1 /* SPDX-License-Identifier: LGPL-2.1+ */
8 #include <linux/loop.h>
10 #include <selinux/selinux.h>
14 #include <sys/personality.h>
15 #include <sys/prctl.h>
16 #include <sys/types.h>
21 #include "sd-daemon.h"
24 #include "alloc-util.h"
26 #include "base-filesystem.h"
27 #include "blkid-util.h"
28 #include "btrfs-util.h"
29 #include "bus-error.h"
32 #include "capability-util.h"
33 #include "cgroup-util.h"
35 #include "cpu-set-util.h"
36 #include "dev-setup.h"
37 #include "dissect-image.h"
42 #include "format-util.h"
45 #include "hexdecoct.h"
46 #include "hostname-util.h"
47 #include "id128-util.h"
49 #include "loop-util.h"
50 #include "loopback-setup.h"
51 #include "machine-image.h"
53 #include "main-func.h"
54 #include "missing_sched.h"
56 #include "mount-util.h"
57 #include "mountpoint-util.h"
58 #include "namespace-util.h"
59 #include "netlink-util.h"
60 #include "nspawn-cgroup.h"
61 #include "nspawn-def.h"
62 #include "nspawn-expose-ports.h"
63 #include "nspawn-mount.h"
64 #include "nspawn-network.h"
65 #include "nspawn-oci.h"
66 #include "nspawn-patch-uid.h"
67 #include "nspawn-register.h"
68 #include "nspawn-seccomp.h"
69 #include "nspawn-settings.h"
70 #include "nspawn-setuid.h"
71 #include "nspawn-stub-pid1.h"
72 #include "nulstr-util.h"
75 #include "parse-util.h"
76 #include "path-util.h"
77 #include "pretty-print.h"
78 #include "process-util.h"
80 #include "random-util.h"
81 #include "raw-clone.h"
82 #include "resolve-util.h"
83 #include "rlimit-util.h"
86 #include "seccomp-util.h"
88 #include "selinux-util.h"
89 #include "signal-util.h"
90 #include "socket-util.h"
91 #include "stat-util.h"
92 #include "stdio-util.h"
93 #include "string-table.h"
94 #include "string-util.h"
96 #include "sysctl-util.h"
97 #include "terminal-util.h"
98 #include "tmpfile-util.h"
99 #include "umask-util.h"
100 #include "unit-name.h"
101 #include "user-util.h"
104 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
105 * nspawn_notify_socket_path is relative to the container
106 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
107 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
109 #define EXIT_FORCE_RESTART 133
111 typedef enum ContainerStatus
{
112 CONTAINER_TERMINATED
,
116 static char *arg_directory
= NULL
;
117 static char *arg_template
= NULL
;
118 static char *arg_chdir
= NULL
;
119 static char *arg_pivot_root_new
= NULL
;
120 static char *arg_pivot_root_old
= NULL
;
121 static char *arg_user
= NULL
;
122 static uid_t arg_uid
= UID_INVALID
;
123 static gid_t arg_gid
= GID_INVALID
;
124 static gid_t
* arg_supplementary_gids
= NULL
;
125 static size_t arg_n_supplementary_gids
= 0;
126 static sd_id128_t arg_uuid
= {};
127 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
128 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
129 static const char *arg_selinux_context
= NULL
;
130 static const char *arg_selinux_apifs_context
= NULL
;
131 static char *arg_slice
= NULL
;
132 static bool arg_private_network
= false;
133 static bool arg_read_only
= false;
134 static StartMode arg_start_mode
= START_PID1
;
135 static bool arg_ephemeral
= false;
136 static LinkJournal arg_link_journal
= LINK_AUTO
;
137 static bool arg_link_journal_try
= false;
138 static uint64_t arg_caps_retain
=
139 (1ULL << CAP_AUDIT_CONTROL
) |
140 (1ULL << CAP_AUDIT_WRITE
) |
141 (1ULL << CAP_CHOWN
) |
142 (1ULL << CAP_DAC_OVERRIDE
) |
143 (1ULL << CAP_DAC_READ_SEARCH
) |
144 (1ULL << CAP_FOWNER
) |
145 (1ULL << CAP_FSETID
) |
146 (1ULL << CAP_IPC_OWNER
) |
148 (1ULL << CAP_LEASE
) |
149 (1ULL << CAP_LINUX_IMMUTABLE
) |
150 (1ULL << CAP_MKNOD
) |
151 (1ULL << CAP_NET_BIND_SERVICE
) |
152 (1ULL << CAP_NET_BROADCAST
) |
153 (1ULL << CAP_NET_RAW
) |
154 (1ULL << CAP_SETFCAP
) |
155 (1ULL << CAP_SETGID
) |
156 (1ULL << CAP_SETPCAP
) |
157 (1ULL << CAP_SETUID
) |
158 (1ULL << CAP_SYS_ADMIN
) |
159 (1ULL << CAP_SYS_BOOT
) |
160 (1ULL << CAP_SYS_CHROOT
) |
161 (1ULL << CAP_SYS_NICE
) |
162 (1ULL << CAP_SYS_PTRACE
) |
163 (1ULL << CAP_SYS_RESOURCE
) |
164 (1ULL << CAP_SYS_TTY_CONFIG
);
165 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
166 static CustomMount
*arg_custom_mounts
= NULL
;
167 static size_t arg_n_custom_mounts
= 0;
168 static char **arg_setenv
= NULL
;
169 static bool arg_quiet
= false;
170 static bool arg_register
= true;
171 static bool arg_keep_unit
= false;
172 static char **arg_network_interfaces
= NULL
;
173 static char **arg_network_macvlan
= NULL
;
174 static char **arg_network_ipvlan
= NULL
;
175 static bool arg_network_veth
= false;
176 static char **arg_network_veth_extra
= NULL
;
177 static char *arg_network_bridge
= NULL
;
178 static char *arg_network_zone
= NULL
;
179 static char *arg_network_namespace_path
= NULL
;
180 static PagerFlags arg_pager_flags
= 0;
181 static unsigned long arg_personality
= PERSONALITY_INVALID
;
182 static char *arg_image
= NULL
;
183 static char *arg_oci_bundle
= NULL
;
184 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
185 static ExposePort
*arg_expose_ports
= NULL
;
186 static char **arg_property
= NULL
;
187 static sd_bus_message
*arg_property_message
= NULL
;
188 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
189 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
190 static bool arg_userns_chown
= false;
191 static int arg_kill_signal
= 0;
192 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
193 static SettingsMask arg_settings_mask
= 0;
194 static int arg_settings_trusted
= -1;
195 static char **arg_parameters
= NULL
;
196 static const char *arg_container_service_name
= "systemd-nspawn";
197 static bool arg_notify_ready
= false;
198 static bool arg_use_cgns
= true;
199 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
200 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
201 static void *arg_root_hash
= NULL
;
202 static size_t arg_root_hash_size
= 0;
203 static char **arg_syscall_whitelist
= NULL
;
204 static char **arg_syscall_blacklist
= NULL
;
206 static scmp_filter_ctx arg_seccomp
= NULL
;
208 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
209 static bool arg_no_new_privileges
= false;
210 static int arg_oom_score_adjust
= 0;
211 static bool arg_oom_score_adjust_set
= false;
212 static CPUSet arg_cpu_set
= {};
213 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
214 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
215 static unsigned arg_console_width
= (unsigned) -1, arg_console_height
= (unsigned) -1;
216 static DeviceNode
* arg_extra_nodes
= NULL
;
217 static size_t arg_n_extra_nodes
= 0;
218 static char **arg_sysctl
= NULL
;
219 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
221 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
222 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
223 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
224 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
225 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
226 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
227 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
228 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
229 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
230 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
231 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
232 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
233 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
243 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_root_hash
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
251 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
253 static int handle_arg_console(const char *arg
) {
254 if (streq(arg
, "help")) {
262 if (streq(arg
, "interactive"))
263 arg_console_mode
= CONSOLE_INTERACTIVE
;
264 else if (streq(arg
, "read-only"))
265 arg_console_mode
= CONSOLE_READ_ONLY
;
266 else if (streq(arg
, "passive"))
267 arg_console_mode
= CONSOLE_PASSIVE
;
268 else if (streq(arg
, "pipe"))
269 arg_console_mode
= CONSOLE_PIPE
;
271 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
273 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
277 static int help(void) {
278 _cleanup_free_
char *link
= NULL
;
281 (void) pager_open(arg_pager_flags
);
283 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
287 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
288 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
289 " -h --help Show this help\n"
290 " --version Print version string\n"
291 " -q --quiet Do not show status information\n"
292 " --no-pager Do not pipe output into a pager\n"
293 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
295 " -D --directory=PATH Root directory for the container\n"
296 " --template=PATH Initialize root directory from template directory,\n"
298 " -x --ephemeral Run container with snapshot of root directory, and\n"
299 " remove it after exit\n"
300 " -i --image=PATH Root file system disk image (or device node) for\n"
302 " --oci-bundle=PATH OCI bundle directory\n"
303 " --read-only Mount the root directory read-only\n"
304 " --volatile[=MODE] Run the system in volatile mode\n"
305 " --root-hash=HASH Specify verity root hash for root disk image\n"
306 " --pivot-root=PATH[:PATH]\n"
307 " Pivot root to given directory in the container\n\n"
308 "%3$sExecution:%4$s\n"
309 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
310 " -b --boot Boot up full system (i.e. invoke init)\n"
311 " --chdir=PATH Set working directory in the container\n"
312 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
313 " -u --user=USER Run the command under specified user or UID\n"
314 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
315 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
316 "%3$sSystem Identity:%4$s\n"
317 " -M --machine=NAME Set the machine name for the container\n"
318 " --hostname=NAME Override the hostname for the container\n"
319 " --uuid=UUID Set a specific machine UUID for the container\n\n"
320 "%3$sProperties:%4$s\n"
321 " -S --slice=SLICE Place the container in the specified slice\n"
322 " --property=NAME=VALUE Set scope unit property\n"
323 " --register=BOOLEAN Register container as machine\n"
324 " --keep-unit Do not register a scope for the machine, reuse\n"
325 " the service unit nspawn is running in\n\n"
326 "%3$sUser Namespacing:%4$s\n"
327 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
328 " --private-users[=UIDBASE[:NUIDS]]\n"
329 " Similar, but with user configured UID/GID range\n"
330 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
331 "%3$sNetworking:%4$s\n"
332 " --private-network Disable network in container\n"
333 " --network-interface=INTERFACE\n"
334 " Assign an existing network interface to the\n"
336 " --network-macvlan=INTERFACE\n"
337 " Create a macvlan network interface based on an\n"
338 " existing network interface to the container\n"
339 " --network-ipvlan=INTERFACE\n"
340 " Create a ipvlan network interface based on an\n"
341 " existing network interface to the container\n"
342 " -n --network-veth Add a virtual Ethernet connection between host\n"
344 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
345 " Add an additional virtual Ethernet link between\n"
346 " host and container\n"
347 " --network-bridge=INTERFACE\n"
348 " Add a virtual Ethernet connection to the container\n"
349 " and attach it to an existing bridge on the host\n"
350 " --network-zone=NAME Similar, but attach the new interface to an\n"
351 " an automatically managed bridge interface\n"
352 " --network-namespace-path=PATH\n"
353 " Set network namespace to the one represented by\n"
354 " the specified kernel namespace file node\n"
355 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
356 " Expose a container IP port on the host\n\n"
357 "%3$sSecurity:%4$s\n"
358 " --capability=CAP In addition to the default, retain specified\n"
360 " --drop-capability=CAP Drop the specified capability from the default set\n"
361 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
362 " --system-call-filter=LIST|~LIST\n"
363 " Permit/prohibit specific system calls\n"
364 " -Z --selinux-context=SECLABEL\n"
365 " Set the SELinux security context to be used by\n"
366 " processes in the container\n"
367 " -L --selinux-apifs-context=SECLABEL\n"
368 " Set the SELinux security context to be used by\n"
369 " API/tmpfs file systems in the container\n\n"
370 "%3$sResources:%4$s\n"
371 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
372 " --oom-score-adjust=VALUE\n"
373 " Adjust the OOM score value for the payload\n"
374 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
375 " --personality=ARCH Pick personality for this container\n\n"
376 "%3$sIntegration:%4$s\n"
377 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
378 " --timezone=MODE Select mode of /etc/localtime initialization\n"
379 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
380 " host, try-guest, try-host\n"
381 " -j Equivalent to --link-journal=try-guest\n\n"
383 " --bind=PATH[:PATH[:OPTIONS]]\n"
384 " Bind mount a file or directory from the host into\n"
386 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
387 " Similar, but creates a read-only bind mount\n"
388 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
390 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
391 " --overlay=PATH[:PATH...]:PATH\n"
392 " Create an overlay mount from the host to \n"
394 " --overlay-ro=PATH[:PATH...]:PATH\n"
395 " Similar, but creates a read-only overlay mount\n\n"
396 "%3$sInput/Output:%4$s\n"
397 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
398 " set up for the container.\n"
399 " -P --pipe Equivalent to --console=pipe\n"
400 "\nSee the %2$s for details.\n"
401 , program_invocation_short_name
403 , ansi_underline(), ansi_normal()
404 , ansi_highlight(), ansi_normal()
410 static int custom_mount_check_all(void) {
413 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
414 CustomMount
*m
= &arg_custom_mounts
[i
];
416 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
417 if (arg_userns_chown
)
418 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
419 "--private-users-chown may not be combined with custom root mounts.");
420 else if (arg_uid_shift
== UID_INVALID
)
421 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
422 "--private-users with automatic UID shift may not be combined with custom root mounts.");
429 static int detect_unified_cgroup_hierarchy_from_environment(void) {
430 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
433 /* Allow the user to control whether the unified hierarchy is used */
437 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
438 var
= "UNIFIED_CGROUP_HIERARCHY";
443 r
= parse_boolean(e
);
445 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
447 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
449 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
455 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
458 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
459 * in the image actually supports. */
460 r
= cg_all_unified();
462 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
464 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
465 * routine only detects 231, so we'll have a false negative here for 230. */
466 r
= systemd_installation_has_version(directory
, 230);
468 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
470 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
472 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
473 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
474 /* Mixed cgroup hierarchy support was added in 233 */
475 r
= systemd_installation_has_version(directory
, 233);
477 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
479 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
481 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
483 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
485 log_debug("Using %s hierarchy for container.",
486 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
487 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
492 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
497 _cleanup_free_
char *t
= NULL
;
499 r
= extract_first_word(&spec
, &t
, ",", 0);
501 return log_error_errno(r
, "Failed to parse capability %s.", t
);
505 if (streq(t
, "help")) {
506 for (int i
= 0; i
< capability_list_length(); i
++) {
509 name
= capability_to_name(i
);
518 mask
= (uint64_t) -1;
520 r
= capability_from_name(t
);
522 return log_error_errno(r
, "Failed to parse capability %s.", t
);
529 return 1; /* continue */
532 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
535 r
= getenv_bool(name
);
539 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
541 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
542 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
546 static int parse_mount_settings_env(void) {
550 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
551 if (r
< 0 && r
!= -ENXIO
)
552 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
554 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
556 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
557 if (streq_ptr(e
, "network"))
558 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
561 r
= parse_boolean(e
);
563 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
565 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
566 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
572 static int parse_environment(void) {
576 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
579 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
582 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
585 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
589 r
= parse_mount_settings_env();
593 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
594 * even if it is supported. If not supported, it has no effect. */
595 if (!cg_ns_supported())
596 arg_use_cgns
= false;
598 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
601 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
605 arg_use_cgns
= r
> 0;
606 arg_settings_mask
|= SETTING_USE_CGNS
;
610 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
612 arg_container_service_name
= e
;
614 return detect_unified_cgroup_hierarchy_from_environment();
617 static int parse_argv(int argc
, char *argv
[]) {
635 ARG_NETWORK_INTERFACE
,
640 ARG_NETWORK_VETH_EXTRA
,
641 ARG_NETWORK_NAMESPACE_PATH
,
651 ARG_PRIVATE_USERS_CHOWN
,
654 ARG_SYSTEM_CALL_FILTER
,
657 ARG_NO_NEW_PRIVILEGES
,
658 ARG_OOM_SCORE_ADJUST
,
668 static const struct option options
[] = {
669 { "help", no_argument
, NULL
, 'h' },
670 { "version", no_argument
, NULL
, ARG_VERSION
},
671 { "directory", required_argument
, NULL
, 'D' },
672 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
673 { "ephemeral", no_argument
, NULL
, 'x' },
674 { "user", required_argument
, NULL
, 'u' },
675 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
676 { "as-pid2", no_argument
, NULL
, 'a' },
677 { "boot", no_argument
, NULL
, 'b' },
678 { "uuid", required_argument
, NULL
, ARG_UUID
},
679 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
680 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
681 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
682 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
683 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
684 { "bind", required_argument
, NULL
, ARG_BIND
},
685 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
686 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
687 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
688 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
689 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
690 { "machine", required_argument
, NULL
, 'M' },
691 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
692 { "slice", required_argument
, NULL
, 'S' },
693 { "setenv", required_argument
, NULL
, 'E' },
694 { "selinux-context", required_argument
, NULL
, 'Z' },
695 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
696 { "quiet", no_argument
, NULL
, 'q' },
697 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
698 { "register", required_argument
, NULL
, ARG_REGISTER
},
699 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
700 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
701 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
702 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
703 { "network-veth", no_argument
, NULL
, 'n' },
704 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
705 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
706 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
707 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
708 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
709 { "image", required_argument
, NULL
, 'i' },
710 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
711 { "port", required_argument
, NULL
, 'p' },
712 { "property", required_argument
, NULL
, ARG_PROPERTY
},
713 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
714 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
715 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
716 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
717 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
718 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
719 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
720 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
721 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
722 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
723 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
724 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
725 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
726 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
727 { "console", required_argument
, NULL
, ARG_CONSOLE
},
728 { "pipe", no_argument
, NULL
, ARG_PIPE
},
729 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
730 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
735 uint64_t plus
= 0, minus
= 0;
736 bool mask_all_settings
= false, mask_no_settings
= false;
741 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
751 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
755 arg_settings_mask
|= SETTING_DIRECTORY
;
759 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
763 arg_settings_mask
|= SETTING_DIRECTORY
;
767 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
771 arg_settings_mask
|= SETTING_DIRECTORY
;
775 r
= parse_path_argument_and_warn(optarg
, false, &arg_oci_bundle
);
782 arg_ephemeral
= true;
783 arg_settings_mask
|= SETTING_EPHEMERAL
;
787 r
= free_and_strdup(&arg_user
, optarg
);
791 arg_settings_mask
|= SETTING_USER
;
794 case ARG_NETWORK_ZONE
: {
797 j
= strjoin("vz-", optarg
);
801 if (!ifname_valid(j
)) {
802 log_error("Network zone name not valid: %s", j
);
807 free_and_replace(arg_network_zone
, j
);
809 arg_network_veth
= true;
810 arg_private_network
= true;
811 arg_settings_mask
|= SETTING_NETWORK
;
815 case ARG_NETWORK_BRIDGE
:
817 if (!ifname_valid(optarg
))
818 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
819 "Bridge interface name not valid: %s", optarg
);
821 r
= free_and_strdup(&arg_network_bridge
, optarg
);
827 arg_network_veth
= true;
828 arg_private_network
= true;
829 arg_settings_mask
|= SETTING_NETWORK
;
832 case ARG_NETWORK_VETH_EXTRA
:
833 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
835 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
837 arg_private_network
= true;
838 arg_settings_mask
|= SETTING_NETWORK
;
841 case ARG_NETWORK_INTERFACE
:
842 if (!ifname_valid(optarg
))
843 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
844 "Network interface name not valid: %s", optarg
);
846 r
= test_network_interface_initialized(optarg
);
850 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
853 arg_private_network
= true;
854 arg_settings_mask
|= SETTING_NETWORK
;
857 case ARG_NETWORK_MACVLAN
:
859 if (!ifname_valid(optarg
))
860 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
861 "MACVLAN network interface name not valid: %s", optarg
);
863 r
= test_network_interface_initialized(optarg
);
867 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
870 arg_private_network
= true;
871 arg_settings_mask
|= SETTING_NETWORK
;
874 case ARG_NETWORK_IPVLAN
:
876 if (!ifname_valid(optarg
))
877 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
878 "IPVLAN network interface name not valid: %s", optarg
);
880 r
= test_network_interface_initialized(optarg
);
884 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
888 case ARG_PRIVATE_NETWORK
:
889 arg_private_network
= true;
890 arg_settings_mask
|= SETTING_NETWORK
;
893 case ARG_NETWORK_NAMESPACE_PATH
:
894 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
898 arg_settings_mask
|= SETTING_NETWORK
;
902 if (arg_start_mode
== START_PID2
)
903 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
904 "--boot and --as-pid2 may not be combined.");
906 arg_start_mode
= START_BOOT
;
907 arg_settings_mask
|= SETTING_START_MODE
;
911 if (arg_start_mode
== START_BOOT
)
912 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
913 "--boot and --as-pid2 may not be combined.");
915 arg_start_mode
= START_PID2
;
916 arg_settings_mask
|= SETTING_START_MODE
;
920 r
= sd_id128_from_string(optarg
, &arg_uuid
);
922 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
924 if (sd_id128_is_null(arg_uuid
))
925 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
926 "Machine UUID may not be all zeroes.");
928 arg_settings_mask
|= SETTING_MACHINE_ID
;
932 _cleanup_free_
char *mangled
= NULL
;
934 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
938 free_and_replace(arg_slice
, mangled
);
939 arg_settings_mask
|= SETTING_SLICE
;
945 arg_machine
= mfree(arg_machine
);
947 if (!machine_name_is_valid(optarg
))
948 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
949 "Invalid machine name: %s", optarg
);
951 r
= free_and_strdup(&arg_machine
, optarg
);
959 arg_hostname
= mfree(arg_hostname
);
961 if (!hostname_is_valid(optarg
, false))
962 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
963 "Invalid hostname: %s", optarg
);
965 r
= free_and_strdup(&arg_hostname
, optarg
);
970 arg_settings_mask
|= SETTING_HOSTNAME
;
974 arg_selinux_context
= optarg
;
978 arg_selinux_apifs_context
= optarg
;
982 arg_read_only
= true;
983 arg_settings_mask
|= SETTING_READ_ONLY
;
987 case ARG_DROP_CAPABILITY
: {
989 r
= parse_capability_spec(optarg
, &m
);
993 if (c
== ARG_CAPABILITY
)
997 arg_settings_mask
|= SETTING_CAPABILITY
;
1000 case ARG_NO_NEW_PRIVILEGES
:
1001 r
= parse_boolean(optarg
);
1003 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1005 arg_no_new_privileges
= r
;
1006 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1010 arg_link_journal
= LINK_GUEST
;
1011 arg_link_journal_try
= true;
1012 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1015 case ARG_LINK_JOURNAL
:
1016 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1018 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1020 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1025 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1027 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1029 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1033 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1035 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1037 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1041 case ARG_OVERLAY_RO
:
1042 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1043 if (r
== -EADDRNOTAVAIL
)
1044 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1046 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1048 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1051 case ARG_INACCESSIBLE
:
1052 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1054 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1056 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1062 if (!env_assignment_is_valid(optarg
))
1063 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1064 "Environment variable assignment '%s' is not valid.", optarg
);
1066 n
= strv_env_set(arg_setenv
, optarg
);
1070 strv_free_and_replace(arg_setenv
, n
);
1071 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1079 case ARG_SHARE_SYSTEM
:
1080 /* We don't officially support this anymore, except for compat reasons. People should use the
1081 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1082 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1083 arg_clone_ns_flags
= 0;
1087 r
= parse_boolean(optarg
);
1089 log_error("Failed to parse --register= argument: %s", optarg
);
1097 arg_keep_unit
= true;
1100 case ARG_PERSONALITY
:
1102 arg_personality
= personality_from_string(optarg
);
1103 if (arg_personality
== PERSONALITY_INVALID
)
1104 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1105 "Unknown or unsupported personality '%s'.", optarg
);
1107 arg_settings_mask
|= SETTING_PERSONALITY
;
1113 arg_volatile_mode
= VOLATILE_YES
;
1114 else if (streq(optarg
, "help")) {
1115 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1120 m
= volatile_mode_from_string(optarg
);
1122 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1123 "Failed to parse --volatile= argument: %s", optarg
);
1125 arg_volatile_mode
= m
;
1128 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1132 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1134 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1136 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1138 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1142 if (strv_extend(&arg_property
, optarg
) < 0)
1147 case ARG_PRIVATE_USERS
: {
1152 else if (!in_charset(optarg
, DIGITS
))
1153 /* do *not* parse numbers as booleans */
1154 boolean
= parse_boolean(optarg
);
1156 if (boolean
== false) {
1157 /* no: User namespacing off */
1158 arg_userns_mode
= USER_NAMESPACE_NO
;
1159 arg_uid_shift
= UID_INVALID
;
1160 arg_uid_range
= UINT32_C(0x10000);
1161 } else if (boolean
== true) {
1162 /* yes: User namespacing on, UID range is read from root dir */
1163 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1164 arg_uid_shift
= UID_INVALID
;
1165 arg_uid_range
= UINT32_C(0x10000);
1166 } else if (streq(optarg
, "pick")) {
1167 /* pick: User namespacing on, UID range is picked randomly */
1168 arg_userns_mode
= USER_NAMESPACE_PICK
;
1169 arg_uid_shift
= UID_INVALID
;
1170 arg_uid_range
= UINT32_C(0x10000);
1172 _cleanup_free_
char *buffer
= NULL
;
1173 const char *range
, *shift
;
1175 /* anything else: User namespacing on, UID range is explicitly configured */
1177 range
= strchr(optarg
, ':');
1179 buffer
= strndup(optarg
, range
- optarg
);
1185 r
= safe_atou32(range
, &arg_uid_range
);
1187 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1191 r
= parse_uid(shift
, &arg_uid_shift
);
1193 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1195 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1198 if (arg_uid_range
<= 0)
1199 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1200 "UID range cannot be 0.");
1202 arg_settings_mask
|= SETTING_USERNS
;
1207 if (userns_supported()) {
1208 arg_userns_mode
= USER_NAMESPACE_PICK
;
1209 arg_uid_shift
= UID_INVALID
;
1210 arg_uid_range
= UINT32_C(0x10000);
1212 arg_settings_mask
|= SETTING_USERNS
;
1217 case ARG_PRIVATE_USERS_CHOWN
:
1218 arg_userns_chown
= true;
1220 arg_settings_mask
|= SETTING_USERNS
;
1223 case ARG_KILL_SIGNAL
:
1224 if (streq(optarg
, "help")) {
1225 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1229 arg_kill_signal
= signal_from_string(optarg
);
1230 if (arg_kill_signal
< 0)
1231 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1232 "Cannot parse signal: %s", optarg
);
1234 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1239 /* no → do not read files
1240 * yes → read files, do not override cmdline, trust only subset
1241 * override → read files, override cmdline, trust only subset
1242 * trusted → read files, do not override cmdline, trust all
1245 r
= parse_boolean(optarg
);
1247 if (streq(optarg
, "trusted")) {
1248 mask_all_settings
= false;
1249 mask_no_settings
= false;
1250 arg_settings_trusted
= true;
1252 } else if (streq(optarg
, "override")) {
1253 mask_all_settings
= false;
1254 mask_no_settings
= true;
1255 arg_settings_trusted
= -1;
1257 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1260 mask_all_settings
= false;
1261 mask_no_settings
= false;
1262 arg_settings_trusted
= -1;
1265 mask_all_settings
= true;
1266 mask_no_settings
= false;
1267 arg_settings_trusted
= false;
1273 if (!path_is_absolute(optarg
))
1274 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1275 "Working directory %s is not an absolute path.", optarg
);
1277 r
= free_and_strdup(&arg_chdir
, optarg
);
1281 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1284 case ARG_PIVOT_ROOT
:
1285 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1287 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1289 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1292 case ARG_NOTIFY_READY
:
1293 r
= parse_boolean(optarg
);
1295 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1296 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1297 arg_notify_ready
= r
;
1298 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1301 case ARG_ROOT_HASH
: {
1305 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1307 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1308 if (l
< sizeof(sd_id128_t
)) {
1310 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1313 free(arg_root_hash
);
1315 arg_root_hash_size
= l
;
1319 case ARG_SYSTEM_CALL_FILTER
: {
1323 negative
= optarg
[0] == '~';
1324 items
= negative
? optarg
+ 1 : optarg
;
1327 _cleanup_free_
char *word
= NULL
;
1329 r
= extract_first_word(&items
, &word
, NULL
, 0);
1335 return log_error_errno(r
, "Failed to parse system call filter: %m");
1338 r
= strv_extend(&arg_syscall_blacklist
, word
);
1340 r
= strv_extend(&arg_syscall_whitelist
, word
);
1345 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1351 _cleanup_free_
char *name
= NULL
;
1354 if (streq(optarg
, "help")) {
1355 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1359 eq
= strchr(optarg
, '=');
1361 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1362 "--rlimit= expects an '=' assignment.");
1364 name
= strndup(optarg
, eq
- optarg
);
1368 rl
= rlimit_from_string_harder(name
);
1370 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1371 "Unknown resource limit: %s", name
);
1373 if (!arg_rlimit
[rl
]) {
1374 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1375 if (!arg_rlimit
[rl
])
1379 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1381 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1383 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1387 case ARG_OOM_SCORE_ADJUST
:
1388 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1390 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1392 arg_oom_score_adjust_set
= true;
1393 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1396 case ARG_CPU_AFFINITY
: {
1399 r
= parse_cpu_set(optarg
, &cpuset
);
1401 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1403 cpu_set_reset(&arg_cpu_set
);
1404 arg_cpu_set
= cpuset
;
1405 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1409 case ARG_RESOLV_CONF
:
1410 if (streq(optarg
, "help")) {
1411 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1415 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1416 if (arg_resolv_conf
< 0)
1417 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1418 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1420 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1424 if (streq(optarg
, "help")) {
1425 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1429 arg_timezone
= timezone_mode_from_string(optarg
);
1430 if (arg_timezone
< 0)
1431 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1432 "Failed to parse /etc/localtime mode: %s", optarg
);
1434 arg_settings_mask
|= SETTING_TIMEZONE
;
1438 r
= handle_arg_console(optarg
);
1445 r
= handle_arg_console("pipe");
1451 arg_pager_flags
|= PAGER_DISABLE
;
1458 assert_not_reached("Unhandled option");
1461 if (argc
> optind
) {
1462 strv_free(arg_parameters
);
1463 arg_parameters
= strv_copy(argv
+ optind
);
1464 if (!arg_parameters
)
1467 arg_settings_mask
|= SETTING_START_MODE
;
1470 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1471 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1472 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1473 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1475 arg_directory
= TAKE_PTR(arg_template
);
1477 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1479 /* Make sure to parse environment before we reset the settings mask below */
1480 r
= parse_environment();
1484 /* Load all settings from .nspawn files */
1485 if (mask_no_settings
)
1486 arg_settings_mask
= 0;
1488 /* Don't load any settings from .nspawn files */
1489 if (mask_all_settings
)
1490 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1495 static int verify_arguments(void) {
1498 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1499 /* If we are running the stub init in the container, we don't need to look at what the init
1500 * in the container supports, because we are not using it. Let's immediately pick the right
1501 * setting based on the host system configuration.
1503 * We only do this, if the user didn't use an environment variable to override the detection.
1506 r
= cg_all_unified();
1508 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1510 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1511 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1512 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1514 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1517 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1518 arg_mount_settings
|= MOUNT_USE_USERNS
;
1520 if (arg_private_network
)
1521 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1523 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1524 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1525 arg_register
= false;
1526 if (arg_start_mode
!= START_PID1
)
1527 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1530 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1531 arg_userns_chown
= true;
1533 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1534 arg_kill_signal
= SIGRTMIN
+3;
1536 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 */
1537 arg_read_only
= true;
1539 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1540 arg_read_only
= true;
1542 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1543 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1544 * The latter is not technically a user session, but we don't need to labour the point. */
1545 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1547 if (arg_directory
&& arg_image
)
1548 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1550 if (arg_template
&& arg_image
)
1551 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1553 if (arg_template
&& !(arg_directory
|| arg_machine
))
1554 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1556 if (arg_ephemeral
&& arg_template
)
1557 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1559 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1560 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1562 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1563 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1565 if (arg_userns_chown
&& arg_read_only
)
1566 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1567 "--read-only and --private-users-chown may not be combined.");
1569 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1570 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1571 * copy-up (in case of overlay) making the entire exercise pointless. */
1572 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1573 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1575 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1576 * we need to error out, to avoid conflicts between different network options. */
1577 if (arg_network_namespace_path
&&
1578 (arg_network_interfaces
|| arg_network_macvlan
||
1579 arg_network_ipvlan
|| arg_network_veth_extra
||
1580 arg_network_bridge
|| arg_network_zone
||
1582 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1584 if (arg_network_bridge
&& arg_network_zone
)
1585 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1586 "--network-bridge= and --network-zone= may not be combined.");
1588 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1589 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1591 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1592 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1594 if (arg_expose_ports
&& !arg_private_network
)
1595 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1598 if (arg_expose_ports
)
1599 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1602 r
= custom_mount_check_all();
1609 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1612 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1615 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1618 if (uid
!= UID_INVALID
) {
1619 uid
+= arg_uid_shift
;
1621 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1625 if (gid
!= GID_INVALID
) {
1626 gid
+= (gid_t
) arg_uid_shift
;
1628 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1632 if (lchown(p
, uid
, gid
) < 0)
1638 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1642 q
= prefix_roota(root
, path
);
1643 r
= mkdir_errno_wrapper(q
, mode
);
1649 return userns_lchown(q
, uid
, gid
);
1652 static const char *timezone_from_path(const char *path
) {
1653 return PATH_STARTSWITH_SET(
1655 "../usr/share/zoneinfo/",
1656 "/usr/share/zoneinfo/");
1659 static bool etc_writable(void) {
1660 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1663 static int setup_timezone(const char *dest
) {
1664 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1665 const char *where
, *check
;
1671 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1672 r
= readlink_malloc("/etc/localtime", &p
);
1673 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1674 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1675 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1676 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1678 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1679 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1683 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1686 } else if (arg_timezone
== TIMEZONE_AUTO
)
1687 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1693 if (m
== TIMEZONE_OFF
)
1696 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1698 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1702 where
= strjoina(etc
, "/localtime");
1706 case TIMEZONE_DELETE
:
1707 if (unlink(where
) < 0)
1708 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1712 case TIMEZONE_SYMLINK
: {
1713 _cleanup_free_
char *q
= NULL
;
1714 const char *z
, *what
;
1716 z
= timezone_from_path(p
);
1718 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1722 r
= readlink_malloc(where
, &q
);
1723 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1724 return 0; /* Already pointing to the right place? Then do nothing .. */
1726 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1727 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1729 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1731 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1732 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1733 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1737 what
= strjoina("../usr/share/zoneinfo/", z
);
1738 if (symlink(what
, where
) < 0) {
1739 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1740 errno
, "Failed to correct timezone of container, ignoring: %m");
1750 case TIMEZONE_BIND
: {
1751 _cleanup_free_
char *resolved
= NULL
;
1754 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1756 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1760 if (found
== 0) /* missing? */
1761 (void) touch(resolved
);
1763 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1765 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1771 /* If mounting failed, try to copy */
1772 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1774 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1775 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1782 assert_not_reached("unexpected mode");
1785 /* Fix permissions of the symlink or file copy we just created */
1786 r
= userns_lchown(where
, 0, 0);
1788 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1793 static int have_resolv_conf(const char *path
) {
1796 if (access(path
, F_OK
) < 0) {
1797 if (errno
== ENOENT
)
1800 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1806 static int resolved_listening(void) {
1807 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1808 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1809 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1812 /* Check if resolved is listening */
1814 r
= sd_bus_open_system(&bus
);
1816 return log_debug_errno(r
, "Failed to open system bus: %m");
1818 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1820 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1824 r
= sd_bus_get_property_string(bus
,
1825 "org.freedesktop.resolve1",
1826 "/org/freedesktop/resolve1",
1827 "org.freedesktop.resolve1.Manager",
1830 &dns_stub_listener_mode
);
1832 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1834 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1837 static int setup_resolv_conf(const char *dest
) {
1838 _cleanup_free_
char *etc
= NULL
;
1839 const char *where
, *what
;
1845 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1846 if (arg_private_network
)
1847 m
= RESOLV_CONF_OFF
;
1848 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1849 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
1850 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1851 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1853 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1856 m
= arg_resolv_conf
;
1858 if (m
== RESOLV_CONF_OFF
)
1861 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1863 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1867 where
= strjoina(etc
, "/resolv.conf");
1869 if (m
== RESOLV_CONF_DELETE
) {
1870 if (unlink(where
) < 0)
1871 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1876 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
1877 what
= PRIVATE_STATIC_RESOLV_CONF
;
1878 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
1879 what
= PRIVATE_UPLINK_RESOLV_CONF
;
1880 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
1881 what
= PRIVATE_STUB_RESOLV_CONF
;
1883 what
= "/etc/resolv.conf";
1885 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
1886 _cleanup_free_
char *resolved
= NULL
;
1889 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1891 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1895 if (found
== 0) /* missing? */
1896 (void) touch(resolved
);
1898 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1900 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1902 /* If that didn't work, let's copy the file */
1905 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
1906 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1908 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
1910 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1911 * resolved or something similar runs inside and the symlink points there.
1913 * If the disk image is read-only, there's also no point in complaining.
1915 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
1916 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1917 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1921 r
= userns_lchown(where
, 0, 0);
1923 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1928 static int setup_boot_id(void) {
1929 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1930 _cleanup_free_
char *path
= NULL
;
1931 sd_id128_t rnd
= SD_ID128_NULL
;
1935 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1937 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
1939 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1941 r
= sd_id128_randomize(&rnd
);
1943 return log_error_errno(r
, "Failed to generate random boot id: %m");
1945 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1947 return log_error_errno(r
, "Failed to write boot id: %m");
1949 from
= TAKE_PTR(path
);
1950 to
= "/proc/sys/kernel/random/boot_id";
1952 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1956 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1959 static int copy_devnodes(const char *dest
) {
1960 static const char devnodes
[] =
1969 _cleanup_umask_ mode_t u
;
1977 /* Create /dev/net, so that we can create /dev/net/tun in it */
1978 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1979 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1981 NULSTR_FOREACH(d
, devnodes
) {
1982 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1985 from
= path_join("/dev/", d
);
1989 to
= path_join(dest
, from
);
1993 if (stat(from
, &st
) < 0) {
1995 if (errno
!= ENOENT
)
1996 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1998 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1999 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2000 "%s is not a char or block device, cannot copy.", from
);
2002 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2004 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2005 /* Explicitly warn the user when /dev is already populated. */
2006 if (errno
== EEXIST
)
2007 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
);
2009 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2011 /* Some systems abusively restrict mknod but allow bind mounts. */
2014 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2015 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2017 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2020 r
= userns_lchown(to
, 0, 0);
2022 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2024 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2028 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2030 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2032 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2035 prefixed
= path_join(dest
, sl
);
2039 t
= path_join("..", d
);
2043 if (symlink(t
, prefixed
) < 0)
2044 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2051 static int make_extra_nodes(const char *dest
) {
2052 _cleanup_umask_ mode_t u
;
2058 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2059 _cleanup_free_
char *path
= NULL
;
2060 DeviceNode
*n
= arg_extra_nodes
+ i
;
2062 path
= path_join(dest
, n
->path
);
2066 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2067 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2069 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2071 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2077 static int setup_pts(const char *dest
) {
2078 _cleanup_free_
char *options
= NULL
;
2083 if (arg_selinux_apifs_context
)
2084 (void) asprintf(&options
,
2085 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2086 arg_uid_shift
+ TTY_GID
,
2087 arg_selinux_apifs_context
);
2090 (void) asprintf(&options
,
2091 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2092 arg_uid_shift
+ TTY_GID
);
2097 /* Mount /dev/pts itself */
2098 p
= prefix_roota(dest
, "/dev/pts");
2099 r
= mkdir_errno_wrapper(p
, 0755);
2101 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2103 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2106 r
= userns_lchown(p
, 0, 0);
2108 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2110 /* Create /dev/ptmx symlink */
2111 p
= prefix_roota(dest
, "/dev/ptmx");
2112 if (symlink("pts/ptmx", p
) < 0)
2113 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2114 r
= userns_lchown(p
, 0, 0);
2116 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2118 /* And fix /dev/pts/ptmx ownership */
2119 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2120 r
= userns_lchown(p
, 0, 0);
2122 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2127 static int setup_stdio_as_dev_console(void) {
2131 terminal
= open_terminal("/dev/console", O_RDWR
);
2133 return log_error_errno(terminal
, "Failed to open console: %m");
2135 /* Make sure we can continue logging to the original stderr, even if
2136 * stderr points elsewhere now */
2137 r
= log_dup_console();
2139 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2141 /* invalidates 'terminal' on success and failure */
2142 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2144 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2149 static int setup_dev_console(const char *console
) {
2150 _cleanup_free_
char *p
= NULL
;
2153 /* Create /dev/console symlink */
2154 r
= path_make_relative("/dev", console
, &p
);
2156 return log_error_errno(r
, "Failed to create relative path: %m");
2158 if (symlink(p
, "/dev/console") < 0)
2159 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2164 static int setup_keyring(void) {
2165 key_serial_t keyring
;
2167 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2168 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2169 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2170 * these system calls let's make sure we don't leak anything into the container. */
2172 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2173 if (keyring
== -1) {
2174 if (errno
== ENOSYS
)
2175 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2176 else if (IN_SET(errno
, EACCES
, EPERM
))
2177 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2179 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2185 static int setup_kmsg(int kmsg_socket
) {
2186 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2187 _cleanup_free_
char *fifo
= NULL
;
2188 _cleanup_close_
int fd
= -1;
2189 _cleanup_umask_ mode_t u
;
2192 assert(kmsg_socket
>= 0);
2196 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2197 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2198 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2199 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2201 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2203 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2205 if (mkfifo(fifo
, 0600) < 0)
2206 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2208 from
= TAKE_PTR(fifo
);
2210 r
= mount_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2214 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2216 return log_error_errno(errno
, "Failed to open fifo: %m");
2218 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2219 r
= send_one_fd(kmsg_socket
, fd
, 0);
2221 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2226 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2227 union in_addr_union
*exposed
= userdata
;
2233 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2237 static int setup_hostname(void) {
2240 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2243 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2245 return log_error_errno(r
, "Failed to set hostname: %m");
2250 static int setup_journal(const char *directory
) {
2251 _cleanup_free_
char *d
= NULL
;
2252 char id
[SD_ID128_STRING_MAX
];
2253 const char *dirname
, *p
, *q
;
2258 /* Don't link journals in ephemeral mode */
2262 if (arg_link_journal
== LINK_NO
)
2265 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2267 r
= sd_id128_get_machine(&this_id
);
2269 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2271 if (sd_id128_equal(arg_uuid
, this_id
)) {
2272 log_full(try ? LOG_WARNING
: LOG_ERR
,
2273 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2279 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2280 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2282 bool ignore
= r
== -EROFS
&& try;
2283 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2284 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2285 return ignore
? 0 : r
;
2289 (void) sd_id128_to_string(arg_uuid
, id
);
2291 p
= strjoina("/var/log/journal/", id
);
2292 q
= prefix_roota(directory
, p
);
2294 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2298 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2299 "%s: already a mount point, refusing to use for journal", p
);
2302 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2306 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2307 "%s: already a mount point, refusing to use for journal", q
);
2310 r
= readlink_and_make_absolute(p
, &d
);
2312 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2315 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2317 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2322 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2323 } else if (r
== -EINVAL
) {
2325 if (arg_link_journal
== LINK_GUEST
&&
2328 if (errno
== ENOTDIR
) {
2329 log_error("%s already exists and is neither a symlink nor a directory", p
);
2332 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2334 } else if (r
!= -ENOENT
)
2335 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2337 if (arg_link_journal
== LINK_GUEST
) {
2339 if (symlink(q
, p
) < 0) {
2341 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2344 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2347 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2349 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2353 if (arg_link_journal
== LINK_HOST
) {
2354 /* don't create parents here — if the host doesn't have
2355 * permanent journal set up, don't force it here */
2357 r
= mkdir_errno_wrapper(p
, 0755);
2358 if (r
< 0 && r
!= -EEXIST
) {
2360 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2363 return log_error_errno(r
, "Failed to create %s: %m", p
);
2366 } else if (access(p
, F_OK
) < 0)
2369 if (dir_is_empty(q
) == 0)
2370 log_warning("%s is not empty, proceeding anyway.", q
);
2372 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2374 return log_error_errno(r
, "Failed to create %s: %m", q
);
2376 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2378 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2383 static int drop_capabilities(uid_t uid
) {
2384 CapabilityQuintet q
;
2386 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2387 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2388 * arg_caps_retain. */
2390 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2391 q
= arg_full_capabilities
;
2393 if (q
.bounding
== (uint64_t) -1)
2394 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2396 if (q
.effective
== (uint64_t) -1)
2397 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2399 if (q
.inheritable
== (uint64_t) -1)
2400 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2402 if (q
.permitted
== (uint64_t) -1)
2403 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2405 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2408 if (capability_quintet_mangle(&q
))
2409 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2412 q
= (CapabilityQuintet
) {
2413 .bounding
= arg_caps_retain
,
2414 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2415 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2416 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2417 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2420 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2421 * in order to maintain the same behavior as systemd < 242. */
2422 if (capability_quintet_mangle(&q
))
2423 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2424 "Some capabilities will not be set because they are not in the current bounding set.");
2428 return capability_quintet_enforce(&q
);
2431 static int reset_audit_loginuid(void) {
2432 _cleanup_free_
char *p
= NULL
;
2435 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2438 r
= read_one_line_file("/proc/self/loginuid", &p
);
2442 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2444 /* Already reset? */
2445 if (streq(p
, "4294967295"))
2448 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2451 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2452 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2453 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2454 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2455 "using systemd-nspawn. Sleeping for 5s... (%m)");
2463 static int setup_propagate(const char *root
) {
2467 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2468 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2469 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2470 (void) mkdir_p(p
, 0600);
2472 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2474 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2476 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2478 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2480 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2482 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2484 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2485 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2489 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2493 /* machined will MS_MOVE into that directory, and that's only
2494 * supported for non-shared mounts. */
2495 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2498 static int setup_machine_id(const char *directory
) {
2499 const char *etc_machine_id
;
2503 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2504 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2505 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2506 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2507 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2508 * container behaves nicely). */
2510 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2512 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2514 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2515 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2517 if (sd_id128_is_null(arg_uuid
)) {
2518 r
= sd_id128_randomize(&arg_uuid
);
2520 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2523 if (sd_id128_is_null(id
))
2524 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2525 "Machine ID in container image is zero, refusing.");
2533 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2538 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2541 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2542 if (r
== -EOPNOTSUPP
)
2543 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.");
2545 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2547 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2549 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2551 log_debug("Patched directory tree to match UID/GID range.");
2558 * < 0 : wait_for_terminate() failed to get the state of the
2559 * container, the container was terminated by a signal, or
2560 * failed for an unknown reason. No change is made to the
2561 * container argument.
2562 * > 0 : The program executed in the container terminated with an
2563 * error. The exit code of the program executed in the
2564 * container is returned. The container argument has been set
2565 * to CONTAINER_TERMINATED.
2566 * 0 : The container is being rebooted, has been shut down or exited
2567 * successfully. The container argument has been set to either
2568 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2570 * That is, success is indicated by a return value of zero, and an
2571 * error is indicated by a non-zero value.
2573 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2577 r
= wait_for_terminate(pid
, &status
);
2579 return log_warning_errno(r
, "Failed to wait for container: %m");
2581 switch (status
.si_code
) {
2584 if (status
.si_status
== 0)
2585 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2587 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2589 *container
= CONTAINER_TERMINATED
;
2590 return status
.si_status
;
2593 if (status
.si_status
== SIGINT
) {
2594 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2595 *container
= CONTAINER_TERMINATED
;
2598 } else if (status
.si_status
== SIGHUP
) {
2599 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2600 *container
= CONTAINER_REBOOTED
;
2606 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2607 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2610 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2611 "Container %s failed due to unknown reason.", arg_machine
);
2615 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2618 pid
= PTR_TO_PID(userdata
);
2620 if (kill(pid
, arg_kill_signal
) >= 0) {
2621 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2622 sd_event_source_set_userdata(s
, NULL
);
2627 sd_event_exit(sd_event_source_get_event(s
), 0);
2631 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2637 pid
= PTR_TO_PID(userdata
);
2642 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2643 return log_error_errno(errno
, "Failed to waitid(): %m");
2644 if (si
.si_pid
== 0) /* No pending children. */
2646 if (si
.si_pid
== pid
) {
2647 /* The main process we care for has exited. Return from
2648 * signal handler but leave the zombie. */
2649 sd_event_exit(sd_event_source_get_event(s
), 0);
2653 /* Reap all other children. */
2654 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2660 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2665 pid
= PTR_TO_PID(userdata
);
2667 if (arg_kill_signal
> 0) {
2668 log_info("Container termination requested. Attempting to halt container.");
2669 (void) kill(pid
, arg_kill_signal
);
2671 log_info("Container termination requested. Exiting.");
2672 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2678 static int determine_names(void) {
2681 if (arg_template
&& !arg_directory
&& arg_machine
) {
2683 /* If --template= was specified then we should not
2684 * search for a machine, but instead create a new one
2685 * in /var/lib/machine. */
2687 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2692 if (!arg_image
&& !arg_directory
) {
2694 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2696 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2698 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2700 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2702 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2703 r
= free_and_strdup(&arg_image
, i
->path
);
2705 r
= free_and_strdup(&arg_directory
, i
->path
);
2710 arg_read_only
= arg_read_only
|| i
->read_only
;
2712 r
= safe_getcwd(&arg_directory
);
2714 return log_error_errno(r
, "Failed to determine current directory: %m");
2717 if (!arg_directory
&& !arg_image
)
2718 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2722 if (arg_directory
&& path_equal(arg_directory
, "/"))
2723 arg_machine
= gethostname_malloc();
2728 arg_machine
= strdup(basename(arg_image
));
2730 /* Truncate suffix if there is one */
2731 e
= endswith(arg_machine
, ".raw");
2735 arg_machine
= strdup(basename(arg_directory
));
2740 hostname_cleanup(arg_machine
);
2741 if (!machine_name_is_valid(arg_machine
))
2742 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2744 if (arg_ephemeral
) {
2747 /* Add a random suffix when this is an
2748 * ephemeral machine, so that we can run many
2749 * instances at once without manually having
2750 * to specify -M each time. */
2752 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2763 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2772 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
2774 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2776 return free_and_replace(*p
, chased
);
2779 static int determine_uid_shift(const char *directory
) {
2782 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2787 if (arg_uid_shift
== UID_INVALID
) {
2790 r
= stat(directory
, &st
);
2792 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2794 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2796 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2797 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2798 "UID and GID base of %s don't match.", directory
);
2800 arg_uid_range
= UINT32_C(0x10000);
2803 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2804 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2805 "UID base too high for UID range.");
2810 static unsigned long effective_clone_ns_flags(void) {
2811 unsigned long flags
= arg_clone_ns_flags
;
2813 if (arg_private_network
)
2814 flags
|= CLONE_NEWNET
;
2816 flags
|= CLONE_NEWCGROUP
;
2817 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2818 flags
|= CLONE_NEWUSER
;
2823 static int patch_sysctl(void) {
2825 /* This table is inspired by runc's sysctl() function */
2826 static const struct {
2829 unsigned long clone_flags
;
2831 { "kernel.hostname", false, CLONE_NEWUTS
},
2832 { "kernel.domainname", false, CLONE_NEWUTS
},
2833 { "kernel.msgmax", false, CLONE_NEWIPC
},
2834 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2835 { "kernel.msgmni", false, CLONE_NEWIPC
},
2836 { "kernel.sem", false, CLONE_NEWIPC
},
2837 { "kernel.shmall", false, CLONE_NEWIPC
},
2838 { "kernel.shmmax", false, CLONE_NEWIPC
},
2839 { "kernel.shmmni", false, CLONE_NEWIPC
},
2840 { "fs.mqueue.", true, CLONE_NEWIPC
},
2841 { "net.", true, CLONE_NEWNET
},
2844 unsigned long flags
;
2848 flags
= effective_clone_ns_flags();
2850 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2854 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2856 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2859 if (safe_sysctl
[i
].prefix
)
2860 good
= startswith(*k
, safe_sysctl
[i
].key
);
2862 good
= streq(*k
, safe_sysctl
[i
].key
);
2869 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2871 r
= sysctl_write(*k
, *v
);
2873 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2879 static int inner_child(
2881 const char *directory
,
2885 int master_pty_socket
,
2888 _cleanup_free_
char *home
= NULL
;
2889 char as_uuid
[ID128_UUID_STRING_MAX
];
2891 const char *envp
[] = {
2892 "PATH=" DEFAULT_PATH_COMPAT
,
2893 NULL
, /* container */
2898 NULL
, /* container_uuid */
2899 NULL
, /* LISTEN_FDS */
2900 NULL
, /* LISTEN_PID */
2901 NULL
, /* NOTIFY_SOCKET */
2904 const char *exec_target
;
2905 _cleanup_strv_free_
char **env_use
= NULL
;
2906 int r
, which_failed
;
2908 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2909 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2910 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2911 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2912 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2913 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2916 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2917 * unshare(). See below. */
2921 assert(kmsg_socket
>= 0);
2923 log_debug("Inner child is initializing.");
2925 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2926 /* Tell the parent, that it now can write the UID map. */
2927 (void) barrier_place(barrier
); /* #1 */
2929 /* Wait until the parent wrote the UID map */
2930 if (!barrier_place_and_sync(barrier
)) /* #2 */
2931 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2932 "Parent died too early");
2935 r
= reset_uid_gid();
2937 return log_error_errno(r
, "Couldn't become new root: %m");
2940 arg_mount_settings
| MOUNT_IN_USERNS
,
2942 arg_selinux_apifs_context
);
2946 if (!arg_network_namespace_path
&& arg_private_network
) {
2947 r
= unshare(CLONE_NEWNET
);
2949 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2951 /* Tell the parent that it can setup network interfaces. */
2952 (void) barrier_place(barrier
); /* #3 */
2955 r
= mount_sysfs(NULL
, arg_mount_settings
);
2959 /* Wait until we are cgroup-ified, so that we
2960 * can mount the right cgroup path writable */
2961 if (!barrier_place_and_sync(barrier
)) /* #4 */
2962 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2963 "Parent died too early");
2966 r
= unshare(CLONE_NEWCGROUP
);
2968 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2971 arg_unified_cgroup_hierarchy
,
2972 arg_userns_mode
!= USER_NAMESPACE_NO
,
2975 arg_selinux_apifs_context
,
2978 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2982 r
= setup_boot_id();
2986 r
= setup_kmsg(kmsg_socket
);
2989 kmsg_socket
= safe_close(kmsg_socket
);
2994 arg_n_custom_mounts
,
2996 arg_selinux_apifs_context
,
2997 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3002 return log_error_errno(errno
, "setsid() failed: %m");
3004 if (arg_private_network
)
3005 (void) loopback_setup();
3007 if (arg_expose_ports
) {
3008 r
= expose_port_send_rtnl(rtnl_socket
);
3011 rtnl_socket
= safe_close(rtnl_socket
);
3014 if (arg_console_mode
!= CONSOLE_PIPE
) {
3015 _cleanup_close_
int master
= -1;
3016 _cleanup_free_
char *console
= NULL
;
3018 /* Allocate a pty and make it available as /dev/console. */
3019 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3021 return log_error_errno(master
, "Failed to allocate a pty: %m");
3023 r
= setup_dev_console(console
);
3025 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3027 r
= send_one_fd(master_pty_socket
, master
, 0);
3029 return log_error_errno(r
, "Failed to send master fd: %m");
3030 master_pty_socket
= safe_close(master_pty_socket
);
3032 r
= setup_stdio_as_dev_console();
3041 if (arg_oom_score_adjust_set
) {
3042 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3044 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3047 if (arg_cpu_set
.set
)
3048 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3049 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3051 (void) setup_hostname();
3053 if (arg_personality
!= PERSONALITY_INVALID
) {
3054 r
= safe_personality(arg_personality
);
3056 return log_error_errno(r
, "personality() failed: %m");
3057 } else if (secondary
) {
3058 r
= safe_personality(PER_LINUX32
);
3060 return log_error_errno(r
, "personality() failed: %m");
3063 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3065 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3070 if (is_seccomp_available()) {
3072 r
= seccomp_load(arg_seccomp
);
3073 if (ERRNO_IS_SECCOMP_FATAL(r
))
3074 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3076 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3081 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
3087 if (arg_selinux_context
)
3088 if (setexeccon(arg_selinux_context
) < 0)
3089 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3092 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3093 * if we need to later on. */
3094 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3095 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3097 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3098 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
3100 r
= change_uid_gid(arg_user
, &home
);
3104 r
= drop_capabilities(getuid());
3106 return log_error_errno(r
, "Dropping capabilities failed: %m");
3108 if (arg_no_new_privileges
)
3109 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3110 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3112 /* LXC sets container=lxc, so follow the scheme here */
3113 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3115 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3119 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3120 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3123 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3124 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3125 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3128 assert(!sd_id128_is_null(arg_uuid
));
3130 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3133 if (fdset_size(fds
) > 0) {
3134 r
= fdset_cloexec(fds
, false);
3136 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3138 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3139 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3142 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3145 env_use
= strv_env_merge(2, envp
, arg_setenv
);
3149 /* Let the parent know that we are ready and
3150 * wait until the parent is ready with the
3152 if (!barrier_place_and_sync(barrier
)) /* #5 */
3153 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3154 "Parent died too early");
3157 if (chdir(arg_chdir
) < 0)
3158 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3160 if (arg_start_mode
== START_PID2
) {
3161 r
= stub_pid1(arg_uuid
);
3166 log_debug("Inner child completed, invoking payload.");
3168 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3169 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3170 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3172 log_set_open_when_needed(true);
3174 (void) fdset_close_others(fds
);
3176 if (arg_start_mode
== START_BOOT
) {
3180 /* Automatically search for the init system */
3182 m
= strv_length(arg_parameters
);
3183 a
= newa(char*, m
+ 2);
3184 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3187 a
[0] = (char*) "/usr/lib/systemd/systemd";
3188 execve(a
[0], a
, env_use
);
3190 a
[0] = (char*) "/lib/systemd/systemd";
3191 execve(a
[0], a
, env_use
);
3193 a
[0] = (char*) "/sbin/init";
3194 execve(a
[0], a
, env_use
);
3196 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3197 } else if (!strv_isempty(arg_parameters
)) {
3198 const char *dollar_path
;
3200 exec_target
= arg_parameters
[0];
3202 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3204 dollar_path
= strv_env_get(env_use
, "PATH");
3206 if (putenv((char*) dollar_path
) != 0)
3207 return log_error_errno(errno
, "Failed to update $PATH: %m");
3210 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3213 /* If we cannot change the directory, we'll end up in /, that is expected. */
3214 (void) chdir(home
?: "/root");
3216 execle("/bin/bash", "-bash", NULL
, env_use
);
3217 execle("/bin/sh", "-sh", NULL
, env_use
);
3219 exec_target
= "/bin/bash, /bin/sh";
3222 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3225 static int setup_sd_notify_child(void) {
3226 _cleanup_close_
int fd
= -1;
3227 union sockaddr_union sa
= {
3228 .un
.sun_family
= AF_UNIX
,
3229 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3233 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3235 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3237 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3238 (void) sockaddr_un_unlink(&sa
.un
);
3240 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3242 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3244 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3246 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3248 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3250 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3255 static int outer_child(
3257 const char *directory
,
3258 DissectedImage
*dissected_image
,
3265 int uid_shift_socket
,
3266 int master_pty_socket
,
3267 int unified_cgroup_hierarchy_socket
,
3271 _cleanup_close_
int fd
= -1;
3277 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3278 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3279 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3280 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3284 assert(pid_socket
>= 0);
3285 assert(uuid_socket
>= 0);
3286 assert(notify_socket
>= 0);
3287 assert(master_pty_socket
>= 0);
3288 assert(kmsg_socket
>= 0);
3290 log_debug("Outer child is initializing.");
3292 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3293 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3295 r
= reset_audit_loginuid();
3299 /* Mark everything as slave, so that we still
3300 * receive mounts from the real root, but don't
3301 * propagate mounts to the real root. */
3302 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3306 if (dissected_image
) {
3307 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3308 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3309 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3310 * makes sure ESP partitions and userns are compatible. */
3312 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3313 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3314 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
)|
3315 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3317 return log_error_errno(r
, "File system check for image failed: %m");
3319 return log_error_errno(r
, "Failed to mount image root file system: %m");
3322 r
= determine_uid_shift(directory
);
3326 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3327 /* Let the parent know which UID shift we read from the image */
3328 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3330 return log_error_errno(errno
, "Failed to send UID shift: %m");
3331 if (l
!= sizeof(arg_uid_shift
))
3332 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3333 "Short write while sending UID shift.");
3335 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3336 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3337 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3338 * not it will pick a different one, and send it back to us. */
3340 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3342 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3343 if (l
!= sizeof(arg_uid_shift
))
3344 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3345 "Short read while receiving UID shift.");
3348 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3349 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3352 if (path_equal(directory
, "/")) {
3353 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3354 * place, so that we can make changes to its mount structure (for example, to implement
3355 * --volatile=) without this interfering with our ability to access files such as
3356 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3357 * (instead of a temporary directory, since we are living in our own mount namspace here
3358 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3359 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3361 r
= mount_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3365 directory
= "/run/systemd/nspawn-root";
3368 r
= setup_pivot_root(
3371 arg_pivot_root_old
);
3375 r
= setup_volatile_mode(
3379 arg_selinux_apifs_context
);
3386 arg_n_custom_mounts
,
3388 arg_selinux_apifs_context
,
3393 /* Make sure we always have a mount that we can move to root later on. */
3394 if (!path_is_mount_point(directory
, NULL
, 0)) {
3395 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3400 if (dissected_image
) {
3401 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3402 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3403 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
));
3405 return log_error_errno(r
, "File system check for image failed: %m");
3407 return log_error_errno(r
, "Failed to mount image file system: %m");
3410 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3411 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3413 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3417 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3419 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3420 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3421 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3422 "Short write while sending cgroup mode.");
3424 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3427 /* Mark everything as shared so our mounts get propagated down. This is
3428 * required to make new bind mounts available in systemd services
3429 * inside the container that create a new mount namespace.
3430 * See https://github.com/systemd/systemd/issues/3860
3431 * Further submounts (such as /dev) done after this will inherit the
3432 * shared propagation mode.
3434 * IMPORTANT: Do not overmount the root directory anymore from now on to
3435 * enable moving the root directory mount to root later on.
3436 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3438 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3442 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3446 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3450 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3451 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3452 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3454 return log_error_errno(r
, "Failed to make tree read-only: %m");
3457 r
= mount_all(directory
,
3460 arg_selinux_apifs_context
);
3464 r
= copy_devnodes(directory
);
3468 r
= make_extra_nodes(directory
);
3472 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3474 p
= prefix_roota(directory
, "/run/systemd");
3475 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3477 r
= setup_pts(directory
);
3481 r
= setup_propagate(directory
);
3485 r
= setup_keyring();
3492 arg_n_custom_mounts
,
3494 arg_selinux_apifs_context
,
3495 MOUNT_NON_ROOT_ONLY
);
3499 r
= setup_timezone(directory
);
3503 r
= setup_resolv_conf(directory
);
3507 r
= setup_machine_id(directory
);
3511 r
= setup_journal(directory
);
3515 if (!arg_use_cgns
) {
3518 arg_unified_cgroup_hierarchy
,
3519 arg_userns_mode
!= USER_NAMESPACE_NO
,
3522 arg_selinux_apifs_context
,
3528 r
= mount_move_root(directory
);
3530 return log_error_errno(r
, "Failed to move root directory: %m");
3532 fd
= setup_sd_notify_child();
3536 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3537 arg_clone_ns_flags
|
3538 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3540 return log_error_errno(errno
, "Failed to fork inner child: %m");
3542 pid_socket
= safe_close(pid_socket
);
3543 uuid_socket
= safe_close(uuid_socket
);
3544 notify_socket
= safe_close(notify_socket
);
3545 uid_shift_socket
= safe_close(uid_shift_socket
);
3547 /* The inner child has all namespaces that are
3548 * requested, so that we all are owned by the user if
3549 * user namespaces are turned on. */
3551 if (arg_network_namespace_path
) {
3552 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3554 return log_error_errno(r
, "Failed to join network namespace: %m");
3557 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
);
3559 _exit(EXIT_FAILURE
);
3561 _exit(EXIT_SUCCESS
);
3564 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3566 return log_error_errno(errno
, "Failed to send PID: %m");
3567 if (l
!= sizeof(pid
))
3568 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3569 "Short write while sending PID.");
3571 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3573 return log_error_errno(errno
, "Failed to send machine ID: %m");
3574 if (l
!= sizeof(arg_uuid
))
3575 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3576 "Short write while sending machine ID.");
3578 l
= send_one_fd(notify_socket
, fd
, 0);
3580 return log_error_errno(l
, "Failed to send notify fd: %m");
3582 pid_socket
= safe_close(pid_socket
);
3583 uuid_socket
= safe_close(uuid_socket
);
3584 notify_socket
= safe_close(notify_socket
);
3585 master_pty_socket
= safe_close(master_pty_socket
);
3586 kmsg_socket
= safe_close(kmsg_socket
);
3587 rtnl_socket
= safe_close(rtnl_socket
);
3588 netns_fd
= safe_close(netns_fd
);
3593 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3594 bool tried_hashed
= false;
3595 unsigned n_tries
= 100;
3600 assert(ret_lock_file
);
3601 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3602 assert(arg_uid_range
== 0x10000U
);
3606 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3609 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3610 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3615 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3617 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3620 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3621 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3622 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3627 /* Make some superficial checks whether the range is currently known in the user database */
3628 if (getpwuid(candidate
))
3630 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3632 if (getgrgid(candidate
))
3634 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3637 *ret_lock_file
= lf
;
3638 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3643 if (arg_machine
&& !tried_hashed
) {
3644 /* Try to hash the base from the container name */
3646 static const uint8_t hash_key
[] = {
3647 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3648 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3651 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3653 tried_hashed
= true;
3655 random_bytes(&candidate
, sizeof(candidate
));
3657 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3658 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3662 static int setup_uid_map(pid_t pid
) {
3663 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3668 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3669 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3670 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3672 return log_error_errno(r
, "Failed to write UID map: %m");
3674 /* We always assign the same UID and GID ranges */
3675 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3676 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3678 return log_error_errno(r
, "Failed to write GID map: %m");
3683 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3684 char buf
[NOTIFY_BUFFER_MAX
+1];
3686 struct iovec iovec
= {
3688 .iov_len
= sizeof(buf
)-1,
3690 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
3691 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
3692 struct msghdr msghdr
= {
3695 .msg_control
= &control
,
3696 .msg_controllen
= sizeof(control
),
3698 struct ucred
*ucred
;
3700 pid_t inner_child_pid
;
3701 _cleanup_strv_free_
char **tags
= NULL
;
3705 inner_child_pid
= PTR_TO_PID(userdata
);
3707 if (revents
!= EPOLLIN
) {
3708 log_warning("Got unexpected poll event for notify fd.");
3712 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3713 if (IN_SET(n
, -EAGAIN
, -EINTR
))
3716 return log_warning_errno(n
, "Couldn't read notification socket: %m");
3718 cmsg_close_all(&msghdr
);
3720 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
3721 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3722 log_debug("Received notify message without valid credentials. Ignoring.");
3726 if ((size_t) n
>= sizeof(buf
)) {
3727 log_warning("Received notify message exceeded maximum size. Ignoring.");
3732 tags
= strv_split(buf
, "\n\r");
3736 if (strv_find(tags
, "READY=1"))
3737 (void) sd_notifyf(false, "READY=1\n");
3739 p
= strv_find_startswith(tags
, "STATUS=");
3741 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3746 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3749 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3751 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3753 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3758 static int merge_settings(Settings
*settings
, const char *path
) {
3764 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3765 * that this steals the fields of the Settings* structure, and hence modifies it. */
3767 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3768 settings
->start_mode
>= 0) {
3769 arg_start_mode
= settings
->start_mode
;
3770 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3773 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3774 arg_ephemeral
= settings
->ephemeral
;
3776 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3779 if (!arg_settings_trusted
)
3780 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3782 free_and_replace(arg_directory
, settings
->root
);
3785 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3786 settings
->pivot_root_new
) {
3787 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3788 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3791 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3792 settings
->working_directory
)
3793 free_and_replace(arg_chdir
, settings
->working_directory
);
3795 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3796 settings
->environment
)
3797 strv_free_and_replace(arg_setenv
, settings
->environment
);
3799 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3802 free_and_replace(arg_user
, settings
->user
);
3804 if (uid_is_valid(settings
->uid
))
3805 arg_uid
= settings
->uid
;
3806 if (gid_is_valid(settings
->gid
))
3807 arg_gid
= settings
->gid
;
3808 if (settings
->n_supplementary_gids
> 0) {
3809 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3810 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3814 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3815 uint64_t plus
, minus
;
3816 uint64_t network_minus
= 0;
3818 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3819 * Settings structure */
3821 plus
= settings
->capability
;
3822 minus
= settings
->drop_capability
;
3824 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3825 if (settings_private_network(settings
))
3826 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3828 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3831 if (!arg_settings_trusted
&& plus
!= 0) {
3832 if (settings
->capability
!= 0)
3833 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3835 arg_caps_retain
&= ~network_minus
;
3836 arg_caps_retain
|= plus
;
3839 arg_caps_retain
&= ~minus
;
3841 /* Copy the full capabilities over too */
3842 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3843 if (!arg_settings_trusted
)
3844 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
3846 arg_full_capabilities
= settings
->full_capabilities
;
3850 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3851 settings
->kill_signal
> 0)
3852 arg_kill_signal
= settings
->kill_signal
;
3854 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3855 settings
->personality
!= PERSONALITY_INVALID
)
3856 arg_personality
= settings
->personality
;
3858 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3859 !sd_id128_is_null(settings
->machine_id
)) {
3861 if (!arg_settings_trusted
)
3862 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3864 arg_uuid
= settings
->machine_id
;
3867 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3868 settings
->read_only
>= 0)
3869 arg_read_only
= settings
->read_only
;
3871 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3872 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3873 arg_volatile_mode
= settings
->volatile_mode
;
3875 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3876 settings
->n_custom_mounts
> 0) {
3878 if (!arg_settings_trusted
)
3879 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3881 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3882 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3883 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3884 settings
->n_custom_mounts
= 0;
3888 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3889 (settings
->private_network
>= 0 ||
3890 settings
->network_veth
>= 0 ||
3891 settings
->network_bridge
||
3892 settings
->network_zone
||
3893 settings
->network_interfaces
||
3894 settings
->network_macvlan
||
3895 settings
->network_ipvlan
||
3896 settings
->network_veth_extra
||
3897 settings
->network_namespace_path
)) {
3899 if (!arg_settings_trusted
)
3900 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3902 arg_network_veth
= settings_network_veth(settings
);
3903 arg_private_network
= settings_private_network(settings
);
3905 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3906 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3907 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3908 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3910 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3911 free_and_replace(arg_network_zone
, settings
->network_zone
);
3913 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3917 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3918 settings
->expose_ports
) {
3920 if (!arg_settings_trusted
)
3921 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3923 expose_port_free_all(arg_expose_ports
);
3924 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3928 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3929 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3931 if (!arg_settings_trusted
)
3932 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3934 arg_userns_mode
= settings
->userns_mode
;
3935 arg_uid_shift
= settings
->uid_shift
;
3936 arg_uid_range
= settings
->uid_range
;
3937 arg_userns_chown
= settings
->userns_chown
;
3941 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3942 arg_notify_ready
= settings
->notify_ready
;
3944 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3946 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_whitelist
))
3947 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3949 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3950 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3954 if (!arg_settings_trusted
&& settings
->seccomp
)
3955 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
3957 seccomp_release(arg_seccomp
);
3958 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
3963 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3964 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3967 if (!settings
->rlimit
[rl
])
3970 if (!arg_settings_trusted
) {
3971 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3975 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3978 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3980 free_and_replace(arg_hostname
, settings
->hostname
);
3982 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3983 settings
->no_new_privileges
>= 0)
3984 arg_no_new_privileges
= settings
->no_new_privileges
;
3986 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3987 settings
->oom_score_adjust_set
) {
3989 if (!arg_settings_trusted
)
3990 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
3992 arg_oom_score_adjust
= settings
->oom_score_adjust
;
3993 arg_oom_score_adjust_set
= true;
3997 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
3998 settings
->cpu_set
.set
) {
4000 if (!arg_settings_trusted
)
4001 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4003 cpu_set_reset(&arg_cpu_set
);
4004 arg_cpu_set
= settings
->cpu_set
;
4005 settings
->cpu_set
= (CPUSet
) {};
4009 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4010 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4011 arg_resolv_conf
= settings
->resolv_conf
;
4013 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4014 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4016 if (!arg_settings_trusted
)
4017 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4019 arg_link_journal
= settings
->link_journal
;
4020 arg_link_journal_try
= settings
->link_journal_try
;
4024 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4025 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4026 arg_timezone
= settings
->timezone
;
4028 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4031 if (!arg_settings_trusted
)
4032 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4034 free_and_replace(arg_slice
, settings
->slice
);
4037 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4038 settings
->use_cgns
>= 0) {
4040 if (!arg_settings_trusted
)
4041 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4043 arg_use_cgns
= settings
->use_cgns
;
4046 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4047 settings
->clone_ns_flags
!= (unsigned long) -1) {
4049 if (!arg_settings_trusted
)
4050 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4052 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4055 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4056 settings
->console_mode
>= 0) {
4058 if (!arg_settings_trusted
)
4059 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4061 arg_console_mode
= settings
->console_mode
;
4064 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4065 * don't consult arg_settings_mask for them. */
4067 sd_bus_message_unref(arg_property_message
);
4068 arg_property_message
= TAKE_PTR(settings
->properties
);
4070 arg_console_width
= settings
->console_width
;
4071 arg_console_height
= settings
->console_height
;
4073 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4074 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4075 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4080 static int load_settings(void) {
4081 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4082 _cleanup_fclose_
FILE *f
= NULL
;
4083 _cleanup_free_
char *p
= NULL
;
4090 /* If all settings are masked, there's no point in looking for
4091 * the settings file */
4092 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
4095 fn
= strjoina(arg_machine
, ".nspawn");
4097 /* We first look in the admin's directories in /etc and /run */
4098 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4099 _cleanup_free_
char *j
= NULL
;
4101 j
= path_join(i
, fn
);
4109 /* By default, we trust configuration from /etc and /run */
4110 if (arg_settings_trusted
< 0)
4111 arg_settings_trusted
= true;
4116 if (errno
!= ENOENT
)
4117 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4121 /* After that, let's look for a file next to the
4122 * actual image we shall boot. */
4125 p
= file_in_same_dir(arg_image
, fn
);
4128 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4129 p
= file_in_same_dir(arg_directory
, fn
);
4136 if (!f
&& errno
!= ENOENT
)
4137 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4139 /* By default, we do not trust configuration from /var/lib/machines */
4140 if (arg_settings_trusted
< 0)
4141 arg_settings_trusted
= false;
4148 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4150 r
= settings_load(f
, p
, &settings
);
4154 return merge_settings(settings
, p
);
4157 static int load_oci_bundle(void) {
4158 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4161 if (!arg_oci_bundle
)
4164 /* By default let's trust OCI bundles */
4165 if (arg_settings_trusted
< 0)
4166 arg_settings_trusted
= true;
4168 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4172 return merge_settings(settings
, arg_oci_bundle
);
4175 static int run_container(
4176 DissectedImage
*dissected_image
,
4179 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4180 union in_addr_union
*exposed
,
4181 int *master
, pid_t
*pid
, int *ret
) {
4183 static const struct sigaction sa
= {
4184 .sa_handler
= nop_signal_handler
,
4185 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4188 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4189 _cleanup_close_
int etc_passwd_lock
= -1;
4190 _cleanup_close_pair_
int
4191 kmsg_socket_pair
[2] = { -1, -1 },
4192 rtnl_socket_pair
[2] = { -1, -1 },
4193 pid_socket_pair
[2] = { -1, -1 },
4194 uuid_socket_pair
[2] = { -1, -1 },
4195 notify_socket_pair
[2] = { -1, -1 },
4196 uid_shift_socket_pair
[2] = { -1, -1 },
4197 master_pty_socket_pair
[2] = { -1, -1 },
4198 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4200 _cleanup_close_
int notify_socket
= -1;
4201 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4202 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4203 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4204 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4205 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4206 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4207 ContainerStatus container_status
= 0;
4211 _cleanup_close_
int child_netns_fd
= -1;
4213 assert_se(sigemptyset(&mask_chld
) == 0);
4214 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4216 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4217 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4218 * check with getpwuid() if the specific user already exists. Note that /etc might be
4219 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4220 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4221 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4224 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4225 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4226 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4229 r
= barrier_create(&barrier
);
4231 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4233 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4234 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4236 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4237 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4239 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4240 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4242 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4243 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4245 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4246 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4248 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4249 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4251 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4252 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4253 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4255 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4256 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4257 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4259 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4260 * parent's blocking calls and give it a chance to call wait() and terminate. */
4261 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4263 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4265 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4267 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4269 if (arg_network_namespace_path
) {
4270 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4271 if (child_netns_fd
< 0)
4272 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4274 r
= fd_is_network_ns(child_netns_fd
);
4276 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
);
4278 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4280 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4281 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4284 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4286 return log_error_errno(errno
, "clone() failed%s: %m",
4288 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4291 /* The outer child only has a file system namespace. */
4292 barrier_set_role(&barrier
, BARRIER_CHILD
);
4294 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4295 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4296 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4297 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4298 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4299 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4300 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4301 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4303 (void) reset_all_signal_handlers();
4304 (void) reset_signal_mask();
4306 r
= outer_child(&barrier
,
4311 uuid_socket_pair
[1],
4312 notify_socket_pair
[1],
4313 kmsg_socket_pair
[1],
4314 rtnl_socket_pair
[1],
4315 uid_shift_socket_pair
[1],
4316 master_pty_socket_pair
[1],
4317 unified_cgroup_hierarchy_socket_pair
[1],
4321 _exit(EXIT_FAILURE
);
4323 _exit(EXIT_SUCCESS
);
4326 barrier_set_role(&barrier
, BARRIER_PARENT
);
4330 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4331 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4332 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4333 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4334 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4335 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4336 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4337 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4339 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4340 /* The child just let us know the UID shift it might have read from the image. */
4341 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4343 return log_error_errno(errno
, "Failed to read UID shift: %m");
4344 if (l
!= sizeof arg_uid_shift
)
4345 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4347 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4348 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4349 * image, but if that's already in use, pick a new one, and report back to the child,
4350 * which one we now picked. */
4352 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4354 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4356 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4358 return log_error_errno(errno
, "Failed to send UID shift: %m");
4359 if (l
!= sizeof arg_uid_shift
)
4360 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4364 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4365 /* The child let us know the support cgroup mode it might have read from the image. */
4366 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4368 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4369 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4370 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4371 l
, l
== 0 ? " The child is most likely dead." : "");
4374 /* Wait for the outer child. */
4375 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4378 if (r
!= EXIT_SUCCESS
)
4381 /* And now retrieve the PID of the inner child. */
4382 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4384 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4385 if (l
!= sizeof *pid
)
4386 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4388 /* We also retrieve container UUID in case it was generated by outer child */
4389 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4391 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4392 if (l
!= sizeof(arg_uuid
))
4393 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4395 /* We also retrieve the socket used for notifications generated by outer child */
4396 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4397 if (notify_socket
< 0)
4398 return log_error_errno(notify_socket
,
4399 "Failed to receive notification socket from the outer child: %m");
4401 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4403 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4404 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4405 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4407 r
= setup_uid_map(*pid
);
4411 (void) barrier_place(&barrier
); /* #2 */
4414 if (arg_private_network
) {
4415 if (!arg_network_namespace_path
) {
4416 /* Wait until the child has unshared its network namespace. */
4417 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4418 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4421 if (child_netns_fd
< 0) {
4422 /* Make sure we have an open file descriptor to the child's network
4423 * namespace so it stays alive even if the child exits. */
4424 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4426 return log_error_errno(r
, "Failed to open child network namespace: %m");
4429 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4433 if (arg_network_veth
) {
4434 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4435 arg_network_bridge
|| arg_network_zone
);
4441 if (arg_network_bridge
) {
4442 /* Add the interface to a bridge */
4443 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4448 } else if (arg_network_zone
) {
4449 /* Add the interface to a bridge, possibly creating it */
4450 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4458 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4462 /* We created the primary and extra veth links now; let's remember this, so that we know to
4463 remove them later on. Note that we don't bother with removing veth links that were created
4464 here when their setup failed half-way, because in that case the kernel should be able to
4465 remove them on its own, since they cannot be referenced by anything yet. */
4466 *veth_created
= true;
4468 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4472 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4477 if (arg_register
|| !arg_keep_unit
) {
4478 r
= sd_bus_default_system(&bus
);
4480 return log_error_errno(r
, "Failed to open system bus: %m");
4482 r
= sd_bus_set_close_on_exit(bus
, false);
4484 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4487 if (!arg_keep_unit
) {
4488 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4489 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4490 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4492 r
= sd_bus_match_signal_async(
4495 "org.freedesktop.systemd1",
4497 "org.freedesktop.systemd1.Scope",
4499 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4501 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4505 r
= register_machine(
4513 arg_custom_mounts
, arg_n_custom_mounts
,
4516 arg_property_message
,
4518 arg_container_service_name
);
4522 } else if (!arg_keep_unit
) {
4528 arg_custom_mounts
, arg_n_custom_mounts
,
4531 arg_property_message
);
4535 } else if (arg_slice
|| arg_property
)
4536 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4538 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4542 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4546 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4550 /* Notify the child that the parent is ready with all
4551 * its setup (including cgroup-ification), and that
4552 * the child can now hand over control to the code to
4553 * run inside the container. */
4554 (void) barrier_place(&barrier
); /* #4 */
4556 /* Block SIGCHLD here, before notifying child.
4557 * process_pty() will handle it with the other signals. */
4558 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4560 /* Reset signal to default */
4561 r
= default_signals(SIGCHLD
, -1);
4563 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4565 r
= sd_event_new(&event
);
4567 return log_error_errno(r
, "Failed to get default event source: %m");
4569 (void) sd_event_set_watchdog(event
, true);
4572 r
= sd_bus_attach_event(bus
, event
, 0);
4574 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4577 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4581 /* Let the child know that we are ready and wait that the child is completely ready now. */
4582 if (!barrier_place_and_sync(&barrier
)) /* #5 */
4583 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4585 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4586 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4587 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4589 (void) sd_notifyf(false,
4590 "STATUS=Container running.\n"
4591 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4592 if (!arg_notify_ready
)
4593 (void) sd_notify(false, "READY=1\n");
4595 if (arg_kill_signal
> 0) {
4596 /* Try to kill the init system on SIGINT or SIGTERM */
4597 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4598 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4600 /* Immediately exit */
4601 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4602 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4605 /* Exit when the child exits */
4606 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4608 if (arg_expose_ports
) {
4609 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4613 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4616 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4618 if (arg_console_mode
!= CONSOLE_PIPE
) {
4619 _cleanup_close_
int fd
= -1;
4620 PTYForwardFlags flags
= 0;
4622 /* Retrieve the master pty allocated by inner child */
4623 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
4625 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
4627 switch (arg_console_mode
) {
4629 case CONSOLE_READ_ONLY
:
4630 flags
|= PTY_FORWARD_READ_ONLY
;
4634 case CONSOLE_INTERACTIVE
:
4635 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
4637 r
= pty_forward_new(event
, fd
, flags
, &forward
);
4639 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4641 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4642 (void) pty_forward_set_width_height(forward
,
4644 arg_console_height
);
4648 assert(arg_console_mode
== CONSOLE_PASSIVE
);
4651 *master
= TAKE_FD(fd
);
4654 r
= sd_event_loop(event
);
4656 return log_error_errno(r
, "Failed to run event loop: %m");
4661 (void) pty_forward_get_last_char(forward
, &last_char
);
4662 forward
= pty_forward_free(forward
);
4664 if (!arg_quiet
&& last_char
!= '\n')
4668 /* Kill if it is not dead yet anyway */
4669 if (!arg_register
&& !arg_keep_unit
&& bus
)
4670 terminate_scope(bus
, arg_machine
);
4672 /* Normally redundant, but better safe than sorry */
4673 (void) kill(*pid
, SIGKILL
);
4675 if (arg_private_network
) {
4676 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
4677 * to avoid having to move the parent to the child network namespace. */
4678 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
4683 _cleanup_close_
int parent_netns_fd
= -1;
4685 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
4687 log_error_errno(r
, "Failed to open parent network namespace: %m");
4688 _exit(EXIT_FAILURE
);
4691 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
4693 log_error_errno(r
, "Failed to enter child network namespace: %m");
4694 _exit(EXIT_FAILURE
);
4697 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
4699 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
4701 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
4705 r
= wait_for_container(*pid
, &container_status
);
4708 /* Tell machined that we are gone. */
4710 (void) unregister_machine(bus
, arg_machine
);
4713 /* We failed to wait for the container, or the container exited abnormally. */
4715 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4716 /* r > 0 → The container exited with a non-zero status.
4717 * As a special case, we need to replace 133 with a different value,
4718 * because 133 is special-cased in the service file to reboot the container.
4719 * otherwise → The container exited with zero status and a reboot was not requested.
4721 if (r
== EXIT_FORCE_RESTART
)
4722 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4724 return 0; /* finito */
4727 /* CONTAINER_REBOOTED, loop again */
4729 if (arg_keep_unit
) {
4730 /* Special handling if we are running as a service: instead of simply
4731 * restarting the machine we want to restart the entire service, so let's
4732 * inform systemd about this with the special exit code 133. The service
4733 * file uses RestartForceExitStatus=133 so that this results in a full
4734 * nspawn restart. This is necessary since we might have cgroup parameters
4735 * set we want to have flushed out. */
4736 *ret
= EXIT_FORCE_RESTART
;
4737 return 0; /* finito */
4740 expose_port_flush(arg_expose_ports
, exposed
);
4742 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4743 *veth_created
= false;
4744 return 1; /* loop again */
4747 static int initialize_rlimits(void) {
4748 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4749 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4750 * container execution environments. */
4752 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4753 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4754 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4755 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4756 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4757 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4758 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4759 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4760 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4761 [RLIMIT_NICE
] = { 0, 0 },
4762 [RLIMIT_NOFILE
] = { 1024, 4096 },
4763 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4764 [RLIMIT_RTPRIO
] = { 0, 0 },
4765 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4766 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4768 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4769 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4770 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4771 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4772 * that PID 1 changes a number of other resource limits during early initialization which is why we
4773 * don't read the other limits from PID 1 but prefer the static table above. */
4778 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4779 /* Let's only fill in what the user hasn't explicitly configured anyway */
4780 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4781 const struct rlimit
*v
;
4782 struct rlimit buffer
;
4784 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4785 /* For these two let's read the limits off PID 1. See above for an explanation. */
4787 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4788 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4792 v
= kernel_defaults
+ rl
;
4794 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4795 if (!arg_rlimit
[rl
])
4799 if (DEBUG_LOGGING
) {
4800 _cleanup_free_
char *k
= NULL
;
4802 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4803 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4810 static int cant_be_in_netns(void) {
4811 union sockaddr_union sa
= {
4813 .sun_family
= AF_UNIX
,
4814 .sun_path
= "/run/udev/control",
4817 char udev_path
[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t
)];
4818 _cleanup_free_
char *udev_ns
= NULL
, *our_ns
= NULL
;
4819 _cleanup_close_
int fd
= -1;
4823 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
4824 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
4827 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
4830 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
4832 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
4834 if (connect(fd
, &sa
.un
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
4836 if (errno
== ENOENT
|| ERRNO_IS_DISCONNECT(errno
))
4837 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
4838 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
4840 return log_error_errno(errno
, "Failed to connect socket to udev control socket: %m");
4843 r
= getpeercred(fd
, &ucred
);
4845 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
4847 xsprintf(udev_path
, "/proc/" PID_FMT
"/ns/net", ucred
.pid
);
4848 r
= readlink_malloc(udev_path
, &udev_ns
);
4850 return log_error_errno(r
, "Failed to read network namespace of udev: %m");
4852 r
= readlink_malloc("/proc/self/ns/net", &our_ns
);
4854 return log_error_errno(r
, "Failed to read our own network namespace: %m");
4856 if (!streq(our_ns
, udev_ns
))
4857 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
4858 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
4862 static int run(int argc
, char *argv
[]) {
4863 bool secondary
= false, remove_directory
= false, remove_image
= false,
4864 veth_created
= false, remove_tmprootdir
= false;
4865 _cleanup_close_
int master
= -1;
4866 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4867 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4868 char veth_name
[IFNAMSIZ
] = "";
4869 union in_addr_union exposed
= {};
4870 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4871 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4872 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4873 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4874 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4877 log_parse_environment();
4880 r
= parse_argv(argc
, argv
);
4888 r
= cant_be_in_netns();
4892 r
= initialize_rlimits();
4896 r
= load_oci_bundle();
4900 r
= determine_names();
4904 r
= load_settings();
4910 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4914 r
= verify_arguments();
4918 /* Reapply environment settings. */
4919 (void) detect_unified_cgroup_hierarchy_from_environment();
4921 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4922 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4923 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4924 (void) ignore_signals(SIGPIPE
, -1);
4926 n_fd_passed
= sd_listen_fds(false);
4927 if (n_fd_passed
> 0) {
4928 r
= fdset_new_listen_fds(&fds
, false);
4930 log_error_errno(r
, "Failed to collect file descriptors: %m");
4935 /* The "default" umask. This is appropriate for most file and directory
4936 * operations performed by nspawn, and is the umask that will be used for
4937 * the child. Functions like copy_devnodes() change the umask temporarily. */
4940 if (arg_directory
) {
4943 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4944 * /var from the host will propagate into container dynamically (because bad things happen if
4945 * two systems write to the same /var). Let's allow it for the special cases where /var is
4946 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4947 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
4948 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4953 if (arg_ephemeral
) {
4954 _cleanup_free_
char *np
= NULL
;
4956 r
= chase_symlinks_and_update(&arg_directory
, 0);
4960 /* If the specified path is a mount point we generate the new snapshot immediately
4961 * inside it under a random name. However if the specified is not a mount point we
4962 * create the new snapshot in the parent directory, just next to it. */
4963 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4965 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4969 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4971 r
= tempfn_random(arg_directory
, "machine.", &np
);
4973 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4977 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4978 * only owned by us and no one else. */
4979 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4981 log_error_errno(r
, "Failed to lock %s: %m", np
);
4986 BLOCK_SIGNALS(SIGINT
);
4987 r
= btrfs_subvol_snapshot(arg_directory
, np
,
4988 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4989 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4990 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4991 BTRFS_SNAPSHOT_RECURSIVE
|
4992 BTRFS_SNAPSHOT_QUOTA
|
4993 BTRFS_SNAPSHOT_SIGINT
);
4996 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5000 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5004 free_and_replace(arg_directory
, np
);
5005 remove_directory
= true;
5007 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5011 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5013 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5017 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5022 r
= chase_symlinks_and_update(&arg_template
, 0);
5027 BLOCK_SIGNALS(SIGINT
);
5028 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5029 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5030 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5031 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5032 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5033 BTRFS_SNAPSHOT_RECURSIVE
|
5034 BTRFS_SNAPSHOT_QUOTA
|
5035 BTRFS_SNAPSHOT_SIGINT
);
5038 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5039 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5040 else if (r
== -EINTR
) {
5041 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5044 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5047 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5048 "Populated %s from template %s.", arg_directory
, arg_template
);
5052 if (arg_start_mode
== START_BOOT
) {
5055 if (arg_pivot_root_new
)
5056 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5060 if (path_is_os_tree(p
) <= 0) {
5061 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5068 if (arg_pivot_root_new
)
5069 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5073 q
= strjoina(p
, "/usr/");
5075 if (laccess(q
, F_OK
) < 0) {
5076 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
5084 assert(!arg_template
);
5086 r
= chase_symlinks_and_update(&arg_image
, 0);
5090 if (arg_ephemeral
) {
5091 _cleanup_free_
char *np
= NULL
;
5093 r
= tempfn_random(arg_image
, "machine.", &np
);
5095 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5099 /* Always take an exclusive lock on our own ephemeral copy. */
5100 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5102 r
= log_error_errno(r
, "Failed to create image lock: %m");
5107 BLOCK_SIGNALS(SIGINT
);
5108 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
);
5111 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5115 r
= log_error_errno(r
, "Failed to copy image file: %m");
5119 free_and_replace(arg_image
, np
);
5120 remove_image
= true;
5122 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5124 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5128 r
= log_error_errno(r
, "Failed to create image lock: %m");
5132 if (!arg_root_hash
) {
5133 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
5135 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
5141 if (!mkdtemp(tmprootdir
)) {
5142 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5146 remove_tmprootdir
= true;
5148 arg_directory
= strdup(tmprootdir
);
5149 if (!arg_directory
) {
5154 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, LO_FLAGS_PARTSCAN
, &loop
);
5156 log_error_errno(r
, "Failed to set up loopback block device: %m");
5160 r
= dissect_image_and_warn(
5163 arg_root_hash
, arg_root_hash_size
,
5164 DISSECT_IMAGE_REQUIRE_ROOT
|DISSECT_IMAGE_RELAX_VAR_CHECK
,
5167 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5168 log_notice("Note that the disk image needs to\n"
5169 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5170 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5171 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5172 " d) or contain a file system without a partition table\n"
5173 "in order to be bootable with systemd-nspawn.");
5179 if (!arg_root_hash
&& dissected_image
->can_verity
)
5180 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5182 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
5186 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5187 if (remove_image
&& unlink(arg_image
) >= 0)
5188 remove_image
= false;
5191 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5195 if (arg_console_mode
< 0)
5197 isatty(STDIN_FILENO
) > 0 &&
5198 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5200 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5204 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5205 arg_machine
, arg_image
?: arg_directory
);
5207 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5209 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5210 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5215 r
= run_container(dissected_image
,
5218 veth_name
, &veth_created
,
5226 (void) sd_notify(false,
5227 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5228 "STOPPING=1\nSTATUS=Terminating...");
5231 (void) kill(pid
, SIGKILL
);
5233 /* Try to flush whatever is still queued in the pty */
5235 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5236 master
= safe_close(master
);
5240 (void) wait_for_terminate(pid
, NULL
);
5244 if (remove_directory
&& arg_directory
) {
5247 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5249 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5252 if (remove_image
&& arg_image
) {
5253 if (unlink(arg_image
) < 0)
5254 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5257 if (remove_tmprootdir
) {
5258 if (rmdir(tmprootdir
) < 0)
5259 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5265 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5266 (void) rm_rf(p
, REMOVE_ROOT
);
5269 expose_port_flush(arg_expose_ports
, &exposed
);
5272 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5273 (void) remove_bridge(arg_network_zone
);
5275 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5276 expose_port_free_all(arg_expose_ports
);
5277 rlimit_free_all(arg_rlimit
);
5278 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5286 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);