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 char *arg_verity_data
= NULL
;
203 static size_t arg_root_hash_size
= 0;
204 static char **arg_syscall_allow_list
= NULL
;
205 static char **arg_syscall_deny_list
= NULL
;
207 static scmp_filter_ctx arg_seccomp
= NULL
;
209 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
210 static bool arg_no_new_privileges
= false;
211 static int arg_oom_score_adjust
= 0;
212 static bool arg_oom_score_adjust_set
= false;
213 static CPUSet arg_cpu_set
= {};
214 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
215 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
216 static unsigned arg_console_width
= (unsigned) -1, arg_console_height
= (unsigned) -1;
217 static DeviceNode
* arg_extra_nodes
= NULL
;
218 static size_t arg_n_extra_nodes
= 0;
219 static char **arg_sysctl
= NULL
;
220 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
222 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
223 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
224 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
225 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
226 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
227 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
228 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
229 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
230 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
231 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
232 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
233 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
244 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_root_hash
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_verity_data
, freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
253 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
255 static int handle_arg_console(const char *arg
) {
256 if (streq(arg
, "help")) {
264 if (streq(arg
, "interactive"))
265 arg_console_mode
= CONSOLE_INTERACTIVE
;
266 else if (streq(arg
, "read-only"))
267 arg_console_mode
= CONSOLE_READ_ONLY
;
268 else if (streq(arg
, "passive"))
269 arg_console_mode
= CONSOLE_PASSIVE
;
270 else if (streq(arg
, "pipe"))
271 arg_console_mode
= CONSOLE_PIPE
;
273 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
275 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
279 static int help(void) {
280 _cleanup_free_
char *link
= NULL
;
283 (void) pager_open(arg_pager_flags
);
285 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
289 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
290 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
291 " -h --help Show this help\n"
292 " --version Print version string\n"
293 " -q --quiet Do not show status information\n"
294 " --no-pager Do not pipe output into a pager\n"
295 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
297 " -D --directory=PATH Root directory for the container\n"
298 " --template=PATH Initialize root directory from template directory,\n"
300 " -x --ephemeral Run container with snapshot of root directory, and\n"
301 " remove it after exit\n"
302 " -i --image=PATH Root file system disk image (or device node) for\n"
304 " --oci-bundle=PATH OCI bundle directory\n"
305 " --read-only Mount the root directory read-only\n"
306 " --volatile[=MODE] Run the system in volatile mode\n"
307 " --root-hash=HASH Specify verity root hash for root disk image\n"
308 " --verity-data=PATH Specify hash device for verity\n"
309 " --pivot-root=PATH[:PATH]\n"
310 " Pivot root to given directory in the container\n\n"
311 "%3$sExecution:%4$s\n"
312 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
313 " -b --boot Boot up full system (i.e. invoke init)\n"
314 " --chdir=PATH Set working directory in the container\n"
315 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
316 " -u --user=USER Run the command under specified user or UID\n"
317 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
318 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
319 "%3$sSystem Identity:%4$s\n"
320 " -M --machine=NAME Set the machine name for the container\n"
321 " --hostname=NAME Override the hostname for the container\n"
322 " --uuid=UUID Set a specific machine UUID for the container\n\n"
323 "%3$sProperties:%4$s\n"
324 " -S --slice=SLICE Place the container in the specified slice\n"
325 " --property=NAME=VALUE Set scope unit property\n"
326 " --register=BOOLEAN Register container as machine\n"
327 " --keep-unit Do not register a scope for the machine, reuse\n"
328 " the service unit nspawn is running in\n\n"
329 "%3$sUser Namespacing:%4$s\n"
330 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
331 " --private-users[=UIDBASE[:NUIDS]]\n"
332 " Similar, but with user configured UID/GID range\n"
333 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
334 "%3$sNetworking:%4$s\n"
335 " --private-network Disable network in container\n"
336 " --network-interface=INTERFACE\n"
337 " Assign an existing network interface to the\n"
339 " --network-macvlan=INTERFACE\n"
340 " Create a macvlan network interface based on an\n"
341 " existing network interface to the container\n"
342 " --network-ipvlan=INTERFACE\n"
343 " Create a ipvlan network interface based on an\n"
344 " existing network interface to the container\n"
345 " -n --network-veth Add a virtual Ethernet connection between host\n"
347 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
348 " Add an additional virtual Ethernet link between\n"
349 " host and container\n"
350 " --network-bridge=INTERFACE\n"
351 " Add a virtual Ethernet connection to the container\n"
352 " and attach it to an existing bridge on the host\n"
353 " --network-zone=NAME Similar, but attach the new interface to an\n"
354 " an automatically managed bridge interface\n"
355 " --network-namespace-path=PATH\n"
356 " Set network namespace to the one represented by\n"
357 " the specified kernel namespace file node\n"
358 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
359 " Expose a container IP port on the host\n\n"
360 "%3$sSecurity:%4$s\n"
361 " --capability=CAP In addition to the default, retain specified\n"
363 " --drop-capability=CAP Drop the specified capability from the default set\n"
364 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
365 " --system-call-filter=LIST|~LIST\n"
366 " Permit/prohibit specific system calls\n"
367 " -Z --selinux-context=SECLABEL\n"
368 " Set the SELinux security context to be used by\n"
369 " processes in the container\n"
370 " -L --selinux-apifs-context=SECLABEL\n"
371 " Set the SELinux security context to be used by\n"
372 " API/tmpfs file systems in the container\n\n"
373 "%3$sResources:%4$s\n"
374 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
375 " --oom-score-adjust=VALUE\n"
376 " Adjust the OOM score value for the payload\n"
377 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
378 " --personality=ARCH Pick personality for this container\n\n"
379 "%3$sIntegration:%4$s\n"
380 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
381 " --timezone=MODE Select mode of /etc/localtime initialization\n"
382 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
383 " host, try-guest, try-host\n"
384 " -j Equivalent to --link-journal=try-guest\n\n"
386 " --bind=PATH[:PATH[:OPTIONS]]\n"
387 " Bind mount a file or directory from the host into\n"
389 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
390 " Similar, but creates a read-only bind mount\n"
391 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
393 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
394 " --overlay=PATH[:PATH...]:PATH\n"
395 " Create an overlay mount from the host to \n"
397 " --overlay-ro=PATH[:PATH...]:PATH\n"
398 " Similar, but creates a read-only overlay mount\n\n"
399 "%3$sInput/Output:%4$s\n"
400 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
401 " set up for the container.\n"
402 " -P --pipe Equivalent to --console=pipe\n"
403 "\nSee the %2$s for details.\n"
404 , program_invocation_short_name
406 , ansi_underline(), ansi_normal()
407 , ansi_highlight(), ansi_normal()
413 static int custom_mount_check_all(void) {
416 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
417 CustomMount
*m
= &arg_custom_mounts
[i
];
419 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
420 if (arg_userns_chown
)
421 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
422 "--private-users-chown may not be combined with custom root mounts.");
423 else if (arg_uid_shift
== UID_INVALID
)
424 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
425 "--private-users with automatic UID shift may not be combined with custom root mounts.");
432 static int detect_unified_cgroup_hierarchy_from_environment(void) {
433 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
436 /* Allow the user to control whether the unified hierarchy is used */
440 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
441 var
= "UNIFIED_CGROUP_HIERARCHY";
446 r
= parse_boolean(e
);
448 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
450 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
452 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
458 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
461 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
462 * in the image actually supports. */
463 r
= cg_all_unified();
465 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
467 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
468 * routine only detects 231, so we'll have a false negative here for 230. */
469 r
= systemd_installation_has_version(directory
, 230);
471 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
473 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
475 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
476 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
477 /* Mixed cgroup hierarchy support was added in 233 */
478 r
= systemd_installation_has_version(directory
, 233);
480 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
482 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
484 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
486 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
488 log_debug("Using %s hierarchy for container.",
489 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
490 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
495 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
500 _cleanup_free_
char *t
= NULL
;
502 r
= extract_first_word(&spec
, &t
, ",", 0);
504 return log_error_errno(r
, "Failed to parse capability %s.", t
);
508 if (streq(t
, "help")) {
509 for (int i
= 0; i
< capability_list_length(); i
++) {
512 name
= capability_to_name(i
);
521 mask
= (uint64_t) -1;
523 r
= capability_from_name(t
);
525 return log_error_errno(r
, "Failed to parse capability %s.", t
);
532 return 1; /* continue */
535 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
538 r
= getenv_bool(name
);
542 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
544 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
545 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
549 static int parse_mount_settings_env(void) {
553 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
554 if (r
< 0 && r
!= -ENXIO
)
555 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
557 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
559 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
560 if (streq_ptr(e
, "network"))
561 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
564 r
= parse_boolean(e
);
566 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
568 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
569 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
575 static int parse_environment(void) {
579 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
582 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
585 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
588 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
592 r
= parse_mount_settings_env();
596 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
597 * even if it is supported. If not supported, it has no effect. */
598 if (!cg_ns_supported())
599 arg_use_cgns
= false;
601 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
604 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
608 arg_use_cgns
= r
> 0;
609 arg_settings_mask
|= SETTING_USE_CGNS
;
613 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
615 arg_container_service_name
= e
;
617 return detect_unified_cgroup_hierarchy_from_environment();
620 static int parse_argv(int argc
, char *argv
[]) {
638 ARG_NETWORK_INTERFACE
,
643 ARG_NETWORK_VETH_EXTRA
,
644 ARG_NETWORK_NAMESPACE_PATH
,
654 ARG_PRIVATE_USERS_CHOWN
,
657 ARG_SYSTEM_CALL_FILTER
,
660 ARG_NO_NEW_PRIVILEGES
,
661 ARG_OOM_SCORE_ADJUST
,
672 static const struct option options
[] = {
673 { "help", no_argument
, NULL
, 'h' },
674 { "version", no_argument
, NULL
, ARG_VERSION
},
675 { "directory", required_argument
, NULL
, 'D' },
676 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
677 { "ephemeral", no_argument
, NULL
, 'x' },
678 { "user", required_argument
, NULL
, 'u' },
679 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
680 { "as-pid2", no_argument
, NULL
, 'a' },
681 { "boot", no_argument
, NULL
, 'b' },
682 { "uuid", required_argument
, NULL
, ARG_UUID
},
683 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
684 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
685 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
686 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
687 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
688 { "bind", required_argument
, NULL
, ARG_BIND
},
689 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
690 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
691 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
692 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
693 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
694 { "machine", required_argument
, NULL
, 'M' },
695 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
696 { "slice", required_argument
, NULL
, 'S' },
697 { "setenv", required_argument
, NULL
, 'E' },
698 { "selinux-context", required_argument
, NULL
, 'Z' },
699 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
700 { "quiet", no_argument
, NULL
, 'q' },
701 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
702 { "register", required_argument
, NULL
, ARG_REGISTER
},
703 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
704 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
705 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
706 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
707 { "network-veth", no_argument
, NULL
, 'n' },
708 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
709 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
710 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
711 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
712 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
713 { "image", required_argument
, NULL
, 'i' },
714 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
715 { "port", required_argument
, NULL
, 'p' },
716 { "property", required_argument
, NULL
, ARG_PROPERTY
},
717 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
718 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
719 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
720 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
721 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
722 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
723 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
724 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
725 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
726 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
727 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
728 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
729 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
730 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
731 { "console", required_argument
, NULL
, ARG_CONSOLE
},
732 { "pipe", no_argument
, NULL
, ARG_PIPE
},
733 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
734 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
735 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
740 uint64_t plus
= 0, minus
= 0;
741 bool mask_all_settings
= false, mask_no_settings
= false;
746 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
756 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
760 arg_settings_mask
|= SETTING_DIRECTORY
;
764 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
768 arg_settings_mask
|= SETTING_DIRECTORY
;
772 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
776 arg_settings_mask
|= SETTING_DIRECTORY
;
780 r
= parse_path_argument_and_warn(optarg
, false, &arg_oci_bundle
);
787 arg_ephemeral
= true;
788 arg_settings_mask
|= SETTING_EPHEMERAL
;
792 r
= free_and_strdup(&arg_user
, optarg
);
796 arg_settings_mask
|= SETTING_USER
;
799 case ARG_NETWORK_ZONE
: {
802 j
= strjoin("vz-", optarg
);
806 if (!ifname_valid(j
)) {
807 log_error("Network zone name not valid: %s", j
);
812 free_and_replace(arg_network_zone
, j
);
814 arg_network_veth
= true;
815 arg_private_network
= true;
816 arg_settings_mask
|= SETTING_NETWORK
;
820 case ARG_NETWORK_BRIDGE
:
822 if (!ifname_valid(optarg
))
823 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
824 "Bridge interface name not valid: %s", optarg
);
826 r
= free_and_strdup(&arg_network_bridge
, optarg
);
832 arg_network_veth
= true;
833 arg_private_network
= true;
834 arg_settings_mask
|= SETTING_NETWORK
;
837 case ARG_NETWORK_VETH_EXTRA
:
838 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
840 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
842 arg_private_network
= true;
843 arg_settings_mask
|= SETTING_NETWORK
;
846 case ARG_NETWORK_INTERFACE
:
847 if (!ifname_valid(optarg
))
848 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
849 "Network interface name not valid: %s", optarg
);
851 r
= test_network_interface_initialized(optarg
);
855 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
858 arg_private_network
= true;
859 arg_settings_mask
|= SETTING_NETWORK
;
862 case ARG_NETWORK_MACVLAN
:
864 if (!ifname_valid(optarg
))
865 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
866 "MACVLAN network interface name not valid: %s", optarg
);
868 r
= test_network_interface_initialized(optarg
);
872 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
875 arg_private_network
= true;
876 arg_settings_mask
|= SETTING_NETWORK
;
879 case ARG_NETWORK_IPVLAN
:
881 if (!ifname_valid(optarg
))
882 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
883 "IPVLAN network interface name not valid: %s", optarg
);
885 r
= test_network_interface_initialized(optarg
);
889 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
893 case ARG_PRIVATE_NETWORK
:
894 arg_private_network
= true;
895 arg_settings_mask
|= SETTING_NETWORK
;
898 case ARG_NETWORK_NAMESPACE_PATH
:
899 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
903 arg_settings_mask
|= SETTING_NETWORK
;
907 if (arg_start_mode
== START_PID2
)
908 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
909 "--boot and --as-pid2 may not be combined.");
911 arg_start_mode
= START_BOOT
;
912 arg_settings_mask
|= SETTING_START_MODE
;
916 if (arg_start_mode
== START_BOOT
)
917 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
918 "--boot and --as-pid2 may not be combined.");
920 arg_start_mode
= START_PID2
;
921 arg_settings_mask
|= SETTING_START_MODE
;
925 r
= sd_id128_from_string(optarg
, &arg_uuid
);
927 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
929 if (sd_id128_is_null(arg_uuid
))
930 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
931 "Machine UUID may not be all zeroes.");
933 arg_settings_mask
|= SETTING_MACHINE_ID
;
937 _cleanup_free_
char *mangled
= NULL
;
939 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
943 free_and_replace(arg_slice
, mangled
);
944 arg_settings_mask
|= SETTING_SLICE
;
950 arg_machine
= mfree(arg_machine
);
952 if (!machine_name_is_valid(optarg
))
953 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
954 "Invalid machine name: %s", optarg
);
956 r
= free_and_strdup(&arg_machine
, optarg
);
964 arg_hostname
= mfree(arg_hostname
);
966 if (!hostname_is_valid(optarg
, false))
967 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
968 "Invalid hostname: %s", optarg
);
970 r
= free_and_strdup(&arg_hostname
, optarg
);
975 arg_settings_mask
|= SETTING_HOSTNAME
;
979 arg_selinux_context
= optarg
;
983 arg_selinux_apifs_context
= optarg
;
987 arg_read_only
= true;
988 arg_settings_mask
|= SETTING_READ_ONLY
;
992 case ARG_DROP_CAPABILITY
: {
994 r
= parse_capability_spec(optarg
, &m
);
998 if (c
== ARG_CAPABILITY
)
1002 arg_settings_mask
|= SETTING_CAPABILITY
;
1005 case ARG_NO_NEW_PRIVILEGES
:
1006 r
= parse_boolean(optarg
);
1008 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1010 arg_no_new_privileges
= r
;
1011 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1015 arg_link_journal
= LINK_GUEST
;
1016 arg_link_journal_try
= true;
1017 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1020 case ARG_LINK_JOURNAL
:
1021 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1023 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1025 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1030 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1032 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1034 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1038 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1040 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1042 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1046 case ARG_OVERLAY_RO
:
1047 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1048 if (r
== -EADDRNOTAVAIL
)
1049 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1051 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1053 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1056 case ARG_INACCESSIBLE
:
1057 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1059 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1061 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1067 if (!env_assignment_is_valid(optarg
))
1068 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1069 "Environment variable assignment '%s' is not valid.", optarg
);
1071 n
= strv_env_set(arg_setenv
, optarg
);
1075 strv_free_and_replace(arg_setenv
, n
);
1076 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1084 case ARG_SHARE_SYSTEM
:
1085 /* We don't officially support this anymore, except for compat reasons. People should use the
1086 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1087 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1088 arg_clone_ns_flags
= 0;
1092 r
= parse_boolean(optarg
);
1094 log_error("Failed to parse --register= argument: %s", optarg
);
1102 arg_keep_unit
= true;
1105 case ARG_PERSONALITY
:
1107 arg_personality
= personality_from_string(optarg
);
1108 if (arg_personality
== PERSONALITY_INVALID
)
1109 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1110 "Unknown or unsupported personality '%s'.", optarg
);
1112 arg_settings_mask
|= SETTING_PERSONALITY
;
1118 arg_volatile_mode
= VOLATILE_YES
;
1119 else if (streq(optarg
, "help")) {
1120 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1125 m
= volatile_mode_from_string(optarg
);
1127 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1128 "Failed to parse --volatile= argument: %s", optarg
);
1130 arg_volatile_mode
= m
;
1133 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1137 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1139 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1141 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1143 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1147 if (strv_extend(&arg_property
, optarg
) < 0)
1152 case ARG_PRIVATE_USERS
: {
1157 else if (!in_charset(optarg
, DIGITS
))
1158 /* do *not* parse numbers as booleans */
1159 boolean
= parse_boolean(optarg
);
1161 if (boolean
== false) {
1162 /* no: User namespacing off */
1163 arg_userns_mode
= USER_NAMESPACE_NO
;
1164 arg_uid_shift
= UID_INVALID
;
1165 arg_uid_range
= UINT32_C(0x10000);
1166 } else if (boolean
== true) {
1167 /* yes: User namespacing on, UID range is read from root dir */
1168 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1169 arg_uid_shift
= UID_INVALID
;
1170 arg_uid_range
= UINT32_C(0x10000);
1171 } else if (streq(optarg
, "pick")) {
1172 /* pick: User namespacing on, UID range is picked randomly */
1173 arg_userns_mode
= USER_NAMESPACE_PICK
;
1174 arg_uid_shift
= UID_INVALID
;
1175 arg_uid_range
= UINT32_C(0x10000);
1177 _cleanup_free_
char *buffer
= NULL
;
1178 const char *range
, *shift
;
1180 /* anything else: User namespacing on, UID range is explicitly configured */
1182 range
= strchr(optarg
, ':');
1184 buffer
= strndup(optarg
, range
- optarg
);
1190 r
= safe_atou32(range
, &arg_uid_range
);
1192 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1196 r
= parse_uid(shift
, &arg_uid_shift
);
1198 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1200 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1203 if (arg_uid_range
<= 0)
1204 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1205 "UID range cannot be 0.");
1207 arg_settings_mask
|= SETTING_USERNS
;
1212 if (userns_supported()) {
1213 arg_userns_mode
= USER_NAMESPACE_PICK
;
1214 arg_uid_shift
= UID_INVALID
;
1215 arg_uid_range
= UINT32_C(0x10000);
1217 arg_settings_mask
|= SETTING_USERNS
;
1222 case ARG_PRIVATE_USERS_CHOWN
:
1223 arg_userns_chown
= true;
1225 arg_settings_mask
|= SETTING_USERNS
;
1228 case ARG_KILL_SIGNAL
:
1229 if (streq(optarg
, "help")) {
1230 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1234 arg_kill_signal
= signal_from_string(optarg
);
1235 if (arg_kill_signal
< 0)
1236 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1237 "Cannot parse signal: %s", optarg
);
1239 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1244 /* no → do not read files
1245 * yes → read files, do not override cmdline, trust only subset
1246 * override → read files, override cmdline, trust only subset
1247 * trusted → read files, do not override cmdline, trust all
1250 r
= parse_boolean(optarg
);
1252 if (streq(optarg
, "trusted")) {
1253 mask_all_settings
= false;
1254 mask_no_settings
= false;
1255 arg_settings_trusted
= true;
1257 } else if (streq(optarg
, "override")) {
1258 mask_all_settings
= false;
1259 mask_no_settings
= true;
1260 arg_settings_trusted
= -1;
1262 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1265 mask_all_settings
= false;
1266 mask_no_settings
= false;
1267 arg_settings_trusted
= -1;
1270 mask_all_settings
= true;
1271 mask_no_settings
= false;
1272 arg_settings_trusted
= false;
1278 if (!path_is_absolute(optarg
))
1279 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1280 "Working directory %s is not an absolute path.", optarg
);
1282 r
= free_and_strdup(&arg_chdir
, optarg
);
1286 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1289 case ARG_PIVOT_ROOT
:
1290 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1292 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1294 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1297 case ARG_NOTIFY_READY
:
1298 r
= parse_boolean(optarg
);
1300 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1301 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1302 arg_notify_ready
= r
;
1303 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1306 case ARG_ROOT_HASH
: {
1310 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1312 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1313 if (l
< sizeof(sd_id128_t
)) {
1315 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1318 free(arg_root_hash
);
1320 arg_root_hash_size
= l
;
1324 case ARG_VERITY_DATA
:
1325 r
= parse_path_argument_and_warn(optarg
, false, &arg_verity_data
);
1330 case ARG_SYSTEM_CALL_FILTER
: {
1334 negative
= optarg
[0] == '~';
1335 items
= negative
? optarg
+ 1 : optarg
;
1338 _cleanup_free_
char *word
= NULL
;
1340 r
= extract_first_word(&items
, &word
, NULL
, 0);
1346 return log_error_errno(r
, "Failed to parse system call filter: %m");
1349 r
= strv_extend(&arg_syscall_deny_list
, word
);
1351 r
= strv_extend(&arg_syscall_allow_list
, word
);
1356 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1362 _cleanup_free_
char *name
= NULL
;
1365 if (streq(optarg
, "help")) {
1366 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1370 eq
= strchr(optarg
, '=');
1372 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1373 "--rlimit= expects an '=' assignment.");
1375 name
= strndup(optarg
, eq
- optarg
);
1379 rl
= rlimit_from_string_harder(name
);
1381 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1382 "Unknown resource limit: %s", name
);
1384 if (!arg_rlimit
[rl
]) {
1385 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1386 if (!arg_rlimit
[rl
])
1390 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1392 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1394 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1398 case ARG_OOM_SCORE_ADJUST
:
1399 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1401 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1403 arg_oom_score_adjust_set
= true;
1404 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1407 case ARG_CPU_AFFINITY
: {
1410 r
= parse_cpu_set(optarg
, &cpuset
);
1412 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1414 cpu_set_reset(&arg_cpu_set
);
1415 arg_cpu_set
= cpuset
;
1416 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1420 case ARG_RESOLV_CONF
:
1421 if (streq(optarg
, "help")) {
1422 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1426 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1427 if (arg_resolv_conf
< 0)
1428 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1429 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1431 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1435 if (streq(optarg
, "help")) {
1436 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1440 arg_timezone
= timezone_mode_from_string(optarg
);
1441 if (arg_timezone
< 0)
1442 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1443 "Failed to parse /etc/localtime mode: %s", optarg
);
1445 arg_settings_mask
|= SETTING_TIMEZONE
;
1449 r
= handle_arg_console(optarg
);
1456 r
= handle_arg_console("pipe");
1462 arg_pager_flags
|= PAGER_DISABLE
;
1469 assert_not_reached("Unhandled option");
1472 if (argc
> optind
) {
1473 strv_free(arg_parameters
);
1474 arg_parameters
= strv_copy(argv
+ optind
);
1475 if (!arg_parameters
)
1478 arg_settings_mask
|= SETTING_START_MODE
;
1481 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1482 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1483 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1484 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1486 arg_directory
= TAKE_PTR(arg_template
);
1488 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1490 /* Make sure to parse environment before we reset the settings mask below */
1491 r
= parse_environment();
1495 /* Load all settings from .nspawn files */
1496 if (mask_no_settings
)
1497 arg_settings_mask
= 0;
1499 /* Don't load any settings from .nspawn files */
1500 if (mask_all_settings
)
1501 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1506 static int verify_arguments(void) {
1509 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1510 /* If we are running the stub init in the container, we don't need to look at what the init
1511 * in the container supports, because we are not using it. Let's immediately pick the right
1512 * setting based on the host system configuration.
1514 * We only do this, if the user didn't use an environment variable to override the detection.
1517 r
= cg_all_unified();
1519 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1521 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1522 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1523 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1525 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1528 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1529 arg_mount_settings
|= MOUNT_USE_USERNS
;
1531 if (arg_private_network
)
1532 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1534 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1535 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1536 arg_register
= false;
1537 if (arg_start_mode
!= START_PID1
)
1538 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1541 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1542 arg_userns_chown
= true;
1544 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1545 arg_kill_signal
= SIGRTMIN
+3;
1547 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 */
1548 arg_read_only
= true;
1550 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1551 arg_read_only
= true;
1553 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1554 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1555 * The latter is not technically a user session, but we don't need to labour the point. */
1556 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1558 if (arg_directory
&& arg_image
)
1559 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1561 if (arg_template
&& arg_image
)
1562 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1564 if (arg_template
&& !(arg_directory
|| arg_machine
))
1565 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1567 if (arg_ephemeral
&& arg_template
)
1568 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1570 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1571 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1573 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1574 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1576 if (arg_userns_chown
&& arg_read_only
)
1577 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1578 "--read-only and --private-users-chown may not be combined.");
1580 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1581 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1582 * copy-up (in case of overlay) making the entire exercise pointless. */
1583 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1584 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1586 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1587 * we need to error out, to avoid conflicts between different network options. */
1588 if (arg_network_namespace_path
&&
1589 (arg_network_interfaces
|| arg_network_macvlan
||
1590 arg_network_ipvlan
|| arg_network_veth_extra
||
1591 arg_network_bridge
|| arg_network_zone
||
1593 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1595 if (arg_network_bridge
&& arg_network_zone
)
1596 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1597 "--network-bridge= and --network-zone= may not be combined.");
1599 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1600 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1602 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1603 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1605 if (arg_expose_ports
&& !arg_private_network
)
1606 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1609 if (arg_expose_ports
)
1610 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1613 r
= custom_mount_check_all();
1620 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1623 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1626 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1629 if (uid
!= UID_INVALID
) {
1630 uid
+= arg_uid_shift
;
1632 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1636 if (gid
!= GID_INVALID
) {
1637 gid
+= (gid_t
) arg_uid_shift
;
1639 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1643 if (lchown(p
, uid
, gid
) < 0)
1649 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1653 q
= prefix_roota(root
, path
);
1654 r
= mkdir_errno_wrapper(q
, mode
);
1660 return userns_lchown(q
, uid
, gid
);
1663 static const char *timezone_from_path(const char *path
) {
1664 return PATH_STARTSWITH_SET(
1666 "../usr/share/zoneinfo/",
1667 "/usr/share/zoneinfo/");
1670 static bool etc_writable(void) {
1671 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1674 static int setup_timezone(const char *dest
) {
1675 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1676 const char *where
, *check
;
1682 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1683 r
= readlink_malloc("/etc/localtime", &p
);
1684 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1685 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1686 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1687 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1689 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1690 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1694 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1697 } else if (arg_timezone
== TIMEZONE_AUTO
)
1698 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1704 if (m
== TIMEZONE_OFF
)
1707 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1709 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1713 where
= strjoina(etc
, "/localtime");
1717 case TIMEZONE_DELETE
:
1718 if (unlink(where
) < 0)
1719 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1723 case TIMEZONE_SYMLINK
: {
1724 _cleanup_free_
char *q
= NULL
;
1725 const char *z
, *what
;
1727 z
= timezone_from_path(p
);
1729 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1733 r
= readlink_malloc(where
, &q
);
1734 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1735 return 0; /* Already pointing to the right place? Then do nothing .. */
1737 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1738 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1740 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1742 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1743 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1744 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1748 what
= strjoina("../usr/share/zoneinfo/", z
);
1749 if (symlink(what
, where
) < 0) {
1750 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1751 errno
, "Failed to correct timezone of container, ignoring: %m");
1761 case TIMEZONE_BIND
: {
1762 _cleanup_free_
char *resolved
= NULL
;
1765 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1767 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1771 if (found
== 0) /* missing? */
1772 (void) touch(resolved
);
1774 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1776 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1782 /* If mounting failed, try to copy */
1783 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1785 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1786 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1793 assert_not_reached("unexpected mode");
1796 /* Fix permissions of the symlink or file copy we just created */
1797 r
= userns_lchown(where
, 0, 0);
1799 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1804 static int have_resolv_conf(const char *path
) {
1807 if (access(path
, F_OK
) < 0) {
1808 if (errno
== ENOENT
)
1811 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1817 static int resolved_listening(void) {
1818 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1819 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1820 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1823 /* Check if resolved is listening */
1825 r
= sd_bus_open_system(&bus
);
1827 return log_debug_errno(r
, "Failed to open system bus: %m");
1829 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1831 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1835 r
= sd_bus_get_property_string(bus
,
1836 "org.freedesktop.resolve1",
1837 "/org/freedesktop/resolve1",
1838 "org.freedesktop.resolve1.Manager",
1841 &dns_stub_listener_mode
);
1843 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1845 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1848 static int setup_resolv_conf(const char *dest
) {
1849 _cleanup_free_
char *etc
= NULL
;
1850 const char *where
, *what
;
1856 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1857 if (arg_private_network
)
1858 m
= RESOLV_CONF_OFF
;
1859 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1860 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
1861 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1862 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1864 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1867 m
= arg_resolv_conf
;
1869 if (m
== RESOLV_CONF_OFF
)
1872 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1874 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1878 where
= strjoina(etc
, "/resolv.conf");
1880 if (m
== RESOLV_CONF_DELETE
) {
1881 if (unlink(where
) < 0)
1882 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1887 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
1888 what
= PRIVATE_STATIC_RESOLV_CONF
;
1889 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
1890 what
= PRIVATE_UPLINK_RESOLV_CONF
;
1891 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
1892 what
= PRIVATE_STUB_RESOLV_CONF
;
1894 what
= "/etc/resolv.conf";
1896 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
1897 _cleanup_free_
char *resolved
= NULL
;
1900 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1902 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1906 if (found
== 0) /* missing? */
1907 (void) touch(resolved
);
1909 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1911 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1913 /* If that didn't work, let's copy the file */
1916 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
1917 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1919 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
1921 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1922 * resolved or something similar runs inside and the symlink points there.
1924 * If the disk image is read-only, there's also no point in complaining.
1926 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
1927 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1928 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1932 r
= userns_lchown(where
, 0, 0);
1934 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1939 static int setup_boot_id(void) {
1940 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1941 _cleanup_free_
char *path
= NULL
;
1942 sd_id128_t rnd
= SD_ID128_NULL
;
1946 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1948 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
1950 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1952 r
= sd_id128_randomize(&rnd
);
1954 return log_error_errno(r
, "Failed to generate random boot id: %m");
1956 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1958 return log_error_errno(r
, "Failed to write boot id: %m");
1960 from
= TAKE_PTR(path
);
1961 to
= "/proc/sys/kernel/random/boot_id";
1963 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1967 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1970 static int copy_devnodes(const char *dest
) {
1971 static const char devnodes
[] =
1980 _cleanup_umask_ mode_t u
;
1988 /* Create /dev/net, so that we can create /dev/net/tun in it */
1989 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1990 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1992 NULSTR_FOREACH(d
, devnodes
) {
1993 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1996 from
= path_join("/dev/", d
);
2000 to
= path_join(dest
, from
);
2004 if (stat(from
, &st
) < 0) {
2006 if (errno
!= ENOENT
)
2007 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2009 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2010 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2011 "%s is not a char or block device, cannot copy.", from
);
2013 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2015 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2016 /* Explicitly warn the user when /dev is already populated. */
2017 if (errno
== EEXIST
)
2018 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
);
2020 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2022 /* Some systems abusively restrict mknod but allow bind mounts. */
2025 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2026 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2028 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2031 r
= userns_lchown(to
, 0, 0);
2033 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2035 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2039 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2041 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2043 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2046 prefixed
= path_join(dest
, sl
);
2050 t
= path_join("..", d
);
2054 if (symlink(t
, prefixed
) < 0)
2055 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2062 static int make_extra_nodes(const char *dest
) {
2063 _cleanup_umask_ mode_t u
;
2069 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2070 _cleanup_free_
char *path
= NULL
;
2071 DeviceNode
*n
= arg_extra_nodes
+ i
;
2073 path
= path_join(dest
, n
->path
);
2077 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2078 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2080 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2082 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2088 static int setup_pts(const char *dest
) {
2089 _cleanup_free_
char *options
= NULL
;
2094 if (arg_selinux_apifs_context
)
2095 (void) asprintf(&options
,
2096 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2097 arg_uid_shift
+ TTY_GID
,
2098 arg_selinux_apifs_context
);
2101 (void) asprintf(&options
,
2102 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2103 arg_uid_shift
+ TTY_GID
);
2108 /* Mount /dev/pts itself */
2109 p
= prefix_roota(dest
, "/dev/pts");
2110 r
= mkdir_errno_wrapper(p
, 0755);
2112 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2114 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2117 r
= userns_lchown(p
, 0, 0);
2119 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2121 /* Create /dev/ptmx symlink */
2122 p
= prefix_roota(dest
, "/dev/ptmx");
2123 if (symlink("pts/ptmx", p
) < 0)
2124 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2125 r
= userns_lchown(p
, 0, 0);
2127 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2129 /* And fix /dev/pts/ptmx ownership */
2130 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2131 r
= userns_lchown(p
, 0, 0);
2133 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2138 static int setup_stdio_as_dev_console(void) {
2142 terminal
= open_terminal("/dev/console", O_RDWR
);
2144 return log_error_errno(terminal
, "Failed to open console: %m");
2146 /* Make sure we can continue logging to the original stderr, even if
2147 * stderr points elsewhere now */
2148 r
= log_dup_console();
2150 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2152 /* invalidates 'terminal' on success and failure */
2153 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2155 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2160 static int setup_dev_console(const char *console
) {
2161 _cleanup_free_
char *p
= NULL
;
2164 /* Create /dev/console symlink */
2165 r
= path_make_relative("/dev", console
, &p
);
2167 return log_error_errno(r
, "Failed to create relative path: %m");
2169 if (symlink(p
, "/dev/console") < 0)
2170 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2175 static int setup_keyring(void) {
2176 key_serial_t keyring
;
2178 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2179 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2180 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2181 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2182 * into the container. */
2184 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2185 if (keyring
== -1) {
2186 if (errno
== ENOSYS
)
2187 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2188 else if (IN_SET(errno
, EACCES
, EPERM
))
2189 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2191 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2197 static int setup_kmsg(int kmsg_socket
) {
2198 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2199 _cleanup_free_
char *fifo
= NULL
;
2200 _cleanup_close_
int fd
= -1;
2201 _cleanup_umask_ mode_t u
;
2204 assert(kmsg_socket
>= 0);
2208 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2209 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2210 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2211 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2213 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2215 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2217 if (mkfifo(fifo
, 0600) < 0)
2218 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2220 from
= TAKE_PTR(fifo
);
2222 r
= mount_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2226 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2228 return log_error_errno(errno
, "Failed to open fifo: %m");
2230 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2231 r
= send_one_fd(kmsg_socket
, fd
, 0);
2233 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2238 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2239 union in_addr_union
*exposed
= userdata
;
2245 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2249 static int setup_hostname(void) {
2252 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2255 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2257 return log_error_errno(r
, "Failed to set hostname: %m");
2262 static int setup_journal(const char *directory
) {
2263 _cleanup_free_
char *d
= NULL
;
2264 char id
[SD_ID128_STRING_MAX
];
2265 const char *dirname
, *p
, *q
;
2270 /* Don't link journals in ephemeral mode */
2274 if (arg_link_journal
== LINK_NO
)
2277 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2279 r
= sd_id128_get_machine(&this_id
);
2281 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2283 if (sd_id128_equal(arg_uuid
, this_id
)) {
2284 log_full(try ? LOG_WARNING
: LOG_ERR
,
2285 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2291 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2292 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2294 bool ignore
= r
== -EROFS
&& try;
2295 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2296 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2297 return ignore
? 0 : r
;
2301 (void) sd_id128_to_string(arg_uuid
, id
);
2303 p
= strjoina("/var/log/journal/", id
);
2304 q
= prefix_roota(directory
, p
);
2306 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2310 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2311 "%s: already a mount point, refusing to use for journal", p
);
2314 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2318 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2319 "%s: already a mount point, refusing to use for journal", q
);
2322 r
= readlink_and_make_absolute(p
, &d
);
2324 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2327 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2329 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2334 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2335 } else if (r
== -EINVAL
) {
2337 if (arg_link_journal
== LINK_GUEST
&&
2340 if (errno
== ENOTDIR
) {
2341 log_error("%s already exists and is neither a symlink nor a directory", p
);
2344 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2346 } else if (r
!= -ENOENT
)
2347 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2349 if (arg_link_journal
== LINK_GUEST
) {
2351 if (symlink(q
, p
) < 0) {
2353 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2356 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2359 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2361 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2365 if (arg_link_journal
== LINK_HOST
) {
2366 /* don't create parents here — if the host doesn't have
2367 * permanent journal set up, don't force it here */
2369 r
= mkdir_errno_wrapper(p
, 0755);
2370 if (r
< 0 && r
!= -EEXIST
) {
2372 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2375 return log_error_errno(r
, "Failed to create %s: %m", p
);
2378 } else if (access(p
, F_OK
) < 0)
2381 if (dir_is_empty(q
) == 0)
2382 log_warning("%s is not empty, proceeding anyway.", q
);
2384 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2386 return log_error_errno(r
, "Failed to create %s: %m", q
);
2388 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2390 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2395 static int drop_capabilities(uid_t uid
) {
2396 CapabilityQuintet q
;
2398 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2399 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2400 * arg_caps_retain. */
2402 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2403 q
= arg_full_capabilities
;
2405 if (q
.bounding
== (uint64_t) -1)
2406 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2408 if (q
.effective
== (uint64_t) -1)
2409 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2411 if (q
.inheritable
== (uint64_t) -1)
2412 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2414 if (q
.permitted
== (uint64_t) -1)
2415 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2417 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2420 if (capability_quintet_mangle(&q
))
2421 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2424 q
= (CapabilityQuintet
) {
2425 .bounding
= arg_caps_retain
,
2426 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2427 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2428 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2429 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2432 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2433 * in order to maintain the same behavior as systemd < 242. */
2434 if (capability_quintet_mangle(&q
))
2435 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2436 "Some capabilities will not be set because they are not in the current bounding set.");
2440 return capability_quintet_enforce(&q
);
2443 static int reset_audit_loginuid(void) {
2444 _cleanup_free_
char *p
= NULL
;
2447 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2450 r
= read_one_line_file("/proc/self/loginuid", &p
);
2454 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2456 /* Already reset? */
2457 if (streq(p
, "4294967295"))
2460 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2463 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2464 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2465 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2466 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2467 "using systemd-nspawn. Sleeping for 5s... (%m)");
2475 static int setup_propagate(const char *root
) {
2479 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2480 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2481 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2482 (void) mkdir_p(p
, 0600);
2484 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2486 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2488 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2490 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2492 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2494 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2496 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2497 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2501 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2505 /* machined will MS_MOVE into that directory, and that's only
2506 * supported for non-shared mounts. */
2507 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2510 static int setup_machine_id(const char *directory
) {
2511 const char *etc_machine_id
;
2515 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2516 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2517 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2518 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2519 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2520 * container behaves nicely). */
2522 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2524 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2526 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2527 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2529 if (sd_id128_is_null(arg_uuid
)) {
2530 r
= sd_id128_randomize(&arg_uuid
);
2532 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2535 if (sd_id128_is_null(id
))
2536 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2537 "Machine ID in container image is zero, refusing.");
2545 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2550 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2553 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2554 if (r
== -EOPNOTSUPP
)
2555 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.");
2557 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2559 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2561 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2563 log_debug("Patched directory tree to match UID/GID range.");
2570 * < 0 : wait_for_terminate() failed to get the state of the
2571 * container, the container was terminated by a signal, or
2572 * failed for an unknown reason. No change is made to the
2573 * container argument.
2574 * > 0 : The program executed in the container terminated with an
2575 * error. The exit code of the program executed in the
2576 * container is returned. The container argument has been set
2577 * to CONTAINER_TERMINATED.
2578 * 0 : The container is being rebooted, has been shut down or exited
2579 * successfully. The container argument has been set to either
2580 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2582 * That is, success is indicated by a return value of zero, and an
2583 * error is indicated by a non-zero value.
2585 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2589 r
= wait_for_terminate(pid
, &status
);
2591 return log_warning_errno(r
, "Failed to wait for container: %m");
2593 switch (status
.si_code
) {
2596 if (status
.si_status
== 0)
2597 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2599 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2601 *container
= CONTAINER_TERMINATED
;
2602 return status
.si_status
;
2605 if (status
.si_status
== SIGINT
) {
2606 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2607 *container
= CONTAINER_TERMINATED
;
2610 } else if (status
.si_status
== SIGHUP
) {
2611 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2612 *container
= CONTAINER_REBOOTED
;
2618 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2619 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2622 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2623 "Container %s failed due to unknown reason.", arg_machine
);
2627 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2630 pid
= PTR_TO_PID(userdata
);
2632 if (kill(pid
, arg_kill_signal
) >= 0) {
2633 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2634 sd_event_source_set_userdata(s
, NULL
);
2639 sd_event_exit(sd_event_source_get_event(s
), 0);
2643 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2649 pid
= PTR_TO_PID(userdata
);
2654 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2655 return log_error_errno(errno
, "Failed to waitid(): %m");
2656 if (si
.si_pid
== 0) /* No pending children. */
2658 if (si
.si_pid
== pid
) {
2659 /* The main process we care for has exited. Return from
2660 * signal handler but leave the zombie. */
2661 sd_event_exit(sd_event_source_get_event(s
), 0);
2665 /* Reap all other children. */
2666 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2672 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2677 pid
= PTR_TO_PID(userdata
);
2679 if (arg_kill_signal
> 0) {
2680 log_info("Container termination requested. Attempting to halt container.");
2681 (void) kill(pid
, arg_kill_signal
);
2683 log_info("Container termination requested. Exiting.");
2684 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2690 static int determine_names(void) {
2693 if (arg_template
&& !arg_directory
&& arg_machine
) {
2695 /* If --template= was specified then we should not
2696 * search for a machine, but instead create a new one
2697 * in /var/lib/machine. */
2699 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2704 if (!arg_image
&& !arg_directory
) {
2706 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2708 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2710 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2712 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2714 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2715 r
= free_and_strdup(&arg_image
, i
->path
);
2717 r
= free_and_strdup(&arg_directory
, i
->path
);
2722 arg_read_only
= arg_read_only
|| i
->read_only
;
2724 r
= safe_getcwd(&arg_directory
);
2726 return log_error_errno(r
, "Failed to determine current directory: %m");
2729 if (!arg_directory
&& !arg_image
)
2730 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2734 if (arg_directory
&& path_equal(arg_directory
, "/"))
2735 arg_machine
= gethostname_malloc();
2740 arg_machine
= strdup(basename(arg_image
));
2742 /* Truncate suffix if there is one */
2743 e
= endswith(arg_machine
, ".raw");
2747 arg_machine
= strdup(basename(arg_directory
));
2752 hostname_cleanup(arg_machine
);
2753 if (!machine_name_is_valid(arg_machine
))
2754 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2756 if (arg_ephemeral
) {
2759 /* Add a random suffix when this is an
2760 * ephemeral machine, so that we can run many
2761 * instances at once without manually having
2762 * to specify -M each time. */
2764 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2775 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2784 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
2786 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2788 return free_and_replace(*p
, chased
);
2791 static int determine_uid_shift(const char *directory
) {
2794 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2799 if (arg_uid_shift
== UID_INVALID
) {
2802 r
= stat(directory
, &st
);
2804 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2806 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2808 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2809 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2810 "UID and GID base of %s don't match.", directory
);
2812 arg_uid_range
= UINT32_C(0x10000);
2815 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2816 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2817 "UID base too high for UID range.");
2822 static unsigned long effective_clone_ns_flags(void) {
2823 unsigned long flags
= arg_clone_ns_flags
;
2825 if (arg_private_network
)
2826 flags
|= CLONE_NEWNET
;
2828 flags
|= CLONE_NEWCGROUP
;
2829 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2830 flags
|= CLONE_NEWUSER
;
2835 static int patch_sysctl(void) {
2837 /* This table is inspired by runc's sysctl() function */
2838 static const struct {
2841 unsigned long clone_flags
;
2843 { "kernel.hostname", false, CLONE_NEWUTS
},
2844 { "kernel.domainname", false, CLONE_NEWUTS
},
2845 { "kernel.msgmax", false, CLONE_NEWIPC
},
2846 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2847 { "kernel.msgmni", false, CLONE_NEWIPC
},
2848 { "kernel.sem", false, CLONE_NEWIPC
},
2849 { "kernel.shmall", false, CLONE_NEWIPC
},
2850 { "kernel.shmmax", false, CLONE_NEWIPC
},
2851 { "kernel.shmmni", false, CLONE_NEWIPC
},
2852 { "fs.mqueue.", true, CLONE_NEWIPC
},
2853 { "net.", true, CLONE_NEWNET
},
2856 unsigned long flags
;
2860 flags
= effective_clone_ns_flags();
2862 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2866 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2868 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2871 if (safe_sysctl
[i
].prefix
)
2872 good
= startswith(*k
, safe_sysctl
[i
].key
);
2874 good
= streq(*k
, safe_sysctl
[i
].key
);
2881 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2883 r
= sysctl_write(*k
, *v
);
2885 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2891 static int inner_child(
2893 const char *directory
,
2897 int master_pty_socket
,
2900 _cleanup_free_
char *home
= NULL
;
2901 char as_uuid
[ID128_UUID_STRING_MAX
];
2903 const char *envp
[] = {
2904 "PATH=" DEFAULT_PATH_COMPAT
,
2905 NULL
, /* container */
2910 NULL
, /* container_uuid */
2911 NULL
, /* LISTEN_FDS */
2912 NULL
, /* LISTEN_PID */
2913 NULL
, /* NOTIFY_SOCKET */
2916 const char *exec_target
;
2917 _cleanup_strv_free_
char **env_use
= NULL
;
2918 int r
, which_failed
;
2920 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2921 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2922 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2923 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2924 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2925 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2928 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2929 * unshare(). See below. */
2933 assert(kmsg_socket
>= 0);
2935 log_debug("Inner child is initializing.");
2937 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2938 /* Tell the parent, that it now can write the UID map. */
2939 (void) barrier_place(barrier
); /* #1 */
2941 /* Wait until the parent wrote the UID map */
2942 if (!barrier_place_and_sync(barrier
)) /* #2 */
2943 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2944 "Parent died too early");
2947 r
= reset_uid_gid();
2949 return log_error_errno(r
, "Couldn't become new root: %m");
2952 arg_mount_settings
| MOUNT_IN_USERNS
,
2954 arg_selinux_apifs_context
);
2958 if (!arg_network_namespace_path
&& arg_private_network
) {
2959 r
= unshare(CLONE_NEWNET
);
2961 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2963 /* Tell the parent that it can setup network interfaces. */
2964 (void) barrier_place(barrier
); /* #3 */
2967 r
= mount_sysfs(NULL
, arg_mount_settings
);
2971 /* Wait until we are cgroup-ified, so that we
2972 * can mount the right cgroup path writable */
2973 if (!barrier_place_and_sync(barrier
)) /* #4 */
2974 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2975 "Parent died too early");
2978 r
= unshare(CLONE_NEWCGROUP
);
2980 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2983 arg_unified_cgroup_hierarchy
,
2984 arg_userns_mode
!= USER_NAMESPACE_NO
,
2987 arg_selinux_apifs_context
,
2990 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2994 r
= setup_boot_id();
2998 r
= setup_kmsg(kmsg_socket
);
3001 kmsg_socket
= safe_close(kmsg_socket
);
3006 arg_n_custom_mounts
,
3008 arg_selinux_apifs_context
,
3009 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3014 return log_error_errno(errno
, "setsid() failed: %m");
3016 if (arg_private_network
)
3017 (void) loopback_setup();
3019 if (arg_expose_ports
) {
3020 r
= expose_port_send_rtnl(rtnl_socket
);
3023 rtnl_socket
= safe_close(rtnl_socket
);
3026 if (arg_console_mode
!= CONSOLE_PIPE
) {
3027 _cleanup_close_
int master
= -1;
3028 _cleanup_free_
char *console
= NULL
;
3030 /* Allocate a pty and make it available as /dev/console. */
3031 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3033 return log_error_errno(master
, "Failed to allocate a pty: %m");
3035 r
= setup_dev_console(console
);
3037 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3039 r
= send_one_fd(master_pty_socket
, master
, 0);
3041 return log_error_errno(r
, "Failed to send master fd: %m");
3042 master_pty_socket
= safe_close(master_pty_socket
);
3044 r
= setup_stdio_as_dev_console();
3053 if (arg_oom_score_adjust_set
) {
3054 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3056 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3059 if (arg_cpu_set
.set
)
3060 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3061 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3063 (void) setup_hostname();
3065 if (arg_personality
!= PERSONALITY_INVALID
) {
3066 r
= safe_personality(arg_personality
);
3068 return log_error_errno(r
, "personality() failed: %m");
3069 } else if (secondary
) {
3070 r
= safe_personality(PER_LINUX32
);
3072 return log_error_errno(r
, "personality() failed: %m");
3075 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3077 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3082 if (is_seccomp_available()) {
3084 r
= seccomp_load(arg_seccomp
);
3085 if (ERRNO_IS_SECCOMP_FATAL(r
))
3086 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3088 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3093 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3099 if (arg_selinux_context
)
3100 if (setexeccon(arg_selinux_context
) < 0)
3101 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3104 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3105 * if we need to later on. */
3106 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3107 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3109 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3110 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
3112 r
= change_uid_gid(arg_user
, &home
);
3116 r
= drop_capabilities(getuid());
3118 return log_error_errno(r
, "Dropping capabilities failed: %m");
3120 if (arg_no_new_privileges
)
3121 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3122 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3124 /* LXC sets container=lxc, so follow the scheme here */
3125 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3127 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3131 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3132 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3135 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3136 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3137 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3140 assert(!sd_id128_is_null(arg_uuid
));
3142 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3145 if (fdset_size(fds
) > 0) {
3146 r
= fdset_cloexec(fds
, false);
3148 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3150 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3151 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3154 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3157 env_use
= strv_env_merge(2, envp
, arg_setenv
);
3161 /* Let the parent know that we are ready and
3162 * wait until the parent is ready with the
3164 if (!barrier_place_and_sync(barrier
)) /* #5 */
3165 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3166 "Parent died too early");
3169 if (chdir(arg_chdir
) < 0)
3170 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3172 if (arg_start_mode
== START_PID2
) {
3173 r
= stub_pid1(arg_uuid
);
3178 log_debug("Inner child completed, invoking payload.");
3180 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3181 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3182 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3184 log_set_open_when_needed(true);
3186 (void) fdset_close_others(fds
);
3188 if (arg_start_mode
== START_BOOT
) {
3192 /* Automatically search for the init system */
3194 m
= strv_length(arg_parameters
);
3195 a
= newa(char*, m
+ 2);
3196 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3199 a
[0] = (char*) "/usr/lib/systemd/systemd";
3200 execve(a
[0], a
, env_use
);
3202 a
[0] = (char*) "/lib/systemd/systemd";
3203 execve(a
[0], a
, env_use
);
3205 a
[0] = (char*) "/sbin/init";
3206 execve(a
[0], a
, env_use
);
3208 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3209 } else if (!strv_isempty(arg_parameters
)) {
3210 const char *dollar_path
;
3212 exec_target
= arg_parameters
[0];
3214 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3216 dollar_path
= strv_env_get(env_use
, "PATH");
3218 if (putenv((char*) dollar_path
) != 0)
3219 return log_error_errno(errno
, "Failed to update $PATH: %m");
3222 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3225 /* If we cannot change the directory, we'll end up in /, that is expected. */
3226 (void) chdir(home
?: "/root");
3228 execle("/bin/bash", "-bash", NULL
, env_use
);
3229 execle("/bin/sh", "-sh", NULL
, env_use
);
3231 exec_target
= "/bin/bash, /bin/sh";
3234 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3237 static int setup_sd_notify_child(void) {
3238 _cleanup_close_
int fd
= -1;
3239 union sockaddr_union sa
= {
3240 .un
.sun_family
= AF_UNIX
,
3241 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3245 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3247 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3249 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3250 (void) sockaddr_un_unlink(&sa
.un
);
3252 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3254 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3256 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3258 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3260 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3262 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3267 static int outer_child(
3269 const char *directory
,
3270 DissectedImage
*dissected_image
,
3277 int uid_shift_socket
,
3278 int master_pty_socket
,
3279 int unified_cgroup_hierarchy_socket
,
3283 _cleanup_close_
int fd
= -1;
3289 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3290 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3291 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3292 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3296 assert(pid_socket
>= 0);
3297 assert(uuid_socket
>= 0);
3298 assert(notify_socket
>= 0);
3299 assert(master_pty_socket
>= 0);
3300 assert(kmsg_socket
>= 0);
3302 log_debug("Outer child is initializing.");
3304 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3305 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3307 r
= reset_audit_loginuid();
3311 /* Mark everything as slave, so that we still
3312 * receive mounts from the real root, but don't
3313 * propagate mounts to the real root. */
3314 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3318 if (dissected_image
) {
3319 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3320 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3321 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3322 * makes sure ESP partitions and userns are compatible. */
3324 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3325 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3326 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
)|
3327 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3329 return log_error_errno(r
, "File system check for image failed: %m");
3331 return log_error_errno(r
, "Failed to mount image root file system: %m");
3334 r
= determine_uid_shift(directory
);
3338 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3339 /* Let the parent know which UID shift we read from the image */
3340 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3342 return log_error_errno(errno
, "Failed to send UID shift: %m");
3343 if (l
!= sizeof(arg_uid_shift
))
3344 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3345 "Short write while sending UID shift.");
3347 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3348 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3349 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3350 * not it will pick a different one, and send it back to us. */
3352 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3354 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3355 if (l
!= sizeof(arg_uid_shift
))
3356 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3357 "Short read while receiving UID shift.");
3360 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3361 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3364 if (path_equal(directory
, "/")) {
3365 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3366 * place, so that we can make changes to its mount structure (for example, to implement
3367 * --volatile=) without this interfering with our ability to access files such as
3368 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3369 * (instead of a temporary directory, since we are living in our own mount namspace here
3370 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3371 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3373 r
= mount_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3377 directory
= "/run/systemd/nspawn-root";
3380 r
= setup_pivot_root(
3383 arg_pivot_root_old
);
3387 r
= setup_volatile_mode(
3391 arg_selinux_apifs_context
);
3398 arg_n_custom_mounts
,
3400 arg_selinux_apifs_context
,
3405 /* Make sure we always have a mount that we can move to root later on. */
3406 if (!path_is_mount_point(directory
, NULL
, 0)) {
3407 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3412 if (dissected_image
) {
3413 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3414 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3415 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
));
3417 return log_error_errno(r
, "File system check for image failed: %m");
3419 return log_error_errno(r
, "Failed to mount image file system: %m");
3422 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3423 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3425 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3429 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3431 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3432 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3433 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3434 "Short write while sending cgroup mode.");
3436 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3439 /* Mark everything as shared so our mounts get propagated down. This is
3440 * required to make new bind mounts available in systemd services
3441 * inside the container that create a new mount namespace.
3442 * See https://github.com/systemd/systemd/issues/3860
3443 * Further submounts (such as /dev) done after this will inherit the
3444 * shared propagation mode.
3446 * IMPORTANT: Do not overmount the root directory anymore from now on to
3447 * enable moving the root directory mount to root later on.
3448 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3450 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3454 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3458 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3462 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3463 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3464 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3466 return log_error_errno(r
, "Failed to make tree read-only: %m");
3469 r
= mount_all(directory
,
3472 arg_selinux_apifs_context
);
3476 r
= copy_devnodes(directory
);
3480 r
= make_extra_nodes(directory
);
3484 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3486 p
= prefix_roota(directory
, "/run");
3487 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3489 r
= setup_pts(directory
);
3493 r
= setup_propagate(directory
);
3497 r
= setup_keyring();
3504 arg_n_custom_mounts
,
3506 arg_selinux_apifs_context
,
3507 MOUNT_NON_ROOT_ONLY
);
3511 r
= setup_timezone(directory
);
3515 r
= setup_resolv_conf(directory
);
3519 r
= setup_machine_id(directory
);
3523 r
= setup_journal(directory
);
3527 if (!arg_use_cgns
) {
3530 arg_unified_cgroup_hierarchy
,
3531 arg_userns_mode
!= USER_NAMESPACE_NO
,
3534 arg_selinux_apifs_context
,
3540 r
= mount_move_root(directory
);
3542 return log_error_errno(r
, "Failed to move root directory: %m");
3544 fd
= setup_sd_notify_child();
3548 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3549 arg_clone_ns_flags
|
3550 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3552 return log_error_errno(errno
, "Failed to fork inner child: %m");
3554 pid_socket
= safe_close(pid_socket
);
3555 uuid_socket
= safe_close(uuid_socket
);
3556 notify_socket
= safe_close(notify_socket
);
3557 uid_shift_socket
= safe_close(uid_shift_socket
);
3559 /* The inner child has all namespaces that are
3560 * requested, so that we all are owned by the user if
3561 * user namespaces are turned on. */
3563 if (arg_network_namespace_path
) {
3564 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3566 return log_error_errno(r
, "Failed to join network namespace: %m");
3569 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
);
3571 _exit(EXIT_FAILURE
);
3573 _exit(EXIT_SUCCESS
);
3576 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3578 return log_error_errno(errno
, "Failed to send PID: %m");
3579 if (l
!= sizeof(pid
))
3580 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3581 "Short write while sending PID.");
3583 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3585 return log_error_errno(errno
, "Failed to send machine ID: %m");
3586 if (l
!= sizeof(arg_uuid
))
3587 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3588 "Short write while sending machine ID.");
3590 l
= send_one_fd(notify_socket
, fd
, 0);
3592 return log_error_errno(l
, "Failed to send notify fd: %m");
3594 pid_socket
= safe_close(pid_socket
);
3595 uuid_socket
= safe_close(uuid_socket
);
3596 notify_socket
= safe_close(notify_socket
);
3597 master_pty_socket
= safe_close(master_pty_socket
);
3598 kmsg_socket
= safe_close(kmsg_socket
);
3599 rtnl_socket
= safe_close(rtnl_socket
);
3600 netns_fd
= safe_close(netns_fd
);
3605 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3606 bool tried_hashed
= false;
3607 unsigned n_tries
= 100;
3612 assert(ret_lock_file
);
3613 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3614 assert(arg_uid_range
== 0x10000U
);
3618 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3621 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3622 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3627 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3629 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3632 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3633 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3634 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3639 /* Make some superficial checks whether the range is currently known in the user database */
3640 if (getpwuid(candidate
))
3642 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3644 if (getgrgid(candidate
))
3646 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3649 *ret_lock_file
= lf
;
3650 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3655 if (arg_machine
&& !tried_hashed
) {
3656 /* Try to hash the base from the container name */
3658 static const uint8_t hash_key
[] = {
3659 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3660 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3663 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3665 tried_hashed
= true;
3667 random_bytes(&candidate
, sizeof(candidate
));
3669 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3670 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3674 static int setup_uid_map(pid_t pid
) {
3675 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3680 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3681 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3682 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3684 return log_error_errno(r
, "Failed to write UID map: %m");
3686 /* We always assign the same UID and GID ranges */
3687 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3688 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3690 return log_error_errno(r
, "Failed to write GID map: %m");
3695 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3696 char buf
[NOTIFY_BUFFER_MAX
+1];
3698 struct iovec iovec
= {
3700 .iov_len
= sizeof(buf
)-1,
3702 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
3703 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
3704 struct msghdr msghdr
= {
3707 .msg_control
= &control
,
3708 .msg_controllen
= sizeof(control
),
3710 struct ucred
*ucred
;
3712 pid_t inner_child_pid
;
3713 _cleanup_strv_free_
char **tags
= NULL
;
3717 inner_child_pid
= PTR_TO_PID(userdata
);
3719 if (revents
!= EPOLLIN
) {
3720 log_warning("Got unexpected poll event for notify fd.");
3724 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3725 if (IN_SET(n
, -EAGAIN
, -EINTR
))
3728 return log_warning_errno(n
, "Couldn't read notification socket: %m");
3730 cmsg_close_all(&msghdr
);
3732 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
3733 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3734 log_debug("Received notify message without valid credentials. Ignoring.");
3738 if ((size_t) n
>= sizeof(buf
)) {
3739 log_warning("Received notify message exceeded maximum size. Ignoring.");
3744 tags
= strv_split(buf
, "\n\r");
3748 if (strv_find(tags
, "READY=1"))
3749 (void) sd_notifyf(false, "READY=1\n");
3751 p
= strv_find_startswith(tags
, "STATUS=");
3753 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3758 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3761 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3763 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3765 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3770 static int merge_settings(Settings
*settings
, const char *path
) {
3776 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3777 * that this steals the fields of the Settings* structure, and hence modifies it. */
3779 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3780 settings
->start_mode
>= 0) {
3781 arg_start_mode
= settings
->start_mode
;
3782 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3785 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3786 arg_ephemeral
= settings
->ephemeral
;
3788 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3791 if (!arg_settings_trusted
)
3792 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3794 free_and_replace(arg_directory
, settings
->root
);
3797 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3798 settings
->pivot_root_new
) {
3799 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3800 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3803 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3804 settings
->working_directory
)
3805 free_and_replace(arg_chdir
, settings
->working_directory
);
3807 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3808 settings
->environment
)
3809 strv_free_and_replace(arg_setenv
, settings
->environment
);
3811 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3814 free_and_replace(arg_user
, settings
->user
);
3816 if (uid_is_valid(settings
->uid
))
3817 arg_uid
= settings
->uid
;
3818 if (gid_is_valid(settings
->gid
))
3819 arg_gid
= settings
->gid
;
3820 if (settings
->n_supplementary_gids
> 0) {
3821 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3822 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3826 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3827 uint64_t plus
, minus
;
3828 uint64_t network_minus
= 0;
3830 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3831 * Settings structure */
3833 plus
= settings
->capability
;
3834 minus
= settings
->drop_capability
;
3836 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3837 if (settings_private_network(settings
))
3838 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3840 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3843 if (!arg_settings_trusted
&& plus
!= 0) {
3844 if (settings
->capability
!= 0)
3845 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3847 arg_caps_retain
&= ~network_minus
;
3848 arg_caps_retain
|= plus
;
3851 arg_caps_retain
&= ~minus
;
3853 /* Copy the full capabilities over too */
3854 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3855 if (!arg_settings_trusted
)
3856 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
3858 arg_full_capabilities
= settings
->full_capabilities
;
3862 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3863 settings
->kill_signal
> 0)
3864 arg_kill_signal
= settings
->kill_signal
;
3866 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3867 settings
->personality
!= PERSONALITY_INVALID
)
3868 arg_personality
= settings
->personality
;
3870 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3871 !sd_id128_is_null(settings
->machine_id
)) {
3873 if (!arg_settings_trusted
)
3874 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3876 arg_uuid
= settings
->machine_id
;
3879 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3880 settings
->read_only
>= 0)
3881 arg_read_only
= settings
->read_only
;
3883 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3884 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3885 arg_volatile_mode
= settings
->volatile_mode
;
3887 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3888 settings
->n_custom_mounts
> 0) {
3890 if (!arg_settings_trusted
)
3891 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3893 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3894 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3895 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3896 settings
->n_custom_mounts
= 0;
3900 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3901 (settings
->private_network
>= 0 ||
3902 settings
->network_veth
>= 0 ||
3903 settings
->network_bridge
||
3904 settings
->network_zone
||
3905 settings
->network_interfaces
||
3906 settings
->network_macvlan
||
3907 settings
->network_ipvlan
||
3908 settings
->network_veth_extra
||
3909 settings
->network_namespace_path
)) {
3911 if (!arg_settings_trusted
)
3912 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3914 arg_network_veth
= settings_network_veth(settings
);
3915 arg_private_network
= settings_private_network(settings
);
3917 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3918 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3919 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3920 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3922 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3923 free_and_replace(arg_network_zone
, settings
->network_zone
);
3925 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3929 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3930 settings
->expose_ports
) {
3932 if (!arg_settings_trusted
)
3933 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3935 expose_port_free_all(arg_expose_ports
);
3936 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3940 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3941 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3943 if (!arg_settings_trusted
)
3944 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3946 arg_userns_mode
= settings
->userns_mode
;
3947 arg_uid_shift
= settings
->uid_shift
;
3948 arg_uid_range
= settings
->uid_range
;
3949 arg_userns_chown
= settings
->userns_chown
;
3953 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3954 arg_notify_ready
= settings
->notify_ready
;
3956 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3958 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
3959 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3961 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
3962 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
3966 if (!arg_settings_trusted
&& settings
->seccomp
)
3967 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
3969 seccomp_release(arg_seccomp
);
3970 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
3975 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3976 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3979 if (!settings
->rlimit
[rl
])
3982 if (!arg_settings_trusted
) {
3983 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3987 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3990 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3992 free_and_replace(arg_hostname
, settings
->hostname
);
3994 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3995 settings
->no_new_privileges
>= 0)
3996 arg_no_new_privileges
= settings
->no_new_privileges
;
3998 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3999 settings
->oom_score_adjust_set
) {
4001 if (!arg_settings_trusted
)
4002 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4004 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4005 arg_oom_score_adjust_set
= true;
4009 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4010 settings
->cpu_set
.set
) {
4012 if (!arg_settings_trusted
)
4013 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4015 cpu_set_reset(&arg_cpu_set
);
4016 arg_cpu_set
= settings
->cpu_set
;
4017 settings
->cpu_set
= (CPUSet
) {};
4021 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4022 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4023 arg_resolv_conf
= settings
->resolv_conf
;
4025 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4026 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4028 if (!arg_settings_trusted
)
4029 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4031 arg_link_journal
= settings
->link_journal
;
4032 arg_link_journal_try
= settings
->link_journal_try
;
4036 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4037 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4038 arg_timezone
= settings
->timezone
;
4040 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4043 if (!arg_settings_trusted
)
4044 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4046 free_and_replace(arg_slice
, settings
->slice
);
4049 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4050 settings
->use_cgns
>= 0) {
4052 if (!arg_settings_trusted
)
4053 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4055 arg_use_cgns
= settings
->use_cgns
;
4058 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4059 settings
->clone_ns_flags
!= (unsigned long) -1) {
4061 if (!arg_settings_trusted
)
4062 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4064 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4067 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4068 settings
->console_mode
>= 0) {
4070 if (!arg_settings_trusted
)
4071 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4073 arg_console_mode
= settings
->console_mode
;
4076 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4077 * don't consult arg_settings_mask for them. */
4079 sd_bus_message_unref(arg_property_message
);
4080 arg_property_message
= TAKE_PTR(settings
->properties
);
4082 arg_console_width
= settings
->console_width
;
4083 arg_console_height
= settings
->console_height
;
4085 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4086 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4087 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4092 static int load_settings(void) {
4093 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4094 _cleanup_fclose_
FILE *f
= NULL
;
4095 _cleanup_free_
char *p
= NULL
;
4102 /* If all settings are masked, there's no point in looking for
4103 * the settings file */
4104 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
4107 fn
= strjoina(arg_machine
, ".nspawn");
4109 /* We first look in the admin's directories in /etc and /run */
4110 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4111 _cleanup_free_
char *j
= NULL
;
4113 j
= path_join(i
, fn
);
4121 /* By default, we trust configuration from /etc and /run */
4122 if (arg_settings_trusted
< 0)
4123 arg_settings_trusted
= true;
4128 if (errno
!= ENOENT
)
4129 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4133 /* After that, let's look for a file next to the
4134 * actual image we shall boot. */
4137 p
= file_in_same_dir(arg_image
, fn
);
4140 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4141 p
= file_in_same_dir(arg_directory
, fn
);
4148 if (!f
&& errno
!= ENOENT
)
4149 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4151 /* By default, we do not trust configuration from /var/lib/machines */
4152 if (arg_settings_trusted
< 0)
4153 arg_settings_trusted
= false;
4160 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4162 r
= settings_load(f
, p
, &settings
);
4166 return merge_settings(settings
, p
);
4169 static int load_oci_bundle(void) {
4170 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4173 if (!arg_oci_bundle
)
4176 /* By default let's trust OCI bundles */
4177 if (arg_settings_trusted
< 0)
4178 arg_settings_trusted
= true;
4180 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4184 return merge_settings(settings
, arg_oci_bundle
);
4187 static int run_container(
4188 DissectedImage
*dissected_image
,
4191 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4192 union in_addr_union
*exposed
,
4193 int *master
, pid_t
*pid
, int *ret
) {
4195 static const struct sigaction sa
= {
4196 .sa_handler
= nop_signal_handler
,
4197 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4200 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4201 _cleanup_close_
int etc_passwd_lock
= -1;
4202 _cleanup_close_pair_
int
4203 kmsg_socket_pair
[2] = { -1, -1 },
4204 rtnl_socket_pair
[2] = { -1, -1 },
4205 pid_socket_pair
[2] = { -1, -1 },
4206 uuid_socket_pair
[2] = { -1, -1 },
4207 notify_socket_pair
[2] = { -1, -1 },
4208 uid_shift_socket_pair
[2] = { -1, -1 },
4209 master_pty_socket_pair
[2] = { -1, -1 },
4210 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4212 _cleanup_close_
int notify_socket
= -1;
4213 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4214 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4215 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4216 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4217 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4218 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4219 ContainerStatus container_status
= 0;
4223 _cleanup_close_
int child_netns_fd
= -1;
4225 assert_se(sigemptyset(&mask_chld
) == 0);
4226 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4228 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4229 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4230 * check with getpwuid() if the specific user already exists. Note that /etc might be
4231 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4232 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4233 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4236 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4237 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4238 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4241 r
= barrier_create(&barrier
);
4243 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4245 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4246 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4248 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4249 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4251 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4252 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4254 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4255 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4257 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4258 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4260 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4261 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4263 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4264 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4265 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4267 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4268 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4269 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4271 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4272 * parent's blocking calls and give it a chance to call wait() and terminate. */
4273 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4275 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4277 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4279 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4281 if (arg_network_namespace_path
) {
4282 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4283 if (child_netns_fd
< 0)
4284 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4286 r
= fd_is_network_ns(child_netns_fd
);
4288 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
);
4290 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4292 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4293 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4296 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4298 return log_error_errno(errno
, "clone() failed%s: %m",
4300 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4303 /* The outer child only has a file system namespace. */
4304 barrier_set_role(&barrier
, BARRIER_CHILD
);
4306 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4307 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4308 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4309 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4310 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4311 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4312 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4313 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4315 (void) reset_all_signal_handlers();
4316 (void) reset_signal_mask();
4318 r
= outer_child(&barrier
,
4323 uuid_socket_pair
[1],
4324 notify_socket_pair
[1],
4325 kmsg_socket_pair
[1],
4326 rtnl_socket_pair
[1],
4327 uid_shift_socket_pair
[1],
4328 master_pty_socket_pair
[1],
4329 unified_cgroup_hierarchy_socket_pair
[1],
4333 _exit(EXIT_FAILURE
);
4335 _exit(EXIT_SUCCESS
);
4338 barrier_set_role(&barrier
, BARRIER_PARENT
);
4342 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4343 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4344 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4345 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4346 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4347 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4348 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4349 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4351 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4352 /* The child just let us know the UID shift it might have read from the image. */
4353 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4355 return log_error_errno(errno
, "Failed to read UID shift: %m");
4356 if (l
!= sizeof arg_uid_shift
)
4357 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4359 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4360 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4361 * image, but if that's already in use, pick a new one, and report back to the child,
4362 * which one we now picked. */
4364 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4366 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4368 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4370 return log_error_errno(errno
, "Failed to send UID shift: %m");
4371 if (l
!= sizeof arg_uid_shift
)
4372 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4376 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4377 /* The child let us know the support cgroup mode it might have read from the image. */
4378 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4380 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4381 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4382 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4383 l
, l
== 0 ? " The child is most likely dead." : "");
4386 /* Wait for the outer child. */
4387 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4390 if (r
!= EXIT_SUCCESS
)
4393 /* And now retrieve the PID of the inner child. */
4394 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4396 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4397 if (l
!= sizeof *pid
)
4398 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4400 /* We also retrieve container UUID in case it was generated by outer child */
4401 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4403 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4404 if (l
!= sizeof(arg_uuid
))
4405 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4407 /* We also retrieve the socket used for notifications generated by outer child */
4408 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4409 if (notify_socket
< 0)
4410 return log_error_errno(notify_socket
,
4411 "Failed to receive notification socket from the outer child: %m");
4413 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4415 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4416 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4417 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4419 r
= setup_uid_map(*pid
);
4423 (void) barrier_place(&barrier
); /* #2 */
4426 if (arg_private_network
) {
4427 if (!arg_network_namespace_path
) {
4428 /* Wait until the child has unshared its network namespace. */
4429 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4430 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4433 if (child_netns_fd
< 0) {
4434 /* Make sure we have an open file descriptor to the child's network
4435 * namespace so it stays alive even if the child exits. */
4436 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4438 return log_error_errno(r
, "Failed to open child network namespace: %m");
4441 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4445 if (arg_network_veth
) {
4446 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4447 arg_network_bridge
|| arg_network_zone
);
4453 if (arg_network_bridge
) {
4454 /* Add the interface to a bridge */
4455 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4460 } else if (arg_network_zone
) {
4461 /* Add the interface to a bridge, possibly creating it */
4462 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4470 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4474 /* We created the primary and extra veth links now; let's remember this, so that we know to
4475 remove them later on. Note that we don't bother with removing veth links that were created
4476 here when their setup failed half-way, because in that case the kernel should be able to
4477 remove them on its own, since they cannot be referenced by anything yet. */
4478 *veth_created
= true;
4480 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4484 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4489 if (arg_register
|| !arg_keep_unit
) {
4490 r
= sd_bus_default_system(&bus
);
4492 return log_error_errno(r
, "Failed to open system bus: %m");
4494 r
= sd_bus_set_close_on_exit(bus
, false);
4496 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4499 if (!arg_keep_unit
) {
4500 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4501 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4502 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4504 r
= sd_bus_match_signal_async(
4507 "org.freedesktop.systemd1",
4509 "org.freedesktop.systemd1.Scope",
4511 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4513 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4517 r
= register_machine(
4525 arg_custom_mounts
, arg_n_custom_mounts
,
4528 arg_property_message
,
4530 arg_container_service_name
);
4534 } else if (!arg_keep_unit
) {
4540 arg_custom_mounts
, arg_n_custom_mounts
,
4543 arg_property_message
);
4547 } else if (arg_slice
|| arg_property
)
4548 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4550 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4554 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4558 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4562 /* Notify the child that the parent is ready with all
4563 * its setup (including cgroup-ification), and that
4564 * the child can now hand over control to the code to
4565 * run inside the container. */
4566 (void) barrier_place(&barrier
); /* #4 */
4568 /* Block SIGCHLD here, before notifying child.
4569 * process_pty() will handle it with the other signals. */
4570 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4572 /* Reset signal to default */
4573 r
= default_signals(SIGCHLD
, -1);
4575 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4577 r
= sd_event_new(&event
);
4579 return log_error_errno(r
, "Failed to get default event source: %m");
4581 (void) sd_event_set_watchdog(event
, true);
4584 r
= sd_bus_attach_event(bus
, event
, 0);
4586 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4589 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4593 /* Let the child know that we are ready and wait that the child is completely ready now. */
4594 if (!barrier_place_and_sync(&barrier
)) /* #5 */
4595 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4597 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4598 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4599 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4601 (void) sd_notifyf(false,
4602 "STATUS=Container running.\n"
4603 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4604 if (!arg_notify_ready
)
4605 (void) sd_notify(false, "READY=1\n");
4607 if (arg_kill_signal
> 0) {
4608 /* Try to kill the init system on SIGINT or SIGTERM */
4609 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4610 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4612 /* Immediately exit */
4613 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4614 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4617 /* Exit when the child exits */
4618 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4620 if (arg_expose_ports
) {
4621 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4625 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4628 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4630 if (arg_console_mode
!= CONSOLE_PIPE
) {
4631 _cleanup_close_
int fd
= -1;
4632 PTYForwardFlags flags
= 0;
4634 /* Retrieve the master pty allocated by inner child */
4635 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
4637 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
4639 switch (arg_console_mode
) {
4641 case CONSOLE_READ_ONLY
:
4642 flags
|= PTY_FORWARD_READ_ONLY
;
4646 case CONSOLE_INTERACTIVE
:
4647 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
4649 r
= pty_forward_new(event
, fd
, flags
, &forward
);
4651 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4653 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4654 (void) pty_forward_set_width_height(forward
,
4656 arg_console_height
);
4660 assert(arg_console_mode
== CONSOLE_PASSIVE
);
4663 *master
= TAKE_FD(fd
);
4666 r
= sd_event_loop(event
);
4668 return log_error_errno(r
, "Failed to run event loop: %m");
4673 (void) pty_forward_get_last_char(forward
, &last_char
);
4674 forward
= pty_forward_free(forward
);
4676 if (!arg_quiet
&& last_char
!= '\n')
4680 /* Kill if it is not dead yet anyway */
4681 if (!arg_register
&& !arg_keep_unit
&& bus
)
4682 terminate_scope(bus
, arg_machine
);
4684 /* Normally redundant, but better safe than sorry */
4685 (void) kill(*pid
, SIGKILL
);
4687 if (arg_private_network
) {
4688 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
4689 * to avoid having to move the parent to the child network namespace. */
4690 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
4695 _cleanup_close_
int parent_netns_fd
= -1;
4697 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
4699 log_error_errno(r
, "Failed to open parent network namespace: %m");
4700 _exit(EXIT_FAILURE
);
4703 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
4705 log_error_errno(r
, "Failed to enter child network namespace: %m");
4706 _exit(EXIT_FAILURE
);
4709 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
4711 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
4713 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
4717 r
= wait_for_container(*pid
, &container_status
);
4720 /* Tell machined that we are gone. */
4722 (void) unregister_machine(bus
, arg_machine
);
4725 /* We failed to wait for the container, or the container exited abnormally. */
4727 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4728 /* r > 0 → The container exited with a non-zero status.
4729 * As a special case, we need to replace 133 with a different value,
4730 * because 133 is special-cased in the service file to reboot the container.
4731 * otherwise → The container exited with zero status and a reboot was not requested.
4733 if (r
== EXIT_FORCE_RESTART
)
4734 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4736 return 0; /* finito */
4739 /* CONTAINER_REBOOTED, loop again */
4741 if (arg_keep_unit
) {
4742 /* Special handling if we are running as a service: instead of simply
4743 * restarting the machine we want to restart the entire service, so let's
4744 * inform systemd about this with the special exit code 133. The service
4745 * file uses RestartForceExitStatus=133 so that this results in a full
4746 * nspawn restart. This is necessary since we might have cgroup parameters
4747 * set we want to have flushed out. */
4748 *ret
= EXIT_FORCE_RESTART
;
4749 return 0; /* finito */
4752 expose_port_flush(arg_expose_ports
, exposed
);
4754 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4755 *veth_created
= false;
4756 return 1; /* loop again */
4759 static int initialize_rlimits(void) {
4760 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4761 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4762 * container execution environments. */
4764 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4765 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4766 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4767 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4768 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4769 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4770 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4771 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4772 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4773 [RLIMIT_NICE
] = { 0, 0 },
4774 [RLIMIT_NOFILE
] = { 1024, 4096 },
4775 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4776 [RLIMIT_RTPRIO
] = { 0, 0 },
4777 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4778 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4780 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4781 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4782 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4783 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4784 * that PID 1 changes a number of other resource limits during early initialization which is why we
4785 * don't read the other limits from PID 1 but prefer the static table above. */
4790 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4791 /* Let's only fill in what the user hasn't explicitly configured anyway */
4792 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4793 const struct rlimit
*v
;
4794 struct rlimit buffer
;
4796 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4797 /* For these two let's read the limits off PID 1. See above for an explanation. */
4799 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4800 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4804 v
= kernel_defaults
+ rl
;
4806 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4807 if (!arg_rlimit
[rl
])
4811 if (DEBUG_LOGGING
) {
4812 _cleanup_free_
char *k
= NULL
;
4814 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4815 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4822 static int cant_be_in_netns(void) {
4823 union sockaddr_union sa
= {
4825 .sun_family
= AF_UNIX
,
4826 .sun_path
= "/run/udev/control",
4829 char udev_path
[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t
)];
4830 _cleanup_free_
char *udev_ns
= NULL
, *our_ns
= NULL
;
4831 _cleanup_close_
int fd
= -1;
4835 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
4836 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
4839 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
4842 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
4844 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
4846 if (connect(fd
, &sa
.un
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
4848 if (errno
== ENOENT
|| ERRNO_IS_DISCONNECT(errno
))
4849 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
4850 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
4852 return log_error_errno(errno
, "Failed to connect socket to udev control socket: %m");
4855 r
= getpeercred(fd
, &ucred
);
4857 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
4859 xsprintf(udev_path
, "/proc/" PID_FMT
"/ns/net", ucred
.pid
);
4860 r
= readlink_malloc(udev_path
, &udev_ns
);
4862 return log_error_errno(r
, "Failed to read network namespace of udev: %m");
4864 r
= readlink_malloc("/proc/self/ns/net", &our_ns
);
4866 return log_error_errno(r
, "Failed to read our own network namespace: %m");
4868 if (!streq(our_ns
, udev_ns
))
4869 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
4870 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
4874 static int run(int argc
, char *argv
[]) {
4875 bool secondary
= false, remove_directory
= false, remove_image
= false,
4876 veth_created
= false, remove_tmprootdir
= false;
4877 _cleanup_close_
int master
= -1;
4878 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4879 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4880 char veth_name
[IFNAMSIZ
] = "";
4881 union in_addr_union exposed
= {};
4882 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4883 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4884 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4885 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4886 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4889 log_parse_environment();
4892 r
= parse_argv(argc
, argv
);
4900 r
= cant_be_in_netns();
4904 r
= initialize_rlimits();
4908 r
= load_oci_bundle();
4912 r
= determine_names();
4916 r
= load_settings();
4922 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4926 r
= verify_arguments();
4930 /* Reapply environment settings. */
4931 (void) detect_unified_cgroup_hierarchy_from_environment();
4933 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4934 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4935 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4936 (void) ignore_signals(SIGPIPE
, -1);
4938 n_fd_passed
= sd_listen_fds(false);
4939 if (n_fd_passed
> 0) {
4940 r
= fdset_new_listen_fds(&fds
, false);
4942 log_error_errno(r
, "Failed to collect file descriptors: %m");
4947 /* The "default" umask. This is appropriate for most file and directory
4948 * operations performed by nspawn, and is the umask that will be used for
4949 * the child. Functions like copy_devnodes() change the umask temporarily. */
4952 if (arg_directory
) {
4955 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4956 * /var from the host will propagate into container dynamically (because bad things happen if
4957 * two systems write to the same /var). Let's allow it for the special cases where /var is
4958 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4959 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
4960 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4965 if (arg_ephemeral
) {
4966 _cleanup_free_
char *np
= NULL
;
4968 r
= chase_symlinks_and_update(&arg_directory
, 0);
4972 /* If the specified path is a mount point we generate the new snapshot immediately
4973 * inside it under a random name. However if the specified is not a mount point we
4974 * create the new snapshot in the parent directory, just next to it. */
4975 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4977 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4981 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4983 r
= tempfn_random(arg_directory
, "machine.", &np
);
4985 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4989 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4990 * only owned by us and no one else. */
4991 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4993 log_error_errno(r
, "Failed to lock %s: %m", np
);
4998 BLOCK_SIGNALS(SIGINT
);
4999 r
= btrfs_subvol_snapshot(arg_directory
, np
,
5000 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5001 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5002 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5003 BTRFS_SNAPSHOT_RECURSIVE
|
5004 BTRFS_SNAPSHOT_QUOTA
|
5005 BTRFS_SNAPSHOT_SIGINT
);
5008 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5012 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5016 free_and_replace(arg_directory
, np
);
5017 remove_directory
= true;
5019 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5023 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5025 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5029 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5034 r
= chase_symlinks_and_update(&arg_template
, 0);
5039 BLOCK_SIGNALS(SIGINT
);
5040 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5041 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5042 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5043 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5044 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5045 BTRFS_SNAPSHOT_RECURSIVE
|
5046 BTRFS_SNAPSHOT_QUOTA
|
5047 BTRFS_SNAPSHOT_SIGINT
);
5050 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5051 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5052 else if (r
== -EINTR
) {
5053 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5056 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5059 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5060 "Populated %s from template %s.", arg_directory
, arg_template
);
5064 if (arg_start_mode
== START_BOOT
) {
5067 if (arg_pivot_root_new
)
5068 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5072 if (path_is_os_tree(p
) <= 0) {
5073 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5080 if (arg_pivot_root_new
)
5081 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5085 q
= strjoina(p
, "/usr/");
5087 if (laccess(q
, F_OK
) < 0) {
5088 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
5095 DissectImageFlags dissect_image_flags
= DISSECT_IMAGE_REQUIRE_ROOT
| DISSECT_IMAGE_RELAX_VAR_CHECK
;
5097 assert(!arg_template
);
5099 r
= chase_symlinks_and_update(&arg_image
, 0);
5103 if (arg_ephemeral
) {
5104 _cleanup_free_
char *np
= NULL
;
5106 r
= tempfn_random(arg_image
, "machine.", &np
);
5108 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5112 /* Always take an exclusive lock on our own ephemeral copy. */
5113 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5115 r
= log_error_errno(r
, "Failed to create image lock: %m");
5120 BLOCK_SIGNALS(SIGINT
);
5121 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
);
5124 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5128 r
= log_error_errno(r
, "Failed to copy image file: %m");
5132 free_and_replace(arg_image
, np
);
5133 remove_image
= true;
5135 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5137 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5141 r
= log_error_errno(r
, "Failed to create image lock: %m");
5145 r
= verity_metadata_load(arg_image
, NULL
, arg_root_hash
? NULL
: &arg_root_hash
, &arg_root_hash_size
,
5146 arg_verity_data
? NULL
: &arg_verity_data
);
5148 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5151 dissect_image_flags
|= arg_verity_data
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0;
5154 if (!mkdtemp(tmprootdir
)) {
5155 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5159 remove_tmprootdir
= true;
5161 arg_directory
= strdup(tmprootdir
);
5162 if (!arg_directory
) {
5167 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, LO_FLAGS_PARTSCAN
, &loop
);
5169 log_error_errno(r
, "Failed to set up loopback block device: %m");
5173 r
= dissect_image_and_warn(
5176 arg_root_hash
, arg_root_hash_size
,
5178 dissect_image_flags
,
5181 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5182 log_notice("Note that the disk image needs to\n"
5183 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5184 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5185 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5186 " d) or contain a file system without a partition table\n"
5187 "in order to be bootable with systemd-nspawn.");
5193 if (!arg_root_hash
&& dissected_image
->can_verity
)
5194 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5196 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, arg_verity_data
, 0, &decrypted_image
);
5200 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5201 if (remove_image
&& unlink(arg_image
) >= 0)
5202 remove_image
= false;
5205 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5209 if (arg_console_mode
< 0)
5211 isatty(STDIN_FILENO
) > 0 &&
5212 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5214 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5218 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5219 arg_machine
, arg_image
?: arg_directory
);
5221 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5223 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5224 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5229 r
= run_container(dissected_image
,
5232 veth_name
, &veth_created
,
5240 (void) sd_notify(false,
5241 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5242 "STOPPING=1\nSTATUS=Terminating...");
5245 (void) kill(pid
, SIGKILL
);
5247 /* Try to flush whatever is still queued in the pty */
5249 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5250 master
= safe_close(master
);
5254 (void) wait_for_terminate(pid
, NULL
);
5258 if (remove_directory
&& arg_directory
) {
5261 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5263 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5266 if (remove_image
&& arg_image
) {
5267 if (unlink(arg_image
) < 0)
5268 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5271 if (remove_tmprootdir
) {
5272 if (rmdir(tmprootdir
) < 0)
5273 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5279 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5280 (void) rm_rf(p
, REMOVE_ROOT
);
5283 expose_port_flush(arg_expose_ports
, &exposed
);
5286 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5287 (void) remove_bridge(arg_network_zone
);
5289 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5290 expose_port_free_all(arg_expose_ports
);
5291 rlimit_free_all(arg_rlimit
);
5292 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5300 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);