a virtual Ethernet connection to the container\n"
336 " and attach it to an existing bridge on the host\n"
337 " --network-zone=NAME Similar, but attach the new interface to an\n"
338 " an automatically managed bridge interface\n"
339 " --network-namespace-path=PATH\n"
340 " Set network namespace to the one represented by\n"
341 " the specified kernel namespace file node\n"
342 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
343 " Expose a container IP port on the host\n\n"
344 "%3$sSecurity:%4$s\n"
345 " --capability=CAP In addition to the default, retain specified\n"
347 " --drop-capability=CAP Drop the specified capability from the default set\n"
348 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
349 " --system-call-filter=LIST|~LIST\n"
350 " Permit/prohibit specific system calls\n"
351 " -Z --selinux-context=SECLABEL\n"
352 " Set the SELinux security context to be used by\n"
353 " processes in the container\n"
354 " -L --selinux-apifs-context=SECLABEL\n"
355 " Set the SELinux security context to be used by\n"
356 " API/tmpfs file systems in the container\n\n"
357 "%3$sResources:%4$s\n"
358 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
359 " --oom-score-adjust=VALUE\n"
360 " Adjust the OOM score value for the payload\n"
361 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
362 " --personality=ARCH Pick personality for this container\n\n"
363 "%3$sIntegration:%4$s\n"
364 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
365 " --timezone=MODE Select mode of /etc/localtime initialization\n"
366 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
367 " host, try-guest, try-host\n"
368 " -j Equivalent to --link-journal=try-guest\n\n"
370 " --bind=PATH[:PATH[:OPTIONS]]\n"
371 " Bind mount a file or directory from the host into\n"
373 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
374 " Similar, but creates a read-only bind mount\n"
375 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
377 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
378 " --overlay=PATH[:PATH...]:PATH\n"
379 " Create an overlay mount from the host to \n"
381 " --overlay-ro=PATH[:PATH...]:PATH\n"
382 " Similar, but creates a read-only overlay mount\n\n"
383 "%3$sInput/Output:%4$s\n"
384 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
385 " set up for the container.\n"
386 " -P --pipe Equivalent to --console=pipe\n"
387 "\nSee the %2$s for details.\n"
388 , program_invocation_short_name
390 , ansi_underline(), ansi_normal());
395 static int custom_mount_check_all(void) {
398 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
399 CustomMount
*m
= &arg_custom_mounts
[i
];
401 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
402 if (arg_userns_chown
)
403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
404 "--private-users-chown may not be combined with custom root mounts.");
405 else if (arg_uid_shift
== UID_INVALID
)
406 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
407 "--private-users with automatic UID shift may not be combined with custom root mounts.");
414 static int detect_unified_cgroup_hierarchy_from_environment(void) {
415 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
418 /* Allow the user to control whether the unified hierarchy is used */
422 static bool warned
= false;
424 var
= "UNIFIED_CGROUP_HIERARCHY";
427 log_info("$UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY.");
433 r
= parse_boolean(e
);
435 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
437 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
439 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
445 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
448 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
449 * in the image actually supports. */
450 r
= cg_all_unified();
452 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
454 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
455 * routine only detects 231, so we'll have a false negative here for 230. */
456 r
= systemd_installation_has_version(directory
, 230);
458 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
460 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
462 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
463 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
464 /* Mixed cgroup hierarchy support was added in 233 */
465 r
= systemd_installation_has_version(directory
, 233);
467 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
469 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
471 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
473 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
475 log_debug("Using %s hierarchy for container.",
476 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
477 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
482 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
485 r
= getenv_bool(name
);
489 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
491 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
492 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
496 static int parse_mount_settings_env(void) {
500 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
501 if (r
< 0 && r
!= -ENXIO
)
502 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
504 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
506 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
507 if (streq_ptr(e
, "network"))
508 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
511 r
= parse_boolean(e
);
513 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
515 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
516 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
522 static int parse_environment(void) {
526 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
529 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
532 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
535 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
539 r
= parse_mount_settings_env();
543 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
544 * even if it is supported. If not supported, it has no effect. */
545 if (!cg_ns_supported())
546 arg_use_cgns
= false;
548 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
551 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
555 arg_use_cgns
= r
> 0;
556 arg_settings_mask
|= SETTING_USE_CGNS
;
560 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
562 arg_container_service_name
= e
;
564 return detect_unified_cgroup_hierarchy_from_environment();
567 static int parse_argv(int argc
, char *argv
[]) {
585 ARG_NETWORK_INTERFACE
,
590 ARG_NETWORK_VETH_EXTRA
,
591 ARG_NETWORK_NAMESPACE_PATH
,
601 ARG_PRIVATE_USERS_CHOWN
,
604 ARG_SYSTEM_CALL_FILTER
,
607 ARG_NO_NEW_PRIVILEGES
,
608 ARG_OOM_SCORE_ADJUST
,
618 static const struct option options
[] = {
619 { "help", no_argument
, NULL
, 'h' },
620 { "version", no_argument
, NULL
, ARG_VERSION
},
621 { "directory", required_argument
, NULL
, 'D' },
622 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
623 { "ephemeral", no_argument
, NULL
, 'x' },
624 { "user", required_argument
, NULL
, 'u' },
625 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
626 { "as-pid2", no_argument
, NULL
, 'a' },
627 { "boot", no_argument
, NULL
, 'b' },
628 { "uuid", required_argument
, NULL
, ARG_UUID
},
629 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
630 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
631 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
632 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
633 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
634 { "bind", required_argument
, NULL
, ARG_BIND
},
635 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
636 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
637 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
638 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
639 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
640 { "machine", required_argument
, NULL
, 'M' },
641 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
642 { "slice", required_argument
, NULL
, 'S' },
643 { "setenv", required_argument
, NULL
, 'E' },
644 { "selinux-context", required_argument
, NULL
, 'Z' },
645 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
646 { "quiet", no_argument
, NULL
, 'q' },
647 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
648 { "register", required_argument
, NULL
, ARG_REGISTER
},
649 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
650 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
651 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
652 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
653 { "network-veth", no_argument
, NULL
, 'n' },
654 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
655 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
656 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
657 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
658 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
659 { "image", required_argument
, NULL
, 'i' },
660 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
661 { "port", required_argument
, NULL
, 'p' },
662 { "property", required_argument
, NULL
, ARG_PROPERTY
},
663 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
664 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
665 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
666 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
667 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
668 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
669 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
670 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
671 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
672 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
673 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
674 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
675 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
676 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
677 { "console", required_argument
, NULL
, ARG_CONSOLE
},
678 { "pipe", no_argument
, NULL
, ARG_PIPE
},
679 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
680 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
686 uint64_t plus
= 0, minus
= 0;
687 bool mask_all_settings
= false, mask_no_settings
= false;
692 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
702 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
706 arg_settings_mask
|= SETTING_DIRECTORY
;
710 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
714 arg_settings_mask
|= SETTING_DIRECTORY
;
718 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
722 arg_settings_mask
|= SETTING_DIRECTORY
;
726 r
= parse_path_argument_and_warn(optarg
, false, &arg_oci_bundle
);
733 arg_ephemeral
= true;
734 arg_settings_mask
|= SETTING_EPHEMERAL
;
738 r
= free_and_strdup(&arg_user
, optarg
);
742 arg_settings_mask
|= SETTING_USER
;
745 case ARG_NETWORK_ZONE
: {
748 j
= strjoin("vz-", optarg
);
752 if (!ifname_valid(j
)) {
753 log_error("Network zone name not valid: %s", j
);
758 free_and_replace(arg_network_zone
, j
);
760 arg_network_veth
= true;
761 arg_private_network
= true;
762 arg_settings_mask
|= SETTING_NETWORK
;
766 case ARG_NETWORK_BRIDGE
:
768 if (!ifname_valid(optarg
))
769 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
770 "Bridge interface name not valid: %s", optarg
);
772 r
= free_and_strdup(&arg_network_bridge
, optarg
);
778 arg_network_veth
= true;
779 arg_private_network
= true;
780 arg_settings_mask
|= SETTING_NETWORK
;
783 case ARG_NETWORK_VETH_EXTRA
:
784 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
786 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
788 arg_private_network
= true;
789 arg_settings_mask
|= SETTING_NETWORK
;
792 case ARG_NETWORK_INTERFACE
:
793 if (!ifname_valid(optarg
))
794 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
795 "Network interface name not valid: %s", optarg
);
797 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
800 arg_private_network
= true;
801 arg_settings_mask
|= SETTING_NETWORK
;
804 case ARG_NETWORK_MACVLAN
:
806 if (!ifname_valid(optarg
))
807 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
808 "MACVLAN network interface name not valid: %s", optarg
);
810 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
813 arg_private_network
= true;
814 arg_settings_mask
|= SETTING_NETWORK
;
817 case ARG_NETWORK_IPVLAN
:
819 if (!ifname_valid(optarg
))
820 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
821 "IPVLAN network interface name not valid: %s", optarg
);
823 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
827 case ARG_PRIVATE_NETWORK
:
828 arg_private_network
= true;
829 arg_settings_mask
|= SETTING_NETWORK
;
832 case ARG_NETWORK_NAMESPACE_PATH
:
833 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
837 arg_settings_mask
|= SETTING_NETWORK
;
841 if (arg_start_mode
== START_PID2
)
842 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
843 "--boot and --as-pid2 may not be combined.");
845 arg_start_mode
= START_BOOT
;
846 arg_settings_mask
|= SETTING_START_MODE
;
850 if (arg_start_mode
== START_BOOT
)
851 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
852 "--boot and --as-pid2 may not be combined.");
854 arg_start_mode
= START_PID2
;
855 arg_settings_mask
|= SETTING_START_MODE
;
859 r
= sd_id128_from_string(optarg
, &arg_uuid
);
861 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
863 if (sd_id128_is_null(arg_uuid
))
864 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
865 "Machine UUID may not be all zeroes.");
867 arg_settings_mask
|= SETTING_MACHINE_ID
;
871 r
= free_and_strdup(&arg_slice
, optarg
);
875 arg_settings_mask
|= SETTING_SLICE
;
880 arg_machine
= mfree(arg_machine
);
882 if (!machine_name_is_valid(optarg
))
883 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
884 "Invalid machine name: %s", optarg
);
886 r
= free_and_strdup(&arg_machine
, optarg
);
894 arg_hostname
= mfree(arg_hostname
);
896 if (!hostname_is_valid(optarg
, false))
897 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
898 "Invalid hostname: %s", optarg
);
900 r
= free_and_strdup(&arg_hostname
, optarg
);
905 arg_settings_mask
|= SETTING_HOSTNAME
;
909 arg_selinux_context
= optarg
;
913 arg_selinux_apifs_context
= optarg
;
917 arg_read_only
= true;
918 arg_settings_mask
|= SETTING_READ_ONLY
;
922 case ARG_DROP_CAPABILITY
: {
925 _cleanup_free_
char *t
= NULL
;
927 r
= extract_first_word(&p
, &t
, ",", 0);
929 return log_error_errno(r
, "Failed to parse capability %s.", t
);
933 if (streq(t
, "all")) {
934 if (c
== ARG_CAPABILITY
)
935 plus
= (uint64_t) -1;
937 minus
= (uint64_t) -1;
939 r
= capability_from_name(t
);
941 return log_error_errno(r
, "Failed to parse capability %s.", t
);
943 if (c
== ARG_CAPABILITY
)
950 arg_settings_mask
|= SETTING_CAPABILITY
;
954 case ARG_NO_NEW_PRIVILEGES
:
955 r
= parse_boolean(optarg
);
957 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
959 arg_no_new_privileges
= r
;
960 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
964 arg_link_journal
= LINK_GUEST
;
965 arg_link_journal_try
= true;
966 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
969 case ARG_LINK_JOURNAL
:
970 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
972 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
974 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
979 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
981 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
983 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
987 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
989 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
991 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
996 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
997 if (r
== -EADDRNOTAVAIL
)
998 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1000 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1002 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1005 case ARG_INACCESSIBLE
:
1006 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1008 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1010 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1016 if (!env_assignment_is_valid(optarg
))
1017 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1018 "Environment variable assignment '%s' is not valid.", optarg
);
1020 n
= strv_env_set(arg_setenv
, optarg
);
1024 strv_free_and_replace(arg_setenv
, n
);
1025 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1033 case ARG_SHARE_SYSTEM
:
1034 /* We don't officially support this anymore, except for compat reasons. People should use the
1035 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1036 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1037 arg_clone_ns_flags
= 0;
1041 r
= parse_boolean(optarg
);
1043 log_error("Failed to parse --register= argument: %s", optarg
);
1051 arg_keep_unit
= true;
1054 case ARG_PERSONALITY
:
1056 arg_personality
= personality_from_string(optarg
);
1057 if (arg_personality
== PERSONALITY_INVALID
)
1058 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1059 "Unknown or unsupported personality '%s'.", optarg
);
1061 arg_settings_mask
|= SETTING_PERSONALITY
;
1067 arg_volatile_mode
= VOLATILE_YES
;
1068 else if (streq(optarg
, "help")) {
1069 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1074 m
= volatile_mode_from_string(optarg
);
1076 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1077 "Failed to parse --volatile= argument: %s", optarg
);
1079 arg_volatile_mode
= m
;
1082 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1086 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1088 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1090 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1092 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1096 if (strv_extend(&arg_property
, optarg
) < 0)
1101 case ARG_PRIVATE_USERS
: {
1106 else if (!in_charset(optarg
, DIGITS
))
1107 /* do *not* parse numbers as booleans */
1108 boolean
= parse_boolean(optarg
);
1110 if (boolean
== false) {
1111 /* no: User namespacing off */
1112 arg_userns_mode
= USER_NAMESPACE_NO
;
1113 arg_uid_shift
= UID_INVALID
;
1114 arg_uid_range
= UINT32_C(0x10000);
1115 } else if (boolean
== true) {
1116 /* yes: User namespacing on, UID range is read from root dir */
1117 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1118 arg_uid_shift
= UID_INVALID
;
1119 arg_uid_range
= UINT32_C(0x10000);
1120 } else if (streq(optarg
, "pick")) {
1121 /* pick: User namespacing on, UID range is picked randomly */
1122 arg_userns_mode
= USER_NAMESPACE_PICK
;
1123 arg_uid_shift
= UID_INVALID
;
1124 arg_uid_range
= UINT32_C(0x10000);
1126 _cleanup_free_
char *buffer
= NULL
;
1127 const char *range
, *shift
;
1129 /* anything else: User namespacing on, UID range is explicitly configured */
1131 range
= strchr(optarg
, ':');
1133 buffer
= strndup(optarg
, range
- optarg
);
1139 r
= safe_atou32(range
, &arg_uid_range
);
1141 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1145 r
= parse_uid(shift
, &arg_uid_shift
);
1147 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1149 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1152 if (arg_uid_range
<= 0)
1153 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1154 "UID range cannot be 0.");
1156 arg_settings_mask
|= SETTING_USERNS
;
1161 if (userns_supported()) {
1162 arg_userns_mode
= USER_NAMESPACE_PICK
;
1163 arg_uid_shift
= UID_INVALID
;
1164 arg_uid_range
= UINT32_C(0x10000);
1166 arg_settings_mask
|= SETTING_USERNS
;
1171 case ARG_PRIVATE_USERS_CHOWN
:
1172 arg_userns_chown
= true;
1174 arg_settings_mask
|= SETTING_USERNS
;
1177 case ARG_KILL_SIGNAL
:
1178 if (streq(optarg
, "help")) {
1179 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1183 arg_kill_signal
= signal_from_string(optarg
);
1184 if (arg_kill_signal
< 0)
1185 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1186 "Cannot parse signal: %s", optarg
);
1188 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1193 /* no → do not read files
1194 * yes → read files, do not override cmdline, trust only subset
1195 * override → read files, override cmdline, trust only subset
1196 * trusted → read files, do not override cmdline, trust all
1199 r
= parse_boolean(optarg
);
1201 if (streq(optarg
, "trusted")) {
1202 mask_all_settings
= false;
1203 mask_no_settings
= false;
1204 arg_settings_trusted
= true;
1206 } else if (streq(optarg
, "override")) {
1207 mask_all_settings
= false;
1208 mask_no_settings
= true;
1209 arg_settings_trusted
= -1;
1211 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1214 mask_all_settings
= false;
1215 mask_no_settings
= false;
1216 arg_settings_trusted
= -1;
1219 mask_all_settings
= true;
1220 mask_no_settings
= false;
1221 arg_settings_trusted
= false;
1227 if (!path_is_absolute(optarg
))
1228 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1229 "Working directory %s is not an absolute path.", optarg
);
1231 r
= free_and_strdup(&arg_chdir
, optarg
);
1235 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1238 case ARG_PIVOT_ROOT
:
1239 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1241 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1243 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1246 case ARG_NOTIFY_READY
:
1247 r
= parse_boolean(optarg
);
1249 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1250 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1251 arg_notify_ready
= r
;
1252 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1255 case ARG_ROOT_HASH
: {
1259 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1261 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1262 if (l
< sizeof(sd_id128_t
)) {
1264 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1267 free(arg_root_hash
);
1269 arg_root_hash_size
= l
;
1273 case ARG_SYSTEM_CALL_FILTER
: {
1277 negative
= optarg
[0] == '~';
1278 items
= negative
? optarg
+ 1 : optarg
;
1281 _cleanup_free_
char *word
= NULL
;
1283 r
= extract_first_word(&items
, &word
, NULL
, 0);
1289 return log_error_errno(r
, "Failed to parse system call filter: %m");
1292 r
= strv_extend(&arg_syscall_blacklist
, word
);
1294 r
= strv_extend(&arg_syscall_whitelist
, word
);
1299 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1305 _cleanup_free_
char *name
= NULL
;
1308 if (streq(optarg
, "help")) {
1309 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1313 eq
= strchr(optarg
, '=');
1315 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1316 "--rlimit= expects an '=' assignment.");
1318 name
= strndup(optarg
, eq
- optarg
);
1322 rl
= rlimit_from_string_harder(name
);
1324 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1325 "Unknown resource limit: %s", name
);
1327 if (!arg_rlimit
[rl
]) {
1328 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1329 if (!arg_rlimit
[rl
])
1333 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1335 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1337 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1341 case ARG_OOM_SCORE_ADJUST
:
1342 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1344 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1346 arg_oom_score_adjust_set
= true;
1347 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1350 case ARG_CPU_AFFINITY
: {
1353 r
= parse_cpu_set(optarg
, &cpuset
);
1355 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1357 cpu_set_reset(&arg_cpu_set
);
1358 arg_cpu_set
= cpuset
;
1359 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1363 case ARG_RESOLV_CONF
:
1364 if (streq(optarg
, "help")) {
1365 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1369 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1370 if (arg_resolv_conf
< 0)
1371 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1372 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1374 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1378 if (streq(optarg
, "help")) {
1379 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1383 arg_timezone
= timezone_mode_from_string(optarg
);
1384 if (arg_timezone
< 0)
1385 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1386 "Failed to parse /etc/localtime mode: %s", optarg
);
1388 arg_settings_mask
|= SETTING_TIMEZONE
;
1392 if (streq(optarg
, "interactive"))
1393 arg_console_mode
= CONSOLE_INTERACTIVE
;
1394 else if (streq(optarg
, "read-only"))
1395 arg_console_mode
= CONSOLE_READ_ONLY
;
1396 else if (streq(optarg
, "passive"))
1397 arg_console_mode
= CONSOLE_PASSIVE
;
1398 else if (streq(optarg
, "pipe"))
1399 arg_console_mode
= CONSOLE_PIPE
;
1400 else if (streq(optarg
, "help"))
1401 puts("interactive\n"
1406 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
1408 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
1413 arg_console_mode
= CONSOLE_PIPE
;
1414 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
1418 arg_pager_flags
|= PAGER_DISABLE
;
1425 assert_not_reached("Unhandled option");
1428 if (argc
> optind
) {
1429 strv_free(arg_parameters
);
1430 arg_parameters
= strv_copy(argv
+ optind
);
1431 if (!arg_parameters
)
1434 arg_settings_mask
|= SETTING_START_MODE
;
1437 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1438 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1439 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1440 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1442 arg_directory
= TAKE_PTR(arg_template
);
1444 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1446 /* Make sure to parse environment before we reset the settings mask below */
1447 r
= parse_environment();
1451 /* Load all settings from .nspawn files */
1452 if (mask_no_settings
)
1453 arg_settings_mask
= 0;
1455 /* Don't load any settings from .nspawn files */
1456 if (mask_all_settings
)
1457 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1462 static int verify_arguments(void) {
1465 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1466 /* If we are running the stub init in the container, we don't need to look at what the init
1467 * in the container supports, because we are not using it. Let's immediately pick the right
1468 * setting based on the host system configuration.
1470 * We only do this, if the user didn't use an environment variable to override the detection.
1473 r
= cg_all_unified();
1475 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1477 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1478 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1479 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1481 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1484 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1485 arg_mount_settings
|= MOUNT_USE_USERNS
;
1487 if (arg_private_network
)
1488 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1490 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1491 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1492 arg_register
= false;
1493 if (arg_start_mode
!= START_PID1
)
1494 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1497 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1498 arg_userns_chown
= true;
1500 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1501 arg_kill_signal
= SIGRTMIN
+3;
1503 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 */
1504 arg_read_only
= true;
1506 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1507 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1508 * The latter is not technically a user session, but we don't need to labour the point. */
1509 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1511 if (arg_directory
&& arg_image
)
1512 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1514 if (arg_template
&& arg_image
)
1515 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1517 if (arg_template
&& !(arg_directory
|| arg_machine
))
1518 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1520 if (arg_ephemeral
&& arg_template
)
1521 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1523 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1524 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1526 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1527 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1529 if (arg_userns_chown
&& arg_read_only
)
1530 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1531 "--read-only and --private-users-chown may not be combined.");
1533 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1534 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1535 * copy-up (in case of overlay) making the entire exercise pointless. */
1536 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1537 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1539 /* If --network-namespace-path is given with any other network-related option, we need to error out,
1540 * to avoid conflicts between different network options. */
1541 if (arg_network_namespace_path
&&
1542 (arg_network_interfaces
|| arg_network_macvlan
||
1543 arg_network_ipvlan
|| arg_network_veth_extra
||
1544 arg_network_bridge
|| arg_network_zone
||
1545 arg_network_veth
|| arg_private_network
))
1546 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1548 if (arg_network_bridge
&& arg_network_zone
)
1549 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1550 "--network-bridge= and --network-zone= may not be combined.");
1552 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1553 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1555 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1556 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1558 if (arg_expose_ports
&& !arg_private_network
)
1559 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1562 if (arg_expose_ports
)
1563 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1566 r
= custom_mount_check_all();
1573 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1576 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1579 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1582 if (uid
!= UID_INVALID
) {
1583 uid
+= arg_uid_shift
;
1585 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1589 if (gid
!= GID_INVALID
) {
1590 gid
+= (gid_t
) arg_uid_shift
;
1592 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1596 if (lchown(p
, uid
, gid
) < 0)
1602 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1606 q
= prefix_roota(root
, path
);
1607 r
= mkdir_errno_wrapper(q
, mode
);
1613 return userns_lchown(q
, uid
, gid
);
1616 static const char *timezone_from_path(const char *path
) {
1617 return PATH_STARTSWITH_SET(
1619 "../usr/share/zoneinfo/",
1620 "/usr/share/zoneinfo/");
1623 static bool etc_writable(void) {
1624 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1627 static int setup_timezone(const char *dest
) {
1628 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1629 const char *where
, *check
;
1635 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1636 r
= readlink_malloc("/etc/localtime", &p
);
1637 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1638 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1639 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1640 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1642 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1643 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1647 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1650 } else if (arg_timezone
== TIMEZONE_AUTO
)
1651 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1657 if (m
== TIMEZONE_OFF
)
1660 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1662 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1666 where
= strjoina(etc
, "/localtime");
1670 case TIMEZONE_DELETE
:
1671 if (unlink(where
) < 0)
1672 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1676 case TIMEZONE_SYMLINK
: {
1677 _cleanup_free_
char *q
= NULL
;
1678 const char *z
, *what
;
1680 z
= timezone_from_path(p
);
1682 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1686 r
= readlink_malloc(where
, &q
);
1687 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1688 return 0; /* Already pointing to the right place? Then do nothing .. */
1690 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1691 r
= chase_symlinks(check
, dest
, 0, NULL
);
1693 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1695 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1696 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1697 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1701 what
= strjoina("../usr/share/zoneinfo/", z
);
1702 if (symlink(what
, where
) < 0) {
1703 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1704 errno
, "Failed to correct timezone of container, ignoring: %m");
1714 case TIMEZONE_BIND
: {
1715 _cleanup_free_
char *resolved
= NULL
;
1718 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1720 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1724 if (found
== 0) /* missing? */
1725 (void) touch(resolved
);
1727 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1729 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1735 /* If mounting failed, try to copy */
1736 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1738 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1739 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1746 assert_not_reached("unexpected mode");
1749 /* Fix permissions of the symlink or file copy we just created */
1750 r
= userns_lchown(where
, 0, 0);
1752 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1757 static int have_resolv_conf(const char *path
) {
1760 if (access(path
, F_OK
) < 0) {
1761 if (errno
== ENOENT
)
1764 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1770 static int resolved_listening(void) {
1771 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1772 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1773 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1776 /* Check if resolved is listening */
1778 r
= sd_bus_open_system(&bus
);
1780 return log_debug_errno(r
, "Failed to open system bus: %m");
1782 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1784 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1788 r
= sd_bus_get_property_string(bus
,
1789 "org.freedesktop.resolve1",
1790 "/org/freedesktop/resolve1",
1791 "org.freedesktop.resolve1.Manager",
1794 &dns_stub_listener_mode
);
1796 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1798 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1801 static int setup_resolv_conf(const char *dest
) {
1802 _cleanup_free_
char *etc
= NULL
;
1803 const char *where
, *what
;
1809 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1810 if (arg_private_network
)
1811 m
= RESOLV_CONF_OFF
;
1812 else if (have_resolv_conf(STATIC_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1813 m
= etc_writable() ? RESOLV_CONF_COPY_STATIC
: RESOLV_CONF_BIND_STATIC
;
1814 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1815 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1817 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1819 m
= arg_resolv_conf
;
1821 if (m
== RESOLV_CONF_OFF
)
1824 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1826 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1830 where
= strjoina(etc
, "/resolv.conf");
1832 if (m
== RESOLV_CONF_DELETE
) {
1833 if (unlink(where
) < 0)
1834 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1839 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_COPY_STATIC
))
1840 what
= STATIC_RESOLV_CONF
;
1842 what
= "/etc/resolv.conf";
1844 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
)) {
1845 _cleanup_free_
char *resolved
= NULL
;
1848 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1850 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1854 if (found
== 0) /* missing? */
1855 (void) touch(resolved
);
1857 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1859 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1862 /* If that didn't work, let's copy the file */
1863 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
1865 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1866 * resolved or something similar runs inside and the symlink points there.
1868 * If the disk image is read-only, there's also no point in complaining.
1870 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
) && IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1871 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1875 r
= userns_lchown(where
, 0, 0);
1877 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1882 static int setup_boot_id(void) {
1883 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1884 _cleanup_free_
char *path
= NULL
;
1885 sd_id128_t rnd
= SD_ID128_NULL
;
1889 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1891 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
1893 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1895 r
= sd_id128_randomize(&rnd
);
1897 return log_error_errno(r
, "Failed to generate random boot id: %m");
1899 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1901 return log_error_errno(r
, "Failed to write boot id: %m");
1903 from
= TAKE_PTR(path
);
1904 to
= "/proc/sys/kernel/random/boot_id";
1906 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1910 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1913 static int copy_devnodes(const char *dest
) {
1914 static const char devnodes
[] =
1923 _cleanup_umask_ mode_t u
;
1931 /* Create /dev/net, so that we can create /dev/net/tun in it */
1932 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1933 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1935 NULSTR_FOREACH(d
, devnodes
) {
1936 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1939 from
= path_join("/dev/", d
);
1943 to
= path_join(dest
, from
);
1947 if (stat(from
, &st
) < 0) {
1949 if (errno
!= ENOENT
)
1950 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1952 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1953 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
1954 "%s is not a char or block device, cannot copy.", from
);
1956 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
1958 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1959 /* Explicitly warn the user when /dev is already populated. */
1960 if (errno
== EEXIST
)
1961 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
);
1963 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
1965 /* Some systems abusively restrict mknod but allow bind mounts. */
1968 return log_error_errno(r
, "touch (%s) failed: %m", to
);
1969 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
1971 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
1974 r
= userns_lchown(to
, 0, 0);
1976 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
1978 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
1982 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
1984 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
1986 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
1989 prefixed
= path_join(dest
, sl
);
1993 t
= path_join("..", d
);
1997 if (symlink(t
, prefixed
) < 0)
1998 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2005 static int make_extra_nodes(const char *dest
) {
2006 _cleanup_umask_ mode_t u
;
2012 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2013 _cleanup_free_
char *path
= NULL
;
2014 DeviceNode
*n
= arg_extra_nodes
+ i
;
2016 path
= path_join(dest
, n
->path
);
2020 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2021 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2023 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2025 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2031 static int setup_pts(const char *dest
) {
2032 _cleanup_free_
char *options
= NULL
;
2037 if (arg_selinux_apifs_context
)
2038 (void) asprintf(&options
,
2039 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2040 arg_uid_shift
+ TTY_GID
,
2041 arg_selinux_apifs_context
);
2044 (void) asprintf(&options
,
2045 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2046 arg_uid_shift
+ TTY_GID
);
2051 /* Mount /dev/pts itself */
2052 p
= prefix_roota(dest
, "/dev/pts");
2053 r
= mkdir_errno_wrapper(p
, 0755);
2055 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2057 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2060 r
= userns_lchown(p
, 0, 0);
2062 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2064 /* Create /dev/ptmx symlink */
2065 p
= prefix_roota(dest
, "/dev/ptmx");
2066 if (symlink("pts/ptmx", p
) < 0)
2067 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2068 r
= userns_lchown(p
, 0, 0);
2070 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2072 /* And fix /dev/pts/ptmx ownership */
2073 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2074 r
= userns_lchown(p
, 0, 0);
2076 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2081 static int setup_stdio_as_dev_console(void) {
2085 terminal
= open_terminal("/dev/console", O_RDWR
);
2087 return log_error_errno(terminal
, "Failed to open console: %m");
2089 /* Make sure we can continue logging to the original stderr, even if
2090 * stderr points elsewhere now */
2091 r
= log_dup_console();
2093 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2095 /* invalidates 'terminal' on success and failure */
2096 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2098 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2103 static int setup_dev_console(const char *console
) {
2104 _cleanup_free_
char *p
= NULL
;
2107 /* Create /dev/console symlink */
2108 r
= path_make_relative("/dev", console
, &p
);
2110 return log_error_errno(r
, "Failed to create relative path: %m");
2112 if (symlink(p
, "/dev/console") < 0)
2113 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2118 static int setup_keyring(void) {
2119 key_serial_t keyring
;
2121 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2122 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2123 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2124 * these system calls let's make sure we don't leak anything into the container. */
2126 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2127 if (keyring
== -1) {
2128 if (errno
== ENOSYS
)
2129 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2130 else if (IN_SET(errno
, EACCES
, EPERM
))
2131 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2133 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2139 static int setup_kmsg(int kmsg_socket
) {
2140 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2141 _cleanup_free_
char *fifo
= NULL
;
2142 _cleanup_close_
int fd
= -1;
2143 _cleanup_umask_ mode_t u
;
2146 assert(kmsg_socket
>= 0);
2150 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2151 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2152 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2153 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2155 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2157 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2159 if (mkfifo(fifo
, 0600) < 0)
2160 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2162 from
= TAKE_PTR(fifo
);
2164 r
= mount_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2168 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2170 return log_error_errno(errno
, "Failed to open fifo: %m");
2172 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2173 r
= send_one_fd(kmsg_socket
, fd
, 0);
2175 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2180 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2181 union in_addr_union
*exposed
= userdata
;
2187 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2191 static int setup_hostname(void) {
2194 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2197 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2199 return log_error_errno(r
, "Failed to set hostname: %m");
2204 static int setup_journal(const char *directory
) {
2205 _cleanup_free_
char *d
= NULL
;
2206 const char *dirname
, *p
, *q
;
2212 /* Don't link journals in ephemeral mode */
2216 if (arg_link_journal
== LINK_NO
)
2219 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2221 r
= sd_id128_get_machine(&this_id
);
2223 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2225 if (sd_id128_equal(arg_uuid
, this_id
)) {
2226 log_full(try ? LOG_WARNING
: LOG_ERR
,
2227 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2233 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2234 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2236 bool ignore
= r
== -EROFS
&& try;
2237 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2238 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2239 return ignore
? 0 : r
;
2243 (void) sd_id128_to_string(arg_uuid
, id
);
2245 p
= strjoina("/var/log/journal/", id
);
2246 q
= prefix_roota(directory
, p
);
2248 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2252 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2253 "%s: already a mount point, refusing to use for journal", p
);
2256 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2260 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2261 "%s: already a mount point, refusing to use for journal", q
);
2264 r
= readlink_and_make_absolute(p
, &d
);
2266 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2269 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2271 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2276 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2277 } else if (r
== -EINVAL
) {
2279 if (arg_link_journal
== LINK_GUEST
&&
2282 if (errno
== ENOTDIR
) {
2283 log_error("%s already exists and is neither a symlink nor a directory", p
);
2286 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2288 } else if (r
!= -ENOENT
)
2289 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2291 if (arg_link_journal
== LINK_GUEST
) {
2293 if (symlink(q
, p
) < 0) {
2295 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2298 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2301 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2303 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2307 if (arg_link_journal
== LINK_HOST
) {
2308 /* don't create parents here — if the host doesn't have
2309 * permanent journal set up, don't force it here */
2311 r
= mkdir_errno_wrapper(p
, 0755);
2312 if (r
< 0 && r
!= -EEXIST
) {
2314 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2317 return log_error_errno(r
, "Failed to create %s: %m", p
);
2320 } else if (access(p
, F_OK
) < 0)
2323 if (dir_is_empty(q
) == 0)
2324 log_warning("%s is not empty, proceeding anyway.", q
);
2326 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2328 return log_error_errno(r
, "Failed to create %s: %m", q
);
2330 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2332 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2337 static int drop_capabilities(uid_t uid
) {
2338 CapabilityQuintet q
;
2340 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2341 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2342 * arg_caps_retain. */
2344 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2345 q
= arg_full_capabilities
;
2347 if (q
.bounding
== (uint64_t) -1)
2348 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2350 if (q
.effective
== (uint64_t) -1)
2351 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2353 if (q
.inheritable
== (uint64_t) -1)
2354 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2356 if (q
.permitted
== (uint64_t) -1)
2357 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2359 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2362 if (capability_quintet_mangle(&q
))
2363 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2366 q
= (CapabilityQuintet
) {
2367 .bounding
= arg_caps_retain
,
2368 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2369 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2370 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2371 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2374 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2375 * in order to maintain the same behavior as systemd < 242. */
2376 if (capability_quintet_mangle(&q
))
2377 log_warning("Some capabilities will not be set because they are not in the current bounding set.");
2381 return capability_quintet_enforce(&q
);
2384 static int reset_audit_loginuid(void) {
2385 _cleanup_free_
char *p
= NULL
;
2388 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2391 r
= read_one_line_file("/proc/self/loginuid", &p
);
2395 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2397 /* Already reset? */
2398 if (streq(p
, "4294967295"))
2401 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2404 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2405 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2406 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2407 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2408 "using systemd-nspawn. Sleeping for 5s... (%m)");
2416 static int setup_propagate(const char *root
) {
2420 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2421 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2422 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2423 (void) mkdir_p(p
, 0600);
2425 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2427 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2429 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2431 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2433 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2435 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2437 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2438 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2442 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2446 /* machined will MS_MOVE into that directory, and that's only
2447 * supported for non-shared mounts. */
2448 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2451 static int setup_machine_id(const char *directory
) {
2452 const char *etc_machine_id
;
2456 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2457 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2458 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2459 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2460 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2461 * container behaves nicely). */
2463 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2465 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2467 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2468 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2470 if (sd_id128_is_null(arg_uuid
)) {
2471 r
= sd_id128_randomize(&arg_uuid
);
2473 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2476 if (sd_id128_is_null(id
))
2477 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2478 "Machine ID in container image is zero, refusing.");
2486 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2491 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2494 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2495 if (r
== -EOPNOTSUPP
)
2496 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.");
2498 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2500 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2502 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2504 log_debug("Patched directory tree to match UID/GID range.");
2511 * < 0 : wait_for_terminate() failed to get the state of the
2512 * container, the container was terminated by a signal, or
2513 * failed for an unknown reason. No change is made to the
2514 * container argument.
2515 * > 0 : The program executed in the container terminated with an
2516 * error. The exit code of the program executed in the
2517 * container is returned. The container argument has been set
2518 * to CONTAINER_TERMINATED.
2519 * 0 : The container is being rebooted, has been shut down or exited
2520 * successfully. The container argument has been set to either
2521 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2523 * That is, success is indicated by a return value of zero, and an
2524 * error is indicated by a non-zero value.
2526 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2530 r
= wait_for_terminate(pid
, &status
);
2532 return log_warning_errno(r
, "Failed to wait for container: %m");
2534 switch (status
.si_code
) {
2537 if (status
.si_status
== 0)
2538 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2540 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2542 *container
= CONTAINER_TERMINATED
;
2543 return status
.si_status
;
2546 if (status
.si_status
== SIGINT
) {
2547 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2548 *container
= CONTAINER_TERMINATED
;
2551 } else if (status
.si_status
== SIGHUP
) {
2552 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2553 *container
= CONTAINER_REBOOTED
;
2559 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2560 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2563 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2564 "Container %s failed due to unknown reason.", arg_machine
);
2568 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2571 pid
= PTR_TO_PID(userdata
);
2573 if (kill(pid
, arg_kill_signal
) >= 0) {
2574 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2575 sd_event_source_set_userdata(s
, NULL
);
2580 sd_event_exit(sd_event_source_get_event(s
), 0);
2584 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2590 pid
= PTR_TO_PID(userdata
);
2595 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2596 return log_error_errno(errno
, "Failed to waitid(): %m");
2597 if (si
.si_pid
== 0) /* No pending children. */
2599 if (si
.si_pid
== pid
) {
2600 /* The main process we care for has exited. Return from
2601 * signal handler but leave the zombie. */
2602 sd_event_exit(sd_event_source_get_event(s
), 0);
2606 /* Reap all other children. */
2607 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2613 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2618 pid
= PTR_TO_PID(userdata
);
2620 if (arg_kill_signal
> 0) {
2621 log_info("Container termination requested. Attempting to halt container.");
2622 (void) kill(pid
, arg_kill_signal
);
2624 log_info("Container termination requested. Exiting.");
2625 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2631 static int determine_names(void) {
2634 if (arg_template
&& !arg_directory
&& arg_machine
) {
2636 /* If --template= was specified then we should not
2637 * search for a machine, but instead create a new one
2638 * in /var/lib/machine. */
2640 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2645 if (!arg_image
&& !arg_directory
) {
2647 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2649 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2651 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2653 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2655 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2656 r
= free_and_strdup(&arg_image
, i
->path
);
2658 r
= free_and_strdup(&arg_directory
, i
->path
);
2663 arg_read_only
= arg_read_only
|| i
->read_only
;
2665 r
= safe_getcwd(&arg_directory
);
2667 return log_error_errno(r
, "Failed to determine current directory: %m");
2670 if (!arg_directory
&& !arg_image
)
2671 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2675 if (arg_directory
&& path_equal(arg_directory
, "/"))
2676 arg_machine
= gethostname_malloc();
2681 arg_machine
= strdup(basename(arg_image
));
2683 /* Truncate suffix if there is one */
2684 e
= endswith(arg_machine
, ".raw");
2688 arg_machine
= strdup(basename(arg_directory
));
2693 hostname_cleanup(arg_machine
);
2694 if (!machine_name_is_valid(arg_machine
))
2695 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2697 if (arg_ephemeral
) {
2700 /* Add a random suffix when this is an
2701 * ephemeral machine, so that we can run many
2702 * instances at once without manually having
2703 * to specify -M each time. */
2705 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2716 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2725 r
= chase_symlinks(*p
, NULL
, flags
, &chased
);
2727 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2729 free_and_replace(*p
, chased
);
2730 return r
; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2733 static int determine_uid_shift(const char *directory
) {
2736 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2741 if (arg_uid_shift
== UID_INVALID
) {
2744 r
= stat(directory
, &st
);
2746 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2748 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2750 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2751 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2752 "UID and GID base of %s don't match.", directory
);
2754 arg_uid_range
= UINT32_C(0x10000);
2757 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2758 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2759 "UID base too high for UID range.");
2764 static unsigned long effective_clone_ns_flags(void) {
2765 unsigned long flags
= arg_clone_ns_flags
;
2767 if (arg_private_network
)
2768 flags
|= CLONE_NEWNET
;
2770 flags
|= CLONE_NEWCGROUP
;
2771 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2772 flags
|= CLONE_NEWUSER
;
2777 static int patch_sysctl(void) {
2779 /* This table is inspired by runc's sysctl() function */
2780 static const struct {
2783 unsigned long clone_flags
;
2785 { "kernel.hostname", false, CLONE_NEWUTS
},
2786 { "kernel.domainname", false, CLONE_NEWUTS
},
2787 { "kernel.msgmax", false, CLONE_NEWIPC
},
2788 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2789 { "kernel.msgmni", false, CLONE_NEWIPC
},
2790 { "kernel.sem", false, CLONE_NEWIPC
},
2791 { "kernel.shmall", false, CLONE_NEWIPC
},
2792 { "kernel.shmmax", false, CLONE_NEWIPC
},
2793 { "kernel.shmmni", false, CLONE_NEWIPC
},
2794 { "fs.mqueue.", true, CLONE_NEWIPC
},
2795 { "net.", true, CLONE_NEWNET
},
2798 unsigned long flags
;
2802 flags
= effective_clone_ns_flags();
2804 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2808 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2810 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2813 if (safe_sysctl
[i
].prefix
)
2814 good
= startswith(*k
, safe_sysctl
[i
].key
);
2816 good
= streq(*k
, safe_sysctl
[i
].key
);
2823 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2825 r
= sysctl_write(*k
, *v
);
2827 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2833 static int inner_child(
2835 const char *directory
,
2839 int master_pty_socket
,
2842 _cleanup_free_
char *home
= NULL
;
2845 const char *envp
[] = {
2846 "PATH=" DEFAULT_PATH_COMPAT
,
2847 NULL
, /* container */
2852 NULL
, /* container_uuid */
2853 NULL
, /* LISTEN_FDS */
2854 NULL
, /* LISTEN_PID */
2855 NULL
, /* NOTIFY_SOCKET */
2858 const char *exec_target
;
2859 _cleanup_strv_free_
char **env_use
= NULL
;
2860 int r
, which_failed
;
2862 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2863 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2864 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2865 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2866 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2867 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2870 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2871 * unshare(). See below. */
2875 assert(kmsg_socket
>= 0);
2877 log_debug("Inner child is initializing.");
2879 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2880 /* Tell the parent, that it now can write the UID map. */
2881 (void) barrier_place(barrier
); /* #1 */
2883 /* Wait until the parent wrote the UID map */
2884 if (!barrier_place_and_sync(barrier
)) /* #2 */
2885 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2886 "Parent died too early");
2889 r
= reset_uid_gid();
2891 return log_error_errno(r
, "Couldn't become new root: %m");
2894 arg_mount_settings
| MOUNT_IN_USERNS
,
2896 arg_selinux_apifs_context
);
2900 if (!arg_network_namespace_path
&& arg_private_network
) {
2901 r
= unshare(CLONE_NEWNET
);
2903 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2905 /* Tell the parent that it can setup network interfaces. */
2906 (void) barrier_place(barrier
); /* #3 */
2909 r
= mount_sysfs(NULL
, arg_mount_settings
);
2913 /* Wait until we are cgroup-ified, so that we
2914 * can mount the right cgroup path writable */
2915 if (!barrier_place_and_sync(barrier
)) /* #4 */
2916 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2917 "Parent died too early");
2920 r
= unshare(CLONE_NEWCGROUP
);
2922 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2925 arg_unified_cgroup_hierarchy
,
2926 arg_userns_mode
!= USER_NAMESPACE_NO
,
2929 arg_selinux_apifs_context
,
2934 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2939 r
= setup_boot_id();
2943 r
= setup_kmsg(kmsg_socket
);
2946 kmsg_socket
= safe_close(kmsg_socket
);
2951 arg_n_custom_mounts
,
2955 arg_selinux_apifs_context
,
2961 return log_error_errno(errno
, "setsid() failed: %m");
2963 if (arg_private_network
)
2966 if (arg_expose_ports
) {
2967 r
= expose_port_send_rtnl(rtnl_socket
);
2970 rtnl_socket
= safe_close(rtnl_socket
);
2973 if (arg_console_mode
!= CONSOLE_PIPE
) {
2974 _cleanup_close_
int master
= -1;
2975 _cleanup_free_
char *console
= NULL
;
2977 /* Allocate a pty and make it available as /dev/console. */
2978 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
2980 return log_error_errno(master
, "Failed to allocate a pty: %m");
2982 r
= setup_dev_console(console
);
2984 return log_error_errno(r
, "Failed to setup /dev/console: %m");
2986 r
= send_one_fd(master_pty_socket
, master
, 0);
2988 return log_error_errno(r
, "Failed to send master fd: %m");
2989 master_pty_socket
= safe_close(master_pty_socket
);
2991 r
= setup_stdio_as_dev_console();
3000 if (arg_oom_score_adjust_set
) {
3001 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3003 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3006 if (arg_cpu_set
.set
)
3007 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3008 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3010 (void) setup_hostname();
3012 if (arg_personality
!= PERSONALITY_INVALID
) {
3013 r
= safe_personality(arg_personality
);
3015 return log_error_errno(r
, "personality() failed: %m");
3016 } else if (secondary
) {
3017 r
= safe_personality(PER_LINUX32
);
3019 return log_error_errno(r
, "personality() failed: %m");
3022 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3024 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3029 if (is_seccomp_available()) {
3031 r
= seccomp_load(arg_seccomp
);
3032 if (ERRNO_IS_SECCOMP_FATAL(r
))
3033 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3035 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3040 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
3046 if (arg_selinux_context
)
3047 if (setexeccon(arg_selinux_context
) < 0)
3048 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3051 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3052 * if we need to later on. */
3053 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3054 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3056 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3057 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
3059 r
= change_uid_gid(arg_user
, &home
);
3063 r
= drop_capabilities(getuid());
3065 return log_error_errno(r
, "Dropping capabilities failed: %m");
3067 if (arg_no_new_privileges
)
3068 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3069 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3071 /* LXC sets container=lxc, so follow the scheme here */
3072 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3074 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3078 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3079 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3082 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3083 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3084 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3087 assert(!sd_id128_is_null(arg_uuid
));
3089 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3092 if (fdset_size(fds
) > 0) {
3093 r
= fdset_cloexec(fds
, false);
3095 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3097 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3098 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3101 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3104 env_use
= strv_env_merge(2, envp
, arg_setenv
);
3108 /* Let the parent know that we are ready and
3109 * wait until the parent is ready with the
3111 if (!barrier_place_and_sync(barrier
)) /* #5 */
3112 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3113 "Parent died too early");
3116 if (chdir(arg_chdir
) < 0)
3117 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3119 if (arg_start_mode
== START_PID2
) {
3120 r
= stub_pid1(arg_uuid
);
3125 log_debug("Inner child completed, invoking payload.");
3127 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3128 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3129 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3131 log_set_open_when_needed(true);
3133 (void) fdset_close_others(fds
);
3135 if (arg_start_mode
== START_BOOT
) {
3139 /* Automatically search for the init system */
3141 m
= strv_length(arg_parameters
);
3142 a
= newa(char*, m
+ 2);
3143 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3146 a
[0] = (char*) "/usr/lib/systemd/systemd";
3147 execve(a
[0], a
, env_use
);
3149 a
[0] = (char*) "/lib/systemd/systemd";
3150 execve(a
[0], a
, env_use
);
3152 a
[0] = (char*) "/sbin/init";
3153 execve(a
[0], a
, env_use
);
3155 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3156 } else if (!strv_isempty(arg_parameters
)) {
3157 const char *dollar_path
;
3159 exec_target
= arg_parameters
[0];
3161 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3163 dollar_path
= strv_env_get(env_use
, "PATH");
3165 if (putenv((char*) dollar_path
) != 0)
3166 return log_error_errno(errno
, "Failed to update $PATH: %m");
3169 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3172 /* If we cannot change the directory, we'll end up in /, that is expected. */
3173 (void) chdir(home
?: "/root");
3175 execle("/bin/bash", "-bash", NULL
, env_use
);
3176 execle("/bin/sh", "-sh", NULL
, env_use
);
3178 exec_target
= "/bin/bash, /bin/sh";
3181 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3184 static int setup_sd_notify_child(void) {
3185 _cleanup_close_
int fd
= -1;
3186 union sockaddr_union sa
= {
3187 .un
.sun_family
= AF_UNIX
,
3188 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3192 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3194 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3196 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3197 (void) sockaddr_un_unlink(&sa
.un
);
3199 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3201 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3203 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3205 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3207 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3209 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3214 static int outer_child(
3216 const char *directory
,
3217 DissectedImage
*dissected_image
,
3224 int uid_shift_socket
,
3225 int master_pty_socket
,
3226 int unified_cgroup_hierarchy_socket
,
3230 _cleanup_close_
int fd
= -1;
3235 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3236 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3237 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3238 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3242 assert(pid_socket
>= 0);
3243 assert(uuid_socket
>= 0);
3244 assert(notify_socket
>= 0);
3245 assert(master_pty_socket
>= 0);
3246 assert(kmsg_socket
>= 0);
3248 log_debug("Outer child is initializing.");
3250 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3251 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3253 r
= reset_audit_loginuid();
3257 /* Mark everything as slave, so that we still
3258 * receive mounts from the real root, but don't
3259 * propagate mounts to the real root. */
3260 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3264 if (dissected_image
) {
3265 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3266 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3267 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3268 * makes sure ESP partitions and userns are compatible. */
3270 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3271 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3272 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0)|
3273 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3278 r
= determine_uid_shift(directory
);
3282 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3283 /* Let the parent know which UID shift we read from the image */
3284 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3286 return log_error_errno(errno
, "Failed to send UID shift: %m");
3287 if (l
!= sizeof(arg_uid_shift
))
3288 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3289 "Short write while sending UID shift.");
3291 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3292 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3293 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3294 * not it will pick a different one, and send it back to us. */
3296 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3298 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3299 if (l
!= sizeof(arg_uid_shift
))
3300 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3301 "Short read while receiving UID shift.");
3304 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3305 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3308 if (path_equal(directory
, "/")) {
3309 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3310 * place, so that we can make changes to its mount structure (for example, to implement
3311 * --volatile=) without this interfering with our ability to access files such as
3312 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3313 * (instead of a temporary directory, since we are living in our own mount namspace here
3314 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3315 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3317 r
= mount_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3321 directory
= "/run/systemd/nspawn-root";
3323 } else if (!dissected_image
) {
3324 /* Turn directory into bind mount (we need that so that we can move the bind mount to root
3326 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3331 r
= setup_pivot_root(
3334 arg_pivot_root_old
);
3338 r
= setup_volatile_mode(
3341 arg_userns_mode
!= USER_NAMESPACE_NO
,
3344 arg_selinux_apifs_context
);
3348 if (dissected_image
) {
3349 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3350 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3351 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
3356 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3357 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3359 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3363 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3365 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3366 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3367 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3368 "Short write while sending cgroup mode.");
3370 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3373 /* Mark everything as shared so our mounts get propagated down. This is
3374 * required to make new bind mounts available in systemd services
3375 * inside the container that create a new mount namespace.
3376 * See https://github.com/systemd/systemd/issues/3860
3377 * Further submounts (such as /dev) done after this will inherit the
3378 * shared propagation mode. */
3379 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3383 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3387 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3391 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
) {
3392 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3394 return log_error_errno(r
, "Failed to make tree read-only: %m");
3397 r
= mount_all(directory
,
3400 arg_selinux_apifs_context
);
3404 r
= copy_devnodes(directory
);
3408 r
= make_extra_nodes(directory
);
3412 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3413 (void) make_inaccessible_nodes(directory
, arg_uid_shift
, arg_uid_shift
);
3415 r
= setup_pts(directory
);
3419 r
= setup_propagate(directory
);
3423 r
= setup_keyring();
3427 r
= setup_timezone(directory
);
3431 r
= setup_resolv_conf(directory
);
3435 r
= setup_machine_id(directory
);
3439 r
= setup_journal(directory
);
3446 arg_n_custom_mounts
,
3447 arg_userns_mode
!= USER_NAMESPACE_NO
,
3450 arg_selinux_apifs_context
,
3455 if (!arg_use_cgns
) {
3458 arg_unified_cgroup_hierarchy
,
3459 arg_userns_mode
!= USER_NAMESPACE_NO
,
3462 arg_selinux_apifs_context
,
3468 r
= mount_move_root(directory
);
3470 return log_error_errno(r
, "Failed to move root directory: %m");
3472 fd
= setup_sd_notify_child();
3476 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3477 arg_clone_ns_flags
|
3478 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3480 return log_error_errno(errno
, "Failed to fork inner child: %m");
3482 pid_socket
= safe_close(pid_socket
);
3483 uuid_socket
= safe_close(uuid_socket
);
3484 notify_socket
= safe_close(notify_socket
);
3485 uid_shift_socket
= safe_close(uid_shift_socket
);
3487 /* The inner child has all namespaces that are
3488 * requested, so that we all are owned by the user if
3489 * user namespaces are turned on. */
3491 if (arg_network_namespace_path
) {
3492 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3494 return log_error_errno(r
, "Failed to join network namespace: %m");
3497 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
);
3499 _exit(EXIT_FAILURE
);
3501 _exit(EXIT_SUCCESS
);
3504 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3506 return log_error_errno(errno
, "Failed to send PID: %m");
3507 if (l
!= sizeof(pid
))
3508 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3509 "Short write while sending PID.");
3511 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3513 return log_error_errno(errno
, "Failed to send machine ID: %m");
3514 if (l
!= sizeof(arg_uuid
))
3515 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3516 "Short write while sending machine ID.");
3518 l
= send_one_fd(notify_socket
, fd
, 0);
3520 return log_error_errno(l
, "Failed to send notify fd: %m");
3522 pid_socket
= safe_close(pid_socket
);
3523 uuid_socket
= safe_close(uuid_socket
);
3524 notify_socket
= safe_close(notify_socket
);
3525 master_pty_socket
= safe_close(master_pty_socket
);
3526 kmsg_socket
= safe_close(kmsg_socket
);
3527 rtnl_socket
= safe_close(rtnl_socket
);
3528 netns_fd
= safe_close(netns_fd
);
3533 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3534 bool tried_hashed
= false;
3535 unsigned n_tries
= 100;
3540 assert(ret_lock_file
);
3541 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3542 assert(arg_uid_range
== 0x10000U
);
3546 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3549 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3550 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3555 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3557 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3560 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3561 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3562 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3567 /* Make some superficial checks whether the range is currently known in the user database */
3568 if (getpwuid(candidate
))
3570 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3572 if (getgrgid(candidate
))
3574 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3577 *ret_lock_file
= lf
;
3578 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3583 if (arg_machine
&& !tried_hashed
) {
3584 /* Try to hash the base from the container name */
3586 static const uint8_t hash_key
[] = {
3587 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3588 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3591 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3593 tried_hashed
= true;
3595 random_bytes(&candidate
, sizeof(candidate
));
3597 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3598 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3602 static int setup_uid_map(pid_t pid
) {
3603 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3608 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3609 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3610 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3612 return log_error_errno(r
, "Failed to write UID map: %m");
3614 /* We always assign the same UID and GID ranges */
3615 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3616 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3618 return log_error_errno(r
, "Failed to write GID map: %m");
3623 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3624 char buf
[NOTIFY_BUFFER_MAX
+1];
3626 struct iovec iovec
= {
3628 .iov_len
= sizeof(buf
)-1,
3631 struct cmsghdr cmsghdr
;
3632 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
3633 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
3635 struct msghdr msghdr
= {
3638 .msg_control
= &control
,
3639 .msg_controllen
= sizeof(control
),
3641 struct cmsghdr
*cmsg
;
3642 struct ucred
*ucred
= NULL
;
3644 pid_t inner_child_pid
;
3645 _cleanup_strv_free_
char **tags
= NULL
;
3649 inner_child_pid
= PTR_TO_PID(userdata
);
3651 if (revents
!= EPOLLIN
) {
3652 log_warning("Got unexpected poll event for notify fd.");
3656 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3658 if (IN_SET(errno
, EAGAIN
, EINTR
))
3661 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
3663 cmsg_close_all(&msghdr
);
3665 CMSG_FOREACH(cmsg
, &msghdr
) {
3666 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
3667 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
3668 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
3670 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
3674 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3675 log_debug("Received notify message without valid credentials. Ignoring.");
3679 if ((size_t) n
>= sizeof(buf
)) {
3680 log_warning("Received notify message exceeded maximum size. Ignoring.");
3685 tags
= strv_split(buf
, "\n\r");
3689 if (strv_find(tags
, "READY=1"))
3690 (void) sd_notifyf(false, "READY=1\n");
3692 p
= strv_find_startswith(tags
, "STATUS=");
3694 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3699 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3702 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3704 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3706 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3711 static int merge_settings(Settings
*settings
, const char *path
) {
3717 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3718 * that this steals the fields of the Settings* structure, and hence modifies it. */
3720 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3721 settings
->start_mode
>= 0) {
3722 arg_start_mode
= settings
->start_mode
;
3723 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3726 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3727 arg_ephemeral
= settings
->ephemeral
;
3729 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3732 if (!arg_settings_trusted
)
3733 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3735 free_and_replace(arg_directory
, settings
->root
);
3738 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3739 settings
->pivot_root_new
) {
3740 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3741 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3744 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3745 settings
->working_directory
)
3746 free_and_replace(arg_chdir
, settings
->working_directory
);
3748 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3749 settings
->environment
)
3750 strv_free_and_replace(arg_setenv
, settings
->environment
);
3752 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3755 free_and_replace(arg_user
, settings
->user
);
3757 if (uid_is_valid(settings
->uid
))
3758 arg_uid
= settings
->uid
;
3759 if (gid_is_valid(settings
->gid
))
3760 arg_gid
= settings
->gid
;
3761 if (settings
->n_supplementary_gids
> 0) {
3762 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3763 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3767 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3768 uint64_t plus
, minus
;
3770 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3771 * Settings structure */
3773 plus
= settings
->capability
;
3774 minus
= settings
->drop_capability
;
3776 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3777 if (settings_private_network(settings
))
3778 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3780 minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3783 if (!arg_settings_trusted
&& plus
!= 0) {
3784 if (settings
->capability
!= 0)
3785 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3787 arg_caps_retain
|= plus
;
3789 arg_caps_retain
&= ~minus
;
3791 /* Copy the full capabilities over too */
3792 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3793 if (!arg_settings_trusted
)
3794 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
3796 arg_full_capabilities
= settings
->full_capabilities
;
3800 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3801 settings
->kill_signal
> 0)
3802 arg_kill_signal
= settings
->kill_signal
;
3804 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3805 settings
->personality
!= PERSONALITY_INVALID
)
3806 arg_personality
= settings
->personality
;
3808 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3809 !sd_id128_is_null(settings
->machine_id
)) {
3811 if (!arg_settings_trusted
)
3812 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3814 arg_uuid
= settings
->machine_id
;
3817 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3818 settings
->read_only
>= 0)
3819 arg_read_only
= settings
->read_only
;
3821 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3822 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3823 arg_volatile_mode
= settings
->volatile_mode
;
3825 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3826 settings
->n_custom_mounts
> 0) {
3828 if (!arg_settings_trusted
)
3829 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3831 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3832 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3833 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3834 settings
->n_custom_mounts
= 0;
3838 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3839 (settings
->private_network
>= 0 ||
3840 settings
->network_veth
>= 0 ||
3841 settings
->network_bridge
||
3842 settings
->network_zone
||
3843 settings
->network_interfaces
||
3844 settings
->network_macvlan
||
3845 settings
->network_ipvlan
||
3846 settings
->network_veth_extra
||
3847 settings
->network_namespace_path
)) {
3849 if (!arg_settings_trusted
)
3850 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3852 arg_network_veth
= settings_network_veth(settings
);
3853 arg_private_network
= settings_private_network(settings
);
3855 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3856 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3857 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3858 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3860 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3861 free_and_replace(arg_network_zone
, settings
->network_zone
);
3863 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3867 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3868 settings
->expose_ports
) {
3870 if (!arg_settings_trusted
)
3871 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3873 expose_port_free_all(arg_expose_ports
);
3874 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3878 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3879 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3881 if (!arg_settings_trusted
)
3882 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3884 arg_userns_mode
= settings
->userns_mode
;
3885 arg_uid_shift
= settings
->uid_shift
;
3886 arg_uid_range
= settings
->uid_range
;
3887 arg_userns_chown
= settings
->userns_chown
;
3891 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3892 arg_notify_ready
= settings
->notify_ready
;
3894 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3896 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_whitelist
))
3897 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3899 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3900 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3904 if (!arg_settings_trusted
&& settings
->seccomp
)
3905 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
3907 seccomp_release(arg_seccomp
);
3908 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
3913 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3914 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3917 if (!settings
->rlimit
[rl
])
3920 if (!arg_settings_trusted
) {
3921 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3925 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3928 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3930 free_and_replace(arg_hostname
, settings
->hostname
);
3932 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3933 settings
->no_new_privileges
>= 0)
3934 arg_no_new_privileges
= settings
->no_new_privileges
;
3936 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3937 settings
->oom_score_adjust_set
) {
3939 if (!arg_settings_trusted
)
3940 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
3942 arg_oom_score_adjust
= settings
->oom_score_adjust
;
3943 arg_oom_score_adjust_set
= true;
3947 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
3948 settings
->cpu_set
.set
) {
3950 if (!arg_settings_trusted
)
3951 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
3953 cpu_set_reset(&arg_cpu_set
);
3954 arg_cpu_set
= settings
->cpu_set
;
3955 settings
->cpu_set
= (CPUSet
) {};
3959 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
3960 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
3961 arg_resolv_conf
= settings
->resolv_conf
;
3963 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
3964 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
3966 if (!arg_settings_trusted
)
3967 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
3969 arg_link_journal
= settings
->link_journal
;
3970 arg_link_journal_try
= settings
->link_journal_try
;
3974 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
3975 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
3976 arg_timezone
= settings
->timezone
;
3978 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
3981 if (!arg_settings_trusted
)
3982 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
3984 free_and_replace(arg_slice
, settings
->slice
);
3987 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
3988 settings
->use_cgns
>= 0) {
3990 if (!arg_settings_trusted
)
3991 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
3993 arg_use_cgns
= settings
->use_cgns
;
3996 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
3997 settings
->clone_ns_flags
!= (unsigned long) -1) {
3999 if (!arg_settings_trusted
)
4000 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4002 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4005 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4006 settings
->console_mode
>= 0) {
4008 if (!arg_settings_trusted
)
4009 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4011 arg_console_mode
= settings
->console_mode
;
4014 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4015 * don't consult arg_settings_mask for them. */
4017 sd_bus_message_unref(arg_property_message
);
4018 arg_property_message
= TAKE_PTR(settings
->properties
);
4020 arg_console_width
= settings
->console_width
;
4021 arg_console_height
= settings
->console_height
;
4023 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4024 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4025 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4030 static int load_settings(void) {
4031 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4032 _cleanup_fclose_
FILE *f
= NULL
;
4033 _cleanup_free_
char *p
= NULL
;
4040 /* If all settings are masked, there's no point in looking for
4041 * the settings file */
4042 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
4045 fn
= strjoina(arg_machine
, ".nspawn");
4047 /* We first look in the admin's directories in /etc and /run */
4048 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4049 _cleanup_free_
char *j
= NULL
;
4051 j
= path_join(i
, fn
);
4059 /* By default, we trust configuration from /etc and /run */
4060 if (arg_settings_trusted
< 0)
4061 arg_settings_trusted
= true;
4066 if (errno
!= ENOENT
)
4067 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4071 /* After that, let's look for a file next to the
4072 * actual image we shall boot. */
4075 p
= file_in_same_dir(arg_image
, fn
);
4078 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4079 p
= file_in_same_dir(arg_directory
, fn
);
4086 if (!f
&& errno
!= ENOENT
)
4087 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4089 /* By default, we do not trust configuration from /var/lib/machines */
4090 if (arg_settings_trusted
< 0)
4091 arg_settings_trusted
= false;
4098 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4100 r
= settings_load(f
, p
, &settings
);
4104 return merge_settings(settings
, p
);
4107 static int load_oci_bundle(void) {
4108 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4111 if (!arg_oci_bundle
)
4114 /* By default let's trust OCI bundles */
4115 if (arg_settings_trusted
< 0)
4116 arg_settings_trusted
= true;
4118 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4122 return merge_settings(settings
, arg_oci_bundle
);
4125 static int run_container(
4126 DissectedImage
*dissected_image
,
4129 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4130 union in_addr_union
*exposed
,
4131 int *master
, pid_t
*pid
, int *ret
) {
4133 static const struct sigaction sa
= {
4134 .sa_handler
= nop_signal_handler
,
4135 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4138 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4139 _cleanup_close_
int etc_passwd_lock
= -1;
4140 _cleanup_close_pair_
int
4141 kmsg_socket_pair
[2] = { -1, -1 },
4142 rtnl_socket_pair
[2] = { -1, -1 },
4143 pid_socket_pair
[2] = { -1, -1 },
4144 uuid_socket_pair
[2] = { -1, -1 },
4145 notify_socket_pair
[2] = { -1, -1 },
4146 uid_shift_socket_pair
[2] = { -1, -1 },
4147 master_pty_socket_pair
[2] = { -1, -1 },
4148 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4150 _cleanup_close_
int notify_socket
= -1;
4151 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4152 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4153 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4154 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4155 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4156 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4157 ContainerStatus container_status
= 0;
4161 _cleanup_close_
int netns_fd
= -1;
4163 assert_se(sigemptyset(&mask_chld
) == 0);
4164 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4166 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4167 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4168 * check with getpwuid() if the specific user already exists. Note that /etc might be
4169 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4170 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4171 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4174 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4175 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4176 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4179 r
= barrier_create(&barrier
);
4181 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4183 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4184 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4186 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4187 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4189 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4190 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4192 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4193 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4195 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4196 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4198 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4199 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4201 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4202 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4203 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4205 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4206 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4207 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4209 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4210 * parent's blocking calls and give it a chance to call wait() and terminate. */
4211 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4213 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4215 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4217 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4219 if (arg_network_namespace_path
) {
4220 netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4222 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4224 r
= fd_is_network_ns(netns_fd
);
4226 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
);
4228 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4230 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4231 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4234 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4236 return log_error_errno(errno
, "clone() failed%s: %m",
4238 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4241 /* The outer child only has a file system namespace. */
4242 barrier_set_role(&barrier
, BARRIER_CHILD
);
4244 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4245 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4246 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4247 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4248 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4249 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4250 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4251 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4253 (void) reset_all_signal_handlers();
4254 (void) reset_signal_mask();
4256 r
= outer_child(&barrier
,
4261 uuid_socket_pair
[1],
4262 notify_socket_pair
[1],
4263 kmsg_socket_pair
[1],
4264 rtnl_socket_pair
[1],
4265 uid_shift_socket_pair
[1],
4266 master_pty_socket_pair
[1],
4267 unified_cgroup_hierarchy_socket_pair
[1],
4271 _exit(EXIT_FAILURE
);
4273 _exit(EXIT_SUCCESS
);
4276 barrier_set_role(&barrier
, BARRIER_PARENT
);
4280 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4281 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4282 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4283 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4284 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4285 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4286 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4287 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4289 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4290 /* The child just let us know the UID shift it might have read from the image. */
4291 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4293 return log_error_errno(errno
, "Failed to read UID shift: %m");
4294 if (l
!= sizeof arg_uid_shift
)
4295 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4297 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4298 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4299 * image, but if that's already in use, pick a new one, and report back to the child,
4300 * which one we now picked. */
4302 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4304 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4306 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4308 return log_error_errno(errno
, "Failed to send UID shift: %m");
4309 if (l
!= sizeof arg_uid_shift
)
4310 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4314 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4315 /* The child let us know the support cgroup mode it might have read from the image. */
4316 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4318 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4319 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4320 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4321 l
, l
== 0 ? " The child is most likely dead." : "");
4324 /* Wait for the outer child. */
4325 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4328 if (r
!= EXIT_SUCCESS
)
4331 /* And now retrieve the PID of the inner child. */
4332 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4334 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4335 if (l
!= sizeof *pid
)
4336 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4338 /* We also retrieve container UUID in case it was generated by outer child */
4339 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4341 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4342 if (l
!= sizeof(arg_uuid
))
4343 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4345 /* We also retrieve the socket used for notifications generated by outer child */
4346 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4347 if (notify_socket
< 0)
4348 return log_error_errno(notify_socket
,
4349 "Failed to receive notification socket from the outer child: %m");
4351 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4353 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4354 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4355 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4357 r
= setup_uid_map(*pid
);
4361 (void) barrier_place(&barrier
); /* #2 */
4364 if (arg_private_network
) {
4365 if (!arg_network_namespace_path
) {
4366 /* Wait until the child has unshared its network namespace. */
4367 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4368 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4371 r
= move_network_interfaces(*pid
, arg_network_interfaces
);
4375 if (arg_network_veth
) {
4376 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4377 arg_network_bridge
|| arg_network_zone
);
4383 if (arg_network_bridge
) {
4384 /* Add the interface to a bridge */
4385 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4390 } else if (arg_network_zone
) {
4391 /* Add the interface to a bridge, possibly creating it */
4392 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4400 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4404 /* We created the primary and extra veth links now; let's remember this, so that we know to
4405 remove them later on. Note that we don't bother with removing veth links that were created
4406 here when their setup failed half-way, because in that case the kernel should be able to
4407 remove them on its own, since they cannot be referenced by anything yet. */
4408 *veth_created
= true;
4410 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4414 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4419 if (arg_register
|| !arg_keep_unit
) {
4420 r
= sd_bus_default_system(&bus
);
4422 return log_error_errno(r
, "Failed to open system bus: %m");
4424 r
= sd_bus_set_close_on_exit(bus
, false);
4426 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4429 if (!arg_keep_unit
) {
4430 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4431 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4432 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4434 r
= sd_bus_match_signal_async(
4437 "org.freedesktop.systemd1",
4439 "org.freedesktop.systemd1.Scope",
4441 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4443 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4447 r
= register_machine(
4455 arg_custom_mounts
, arg_n_custom_mounts
,
4458 arg_property_message
,
4460 arg_container_service_name
);
4464 } else if (!arg_keep_unit
) {
4470 arg_custom_mounts
, arg_n_custom_mounts
,
4473 arg_property_message
);
4477 } else if (arg_slice
|| arg_property
)
4478 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4480 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4484 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4488 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4492 /* Notify the child that the parent is ready with all
4493 * its setup (including cgroup-ification), and that
4494 * the child can now hand over control to the code to
4495 * run inside the container. */
4496 (void) barrier_place(&barrier
); /* #4 */
4498 /* Block SIGCHLD here, before notifying child.
4499 * process_pty() will handle it with the other signals. */
4500 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4502 /* Reset signal to default */
4503 r
= default_signals(SIGCHLD
, -1);
4505 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4507 r
= sd_event_new(&event
);
4509 return log_error_errno(r
, "Failed to get default event source: %m");
4511 (void) sd_event_set_watchdog(event
, true);
4514 r
= sd_bus_attach_event(bus
, event
, 0);
4516 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4519 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4523 /* Let the child know that we are ready and wait that the child is completely ready now. */
4524 if (!barrier_place_and_sync(&barrier
)) /* #5 */
4525 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4527 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4528 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4529 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4531 (void) sd_notifyf(false,
4532 "STATUS=Container running.\n"
4533 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4534 if (!arg_notify_ready
)
4535 (void) sd_notify(false, "READY=1\n");
4537 if (arg_kill_signal
> 0) {
4538 /* Try to kill the init system on SIGINT or SIGTERM */
4539 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4540 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4542 /* Immediately exit */
4543 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4544 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4547 /* Exit when the child exits */
4548 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4550 if (arg_expose_ports
) {
4551 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4555 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4558 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4560 if (arg_console_mode
!= CONSOLE_PIPE
) {
4561 _cleanup_close_
int fd
= -1;
4562 PTYForwardFlags flags
= 0;
4564 /* Retrieve the master pty allocated by inner child */
4565 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
4567 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
4569 switch (arg_console_mode
) {
4571 case CONSOLE_READ_ONLY
:
4572 flags
|= PTY_FORWARD_READ_ONLY
;
4576 case CONSOLE_INTERACTIVE
:
4577 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
4579 r
= pty_forward_new(event
, fd
, flags
, &forward
);
4581 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4583 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4584 (void) pty_forward_set_width_height(forward
,
4586 arg_console_height
);
4590 assert(arg_console_mode
== CONSOLE_PASSIVE
);
4593 *master
= TAKE_FD(fd
);
4596 r
= sd_event_loop(event
);
4598 return log_error_errno(r
, "Failed to run event loop: %m");
4603 (void) pty_forward_get_last_char(forward
, &last_char
);
4604 forward
= pty_forward_free(forward
);
4606 if (!arg_quiet
&& last_char
!= '\n')
4610 /* Kill if it is not dead yet anyway */
4613 terminate_machine(bus
, arg_machine
);
4614 else if (!arg_keep_unit
)
4615 terminate_scope(bus
, arg_machine
);
4618 /* Normally redundant, but better safe than sorry */
4619 (void) kill(*pid
, SIGKILL
);
4621 r
= wait_for_container(*pid
, &container_status
);
4625 /* We failed to wait for the container, or the container exited abnormally. */
4627 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4628 /* r > 0 → The container exited with a non-zero status.
4629 * As a special case, we need to replace 133 with a different value,
4630 * because 133 is special-cased in the service file to reboot the container.
4631 * otherwise → The container exited with zero status and a reboot was not requested.
4633 if (r
== EXIT_FORCE_RESTART
)
4634 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4636 return 0; /* finito */
4639 /* CONTAINER_REBOOTED, loop again */
4641 if (arg_keep_unit
) {
4642 /* Special handling if we are running as a service: instead of simply
4643 * restarting the machine we want to restart the entire service, so let's
4644 * inform systemd about this with the special exit code 133. The service
4645 * file uses RestartForceExitStatus=133 so that this results in a full
4646 * nspawn restart. This is necessary since we might have cgroup parameters
4647 * set we want to have flushed out. */
4648 *ret
= EXIT_FORCE_RESTART
;
4649 return 0; /* finito */
4652 expose_port_flush(arg_expose_ports
, exposed
);
4654 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4655 *veth_created
= false;
4656 return 1; /* loop again */
4659 static int initialize_rlimits(void) {
4660 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4661 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4662 * container execution environments. */
4664 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4665 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4666 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4667 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4668 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4669 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4670 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4671 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4672 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4673 [RLIMIT_NICE
] = { 0, 0 },
4674 [RLIMIT_NOFILE
] = { 1024, 4096 },
4675 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4676 [RLIMIT_RTPRIO
] = { 0, 0 },
4677 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4678 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4680 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4681 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4682 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4683 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4684 * that PID 1 changes a number of other resource limits during early initialization which is why we
4685 * don't read the other limits from PID 1 but prefer the static table above. */
4690 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4691 /* Let's only fill in what the user hasn't explicitly configured anyway */
4692 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4693 const struct rlimit
*v
;
4694 struct rlimit buffer
;
4696 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4697 /* For these two let's read the limits off PID 1. See above for an explanation. */
4699 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4700 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4704 v
= kernel_defaults
+ rl
;
4706 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4707 if (!arg_rlimit
[rl
])
4711 if (DEBUG_LOGGING
) {
4712 _cleanup_free_
char *k
= NULL
;
4714 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4715 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4722 static int run(int argc
, char *argv
[]) {
4723 bool secondary
= false, remove_directory
= false, remove_image
= false,
4724 veth_created
= false, remove_tmprootdir
= false;
4725 _cleanup_close_
int master
= -1;
4726 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4727 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4728 char veth_name
[IFNAMSIZ
] = "";
4729 union in_addr_union exposed
= {};
4730 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4731 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4732 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4733 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4734 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4737 log_parse_environment();
4740 r
= parse_argv(argc
, argv
);
4748 r
= initialize_rlimits();
4752 r
= load_oci_bundle();
4756 r
= determine_names();
4760 r
= load_settings();
4766 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4770 r
= verify_arguments();
4774 /* Reapply environment settings. */
4775 (void) detect_unified_cgroup_hierarchy_from_environment();
4777 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4778 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4779 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4780 (void) ignore_signals(SIGPIPE
, -1);
4782 n_fd_passed
= sd_listen_fds(false);
4783 if (n_fd_passed
> 0) {
4784 r
= fdset_new_listen_fds(&fds
, false);
4786 log_error_errno(r
, "Failed to collect file descriptors: %m");
4791 /* The "default" umask. This is appropriate for most file and directory
4792 * operations performed by nspawn, and is the umask that will be used for
4793 * the child. Functions like copy_devnodes() change the umask temporarily. */
4796 if (arg_directory
) {
4799 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4800 * /var from the host will propagate into container dynamically (because bad things happen if
4801 * two systems write to the same /var). Let's allow it for the special cases where /var is
4802 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4803 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
4804 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4809 if (arg_ephemeral
) {
4810 _cleanup_free_
char *np
= NULL
;
4812 r
= chase_symlinks_and_update(&arg_directory
, 0);
4816 /* If the specified path is a mount point we generate the new snapshot immediately
4817 * inside it under a random name. However if the specified is not a mount point we
4818 * create the new snapshot in the parent directory, just next to it. */
4819 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4821 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4825 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4827 r
= tempfn_random(arg_directory
, "machine.", &np
);
4829 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4833 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4834 * only owned by us and noone else. */
4835 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4837 log_error_errno(r
, "Failed to lock %s: %m", np
);
4842 BLOCK_SIGNALS(SIGINT
);
4843 r
= btrfs_subvol_snapshot(arg_directory
, np
,
4844 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4845 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4846 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4847 BTRFS_SNAPSHOT_RECURSIVE
|
4848 BTRFS_SNAPSHOT_QUOTA
|
4849 BTRFS_SNAPSHOT_SIGINT
);
4852 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
4856 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
4860 free_and_replace(arg_directory
, np
);
4861 remove_directory
= true;
4863 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
4867 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4869 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
4873 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
4878 r
= chase_symlinks_and_update(&arg_template
, 0);
4883 BLOCK_SIGNALS(SIGINT
);
4884 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
4885 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4886 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4887 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4888 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
4889 BTRFS_SNAPSHOT_RECURSIVE
|
4890 BTRFS_SNAPSHOT_QUOTA
|
4891 BTRFS_SNAPSHOT_SIGINT
);
4894 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4895 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
4896 else if (r
== -EINTR
) {
4897 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
4900 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
4903 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4904 "Populated %s from template %s.", arg_directory
, arg_template
);
4908 if (arg_start_mode
== START_BOOT
) {
4911 if (arg_pivot_root_new
)
4912 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4916 if (path_is_os_tree(p
) <= 0) {
4917 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
4924 if (arg_pivot_root_new
)
4925 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4929 q
= strjoina(p
, "/usr/");
4931 if (laccess(q
, F_OK
) < 0) {
4932 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
4940 assert(!arg_template
);
4942 r
= chase_symlinks_and_update(&arg_image
, 0);
4946 if (arg_ephemeral
) {
4947 _cleanup_free_
char *np
= NULL
;
4949 r
= tempfn_random(arg_image
, "machine.", &np
);
4951 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
4955 /* Always take an exclusive lock on our own ephemeral copy. */
4956 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4958 r
= log_error_errno(r
, "Failed to create image lock: %m");
4963 BLOCK_SIGNALS(SIGINT
);
4964 r
= copy_file(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600, FS_NOCOW_FL
, FS_NOCOW_FL
, COPY_REFLINK
|COPY_CRTIME
|COPY_SIGINT
);
4967 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
4971 r
= log_error_errno(r
, "Failed to copy image file: %m");
4975 free_and_replace(arg_image
, np
);
4976 remove_image
= true;
4978 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4980 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
4984 r
= log_error_errno(r
, "Failed to create image lock: %m");
4988 if (!arg_root_hash
) {
4989 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
4991 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
4997 if (!mkdtemp(tmprootdir
)) {
4998 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5002 remove_tmprootdir
= true;
5004 arg_directory
= strdup(tmprootdir
);
5005 if (!arg_directory
) {
5010 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, &loop
);
5012 log_error_errno(r
, "Failed to set up loopback block device: %m");
5016 r
= dissect_image_and_warn(
5019 arg_root_hash
, arg_root_hash_size
,
5020 DISSECT_IMAGE_REQUIRE_ROOT
,
5023 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5024 log_notice("Note that the disk image needs to\n"
5025 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5026 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5027 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
5028 " d) or contain a file system without a partition table\n"
5029 "in order to be bootable with systemd-nspawn.");
5035 if (!arg_root_hash
&& dissected_image
->can_verity
)
5036 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5038 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
5042 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5043 if (remove_image
&& unlink(arg_image
) >= 0)
5044 remove_image
= false;
5047 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5051 if (arg_console_mode
< 0)
5053 isatty(STDIN_FILENO
) > 0 &&
5054 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5056 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5060 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5061 arg_machine
, arg_image
?: arg_directory
);
5063 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5065 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5066 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5071 r
= run_container(dissected_image
,
5074 veth_name
, &veth_created
,
5082 (void) sd_notify(false,
5083 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5084 "STOPPING=1\nSTATUS=Terminating...");
5087 (void) kill(pid
, SIGKILL
);
5089 /* Try to flush whatever is still queued in the pty */
5091 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5092 master
= safe_close(master
);
5096 (void) wait_for_terminate(pid
, NULL
);
5100 if (remove_directory
&& arg_directory
) {
5103 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5105 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5108 if (remove_image
&& arg_image
) {
5109 if (unlink(arg_image
) < 0)
5110 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5113 if (remove_tmprootdir
) {
5114 if (rmdir(tmprootdir
) < 0)
5115 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5121 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5122 (void) rm_rf(p
, REMOVE_ROOT
);
5125 expose_port_flush(arg_expose_ports
, &exposed
);
5128 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5129 (void) remove_bridge(arg_network_zone
);
5131 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5132 expose_port_free_all(arg_expose_ports
);
5133 rlimit_free_all(arg_rlimit
);
5134 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5142 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);