1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
8 #include <linux/loop.h>
10 #include <selinux/selinux.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.h>
23 #include "sd-daemon.h"
26 #include "alloc-util.h"
28 #include "base-filesystem.h"
29 #include "blkid-util.h"
30 #include "btrfs-util.h"
31 #include "bus-error.h"
34 #include "capability-util.h"
35 #include "cgroup-util.h"
36 #include "chase-symlinks.h"
38 #include "cpu-set-util.h"
39 #include "creds-util.h"
40 #include "dev-setup.h"
41 #include "discover-image.h"
42 #include "dissect-image.h"
48 #include "format-util.h"
51 #include "hexdecoct.h"
52 #include "hostname-setup.h"
53 #include "hostname-util.h"
54 #include "id128-util.h"
57 #include "loop-util.h"
58 #include "loopback-setup.h"
60 #include "main-func.h"
61 #include "missing_sched.h"
63 #include "mount-util.h"
64 #include "mountpoint-util.h"
65 #include "namespace-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-bind-user.h"
68 #include "nspawn-cgroup.h"
69 #include "nspawn-creds.h"
70 #include "nspawn-def.h"
71 #include "nspawn-expose-ports.h"
72 #include "nspawn-mount.h"
73 #include "nspawn-network.h"
74 #include "nspawn-oci.h"
75 #include "nspawn-patch-uid.h"
76 #include "nspawn-register.h"
77 #include "nspawn-seccomp.h"
78 #include "nspawn-settings.h"
79 #include "nspawn-setuid.h"
80 #include "nspawn-stub-pid1.h"
81 #include "nspawn-util.h"
83 #include "nulstr-util.h"
86 #include "parse-argument.h"
87 #include "parse-util.h"
88 #include "pretty-print.h"
89 #include "process-util.h"
91 #include "random-util.h"
92 #include "raw-clone.h"
93 #include "resolve-util.h"
94 #include "rlimit-util.h"
97 #include "seccomp-util.h"
99 #include "selinux-util.h"
100 #include "signal-util.h"
101 #include "socket-util.h"
102 #include "stat-util.h"
103 #include "stdio-util.h"
104 #include "string-table.h"
105 #include "string-util.h"
107 #include "sysctl-util.h"
108 #include "terminal-util.h"
109 #include "tmpfile-util.h"
110 #include "umask-util.h"
111 #include "unit-name.h"
112 #include "user-util.h"
115 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
116 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
118 #define EXIT_FORCE_RESTART 133
120 typedef enum ContainerStatus
{
121 CONTAINER_TERMINATED
,
125 static char *arg_directory
= NULL
;
126 static char *arg_template
= NULL
;
127 static char *arg_chdir
= NULL
;
128 static char *arg_pivot_root_new
= NULL
;
129 static char *arg_pivot_root_old
= NULL
;
130 static char *arg_user
= NULL
;
131 static uid_t arg_uid
= UID_INVALID
;
132 static gid_t arg_gid
= GID_INVALID
;
133 static gid_t
* arg_supplementary_gids
= NULL
;
134 static size_t arg_n_supplementary_gids
= 0;
135 static sd_id128_t arg_uuid
= {};
136 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
137 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
138 static const char *arg_selinux_context
= NULL
;
139 static const char *arg_selinux_apifs_context
= NULL
;
140 static char *arg_slice
= NULL
;
141 static bool arg_private_network
= false;
142 static bool arg_read_only
= false;
143 static StartMode arg_start_mode
= START_PID1
;
144 static bool arg_ephemeral
= false;
145 static LinkJournal arg_link_journal
= LINK_AUTO
;
146 static bool arg_link_journal_try
= false;
147 static uint64_t arg_caps_retain
=
148 (1ULL << CAP_AUDIT_CONTROL
) |
149 (1ULL << CAP_AUDIT_WRITE
) |
150 (1ULL << CAP_CHOWN
) |
151 (1ULL << CAP_DAC_OVERRIDE
) |
152 (1ULL << CAP_DAC_READ_SEARCH
) |
153 (1ULL << CAP_FOWNER
) |
154 (1ULL << CAP_FSETID
) |
155 (1ULL << CAP_IPC_OWNER
) |
157 (1ULL << CAP_LEASE
) |
158 (1ULL << CAP_LINUX_IMMUTABLE
) |
159 (1ULL << CAP_MKNOD
) |
160 (1ULL << CAP_NET_BIND_SERVICE
) |
161 (1ULL << CAP_NET_BROADCAST
) |
162 (1ULL << CAP_NET_RAW
) |
163 (1ULL << CAP_SETFCAP
) |
164 (1ULL << CAP_SETGID
) |
165 (1ULL << CAP_SETPCAP
) |
166 (1ULL << CAP_SETUID
) |
167 (1ULL << CAP_SYS_ADMIN
) |
168 (1ULL << CAP_SYS_BOOT
) |
169 (1ULL << CAP_SYS_CHROOT
) |
170 (1ULL << CAP_SYS_NICE
) |
171 (1ULL << CAP_SYS_PTRACE
) |
172 (1ULL << CAP_SYS_RESOURCE
) |
173 (1ULL << CAP_SYS_TTY_CONFIG
);
174 static uint64_t arg_caps_ambient
= 0;
175 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
176 static CustomMount
*arg_custom_mounts
= NULL
;
177 static size_t arg_n_custom_mounts
= 0;
178 static char **arg_setenv
= NULL
;
179 static bool arg_quiet
= false;
180 static bool arg_register
= true;
181 static bool arg_keep_unit
= false;
182 static char **arg_network_interfaces
= NULL
;
183 static char **arg_network_macvlan
= NULL
;
184 static char **arg_network_ipvlan
= NULL
;
185 static bool arg_network_veth
= false;
186 static char **arg_network_veth_extra
= NULL
;
187 static char *arg_network_bridge
= NULL
;
188 static char *arg_network_zone
= NULL
;
189 static char *arg_network_namespace_path
= NULL
;
190 static PagerFlags arg_pager_flags
= 0;
191 static unsigned long arg_personality
= PERSONALITY_INVALID
;
192 static char *arg_image
= NULL
;
193 static char *arg_oci_bundle
= NULL
;
194 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
195 static ExposePort
*arg_expose_ports
= NULL
;
196 static char **arg_property
= NULL
;
197 static sd_bus_message
*arg_property_message
= NULL
;
198 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
199 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
200 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
201 static int arg_kill_signal
= 0;
202 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
203 static SettingsMask arg_settings_mask
= 0;
204 static int arg_settings_trusted
= -1;
205 static char **arg_parameters
= NULL
;
206 static const char *arg_container_service_name
= "systemd-nspawn";
207 static bool arg_notify_ready
= false;
208 static bool arg_use_cgns
= true;
209 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
210 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
211 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
212 static char **arg_syscall_allow_list
= NULL
;
213 static char **arg_syscall_deny_list
= NULL
;
215 static scmp_filter_ctx arg_seccomp
= NULL
;
217 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
218 static bool arg_no_new_privileges
= false;
219 static int arg_oom_score_adjust
= 0;
220 static bool arg_oom_score_adjust_set
= false;
221 static CPUSet arg_cpu_set
= {};
222 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
223 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
224 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
225 static DeviceNode
* arg_extra_nodes
= NULL
;
226 static size_t arg_n_extra_nodes
= 0;
227 static char **arg_sysctl
= NULL
;
228 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
229 static Credential
*arg_credentials
= NULL
;
230 static size_t arg_n_credentials
= 0;
231 static char **arg_bind_user
= NULL
;
232 static bool arg_suppress_sync
= false;
234 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
256 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
258 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
264 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
265 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
267 static int handle_arg_console(const char *arg
) {
268 if (streq(arg
, "help")) {
277 if (streq(arg
, "interactive"))
278 arg_console_mode
= CONSOLE_INTERACTIVE
;
279 else if (streq(arg
, "read-only"))
280 arg_console_mode
= CONSOLE_READ_ONLY
;
281 else if (streq(arg
, "passive"))
282 arg_console_mode
= CONSOLE_PASSIVE
;
283 else if (streq(arg
, "pipe")) {
284 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
285 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
286 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
287 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
288 "Proceeding anyway.");
290 arg_console_mode
= CONSOLE_PIPE
;
291 } else if (streq(arg
, "autopipe")) {
292 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
293 arg_console_mode
= CONSOLE_INTERACTIVE
;
295 arg_console_mode
= CONSOLE_PIPE
;
297 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
299 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
303 static int help(void) {
304 _cleanup_free_
char *link
= NULL
;
307 pager_open(arg_pager_flags
);
309 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
313 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
314 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
315 " -h --help Show this help\n"
316 " --version Print version string\n"
317 " -q --quiet Do not show status information\n"
318 " --no-pager Do not pipe output into a pager\n"
319 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
321 " -D --directory=PATH Root directory for the container\n"
322 " --template=PATH Initialize root directory from template directory,\n"
324 " -x --ephemeral Run container with snapshot of root directory, and\n"
325 " remove it after exit\n"
326 " -i --image=PATH Root file system disk image (or device node) for\n"
328 " --oci-bundle=PATH OCI bundle directory\n"
329 " --read-only Mount the root directory read-only\n"
330 " --volatile[=MODE] Run the system in volatile mode\n"
331 " --root-hash=HASH Specify verity root hash for root disk image\n"
332 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
333 " as a DER encoded PKCS7, either as a path to a file\n"
334 " or as an ASCII base64 encoded string prefixed by\n"
336 " --verity-data=PATH Specify hash device for verity\n"
337 " --pivot-root=PATH[:PATH]\n"
338 " Pivot root to given directory in the container\n\n"
339 "%3$sExecution:%4$s\n"
340 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
341 " -b --boot Boot up full system (i.e. invoke init)\n"
342 " --chdir=PATH Set working directory in the container\n"
343 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
344 " -u --user=USER Run the command under specified user or UID\n"
345 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
346 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
347 " --suppress-sync=BOOLEAN\n"
348 " Suppress any form of disk data synchronization\n\n"
349 "%3$sSystem Identity:%4$s\n"
350 " -M --machine=NAME Set the machine name for the container\n"
351 " --hostname=NAME Override the hostname for the container\n"
352 " --uuid=UUID Set a specific machine UUID for the container\n\n"
353 "%3$sProperties:%4$s\n"
354 " -S --slice=SLICE Place the container in the specified slice\n"
355 " --property=NAME=VALUE Set scope unit property\n"
356 " --register=BOOLEAN Register container as machine\n"
357 " --keep-unit Do not register a scope for the machine, reuse\n"
358 " the service unit nspawn is running in\n\n"
359 "%3$sUser Namespacing:%4$s\n"
360 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
361 " --private-users[=UIDBASE[:NUIDS]]\n"
362 " Similar, but with user configured UID/GID range\n"
363 " --private-users-ownership=MODE\n"
364 " Adjust ('chown') or map ('map') OS tree ownership\n"
365 " to private UID/GID range\n\n"
366 "%3$sNetworking:%4$s\n"
367 " --private-network Disable network in container\n"
368 " --network-interface=INTERFACE\n"
369 " Assign an existing network interface to the\n"
371 " --network-macvlan=INTERFACE\n"
372 " Create a macvlan network interface based on an\n"
373 " existing network interface to the container\n"
374 " --network-ipvlan=INTERFACE\n"
375 " Create an ipvlan network interface based on an\n"
376 " existing network interface to the container\n"
377 " -n --network-veth Add a virtual Ethernet connection between host\n"
379 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
380 " Add an additional virtual Ethernet link between\n"
381 " host and container\n"
382 " --network-bridge=INTERFACE\n"
383 " Add a virtual Ethernet connection to the container\n"
384 " and attach it to an existing bridge on the host\n"
385 " --network-zone=NAME Similar, but attach the new interface to an\n"
386 " an automatically managed bridge interface\n"
387 " --network-namespace-path=PATH\n"
388 " Set network namespace to the one represented by\n"
389 " the specified kernel namespace file node\n"
390 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
391 " Expose a container IP port on the host\n\n"
392 "%3$sSecurity:%4$s\n"
393 " --capability=CAP In addition to the default, retain specified\n"
395 " --drop-capability=CAP Drop the specified capability from the default set\n"
396 " --ambient-capability=CAP\n"
397 " Sets the specified capability for the started\n"
398 " process. Not useful if booting a machine.\n"
399 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
400 " --system-call-filter=LIST|~LIST\n"
401 " Permit/prohibit specific system calls\n"
402 " -Z --selinux-context=SECLABEL\n"
403 " Set the SELinux security context to be used by\n"
404 " processes in the container\n"
405 " -L --selinux-apifs-context=SECLABEL\n"
406 " Set the SELinux security context to be used by\n"
407 " API/tmpfs file systems in the container\n\n"
408 "%3$sResources:%4$s\n"
409 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
410 " --oom-score-adjust=VALUE\n"
411 " Adjust the OOM score value for the payload\n"
412 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
413 " --personality=ARCH Pick personality for this container\n\n"
414 "%3$sIntegration:%4$s\n"
415 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
416 " --timezone=MODE Select mode of /etc/localtime initialization\n"
417 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
418 " host, try-guest, try-host\n"
419 " -j Equivalent to --link-journal=try-guest\n\n"
421 " --bind=PATH[:PATH[:OPTIONS]]\n"
422 " Bind mount a file or directory from the host into\n"
424 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
425 " Similar, but creates a read-only bind mount\n"
426 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
428 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
429 " --overlay=PATH[:PATH...]:PATH\n"
430 " Create an overlay mount from the host to \n"
432 " --overlay-ro=PATH[:PATH...]:PATH\n"
433 " Similar, but creates a read-only overlay mount\n"
434 " --bind-user=NAME Bind user from host to container\n\n"
435 "%3$sInput/Output:%4$s\n"
436 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
437 " set up for the container.\n"
438 " -P --pipe Equivalent to --console=pipe\n\n"
439 "%3$sCredentials:%4$s\n"
440 " --set-credential=ID:VALUE\n"
441 " Pass a credential with literal value to container.\n"
442 " --load-credential=ID:PATH\n"
443 " Load credential to pass to container from file or\n"
444 " AF_UNIX stream socket.\n"
445 "\nSee the %2$s for details.\n",
446 program_invocation_short_name
,
456 static int custom_mount_check_all(void) {
459 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
460 CustomMount
*m
= &arg_custom_mounts
[i
];
462 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
463 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
464 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
465 "--private-users-ownership=own may not be combined with custom root mounts.");
466 if (arg_uid_shift
== UID_INVALID
)
467 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
468 "--private-users with automatic UID shift may not be combined with custom root mounts.");
475 static int detect_unified_cgroup_hierarchy_from_environment(void) {
476 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
479 /* Allow the user to control whether the unified hierarchy is used */
483 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
484 var
= "UNIFIED_CGROUP_HIERARCHY";
489 r
= parse_boolean(e
);
491 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
493 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
495 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
501 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
504 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
505 * in the image actually supports. */
506 r
= cg_all_unified();
508 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
510 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
511 * routine only detects 231, so we'll have a false negative here for 230. */
512 r
= systemd_installation_has_version(directory
, "230");
514 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
516 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
518 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
519 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
520 /* Mixed cgroup hierarchy support was added in 233 */
521 r
= systemd_installation_has_version(directory
, "233");
523 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
525 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
527 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
529 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
531 log_debug("Using %s hierarchy for container.",
532 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
533 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
538 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
543 _cleanup_free_
char *t
= NULL
;
545 r
= extract_first_word(&spec
, &t
, ",", 0);
547 return log_error_errno(r
, "Failed to parse capability %s.", t
);
551 if (streq(t
, "help")) {
552 for (int i
= 0; i
< capability_list_length(); i
++) {
555 name
= capability_to_name(i
);
566 r
= capability_from_name(t
);
568 return log_error_errno(r
, "Failed to parse capability %s.", t
);
575 return 1; /* continue */
578 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
581 r
= getenv_bool(name
);
585 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
587 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
588 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
592 static int parse_mount_settings_env(void) {
596 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
597 if (r
< 0 && r
!= -ENXIO
)
598 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
600 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
602 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
603 if (streq_ptr(e
, "network"))
604 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
607 r
= parse_boolean(e
);
609 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
611 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
612 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
618 static int parse_environment(void) {
622 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
625 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
628 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
631 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
635 r
= parse_mount_settings_env();
639 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
640 * even if it is supported. If not supported, it has no effect. */
641 if (!cg_ns_supported())
642 arg_use_cgns
= false;
644 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
647 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
651 arg_use_cgns
= r
> 0;
652 arg_settings_mask
|= SETTING_USE_CGNS
;
656 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
658 arg_container_service_name
= e
;
660 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
662 arg_suppress_sync
= r
;
663 else if (r
!= -ENXIO
)
664 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
666 return detect_unified_cgroup_hierarchy_from_environment();
669 static int parse_argv(int argc
, char *argv
[]) {
676 ARG_AMBIENT_CAPABILITY
,
688 ARG_NETWORK_INTERFACE
,
693 ARG_NETWORK_VETH_EXTRA
,
694 ARG_NETWORK_NAMESPACE_PATH
,
704 ARG_PRIVATE_USERS_CHOWN
,
705 ARG_PRIVATE_USERS_OWNERSHIP
,
710 ARG_SYSTEM_CALL_FILTER
,
713 ARG_NO_NEW_PRIVILEGES
,
714 ARG_OOM_SCORE_ADJUST
,
728 static const struct option options
[] = {
729 { "help", no_argument
, NULL
, 'h' },
730 { "version", no_argument
, NULL
, ARG_VERSION
},
731 { "directory", required_argument
, NULL
, 'D' },
732 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
733 { "ephemeral", no_argument
, NULL
, 'x' },
734 { "user", required_argument
, NULL
, 'u' },
735 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
736 { "as-pid2", no_argument
, NULL
, 'a' },
737 { "boot", no_argument
, NULL
, 'b' },
738 { "uuid", required_argument
, NULL
, ARG_UUID
},
739 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
740 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
741 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
742 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
743 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
744 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
745 { "bind", required_argument
, NULL
, ARG_BIND
},
746 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
747 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
748 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
749 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
750 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
751 { "machine", required_argument
, NULL
, 'M' },
752 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
753 { "slice", required_argument
, NULL
, 'S' },
754 { "setenv", required_argument
, NULL
, 'E' },
755 { "selinux-context", required_argument
, NULL
, 'Z' },
756 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
757 { "quiet", no_argument
, NULL
, 'q' },
758 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
759 { "register", required_argument
, NULL
, ARG_REGISTER
},
760 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
761 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
762 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
763 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
764 { "network-veth", no_argument
, NULL
, 'n' },
765 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
766 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
767 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
768 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
769 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
770 { "image", required_argument
, NULL
, 'i' },
771 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
772 { "port", required_argument
, NULL
, 'p' },
773 { "property", required_argument
, NULL
, ARG_PROPERTY
},
774 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
775 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
776 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
777 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
778 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
779 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
780 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
781 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
782 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
783 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
784 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
785 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
786 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
787 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
788 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
789 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
790 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
791 { "console", required_argument
, NULL
, ARG_CONSOLE
},
792 { "pipe", no_argument
, NULL
, ARG_PIPE
},
793 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
794 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
795 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
796 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
797 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
798 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
803 uint64_t plus
= 0, minus
= 0;
804 bool mask_all_settings
= false, mask_no_settings
= false;
809 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
819 r
= parse_path_argument(optarg
, false, &arg_directory
);
823 arg_settings_mask
|= SETTING_DIRECTORY
;
827 r
= parse_path_argument(optarg
, false, &arg_template
);
831 arg_settings_mask
|= SETTING_DIRECTORY
;
835 r
= parse_path_argument(optarg
, false, &arg_image
);
839 arg_settings_mask
|= SETTING_DIRECTORY
;
843 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
850 arg_ephemeral
= true;
851 arg_settings_mask
|= SETTING_EPHEMERAL
;
855 r
= free_and_strdup(&arg_user
, optarg
);
859 arg_settings_mask
|= SETTING_USER
;
862 case ARG_NETWORK_ZONE
: {
865 j
= strjoin("vz-", optarg
);
869 if (!ifname_valid(j
)) {
870 log_error("Network zone name not valid: %s", j
);
875 free_and_replace(arg_network_zone
, j
);
877 arg_network_veth
= true;
878 arg_private_network
= true;
879 arg_settings_mask
|= SETTING_NETWORK
;
883 case ARG_NETWORK_BRIDGE
:
885 if (!ifname_valid(optarg
))
886 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
887 "Bridge interface name not valid: %s", optarg
);
889 r
= free_and_strdup(&arg_network_bridge
, optarg
);
895 arg_network_veth
= true;
896 arg_private_network
= true;
897 arg_settings_mask
|= SETTING_NETWORK
;
900 case ARG_NETWORK_VETH_EXTRA
:
901 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
903 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
905 arg_private_network
= true;
906 arg_settings_mask
|= SETTING_NETWORK
;
909 case ARG_NETWORK_INTERFACE
:
910 if (!ifname_valid(optarg
))
911 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
912 "Network interface name not valid: %s", optarg
);
914 r
= test_network_interface_initialized(optarg
);
918 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
921 arg_private_network
= true;
922 arg_settings_mask
|= SETTING_NETWORK
;
925 case ARG_NETWORK_MACVLAN
:
927 if (!ifname_valid(optarg
))
928 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
929 "MACVLAN network interface name not valid: %s", optarg
);
931 r
= test_network_interface_initialized(optarg
);
935 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
938 arg_private_network
= true;
939 arg_settings_mask
|= SETTING_NETWORK
;
942 case ARG_NETWORK_IPVLAN
:
944 if (!ifname_valid(optarg
))
945 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
946 "IPVLAN network interface name not valid: %s", optarg
);
948 r
= test_network_interface_initialized(optarg
);
952 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
956 case ARG_PRIVATE_NETWORK
:
957 arg_private_network
= true;
958 arg_settings_mask
|= SETTING_NETWORK
;
961 case ARG_NETWORK_NAMESPACE_PATH
:
962 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
966 arg_settings_mask
|= SETTING_NETWORK
;
970 if (arg_start_mode
== START_PID2
)
971 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
972 "--boot and --as-pid2 may not be combined.");
974 arg_start_mode
= START_BOOT
;
975 arg_settings_mask
|= SETTING_START_MODE
;
979 if (arg_start_mode
== START_BOOT
)
980 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
981 "--boot and --as-pid2 may not be combined.");
983 arg_start_mode
= START_PID2
;
984 arg_settings_mask
|= SETTING_START_MODE
;
988 r
= sd_id128_from_string(optarg
, &arg_uuid
);
990 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
992 if (sd_id128_is_null(arg_uuid
))
993 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
994 "Machine UUID may not be all zeroes.");
996 arg_settings_mask
|= SETTING_MACHINE_ID
;
1000 _cleanup_free_
char *mangled
= NULL
;
1002 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1006 free_and_replace(arg_slice
, mangled
);
1007 arg_settings_mask
|= SETTING_SLICE
;
1012 if (isempty(optarg
))
1013 arg_machine
= mfree(arg_machine
);
1015 if (!hostname_is_valid(optarg
, 0))
1016 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1017 "Invalid machine name: %s", optarg
);
1019 r
= free_and_strdup(&arg_machine
, optarg
);
1026 if (isempty(optarg
))
1027 arg_hostname
= mfree(arg_hostname
);
1029 if (!hostname_is_valid(optarg
, 0))
1030 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1031 "Invalid hostname: %s", optarg
);
1033 r
= free_and_strdup(&arg_hostname
, optarg
);
1038 arg_settings_mask
|= SETTING_HOSTNAME
;
1042 arg_selinux_context
= optarg
;
1046 arg_selinux_apifs_context
= optarg
;
1050 arg_read_only
= true;
1051 arg_settings_mask
|= SETTING_READ_ONLY
;
1054 case ARG_AMBIENT_CAPABILITY
: {
1056 r
= parse_capability_spec(optarg
, &m
);
1059 arg_caps_ambient
|= m
;
1060 arg_settings_mask
|= SETTING_CAPABILITY
;
1063 case ARG_CAPABILITY
:
1064 case ARG_DROP_CAPABILITY
: {
1066 r
= parse_capability_spec(optarg
, &m
);
1070 if (c
== ARG_CAPABILITY
)
1074 arg_settings_mask
|= SETTING_CAPABILITY
;
1077 case ARG_NO_NEW_PRIVILEGES
:
1078 r
= parse_boolean(optarg
);
1080 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1082 arg_no_new_privileges
= r
;
1083 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1087 arg_link_journal
= LINK_GUEST
;
1088 arg_link_journal_try
= true;
1089 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1092 case ARG_LINK_JOURNAL
:
1093 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1095 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1097 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1102 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1104 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1106 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1110 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1112 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1114 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1118 case ARG_OVERLAY_RO
:
1119 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1120 if (r
== -EADDRNOTAVAIL
)
1121 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1123 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1125 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1128 case ARG_INACCESSIBLE
:
1129 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1131 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1133 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1137 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1139 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1141 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1148 case ARG_SHARE_SYSTEM
:
1149 /* We don't officially support this anymore, except for compat reasons. People should use the
1150 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1151 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1152 arg_clone_ns_flags
= 0;
1156 r
= parse_boolean(optarg
);
1158 log_error("Failed to parse --register= argument: %s", optarg
);
1166 arg_keep_unit
= true;
1169 case ARG_PERSONALITY
:
1171 arg_personality
= personality_from_string(optarg
);
1172 if (arg_personality
== PERSONALITY_INVALID
)
1173 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1174 "Unknown or unsupported personality '%s'.", optarg
);
1176 arg_settings_mask
|= SETTING_PERSONALITY
;
1182 arg_volatile_mode
= VOLATILE_YES
;
1183 else if (streq(optarg
, "help")) {
1184 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1189 m
= volatile_mode_from_string(optarg
);
1191 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1192 "Failed to parse --volatile= argument: %s", optarg
);
1194 arg_volatile_mode
= m
;
1197 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1201 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1203 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1205 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1207 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1211 if (strv_extend(&arg_property
, optarg
) < 0)
1216 case ARG_PRIVATE_USERS
: {
1221 else if (!in_charset(optarg
, DIGITS
))
1222 /* do *not* parse numbers as booleans */
1223 boolean
= parse_boolean(optarg
);
1228 /* no: User namespacing off */
1229 arg_userns_mode
= USER_NAMESPACE_NO
;
1230 arg_uid_shift
= UID_INVALID
;
1231 arg_uid_range
= UINT32_C(0x10000);
1232 } else if (boolean
> 0) {
1233 /* yes: User namespacing on, UID range is read from root dir */
1234 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1235 arg_uid_shift
= UID_INVALID
;
1236 arg_uid_range
= UINT32_C(0x10000);
1237 } else if (streq(optarg
, "pick")) {
1238 /* pick: User namespacing on, UID range is picked randomly */
1239 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1240 * implied by USER_NAMESPACE_PICK
1242 arg_uid_shift
= UID_INVALID
;
1243 arg_uid_range
= UINT32_C(0x10000);
1245 } else if (streq(optarg
, "identity")) {
1246 /* identitiy: User namespaces on, UID range is map the 0…0xFFFF range to
1247 * itself, i.e. we don't actually map anything, but do take benefit of
1248 * isolation of capability sets. */
1249 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1251 arg_uid_range
= UINT32_C(0x10000);
1253 _cleanup_free_
char *buffer
= NULL
;
1254 const char *range
, *shift
;
1256 /* anything else: User namespacing on, UID range is explicitly configured */
1258 range
= strchr(optarg
, ':');
1260 buffer
= strndup(optarg
, range
- optarg
);
1266 r
= safe_atou32(range
, &arg_uid_range
);
1268 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1272 r
= parse_uid(shift
, &arg_uid_shift
);
1274 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1276 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1278 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1279 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1282 arg_settings_mask
|= SETTING_USERNS
;
1287 if (userns_supported()) {
1288 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1289 * implied by USER_NAMESPACE_PICK
1291 arg_uid_shift
= UID_INVALID
;
1292 arg_uid_range
= UINT32_C(0x10000);
1294 arg_settings_mask
|= SETTING_USERNS
;
1299 case ARG_PRIVATE_USERS_CHOWN
:
1300 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1302 arg_settings_mask
|= SETTING_USERNS
;
1305 case ARG_PRIVATE_USERS_OWNERSHIP
:
1306 if (streq(optarg
, "help")) {
1307 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1311 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1312 if (arg_userns_ownership
< 0)
1313 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1315 arg_settings_mask
|= SETTING_USERNS
;
1318 case ARG_KILL_SIGNAL
:
1319 if (streq(optarg
, "help")) {
1320 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1324 arg_kill_signal
= signal_from_string(optarg
);
1325 if (arg_kill_signal
< 0)
1326 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1328 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1333 /* no → do not read files
1334 * yes → read files, do not override cmdline, trust only subset
1335 * override → read files, override cmdline, trust only subset
1336 * trusted → read files, do not override cmdline, trust all
1339 r
= parse_boolean(optarg
);
1341 if (streq(optarg
, "trusted")) {
1342 mask_all_settings
= false;
1343 mask_no_settings
= false;
1344 arg_settings_trusted
= true;
1346 } else if (streq(optarg
, "override")) {
1347 mask_all_settings
= false;
1348 mask_no_settings
= true;
1349 arg_settings_trusted
= -1;
1351 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1354 mask_all_settings
= false;
1355 mask_no_settings
= false;
1356 arg_settings_trusted
= -1;
1359 mask_all_settings
= true;
1360 mask_no_settings
= false;
1361 arg_settings_trusted
= false;
1367 if (!path_is_absolute(optarg
))
1368 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1369 "Working directory %s is not an absolute path.", optarg
);
1371 r
= free_and_strdup(&arg_chdir
, optarg
);
1375 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1378 case ARG_PIVOT_ROOT
:
1379 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1381 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1383 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1386 case ARG_NOTIFY_READY
:
1387 r
= parse_boolean(optarg
);
1389 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1390 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1391 arg_notify_ready
= r
;
1392 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1395 case ARG_ROOT_HASH
: {
1396 _cleanup_free_
void *k
= NULL
;
1399 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1401 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1402 if (l
< sizeof(sd_id128_t
))
1403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1405 free_and_replace(arg_verity_settings
.root_hash
, k
);
1406 arg_verity_settings
.root_hash_size
= l
;
1410 case ARG_ROOT_HASH_SIG
: {
1415 if ((value
= startswith(optarg
, "base64:"))) {
1416 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1418 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1421 r
= read_full_file(optarg
, (char**) &p
, &l
);
1423 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1426 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1427 arg_verity_settings
.root_hash_sig_size
= l
;
1431 case ARG_VERITY_DATA
:
1432 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1437 case ARG_SYSTEM_CALL_FILTER
: {
1441 negative
= optarg
[0] == '~';
1442 items
= negative
? optarg
+ 1 : optarg
;
1445 _cleanup_free_
char *word
= NULL
;
1447 r
= extract_first_word(&items
, &word
, NULL
, 0);
1453 return log_error_errno(r
, "Failed to parse system call filter: %m");
1456 r
= strv_extend(&arg_syscall_deny_list
, word
);
1458 r
= strv_extend(&arg_syscall_allow_list
, word
);
1463 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1469 _cleanup_free_
char *name
= NULL
;
1472 if (streq(optarg
, "help")) {
1473 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1477 eq
= strchr(optarg
, '=');
1479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1480 "--rlimit= expects an '=' assignment.");
1482 name
= strndup(optarg
, eq
- optarg
);
1486 rl
= rlimit_from_string_harder(name
);
1488 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1490 if (!arg_rlimit
[rl
]) {
1491 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1492 if (!arg_rlimit
[rl
])
1496 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1498 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1500 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1504 case ARG_OOM_SCORE_ADJUST
:
1505 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1507 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1509 arg_oom_score_adjust_set
= true;
1510 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1513 case ARG_CPU_AFFINITY
: {
1516 r
= parse_cpu_set(optarg
, &cpuset
);
1518 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1520 cpu_set_reset(&arg_cpu_set
);
1521 arg_cpu_set
= cpuset
;
1522 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1526 case ARG_RESOLV_CONF
:
1527 if (streq(optarg
, "help")) {
1528 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1532 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1533 if (arg_resolv_conf
< 0)
1534 return log_error_errno(arg_resolv_conf
,
1535 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1537 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1541 if (streq(optarg
, "help")) {
1542 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1546 arg_timezone
= timezone_mode_from_string(optarg
);
1547 if (arg_timezone
< 0)
1548 return log_error_errno(arg_timezone
,
1549 "Failed to parse /etc/localtime mode: %s", optarg
);
1551 arg_settings_mask
|= SETTING_TIMEZONE
;
1555 r
= handle_arg_console(optarg
);
1562 r
= handle_arg_console("pipe");
1568 arg_pager_flags
|= PAGER_DISABLE
;
1571 case ARG_SET_CREDENTIAL
: {
1572 _cleanup_free_
char *word
= NULL
, *data
= NULL
;
1573 const char *p
= optarg
;
1577 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1581 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1583 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1585 if (!credential_name_valid(word
))
1586 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1588 for (size_t i
= 0; i
< arg_n_credentials
; i
++)
1589 if (streq(arg_credentials
[i
].id
, word
))
1590 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1592 l
= cunescape(p
, UNESCAPE_ACCEPT_NUL
, &data
);
1594 return log_error_errno(l
, "Failed to unescape credential data: %s", p
);
1596 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1600 a
[arg_n_credentials
++] = (Credential
) {
1601 .id
= TAKE_PTR(word
),
1602 .data
= TAKE_PTR(data
),
1606 arg_credentials
= a
;
1608 arg_settings_mask
|= SETTING_CREDENTIALS
;
1612 case ARG_LOAD_CREDENTIAL
: {
1613 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
1614 _cleanup_(erase_and_freep
) char *data
= NULL
;
1615 _cleanup_free_
char *word
= NULL
, *j
= NULL
;
1616 const char *p
= optarg
;
1620 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1624 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1626 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1628 if (!credential_name_valid(word
))
1629 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1631 for (i
= 0; i
< arg_n_credentials
; i
++)
1632 if (streq(arg_credentials
[i
].id
, word
))
1633 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1635 if (path_is_absolute(p
))
1636 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
1640 r
= get_credentials_dir(&e
);
1642 return log_error_errno(r
, "Credential not available (no credentials passed at all): %s", word
);
1644 j
= path_join(e
, p
);
1649 r
= read_full_file_full(AT_FDCWD
, j
?: p
, UINT64_MAX
, SIZE_MAX
,
1654 return log_error_errno(r
, "Failed to read credential '%s': %m", j
?: p
);
1656 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1660 a
[arg_n_credentials
++] = (Credential
) {
1661 .id
= TAKE_PTR(word
),
1662 .data
= TAKE_PTR(data
),
1666 arg_credentials
= a
;
1668 arg_settings_mask
|= SETTING_CREDENTIALS
;
1673 if (!valid_user_group_name(optarg
, 0))
1674 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1676 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1679 arg_settings_mask
|= SETTING_BIND_USER
;
1682 case ARG_SUPPRESS_SYNC
:
1683 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1687 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1694 assert_not_reached();
1697 if (argc
> optind
) {
1698 strv_free(arg_parameters
);
1699 arg_parameters
= strv_copy(argv
+ optind
);
1700 if (!arg_parameters
)
1703 arg_settings_mask
|= SETTING_START_MODE
;
1706 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1707 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1708 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1709 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1711 arg_directory
= TAKE_PTR(arg_template
);
1713 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1715 /* Make sure to parse environment before we reset the settings mask below */
1716 r
= parse_environment();
1720 /* Load all settings from .nspawn files */
1721 if (mask_no_settings
)
1722 arg_settings_mask
= 0;
1724 /* Don't load any settings from .nspawn files */
1725 if (mask_all_settings
)
1726 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1731 static int verify_arguments(void) {
1734 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1735 /* If we are running the stub init in the container, we don't need to look at what the init
1736 * in the container supports, because we are not using it. Let's immediately pick the right
1737 * setting based on the host system configuration.
1739 * We only do this, if the user didn't use an environment variable to override the detection.
1742 r
= cg_all_unified();
1744 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1746 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1747 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1748 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1750 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1753 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1754 arg_mount_settings
|= MOUNT_USE_USERNS
;
1756 if (arg_private_network
)
1757 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1759 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1760 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1761 arg_register
= false;
1762 if (arg_start_mode
!= START_PID1
)
1763 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1766 if (arg_userns_ownership
< 0)
1767 arg_userns_ownership
=
1768 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1769 USER_NAMESPACE_OWNERSHIP_OFF
;
1771 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1772 arg_kill_signal
= SIGRTMIN
+3;
1774 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 */
1775 arg_read_only
= true;
1777 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1778 arg_read_only
= true;
1780 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1781 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1782 * The latter is not technically a user session, but we don't need to labour the point. */
1783 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1785 if (arg_directory
&& arg_image
)
1786 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1788 if (arg_template
&& arg_image
)
1789 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1791 if (arg_template
&& !(arg_directory
|| arg_machine
))
1792 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1794 if (arg_ephemeral
&& arg_template
)
1795 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1797 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1798 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1800 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1801 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1803 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1804 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1805 "--read-only and --private-users-ownership=chown may not be combined.");
1807 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1808 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1809 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1810 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1811 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1813 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1814 * we need to error out, to avoid conflicts between different network options. */
1815 if (arg_network_namespace_path
&&
1816 (arg_network_interfaces
|| arg_network_macvlan
||
1817 arg_network_ipvlan
|| arg_network_veth_extra
||
1818 arg_network_bridge
|| arg_network_zone
||
1820 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1822 if (arg_network_bridge
&& arg_network_zone
)
1823 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1824 "--network-bridge= and --network-zone= may not be combined.");
1826 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1827 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1829 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1830 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1832 if (arg_expose_ports
&& !arg_private_network
)
1833 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1835 if (arg_caps_ambient
) {
1836 if (arg_caps_ambient
== UINT64_MAX
)
1837 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1839 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1840 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1842 if (arg_start_mode
== START_BOOT
)
1843 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1846 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1847 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1849 /* Drop duplicate --bind-user= entries */
1850 strv_uniq(arg_bind_user
);
1852 r
= custom_mount_check_all();
1859 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1862 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1865 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1868 if (uid
!= UID_INVALID
) {
1869 uid
+= arg_uid_shift
;
1871 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1875 if (gid
!= GID_INVALID
) {
1876 gid
+= (gid_t
) arg_uid_shift
;
1878 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1882 return RET_NERRNO(lchown(p
, uid
, gid
));
1885 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1889 q
= prefix_roota(root
, path
);
1890 r
= RET_NERRNO(mkdir(q
, mode
));
1896 return userns_lchown(q
, uid
, gid
);
1899 static const char *timezone_from_path(const char *path
) {
1900 return PATH_STARTSWITH_SET(
1902 "../usr/share/zoneinfo/",
1903 "/usr/share/zoneinfo/");
1906 static bool etc_writable(void) {
1907 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1910 static int setup_timezone(const char *dest
) {
1911 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1912 const char *where
, *check
;
1918 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1919 r
= readlink_malloc("/etc/localtime", &p
);
1920 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1921 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1922 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1923 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1925 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1926 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1930 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1933 } else if (arg_timezone
== TIMEZONE_AUTO
)
1934 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1940 if (m
== TIMEZONE_OFF
)
1943 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1945 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1949 where
= strjoina(etc
, "/localtime");
1953 case TIMEZONE_DELETE
:
1954 if (unlink(where
) < 0)
1955 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1959 case TIMEZONE_SYMLINK
: {
1960 _cleanup_free_
char *q
= NULL
;
1961 const char *z
, *what
;
1963 z
= timezone_from_path(p
);
1965 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1969 r
= readlink_malloc(where
, &q
);
1970 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1971 return 0; /* Already pointing to the right place? Then do nothing .. */
1973 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1974 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1976 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1978 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1979 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1980 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1984 what
= strjoina("../usr/share/zoneinfo/", z
);
1985 if (symlink(what
, where
) < 0) {
1986 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1987 errno
, "Failed to correct timezone of container, ignoring: %m");
1997 case TIMEZONE_BIND
: {
1998 _cleanup_free_
char *resolved
= NULL
;
2001 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2003 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
2007 if (found
== 0) /* missing? */
2008 (void) touch(resolved
);
2010 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
2012 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2018 /* If mounting failed, try to copy */
2019 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2021 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2022 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
2029 assert_not_reached();
2032 /* Fix permissions of the symlink or file copy we just created */
2033 r
= userns_lchown(where
, 0, 0);
2035 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
2040 static int have_resolv_conf(const char *path
) {
2043 if (access(path
, F_OK
) < 0) {
2044 if (errno
== ENOENT
)
2047 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
2053 static int resolved_listening(void) {
2054 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2055 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
2056 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
2059 /* Check if resolved is listening */
2061 r
= sd_bus_open_system(&bus
);
2063 return log_debug_errno(r
, "Failed to open system bus: %m");
2065 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2067 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2071 r
= sd_bus_get_property_string(bus
,
2072 "org.freedesktop.resolve1",
2073 "/org/freedesktop/resolve1",
2074 "org.freedesktop.resolve1.Manager",
2077 &dns_stub_listener_mode
);
2079 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2081 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2084 static int setup_resolv_conf(const char *dest
) {
2085 _cleanup_free_
char *etc
= NULL
;
2086 const char *where
, *what
;
2092 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2093 if (arg_private_network
)
2094 m
= RESOLV_CONF_OFF
;
2095 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2096 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2097 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2098 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2100 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2103 m
= arg_resolv_conf
;
2105 if (m
== RESOLV_CONF_OFF
)
2108 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2110 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2114 where
= strjoina(etc
, "/resolv.conf");
2116 if (m
== RESOLV_CONF_DELETE
) {
2117 if (unlink(where
) < 0)
2118 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2123 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2124 what
= PRIVATE_STATIC_RESOLV_CONF
;
2125 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2126 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2127 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2128 what
= PRIVATE_STUB_RESOLV_CONF
;
2130 what
= "/etc/resolv.conf";
2132 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2133 _cleanup_free_
char *resolved
= NULL
;
2136 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2138 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2142 if (found
== 0) /* missing? */
2143 (void) touch(resolved
);
2145 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2147 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2149 /* If that didn't work, let's copy the file */
2152 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2153 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2155 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
2157 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2158 * resolved or something similar runs inside and the symlink points there.
2160 * If the disk image is read-only, there's also no point in complaining.
2162 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2163 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2164 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2168 r
= userns_lchown(where
, 0, 0);
2170 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2175 static int setup_boot_id(void) {
2176 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2177 _cleanup_free_
char *path
= NULL
;
2178 sd_id128_t rnd
= SD_ID128_NULL
;
2182 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2184 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2186 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2188 r
= sd_id128_randomize(&rnd
);
2190 return log_error_errno(r
, "Failed to generate random boot id: %m");
2192 r
= id128_write(path
, ID128_UUID
, rnd
, false);
2194 return log_error_errno(r
, "Failed to write boot id: %m");
2196 from
= TAKE_PTR(path
);
2197 to
= "/proc/sys/kernel/random/boot_id";
2199 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2203 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2206 static int copy_devnodes(const char *dest
) {
2207 static const char devnodes
[] =
2221 BLOCK_WITH_UMASK(0000);
2223 /* Create /dev/net, so that we can create /dev/net/tun in it */
2224 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2225 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2227 NULSTR_FOREACH(d
, devnodes
) {
2228 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2231 from
= path_join("/dev/", d
);
2235 to
= path_join(dest
, from
);
2239 if (stat(from
, &st
) < 0) {
2241 if (errno
!= ENOENT
)
2242 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2244 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2245 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2246 "%s is not a char or block device, cannot copy.", from
);
2248 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2250 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2251 /* Explicitly warn the user when /dev is already populated. */
2252 if (errno
== EEXIST
)
2253 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
);
2255 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2257 /* Some systems abusively restrict mknod but allow bind mounts. */
2260 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2261 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2263 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2266 r
= userns_lchown(to
, 0, 0);
2268 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2270 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2274 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2276 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2278 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2281 prefixed
= path_join(dest
, sl
);
2285 t
= path_join("..", d
);
2289 if (symlink(t
, prefixed
) < 0)
2290 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2297 static int make_extra_nodes(const char *dest
) {
2301 BLOCK_WITH_UMASK(0000);
2303 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2304 _cleanup_free_
char *path
= NULL
;
2305 DeviceNode
*n
= arg_extra_nodes
+ i
;
2307 path
= path_join(dest
, n
->path
);
2311 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2312 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2314 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2316 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2322 static int setup_pts(const char *dest
) {
2323 _cleanup_free_
char *options
= NULL
;
2328 if (arg_selinux_apifs_context
)
2329 (void) asprintf(&options
,
2330 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2331 arg_uid_shift
+ TTY_GID
,
2332 arg_selinux_apifs_context
);
2335 (void) asprintf(&options
,
2336 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2337 arg_uid_shift
+ TTY_GID
);
2342 /* Mount /dev/pts itself */
2343 p
= prefix_roota(dest
, "/dev/pts");
2344 r
= RET_NERRNO(mkdir(p
, 0755));
2346 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2348 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2351 r
= userns_lchown(p
, 0, 0);
2353 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2355 /* Create /dev/ptmx symlink */
2356 p
= prefix_roota(dest
, "/dev/ptmx");
2357 if (symlink("pts/ptmx", p
) < 0)
2358 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2359 r
= userns_lchown(p
, 0, 0);
2361 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2363 /* And fix /dev/pts/ptmx ownership */
2364 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2365 r
= userns_lchown(p
, 0, 0);
2367 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2372 static int setup_stdio_as_dev_console(void) {
2373 _cleanup_close_
int terminal
= -1;
2376 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2377 * explicitly, if we are configured to. */
2378 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2380 return log_error_errno(terminal
, "Failed to open console: %m");
2382 /* Make sure we can continue logging to the original stderr, even if
2383 * stderr points elsewhere now */
2384 r
= log_dup_console();
2386 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2388 /* invalidates 'terminal' on success and failure */
2389 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2392 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2397 static int setup_dev_console(const char *console
) {
2398 _cleanup_free_
char *p
= NULL
;
2401 /* Create /dev/console symlink */
2402 r
= path_make_relative("/dev", console
, &p
);
2404 return log_error_errno(r
, "Failed to create relative path: %m");
2406 if (symlink(p
, "/dev/console") < 0)
2407 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2412 static int setup_keyring(void) {
2413 key_serial_t keyring
;
2415 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2416 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2417 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2418 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2419 * into the container. */
2421 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2422 if (keyring
== -1) {
2423 if (errno
== ENOSYS
)
2424 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2425 else if (ERRNO_IS_PRIVILEGE(errno
))
2426 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2428 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2434 static int setup_credentials(const char *root
) {
2438 if (arg_n_credentials
<= 0)
2441 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2443 return log_error_errno(r
, "Failed to create /run/host: %m");
2445 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2447 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2449 q
= prefix_roota(root
, "/run/host/credentials");
2450 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2454 for (size_t i
= 0; i
< arg_n_credentials
; i
++) {
2455 _cleanup_free_
char *j
= NULL
;
2456 _cleanup_close_
int fd
= -1;
2458 j
= path_join(q
, arg_credentials
[i
].id
);
2462 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2464 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2466 r
= loop_write(fd
, arg_credentials
[i
].data
, arg_credentials
[i
].size
, /* do_poll= */ false);
2468 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2470 if (fchmod(fd
, 0400) < 0)
2471 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2473 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2474 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2475 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2479 if (chmod(q
, 0500) < 0)
2480 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2482 r
= userns_lchown(q
, 0, 0);
2486 /* Make both mount and superblock read-only now */
2487 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2491 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2494 static int setup_kmsg(int kmsg_socket
) {
2495 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2496 _cleanup_free_
char *fifo
= NULL
;
2497 _cleanup_close_
int fd
= -1;
2500 assert(kmsg_socket
>= 0);
2502 BLOCK_WITH_UMASK(0000);
2504 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2505 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2506 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2507 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2509 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2511 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2513 if (mkfifo(fifo
, 0600) < 0)
2514 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2516 from
= TAKE_PTR(fifo
);
2518 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2522 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2524 return log_error_errno(errno
, "Failed to open fifo: %m");
2526 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2527 r
= send_one_fd(kmsg_socket
, fd
, 0);
2529 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2535 union in_addr_union address4
;
2536 union in_addr_union address6
;
2537 struct FirewallContext
*fw_ctx
;
2540 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2541 struct ExposeArgs
*args
= userdata
;
2547 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2548 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2552 static int setup_hostname(void) {
2555 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2558 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2560 return log_error_errno(r
, "Failed to set hostname: %m");
2565 static int setup_journal(const char *directory
) {
2566 _cleanup_free_
char *d
= NULL
;
2572 /* Don't link journals in ephemeral mode */
2576 if (arg_link_journal
== LINK_NO
)
2579 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2581 r
= sd_id128_get_machine(&this_id
);
2583 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2585 if (sd_id128_equal(arg_uuid
, this_id
)) {
2586 log_full(try ? LOG_WARNING
: LOG_ERR
,
2587 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2593 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2594 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2596 bool ignore
= r
== -EROFS
&& try;
2597 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2598 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2599 return ignore
? 0 : r
;
2603 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2604 q
= prefix_roota(directory
, p
);
2606 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2610 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2611 "%s: already a mount point, refusing to use for journal", p
);
2614 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2618 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2619 "%s: already a mount point, refusing to use for journal", q
);
2622 r
= readlink_and_make_absolute(p
, &d
);
2624 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2627 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2629 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2634 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2635 } else if (r
== -EINVAL
) {
2637 if (arg_link_journal
== LINK_GUEST
&&
2640 if (errno
== ENOTDIR
) {
2641 log_error("%s already exists and is neither a symlink nor a directory", p
);
2644 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2646 } else if (r
!= -ENOENT
)
2647 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2649 if (arg_link_journal
== LINK_GUEST
) {
2651 if (symlink(q
, p
) < 0) {
2653 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2656 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2659 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2661 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2665 if (arg_link_journal
== LINK_HOST
) {
2666 /* don't create parents here — if the host doesn't have
2667 * permanent journal set up, don't force it here */
2669 r
= RET_NERRNO(mkdir(p
, 0755));
2670 if (r
< 0 && r
!= -EEXIST
) {
2672 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2675 return log_error_errno(r
, "Failed to create %s: %m", p
);
2678 } else if (access(p
, F_OK
) < 0)
2681 if (dir_is_empty(q
) == 0)
2682 log_warning("%s is not empty, proceeding anyway.", q
);
2684 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2686 return log_error_errno(r
, "Failed to create %s: %m", q
);
2688 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2690 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2695 static int drop_capabilities(uid_t uid
) {
2696 CapabilityQuintet q
;
2698 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2699 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2700 * arg_caps_retain. */
2702 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2703 q
= arg_full_capabilities
;
2705 if (q
.bounding
== UINT64_MAX
)
2706 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2708 if (q
.effective
== UINT64_MAX
)
2709 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2711 if (q
.inheritable
== UINT64_MAX
)
2712 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2714 if (q
.permitted
== UINT64_MAX
)
2715 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2717 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2718 q
.ambient
= arg_caps_ambient
;
2720 if (capability_quintet_mangle(&q
))
2721 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2724 q
= (CapabilityQuintet
) {
2725 .bounding
= arg_caps_retain
,
2726 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2727 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2728 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2729 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2732 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2733 * in order to maintain the same behavior as systemd < 242. */
2734 if (capability_quintet_mangle(&q
))
2735 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2736 "Some capabilities will not be set because they are not in the current bounding set.");
2740 return capability_quintet_enforce(&q
);
2743 static int reset_audit_loginuid(void) {
2744 _cleanup_free_
char *p
= NULL
;
2747 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2750 r
= read_one_line_file("/proc/self/loginuid", &p
);
2754 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2756 /* Already reset? */
2757 if (streq(p
, "4294967295"))
2760 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2763 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2764 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2765 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2766 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2767 "using systemd-nspawn. Sleeping for 5s... (%m)");
2775 static int setup_propagate(const char *root
) {
2779 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2780 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2781 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2782 (void) mkdir_p(p
, 0600);
2784 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2786 return log_error_errno(r
, "Failed to create /run/host: %m");
2788 r
= userns_mkdir(root
, "/run/host/incoming", 0600, 0, 0);
2790 return log_error_errno(r
, "Failed to create /run/host/incoming: %m");
2792 q
= prefix_roota(root
, "/run/host/incoming");
2793 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2797 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2801 /* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
2802 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2805 static int setup_machine_id(const char *directory
) {
2806 const char *etc_machine_id
;
2810 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2811 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2812 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2813 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2814 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2815 * container behaves nicely). */
2817 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2819 r
= id128_read(etc_machine_id
, ID128_PLAIN_OR_UNINIT
, &id
);
2821 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2822 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2824 if (sd_id128_is_null(arg_uuid
)) {
2825 r
= sd_id128_randomize(&arg_uuid
);
2827 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2830 if (sd_id128_is_null(id
))
2831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2832 "Machine ID in container image is zero, refusing.");
2840 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2845 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2848 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2849 if (r
== -EOPNOTSUPP
)
2850 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.");
2852 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2854 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2856 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2858 log_debug("Patched directory tree to match UID/GID range.");
2865 * < 0 : wait_for_terminate() failed to get the state of the
2866 * container, the container was terminated by a signal, or
2867 * failed for an unknown reason. No change is made to the
2868 * container argument.
2869 * > 0 : The program executed in the container terminated with an
2870 * error. The exit code of the program executed in the
2871 * container is returned. The container argument has been set
2872 * to CONTAINER_TERMINATED.
2873 * 0 : The container is being rebooted, has been shut down or exited
2874 * successfully. The container argument has been set to either
2875 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2877 * That is, success is indicated by a return value of zero, and an
2878 * error is indicated by a non-zero value.
2880 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2884 r
= wait_for_terminate(pid
, &status
);
2886 return log_warning_errno(r
, "Failed to wait for container: %m");
2888 switch (status
.si_code
) {
2891 if (status
.si_status
== 0)
2892 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2894 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2896 *container
= CONTAINER_TERMINATED
;
2897 return status
.si_status
;
2900 if (status
.si_status
== SIGINT
) {
2901 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2902 *container
= CONTAINER_TERMINATED
;
2905 } else if (status
.si_status
== SIGHUP
) {
2906 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2907 *container
= CONTAINER_REBOOTED
;
2913 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2914 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2917 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2918 "Container %s failed due to unknown reason.", arg_machine
);
2922 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2925 pid
= PTR_TO_PID(userdata
);
2927 if (kill(pid
, arg_kill_signal
) >= 0) {
2928 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2929 sd_event_source_set_userdata(s
, NULL
);
2934 sd_event_exit(sd_event_source_get_event(s
), 0);
2938 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2944 pid
= PTR_TO_PID(userdata
);
2949 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2950 return log_error_errno(errno
, "Failed to waitid(): %m");
2951 if (si
.si_pid
== 0) /* No pending children. */
2953 if (si
.si_pid
== pid
) {
2954 /* The main process we care for has exited. Return from
2955 * signal handler but leave the zombie. */
2956 sd_event_exit(sd_event_source_get_event(s
), 0);
2960 /* Reap all other children. */
2961 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2967 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2972 pid
= PTR_TO_PID(userdata
);
2974 if (arg_kill_signal
> 0) {
2975 log_info("Container termination requested. Attempting to halt container.");
2976 (void) kill(pid
, arg_kill_signal
);
2978 log_info("Container termination requested. Exiting.");
2979 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2985 static int determine_names(void) {
2988 if (arg_template
&& !arg_directory
&& arg_machine
) {
2990 /* If --template= was specified then we should not
2991 * search for a machine, but instead create a new one
2992 * in /var/lib/machine. */
2994 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2999 if (!arg_image
&& !arg_directory
) {
3001 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3003 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3005 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3007 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3009 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3010 r
= free_and_strdup(&arg_image
, i
->path
);
3012 r
= free_and_strdup(&arg_directory
, i
->path
);
3017 arg_read_only
= arg_read_only
|| i
->read_only
;
3019 r
= safe_getcwd(&arg_directory
);
3021 return log_error_errno(r
, "Failed to determine current directory: %m");
3024 if (!arg_directory
&& !arg_image
)
3025 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3029 if (arg_directory
&& path_equal(arg_directory
, "/"))
3030 arg_machine
= gethostname_malloc();
3031 else if (arg_image
) {
3034 arg_machine
= strdup(basename(arg_image
));
3036 /* Truncate suffix if there is one */
3037 e
= endswith(arg_machine
, ".raw");
3041 arg_machine
= strdup(basename(arg_directory
));
3045 hostname_cleanup(arg_machine
);
3046 if (!hostname_is_valid(arg_machine
, 0))
3047 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3049 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3050 * instances at once without manually having to specify -M each time. */
3052 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3059 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
3068 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
3070 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3072 return free_and_replace(*p
, chased
);
3075 static int determine_uid_shift(const char *directory
) {
3077 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3082 if (arg_uid_shift
== UID_INVALID
) {
3085 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3087 if (stat(directory
, &st
) < 0)
3088 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3090 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3092 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3093 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3094 "UID and GID base of %s don't match.", directory
);
3096 arg_uid_range
= UINT32_C(0x10000);
3098 if (arg_uid_shift
!= 0) {
3099 /* If the image is shifted already, then we'll fall back to classic chowning, for
3100 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3102 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3103 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3104 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3105 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3106 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3107 "UID base of %s is not zero, UID mapping not supported.", directory
);
3111 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3112 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3117 static unsigned long effective_clone_ns_flags(void) {
3118 unsigned long flags
= arg_clone_ns_flags
;
3120 if (arg_private_network
)
3121 flags
|= CLONE_NEWNET
;
3123 flags
|= CLONE_NEWCGROUP
;
3124 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3125 flags
|= CLONE_NEWUSER
;
3130 static int patch_sysctl(void) {
3132 /* This table is inspired by runc's sysctl() function */
3133 static const struct {
3136 unsigned long clone_flags
;
3138 { "kernel.hostname", false, CLONE_NEWUTS
},
3139 { "kernel.domainname", false, CLONE_NEWUTS
},
3140 { "kernel.msgmax", false, CLONE_NEWIPC
},
3141 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3142 { "kernel.msgmni", false, CLONE_NEWIPC
},
3143 { "kernel.sem", false, CLONE_NEWIPC
},
3144 { "kernel.shmall", false, CLONE_NEWIPC
},
3145 { "kernel.shmmax", false, CLONE_NEWIPC
},
3146 { "kernel.shmmni", false, CLONE_NEWIPC
},
3147 { "fs.mqueue.", true, CLONE_NEWIPC
},
3148 { "net.", true, CLONE_NEWNET
},
3151 unsigned long flags
;
3154 flags
= effective_clone_ns_flags();
3156 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3160 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3162 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3165 if (safe_sysctl
[i
].prefix
)
3166 good
= startswith(*k
, safe_sysctl
[i
].key
);
3168 good
= streq(*k
, safe_sysctl
[i
].key
);
3175 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3177 r
= sysctl_write(*k
, *v
);
3179 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3185 static int inner_child(
3187 const char *directory
,
3191 int master_pty_socket
,
3193 char **os_release_pairs
) {
3195 _cleanup_free_
char *home
= NULL
;
3198 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3199 NULL
, /* container */
3204 NULL
, /* container_uuid */
3205 NULL
, /* LISTEN_FDS */
3206 NULL
, /* LISTEN_PID */
3207 NULL
, /* NOTIFY_SOCKET */
3208 NULL
, /* CREDENTIALS_DIRECTORY */
3211 const char *exec_target
;
3212 _cleanup_strv_free_
char **env_use
= NULL
;
3213 int r
, which_failed
;
3215 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3216 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3217 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3218 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3219 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3220 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3223 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3224 * unshare(). See below. */
3228 assert(kmsg_socket
>= 0);
3230 log_debug("Inner child is initializing.");
3232 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3233 /* Tell the parent, that it now can write the UID map. */
3234 (void) barrier_place(barrier
); /* #1 */
3236 /* Wait until the parent wrote the UID map */
3237 if (!barrier_place_and_sync(barrier
)) /* #2 */
3238 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3240 /* Become the new root user inside our namespace */
3241 r
= reset_uid_gid();
3243 return log_error_errno(r
, "Couldn't become new root: %m");
3245 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3246 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3247 * propagation, but simply create new peer groups for all our mounts). */
3248 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3254 arg_mount_settings
| MOUNT_IN_USERNS
,
3256 arg_selinux_apifs_context
);
3260 if (!arg_network_namespace_path
&& arg_private_network
) {
3261 r
= unshare(CLONE_NEWNET
);
3263 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3265 /* Tell the parent that it can setup network interfaces. */
3266 (void) barrier_place(barrier
); /* #3 */
3269 r
= mount_sysfs(NULL
, arg_mount_settings
);
3273 /* Wait until we are cgroup-ified, so that we
3274 * can mount the right cgroup path writable */
3275 if (!barrier_place_and_sync(barrier
)) /* #4 */
3276 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3277 "Parent died too early");
3280 r
= unshare(CLONE_NEWCGROUP
);
3282 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3285 arg_unified_cgroup_hierarchy
,
3286 arg_userns_mode
!= USER_NAMESPACE_NO
,
3289 arg_selinux_apifs_context
,
3292 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3296 r
= setup_boot_id();
3300 r
= setup_kmsg(kmsg_socket
);
3303 kmsg_socket
= safe_close(kmsg_socket
);
3308 arg_n_custom_mounts
,
3311 arg_selinux_apifs_context
,
3312 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3317 return log_error_errno(errno
, "setsid() failed: %m");
3319 if (arg_private_network
)
3320 (void) loopback_setup();
3322 if (arg_expose_ports
) {
3323 r
= expose_port_send_rtnl(rtnl_socket
);
3326 rtnl_socket
= safe_close(rtnl_socket
);
3329 if (arg_console_mode
!= CONSOLE_PIPE
) {
3330 _cleanup_close_
int master
= -1;
3331 _cleanup_free_
char *console
= NULL
;
3333 /* Allocate a pty and make it available as /dev/console. */
3334 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3336 return log_error_errno(master
, "Failed to allocate a pty: %m");
3338 r
= setup_dev_console(console
);
3340 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3342 r
= send_one_fd(master_pty_socket
, master
, 0);
3344 return log_error_errno(r
, "Failed to send master fd: %m");
3345 master_pty_socket
= safe_close(master_pty_socket
);
3347 r
= setup_stdio_as_dev_console();
3356 if (arg_oom_score_adjust_set
) {
3357 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3359 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3362 if (arg_cpu_set
.set
)
3363 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3364 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3366 (void) setup_hostname();
3368 if (arg_personality
!= PERSONALITY_INVALID
) {
3369 r
= safe_personality(arg_personality
);
3371 return log_error_errno(r
, "personality() failed: %m");
3372 } else if (secondary
) {
3373 r
= safe_personality(PER_LINUX32
);
3375 return log_error_errno(r
, "personality() failed: %m");
3378 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3380 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3385 if (is_seccomp_available()) {
3387 r
= seccomp_load(arg_seccomp
);
3388 if (ERRNO_IS_SECCOMP_FATAL(r
))
3389 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3391 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3396 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3401 if (arg_suppress_sync
) {
3403 r
= seccomp_suppress_sync();
3405 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3407 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3412 if (arg_selinux_context
)
3413 if (setexeccon(arg_selinux_context
) < 0)
3414 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3417 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3418 * if we need to later on. */
3419 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3420 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3422 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3423 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3425 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3429 r
= drop_capabilities(getuid());
3431 return log_error_errno(r
, "Dropping capabilities failed: %m");
3433 if (arg_no_new_privileges
)
3434 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3435 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3437 /* LXC sets container=lxc, so follow the scheme here */
3438 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3440 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3444 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3445 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3448 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3449 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3450 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3453 assert(!sd_id128_is_null(arg_uuid
));
3455 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3458 if (fdset_size(fds
) > 0) {
3459 r
= fdset_cloexec(fds
, false);
3461 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3463 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3464 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3467 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3470 if (arg_n_credentials
> 0) {
3471 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3477 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3481 /* Let the parent know that we are ready and
3482 * wait until the parent is ready with the
3484 if (!barrier_place_and_sync(barrier
)) /* #5 */
3485 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3488 if (chdir(arg_chdir
) < 0)
3489 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3491 if (arg_start_mode
== START_PID2
) {
3492 r
= stub_pid1(arg_uuid
);
3497 if (arg_console_mode
!= CONSOLE_PIPE
) {
3498 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3499 * are configured for that. Acquire it as controlling tty. */
3500 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3501 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3504 log_debug("Inner child completed, invoking payload.");
3506 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3507 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3508 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3510 log_set_open_when_needed(true);
3512 (void) fdset_close_others(fds
);
3514 if (arg_start_mode
== START_BOOT
) {
3518 /* Automatically search for the init system */
3520 m
= strv_length(arg_parameters
);
3521 a
= newa(char*, m
+ 2);
3522 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3525 FOREACH_STRING(init
,
3526 "/usr/lib/systemd/systemd",
3527 "/lib/systemd/systemd",
3529 a
[0] = (char*) init
;
3530 execve(a
[0], a
, env_use
);
3533 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3534 } else if (!strv_isempty(arg_parameters
)) {
3535 const char *dollar_path
;
3537 exec_target
= arg_parameters
[0];
3539 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3541 dollar_path
= strv_env_get(env_use
, "PATH");
3543 if (setenv("PATH", dollar_path
, 1) < 0)
3544 return log_error_errno(errno
, "Failed to update $PATH: %m");
3547 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3550 /* If we cannot change the directory, we'll end up in /, that is expected. */
3551 (void) chdir(home
?: "/root");
3553 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3554 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3555 execle("/bin/bash", "-bash", NULL
, env_use
);
3556 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3557 execle("/bin/sh", "-sh", NULL
, env_use
);
3559 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3562 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3565 static int setup_notify_child(void) {
3566 _cleanup_close_
int fd
= -1;
3567 static const union sockaddr_union sa
= {
3568 .un
.sun_family
= AF_UNIX
,
3569 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3573 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3575 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3577 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3578 (void) sockaddr_un_unlink(&sa
.un
);
3580 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3582 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3584 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3586 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3588 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3590 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3595 static int outer_child(
3597 const char *directory
,
3598 DissectedImage
*dissected_image
,
3605 int uid_shift_socket
,
3606 int master_pty_socket
,
3607 int unified_cgroup_hierarchy_socket
,
3611 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3612 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3613 _cleanup_close_
int fd
= -1;
3620 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3621 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3622 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3623 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3624 * forked off it, and it exits. */
3628 assert(pid_socket
>= 0);
3629 assert(uuid_socket
>= 0);
3630 assert(notify_socket
>= 0);
3631 assert(master_pty_socket
>= 0);
3632 assert(kmsg_socket
>= 0);
3634 log_debug("Outer child is initializing.");
3636 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3638 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3640 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3641 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3643 r
= reset_audit_loginuid();
3647 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3648 * mounts to the real root. */
3649 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3653 if (dissected_image
) {
3654 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3655 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3656 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3657 * right place right away. This makes sure ESP partitions and userns are compatible. */
3659 r
= dissected_image_mount_and_warn(
3664 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3665 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3666 DISSECT_IMAGE_USR_NO_ROOT
|
3667 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3668 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3673 r
= determine_uid_shift(directory
);
3677 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3678 /* Let the parent know which UID shift we read from the image */
3679 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3681 return log_error_errno(errno
, "Failed to send UID shift: %m");
3682 if (l
!= sizeof(arg_uid_shift
))
3683 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3684 "Short write while sending UID shift.");
3686 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3687 /* When we are supposed to pick the UID shift, the parent will check now whether the
3688 * UID shift we just read from the image is available. If yes, it will send the UID
3689 * shift back to us, if not it will pick a different one, and send it back to us. */
3691 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3693 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3694 if (l
!= sizeof(arg_uid_shift
))
3695 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3696 "Short read while receiving UID shift.");
3699 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3700 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3703 if (path_equal(directory
, "/")) {
3704 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3705 * place, so that we can make changes to its mount structure (for example, to implement
3706 * --volatile=) without this interfering with our ability to access files such as
3707 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3708 * (instead of a temporary directory, since we are living in our own mount namspace here
3709 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3710 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3712 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3716 directory
= "/run/systemd/nspawn-root";
3719 r
= setup_pivot_root(
3722 arg_pivot_root_old
);
3726 r
= setup_volatile_mode(
3730 arg_selinux_apifs_context
);
3734 r
= bind_user_prepare(
3739 &arg_custom_mounts
, &arg_n_custom_mounts
,
3740 &bind_user_context
);
3744 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3745 /* Send the user maps we determined to the parent, so that it installs it in our user
3746 * namespace UID map table */
3748 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3750 bind_user_context
->data
[i
].payload_user
->uid
,
3751 bind_user_context
->data
[i
].host_user
->uid
,
3752 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3753 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3756 l
= send(uid_shift_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3758 return log_error_errno(errno
, "Failed to send user UID map: %m");
3759 if (l
!= sizeof(map
))
3760 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3761 "Short write while sending user UID map.");
3768 arg_n_custom_mounts
,
3771 arg_selinux_apifs_context
,
3776 /* Make sure we always have a mount that we can move to root later on. */
3777 r
= make_mount_point(directory
);
3781 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3782 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3783 arg_uid_shift
!= 0) {
3785 r
= remount_idmap(directory
, arg_uid_shift
, arg_uid_range
, REMOUNT_IDMAP_HOST_ROOT
);
3786 if (r
== -EINVAL
|| ERRNO_IS_NOT_SUPPORTED(r
)) {
3787 /* This might fail because the kernel or file system doesn't support idmapping. We
3788 * can't really distinguish this nicely, nor do we have any guarantees about the
3789 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3790 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3791 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3792 "ID mapped mounts are apparently not available, sorry.");
3794 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3795 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3797 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3799 log_debug("ID mapped mounts available, making use of them.");
3804 if (dissected_image
) {
3805 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3806 r
= dissected_image_mount(
3811 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3812 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3813 DISSECT_IMAGE_USR_NO_ROOT
|
3814 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3815 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3817 return log_error_errno(r
, "File system check for image failed: %m");
3819 return log_error_errno(r
, "Failed to mount image file system: %m");
3822 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3823 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3825 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3829 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3831 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3832 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3833 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3834 "Short write while sending cgroup mode.");
3836 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3839 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3840 * mounts available in systemd services inside the container that create a new mount namespace. See
3841 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3842 * will inherit the shared propagation mode.
3844 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3845 * directory mount to root later on.
3846 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3848 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3852 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3856 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3860 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3861 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3862 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3864 return log_error_errno(r
, "Failed to make tree read-only: %m");
3867 r
= mount_all(directory
,
3870 arg_selinux_apifs_context
);
3874 r
= copy_devnodes(directory
);
3878 r
= make_extra_nodes(directory
);
3882 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3884 p
= prefix_roota(directory
, "/run/host");
3885 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3887 r
= setup_pts(directory
);
3891 r
= setup_propagate(directory
);
3895 r
= setup_keyring();
3899 r
= setup_credentials(directory
);
3903 r
= bind_user_setup(bind_user_context
, directory
);
3910 arg_n_custom_mounts
,
3913 arg_selinux_apifs_context
,
3914 MOUNT_NON_ROOT_ONLY
);
3918 r
= setup_timezone(directory
);
3922 r
= setup_resolv_conf(directory
);
3926 r
= setup_machine_id(directory
);
3930 r
= setup_journal(directory
);
3934 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3935 p
= prefix_roota(directory
, "/run/host/container-manager");
3936 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
);
3938 /* The same stuff as the $container_uuid env var */
3939 p
= prefix_roota(directory
, "/run/host/container-uuid");
3940 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
3942 if (!arg_use_cgns
) {
3945 arg_unified_cgroup_hierarchy
,
3946 arg_userns_mode
!= USER_NAMESPACE_NO
,
3949 arg_selinux_apifs_context
,
3955 r
= mount_move_root(directory
);
3957 return log_error_errno(r
, "Failed to move root directory: %m");
3959 fd
= setup_notify_child();
3963 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3964 arg_clone_ns_flags
|
3965 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3967 return log_error_errno(errno
, "Failed to fork inner child: %m");
3969 pid_socket
= safe_close(pid_socket
);
3970 uuid_socket
= safe_close(uuid_socket
);
3971 notify_socket
= safe_close(notify_socket
);
3972 uid_shift_socket
= safe_close(uid_shift_socket
);
3974 /* The inner child has all namespaces that are requested, so that we all are owned by the
3975 * user if user namespaces are turned on. */
3977 if (arg_network_namespace_path
) {
3978 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3980 return log_error_errno(r
, "Failed to join network namespace: %m");
3983 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
, os_release_pairs
);
3985 _exit(EXIT_FAILURE
);
3987 _exit(EXIT_SUCCESS
);
3990 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3992 return log_error_errno(errno
, "Failed to send PID: %m");
3993 if (l
!= sizeof(pid
))
3994 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3995 "Short write while sending PID.");
3997 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3999 return log_error_errno(errno
, "Failed to send machine ID: %m");
4000 if (l
!= sizeof(arg_uuid
))
4001 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4002 "Short write while sending machine ID.");
4004 l
= send_one_fd(notify_socket
, fd
, 0);
4006 return log_error_errno(l
, "Failed to send notify fd: %m");
4008 pid_socket
= safe_close(pid_socket
);
4009 uuid_socket
= safe_close(uuid_socket
);
4010 notify_socket
= safe_close(notify_socket
);
4011 master_pty_socket
= safe_close(master_pty_socket
);
4012 kmsg_socket
= safe_close(kmsg_socket
);
4013 rtnl_socket
= safe_close(rtnl_socket
);
4014 netns_fd
= safe_close(netns_fd
);
4019 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4020 bool tried_hashed
= false;
4021 unsigned n_tries
= 100;
4026 assert(ret_lock_file
);
4027 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4028 assert(arg_uid_range
== 0x10000U
);
4032 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4035 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4036 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4041 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4043 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4046 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4047 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4048 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4053 /* Make some superficial checks whether the range is currently known in the user database */
4054 if (getpwuid(candidate
))
4056 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4058 if (getgrgid(candidate
))
4060 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4063 *ret_lock_file
= lf
;
4064 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4069 if (arg_machine
&& !tried_hashed
) {
4070 /* Try to hash the base from the container name */
4072 static const uint8_t hash_key
[] = {
4073 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4074 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4077 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4079 tried_hashed
= true;
4081 random_bytes(&candidate
, sizeof(candidate
));
4083 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4084 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4088 static int add_one_uid_map(
4090 uid_t container_uid
,
4094 return strextendf(p
,
4095 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4096 container_uid
, host_uid
, range
);
4099 static int make_uid_map_string(
4100 const uid_t bind_user_uid
[],
4101 size_t n_bind_user_uid
,
4105 _cleanup_free_
char *s
= NULL
;
4106 uid_t previous_uid
= 0;
4109 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4110 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4113 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4114 * quadruplet, consisting of host and container UID + GID. */
4116 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4117 uid_t payload_uid
= bind_user_uid
[i
*2+offset
],
4118 host_uid
= bind_user_uid
[i
*2+offset
+1];
4120 assert(previous_uid
<= payload_uid
);
4121 assert(payload_uid
< arg_uid_range
);
4123 /* Add a range to close the gap to previous entry */
4124 if (payload_uid
> previous_uid
) {
4125 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4130 /* Map this specific user */
4131 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4135 previous_uid
= payload_uid
+ 1;
4138 /* And add a range to close the gap to finish the range */
4139 if (arg_uid_range
> previous_uid
) {
4140 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4151 static int setup_uid_map(
4153 const uid_t bind_user_uid
[],
4154 size_t n_bind_user_uid
) {
4156 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4157 _cleanup_free_
char *s
= NULL
;
4162 /* Build the UID map string */
4163 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4166 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4167 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4169 return log_error_errno(r
, "Failed to write UID map: %m");
4171 /* And now build the GID map string */
4173 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4176 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4177 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4179 return log_error_errno(r
, "Failed to write GID map: %m");
4184 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4185 char buf
[NOTIFY_BUFFER_MAX
+1];
4187 struct iovec iovec
= {
4189 .iov_len
= sizeof(buf
)-1,
4191 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4192 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4193 struct msghdr msghdr
= {
4196 .msg_control
= &control
,
4197 .msg_controllen
= sizeof(control
),
4199 struct ucred
*ucred
;
4201 pid_t inner_child_pid
;
4202 _cleanup_strv_free_
char **tags
= NULL
;
4207 inner_child_pid
= PTR_TO_PID(userdata
);
4209 if (revents
!= EPOLLIN
) {
4210 log_warning("Got unexpected poll event for notify fd.");
4214 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4216 if (ERRNO_IS_TRANSIENT(n
))
4219 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4222 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4225 cmsg_close_all(&msghdr
);
4227 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4228 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4229 log_debug("Received notify message without valid credentials. Ignoring.");
4233 if ((size_t) n
>= sizeof(buf
)) {
4234 log_warning("Received notify message exceeded maximum size. Ignoring.");
4239 tags
= strv_split(buf
, "\n\r");
4243 if (strv_contains(tags
, "READY=1")) {
4244 r
= sd_notify(false, "READY=1\n");
4246 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4249 p
= strv_find_startswith(tags
, "STATUS=");
4251 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4256 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4259 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4261 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4263 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4268 static int merge_settings(Settings
*settings
, const char *path
) {
4274 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4275 * that this steals the fields of the Settings* structure, and hence modifies it. */
4277 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4278 settings
->start_mode
>= 0) {
4279 arg_start_mode
= settings
->start_mode
;
4280 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4283 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4284 settings
->ephemeral
>= 0)
4285 arg_ephemeral
= settings
->ephemeral
;
4287 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4290 if (!arg_settings_trusted
)
4291 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4293 free_and_replace(arg_directory
, settings
->root
);
4296 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4297 settings
->pivot_root_new
) {
4298 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4299 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4302 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4303 settings
->working_directory
)
4304 free_and_replace(arg_chdir
, settings
->working_directory
);
4306 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4307 settings
->environment
)
4308 strv_free_and_replace(arg_setenv
, settings
->environment
);
4310 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4313 free_and_replace(arg_user
, settings
->user
);
4315 if (uid_is_valid(settings
->uid
))
4316 arg_uid
= settings
->uid
;
4317 if (gid_is_valid(settings
->gid
))
4318 arg_gid
= settings
->gid
;
4319 if (settings
->n_supplementary_gids
> 0) {
4320 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4321 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4325 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4326 uint64_t plus
, minus
;
4327 uint64_t network_minus
= 0;
4330 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4331 * Settings structure */
4333 plus
= settings
->capability
;
4334 minus
= settings
->drop_capability
;
4336 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4337 settings_network_configured(settings
)) {
4338 if (settings_private_network(settings
))
4339 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4341 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4344 if (!arg_settings_trusted
&& plus
!= 0) {
4345 if (settings
->capability
!= 0)
4346 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4348 arg_caps_retain
&= ~network_minus
;
4349 arg_caps_retain
|= plus
;
4352 arg_caps_retain
&= ~minus
;
4354 /* Copy the full capabilities over too */
4355 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4356 if (!arg_settings_trusted
)
4357 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4359 arg_full_capabilities
= settings
->full_capabilities
;
4362 ambient
= settings
->ambient_capability
;
4363 if (!arg_settings_trusted
&& ambient
!= 0)
4364 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4366 arg_caps_ambient
|= ambient
;
4369 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4370 settings
->kill_signal
> 0)
4371 arg_kill_signal
= settings
->kill_signal
;
4373 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4374 settings
->personality
!= PERSONALITY_INVALID
)
4375 arg_personality
= settings
->personality
;
4377 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4378 !sd_id128_is_null(settings
->machine_id
)) {
4380 if (!arg_settings_trusted
)
4381 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4383 arg_uuid
= settings
->machine_id
;
4386 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4387 settings
->read_only
>= 0)
4388 arg_read_only
= settings
->read_only
;
4390 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4391 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4392 arg_volatile_mode
= settings
->volatile_mode
;
4394 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4395 settings
->n_custom_mounts
> 0) {
4397 if (!arg_settings_trusted
)
4398 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4400 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4401 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4402 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4403 settings
->n_custom_mounts
= 0;
4407 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4408 settings_network_configured(settings
)) {
4410 if (!arg_settings_trusted
)
4411 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4413 arg_network_veth
= settings_network_veth(settings
);
4414 arg_private_network
= settings_private_network(settings
);
4416 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4417 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4418 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4419 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4421 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4422 free_and_replace(arg_network_zone
, settings
->network_zone
);
4424 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4428 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4429 settings
->expose_ports
) {
4431 if (!arg_settings_trusted
)
4432 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4434 expose_port_free_all(arg_expose_ports
);
4435 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4439 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4440 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4442 if (!arg_settings_trusted
)
4443 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4445 arg_userns_mode
= settings
->userns_mode
;
4446 arg_uid_shift
= settings
->uid_shift
;
4447 arg_uid_range
= settings
->uid_range
;
4448 arg_userns_ownership
= settings
->userns_ownership
;
4452 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4453 !strv_isempty(settings
->bind_user
))
4454 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4456 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4457 settings
->notify_ready
>= 0)
4458 arg_notify_ready
= settings
->notify_ready
;
4460 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4462 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4463 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4464 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4466 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4467 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4472 if (settings
->seccomp
) {
4473 if (!arg_settings_trusted
)
4474 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4476 seccomp_release(arg_seccomp
);
4477 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4483 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4484 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4487 if (!settings
->rlimit
[rl
])
4490 if (!arg_settings_trusted
) {
4491 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4495 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4498 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4500 free_and_replace(arg_hostname
, settings
->hostname
);
4502 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4503 settings
->no_new_privileges
>= 0)
4504 arg_no_new_privileges
= settings
->no_new_privileges
;
4506 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4507 settings
->oom_score_adjust_set
) {
4509 if (!arg_settings_trusted
)
4510 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4512 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4513 arg_oom_score_adjust_set
= true;
4517 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4518 settings
->cpu_set
.set
) {
4520 if (!arg_settings_trusted
)
4521 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4523 cpu_set_reset(&arg_cpu_set
);
4524 arg_cpu_set
= settings
->cpu_set
;
4525 settings
->cpu_set
= (CPUSet
) {};
4529 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4530 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4531 arg_resolv_conf
= settings
->resolv_conf
;
4533 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4534 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4536 if (!arg_settings_trusted
)
4537 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4539 arg_link_journal
= settings
->link_journal
;
4540 arg_link_journal_try
= settings
->link_journal_try
;
4544 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4545 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4546 arg_timezone
= settings
->timezone
;
4548 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4551 if (!arg_settings_trusted
)
4552 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4554 free_and_replace(arg_slice
, settings
->slice
);
4557 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4558 settings
->use_cgns
>= 0) {
4560 if (!arg_settings_trusted
)
4561 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4563 arg_use_cgns
= settings
->use_cgns
;
4566 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4567 settings
->clone_ns_flags
!= ULONG_MAX
) {
4569 if (!arg_settings_trusted
)
4570 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4572 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4575 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4576 settings
->console_mode
>= 0) {
4578 if (!arg_settings_trusted
)
4579 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4581 arg_console_mode
= settings
->console_mode
;
4584 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4585 settings
->suppress_sync
>= 0)
4586 arg_suppress_sync
= settings
->suppress_sync
;
4588 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4589 * don't consult arg_settings_mask for them. */
4591 sd_bus_message_unref(arg_property_message
);
4592 arg_property_message
= TAKE_PTR(settings
->properties
);
4594 arg_console_width
= settings
->console_width
;
4595 arg_console_height
= settings
->console_height
;
4597 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4598 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4599 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4604 static int load_settings(void) {
4605 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4606 _cleanup_fclose_
FILE *f
= NULL
;
4607 _cleanup_free_
char *p
= NULL
;
4614 /* If all settings are masked, there's no point in looking for
4615 * the settings file */
4616 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4619 fn
= strjoina(arg_machine
, ".nspawn");
4621 /* We first look in the admin's directories in /etc and /run */
4622 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4623 _cleanup_free_
char *j
= NULL
;
4625 j
= path_join(i
, fn
);
4633 /* By default, we trust configuration from /etc and /run */
4634 if (arg_settings_trusted
< 0)
4635 arg_settings_trusted
= true;
4640 if (errno
!= ENOENT
)
4641 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4645 /* After that, let's look for a file next to the
4646 * actual image we shall boot. */
4649 p
= file_in_same_dir(arg_image
, fn
);
4652 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4653 p
= file_in_same_dir(arg_directory
, fn
);
4660 if (!f
&& errno
!= ENOENT
)
4661 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4663 /* By default, we do not trust configuration from /var/lib/machines */
4664 if (arg_settings_trusted
< 0)
4665 arg_settings_trusted
= false;
4672 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4674 r
= settings_load(f
, p
, &settings
);
4678 return merge_settings(settings
, p
);
4681 static int load_oci_bundle(void) {
4682 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4685 if (!arg_oci_bundle
)
4688 /* By default let's trust OCI bundles */
4689 if (arg_settings_trusted
< 0)
4690 arg_settings_trusted
= true;
4692 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4696 return merge_settings(settings
, arg_oci_bundle
);
4699 static int run_container(
4700 DissectedImage
*dissected_image
,
4703 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4704 struct ExposeArgs
*expose_args
,
4705 int *master
, pid_t
*pid
, int *ret
) {
4707 static const struct sigaction sa
= {
4708 .sa_handler
= nop_signal_handler
,
4709 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4712 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4713 _cleanup_close_
int etc_passwd_lock
= -1;
4714 _cleanup_close_pair_
int
4715 kmsg_socket_pair
[2] = { -1, -1 },
4716 rtnl_socket_pair
[2] = { -1, -1 },
4717 pid_socket_pair
[2] = { -1, -1 },
4718 uuid_socket_pair
[2] = { -1, -1 },
4719 notify_socket_pair
[2] = { -1, -1 },
4720 uid_shift_socket_pair
[2] = { -1, -1 },
4721 master_pty_socket_pair
[2] = { -1, -1 },
4722 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4724 _cleanup_close_
int notify_socket
= -1;
4725 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4726 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4727 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4728 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4729 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4730 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4731 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4732 size_t n_bind_user_uid
= 0;
4733 ContainerStatus container_status
= 0;
4737 _cleanup_close_
int child_netns_fd
= -1;
4739 assert_se(sigemptyset(&mask_chld
) == 0);
4740 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4742 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4743 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4744 * check with getpwuid() if the specific user already exists. Note that /etc might be
4745 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4746 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4747 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4750 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4751 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4752 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4755 r
= barrier_create(&barrier
);
4757 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4759 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4760 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4762 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4763 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4765 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4766 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4768 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4769 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4771 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4772 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4774 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4775 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4777 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4778 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4779 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4781 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4782 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4783 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4785 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4786 * parent's blocking calls and give it a chance to call wait() and terminate. */
4787 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4789 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4791 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4793 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4795 if (arg_network_namespace_path
) {
4796 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4797 if (child_netns_fd
< 0)
4798 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4800 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4802 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
);
4804 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4806 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4807 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4810 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4812 return log_error_errno(errno
, "clone() failed%s: %m",
4814 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4817 /* The outer child only has a file system namespace. */
4818 barrier_set_role(&barrier
, BARRIER_CHILD
);
4820 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4821 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4822 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4823 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4824 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4825 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4826 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4827 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4829 (void) reset_all_signal_handlers();
4830 (void) reset_signal_mask();
4832 r
= outer_child(&barrier
,
4837 uuid_socket_pair
[1],
4838 notify_socket_pair
[1],
4839 kmsg_socket_pair
[1],
4840 rtnl_socket_pair
[1],
4841 uid_shift_socket_pair
[1],
4842 master_pty_socket_pair
[1],
4843 unified_cgroup_hierarchy_socket_pair
[1],
4847 _exit(EXIT_FAILURE
);
4849 _exit(EXIT_SUCCESS
);
4852 barrier_set_role(&barrier
, BARRIER_PARENT
);
4856 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4857 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4858 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4859 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4860 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4861 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4862 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4863 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4865 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4866 /* The child just let us know the UID shift it might have read from the image. */
4867 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4869 return log_error_errno(errno
, "Failed to read UID shift: %m");
4870 if (l
!= sizeof arg_uid_shift
)
4871 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4873 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4874 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4875 * image, but if that's already in use, pick a new one, and report back to the child,
4876 * which one we now picked. */
4878 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4880 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4882 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4884 return log_error_errno(errno
, "Failed to send UID shift: %m");
4885 if (l
!= sizeof arg_uid_shift
)
4886 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4889 n_bind_user_uid
= strv_length(arg_bind_user
);
4890 if (n_bind_user_uid
> 0) {
4891 /* Right after the UID shift, we'll receive the list of UID mappings for the
4892 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4894 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4898 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4899 l
= recv(uid_shift_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4901 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4902 if (l
!= sizeof(uid_t
)*4)
4903 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4904 SYNTHETIC_ERRNO(EIO
),
4905 "Short read while reading bind user UID pairs.");
4910 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4911 /* The child let us know the support cgroup mode it might have read from the image. */
4912 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4914 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4915 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4916 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4917 l
, l
== 0 ? " The child is most likely dead." : "");
4920 /* Wait for the outer child. */
4921 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4924 if (r
!= EXIT_SUCCESS
)
4927 /* And now retrieve the PID of the inner child. */
4928 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4930 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4931 if (l
!= sizeof *pid
)
4932 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4934 /* We also retrieve container UUID in case it was generated by outer child */
4935 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4937 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4938 if (l
!= sizeof(arg_uuid
))
4939 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4941 /* We also retrieve the socket used for notifications generated by outer child */
4942 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4943 if (notify_socket
< 0)
4944 return log_error_errno(notify_socket
,
4945 "Failed to receive notification socket from the outer child: %m");
4947 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4949 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4950 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4951 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4953 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4957 (void) barrier_place(&barrier
); /* #2 */
4960 if (arg_private_network
) {
4961 if (!arg_network_namespace_path
) {
4962 /* Wait until the child has unshared its network namespace. */
4963 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4964 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4967 if (child_netns_fd
< 0) {
4968 /* Make sure we have an open file descriptor to the child's network
4969 * namespace so it stays alive even if the child exits. */
4970 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4972 return log_error_errno(r
, "Failed to open child network namespace: %m");
4975 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4979 if (arg_network_veth
) {
4980 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4981 arg_network_bridge
|| arg_network_zone
);
4987 if (arg_network_bridge
) {
4988 /* Add the interface to a bridge */
4989 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4994 } else if (arg_network_zone
) {
4995 /* Add the interface to a bridge, possibly creating it */
4996 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5004 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5008 /* We created the primary and extra veth links now; let's remember this, so that we know to
5009 remove them later on. Note that we don't bother with removing veth links that were created
5010 here when their setup failed half-way, because in that case the kernel should be able to
5011 remove them on its own, since they cannot be referenced by anything yet. */
5012 *veth_created
= true;
5014 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5018 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5023 if (arg_register
|| !arg_keep_unit
) {
5024 r
= sd_bus_default_system(&bus
);
5026 return log_error_errno(r
, "Failed to open system bus: %m");
5028 r
= sd_bus_set_close_on_exit(bus
, false);
5030 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5033 if (!arg_keep_unit
) {
5034 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5035 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5036 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5038 r
= sd_bus_match_signal_async(
5041 "org.freedesktop.systemd1",
5043 "org.freedesktop.systemd1.Scope",
5045 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5047 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5051 r
= register_machine(
5059 arg_custom_mounts
, arg_n_custom_mounts
,
5062 arg_property_message
,
5064 arg_container_service_name
);
5068 } else if (!arg_keep_unit
) {
5074 arg_custom_mounts
, arg_n_custom_mounts
,
5077 arg_property_message
);
5081 } else if (arg_slice
|| arg_property
)
5082 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5084 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5088 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5092 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5096 /* Notify the child that the parent is ready with all
5097 * its setup (including cgroup-ification), and that
5098 * the child can now hand over control to the code to
5099 * run inside the container. */
5100 (void) barrier_place(&barrier
); /* #4 */
5102 /* Block SIGCHLD here, before notifying child.
5103 * process_pty() will handle it with the other signals. */
5104 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5106 /* Reset signal to default */
5107 r
= default_signals(SIGCHLD
);
5109 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5111 r
= sd_event_new(&event
);
5113 return log_error_errno(r
, "Failed to get default event source: %m");
5115 (void) sd_event_set_watchdog(event
, true);
5118 r
= sd_bus_attach_event(bus
, event
, 0);
5120 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5123 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5127 /* Let the child know that we are ready and wait that the child is completely ready now. */
5128 if (!barrier_place_and_sync(&barrier
)) /* #5 */
5129 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5131 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5132 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5133 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5135 (void) sd_notifyf(false,
5136 "STATUS=Container running.\n"
5137 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5138 if (!arg_notify_ready
) {
5139 r
= sd_notify(false, "READY=1\n");
5141 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5144 if (arg_kill_signal
> 0) {
5145 /* Try to kill the init system on SIGINT or SIGTERM */
5146 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5147 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5149 /* Immediately exit */
5150 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5151 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5154 /* Exit when the child exits */
5155 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5157 if (arg_expose_ports
) {
5158 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5162 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5163 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5166 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
5168 if (arg_console_mode
!= CONSOLE_PIPE
) {
5169 _cleanup_close_
int fd
= -1;
5170 PTYForwardFlags flags
= 0;
5172 /* Retrieve the master pty allocated by inner child */
5173 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
5175 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5177 switch (arg_console_mode
) {
5179 case CONSOLE_READ_ONLY
:
5180 flags
|= PTY_FORWARD_READ_ONLY
;
5184 case CONSOLE_INTERACTIVE
:
5185 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5187 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5189 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5191 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5192 (void) pty_forward_set_width_height(forward
,
5194 arg_console_height
);
5198 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5201 *master
= TAKE_FD(fd
);
5204 r
= sd_event_loop(event
);
5206 return log_error_errno(r
, "Failed to run event loop: %m");
5211 (void) pty_forward_get_last_char(forward
, &last_char
);
5212 forward
= pty_forward_free(forward
);
5214 if (!arg_quiet
&& last_char
!= '\n')
5218 /* Kill if it is not dead yet anyway */
5219 if (!arg_register
&& !arg_keep_unit
&& bus
)
5220 terminate_scope(bus
, arg_machine
);
5222 /* Normally redundant, but better safe than sorry */
5223 (void) kill(*pid
, SIGKILL
);
5225 if (arg_private_network
) {
5226 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5227 * to avoid having to move the parent to the child network namespace. */
5228 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
5233 _cleanup_close_
int parent_netns_fd
= -1;
5235 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5237 log_error_errno(r
, "Failed to open parent network namespace: %m");
5238 _exit(EXIT_FAILURE
);
5241 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5243 log_error_errno(r
, "Failed to enter child network namespace: %m");
5244 _exit(EXIT_FAILURE
);
5247 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5249 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5251 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5255 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5257 /* Tell machined that we are gone. */
5259 (void) unregister_machine(bus
, arg_machine
);
5262 /* We failed to wait for the container, or the container exited abnormally. */
5264 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5265 /* r > 0 → The container exited with a non-zero status.
5266 * As a special case, we need to replace 133 with a different value,
5267 * because 133 is special-cased in the service file to reboot the container.
5268 * otherwise → The container exited with zero status and a reboot was not requested.
5270 if (r
== EXIT_FORCE_RESTART
)
5271 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5273 return 0; /* finito */
5276 /* CONTAINER_REBOOTED, loop again */
5278 if (arg_keep_unit
) {
5279 /* Special handling if we are running as a service: instead of simply
5280 * restarting the machine we want to restart the entire service, so let's
5281 * inform systemd about this with the special exit code 133. The service
5282 * file uses RestartForceExitStatus=133 so that this results in a full
5283 * nspawn restart. This is necessary since we might have cgroup parameters
5284 * set we want to have flushed out. */
5285 *ret
= EXIT_FORCE_RESTART
;
5286 return 0; /* finito */
5289 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5290 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5292 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5293 *veth_created
= false;
5294 return 1; /* loop again */
5297 static int initialize_rlimits(void) {
5298 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5299 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5300 * container execution environments. */
5302 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5303 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5304 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5305 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5306 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5307 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5308 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5309 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5310 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5311 [RLIMIT_NICE
] = { 0, 0 },
5312 [RLIMIT_NOFILE
] = { 1024, 4096 },
5313 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5314 [RLIMIT_RTPRIO
] = { 0, 0 },
5315 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5316 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5318 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5319 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5320 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5321 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5322 * that PID 1 changes a number of other resource limits during early initialization which is why we
5323 * don't read the other limits from PID 1 but prefer the static table above. */
5328 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5329 /* Let's only fill in what the user hasn't explicitly configured anyway */
5330 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5331 const struct rlimit
*v
;
5332 struct rlimit buffer
;
5334 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5335 /* For these two let's read the limits off PID 1. See above for an explanation. */
5337 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
5338 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5341 } else if (rl
== RLIMIT_NOFILE
) {
5342 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5343 * userspace. Given that nspawn containers are often run without our PID 1,
5344 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5345 * so that container userspace gets similar resources as host userspace
5347 buffer
= kernel_defaults
[rl
];
5348 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5351 v
= kernel_defaults
+ rl
;
5353 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5354 if (!arg_rlimit
[rl
])
5358 if (DEBUG_LOGGING
) {
5359 _cleanup_free_
char *k
= NULL
;
5361 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5362 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5369 static int cant_be_in_netns(void) {
5370 union sockaddr_union sa
= {
5372 .sun_family
= AF_UNIX
,
5373 .sun_path
= "/run/udev/control",
5376 char udev_path
[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t
)];
5377 _cleanup_free_
char *udev_ns
= NULL
, *our_ns
= NULL
;
5378 _cleanup_close_
int fd
= -1;
5382 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5383 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5386 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5389 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5391 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5393 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
5395 if (errno
== ENOENT
|| ERRNO_IS_DISCONNECT(errno
))
5396 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5397 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5399 return log_error_errno(errno
, "Failed to connect socket to udev control socket: %m");
5402 r
= getpeercred(fd
, &ucred
);
5404 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5406 xsprintf(udev_path
, "/proc/" PID_FMT
"/ns/net", ucred
.pid
);
5407 r
= readlink_malloc(udev_path
, &udev_ns
);
5409 return log_error_errno(r
, "Failed to read network namespace of udev: %m");
5411 r
= readlink_malloc("/proc/self/ns/net", &our_ns
);
5413 return log_error_errno(r
, "Failed to read our own network namespace: %m");
5415 if (!streq(our_ns
, udev_ns
))
5416 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5417 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5421 static int run(int argc
, char *argv
[]) {
5422 bool secondary
= false, remove_directory
= false, remove_image
= false,
5423 veth_created
= false, remove_tmprootdir
= false;
5424 _cleanup_close_
int master
= -1;
5425 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5426 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5427 char veth_name
[IFNAMSIZ
] = "";
5428 struct ExposeArgs expose_args
= {};
5429 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5430 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5431 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5432 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
5433 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5434 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5437 log_parse_environment();
5440 r
= parse_argv(argc
, argv
);
5444 if (geteuid() != 0) {
5445 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5446 argc
>= 2 ? "Need to be root." :
5447 "Need to be root (and some arguments are usually required).\nHint: try --help");
5451 r
= cant_be_in_netns();
5455 r
= initialize_rlimits();
5459 r
= load_oci_bundle();
5463 r
= determine_names();
5467 r
= load_settings();
5473 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5477 r
= verify_arguments();
5481 /* Reapply environment settings. */
5482 (void) detect_unified_cgroup_hierarchy_from_environment();
5484 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5485 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5486 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5487 (void) ignore_signals(SIGPIPE
);
5489 n_fd_passed
= sd_listen_fds(false);
5490 if (n_fd_passed
> 0) {
5491 r
= fdset_new_listen_fds(&fds
, false);
5493 log_error_errno(r
, "Failed to collect file descriptors: %m");
5498 /* The "default" umask. This is appropriate for most file and directory
5499 * operations performed by nspawn, and is the umask that will be used for
5500 * the child. Functions like copy_devnodes() change the umask temporarily. */
5503 if (arg_directory
) {
5506 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5507 * /var from the host will propagate into container dynamically (because bad things happen if
5508 * two systems write to the same /var). Let's allow it for the special cases where /var is
5509 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5510 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5511 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5516 if (arg_ephemeral
) {
5517 _cleanup_free_
char *np
= NULL
;
5519 r
= chase_symlinks_and_update(&arg_directory
, 0);
5523 /* If the specified path is a mount point we generate the new snapshot immediately
5524 * inside it under a random name. However if the specified is not a mount point we
5525 * create the new snapshot in the parent directory, just next to it. */
5526 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5528 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5532 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5534 r
= tempfn_random(arg_directory
, "machine.", &np
);
5536 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5540 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5541 * only owned by us and no one else. */
5542 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5544 log_error_errno(r
, "Failed to lock %s: %m", np
);
5549 BLOCK_SIGNALS(SIGINT
);
5550 r
= btrfs_subvol_snapshot(arg_directory
, np
,
5551 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5552 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5553 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5554 BTRFS_SNAPSHOT_RECURSIVE
|
5555 BTRFS_SNAPSHOT_QUOTA
|
5556 BTRFS_SNAPSHOT_SIGINT
);
5559 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5563 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5567 free_and_replace(arg_directory
, np
);
5568 remove_directory
= true;
5570 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5574 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5576 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5580 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5585 r
= chase_symlinks_and_update(&arg_template
, 0);
5590 BLOCK_SIGNALS(SIGINT
);
5591 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5592 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5593 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5594 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5595 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5596 BTRFS_SNAPSHOT_RECURSIVE
|
5597 BTRFS_SNAPSHOT_QUOTA
|
5598 BTRFS_SNAPSHOT_SIGINT
);
5601 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5602 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5603 else if (r
== -EINTR
) {
5604 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5607 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5610 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5611 "Populated %s from template %s.", arg_directory
, arg_template
);
5615 if (arg_start_mode
== START_BOOT
) {
5616 _cleanup_free_
char *b
= NULL
;
5619 if (arg_pivot_root_new
) {
5620 b
= path_join(arg_directory
, arg_pivot_root_new
);
5628 if (path_is_os_tree(p
) <= 0) {
5629 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5630 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5634 _cleanup_free_
char *p
= NULL
;
5636 if (arg_pivot_root_new
)
5637 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5639 p
= path_join(arg_directory
, "/usr/");
5643 if (laccess(p
, F_OK
) < 0) {
5644 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5645 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5651 DissectImageFlags dissect_image_flags
=
5652 DISSECT_IMAGE_GENERIC_ROOT
|
5653 DISSECT_IMAGE_REQUIRE_ROOT
|
5654 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5655 DISSECT_IMAGE_USR_NO_ROOT
;
5657 assert(!arg_template
);
5659 r
= chase_symlinks_and_update(&arg_image
, 0);
5663 if (arg_ephemeral
) {
5664 _cleanup_free_
char *np
= NULL
;
5666 r
= tempfn_random(arg_image
, "machine.", &np
);
5668 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5672 /* Always take an exclusive lock on our own ephemeral copy. */
5673 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5675 r
= log_error_errno(r
, "Failed to create image lock: %m");
5680 BLOCK_SIGNALS(SIGINT
);
5681 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
);
5684 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5688 r
= log_error_errno(r
, "Failed to copy image file: %m");
5692 free_and_replace(arg_image
, np
);
5693 remove_image
= true;
5695 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5697 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5701 r
= log_error_errno(r
, "Failed to create image lock: %m");
5705 r
= verity_settings_load(
5706 &arg_verity_settings
,
5707 arg_image
, NULL
, NULL
);
5709 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5713 if (arg_verity_settings
.data_path
)
5714 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5717 if (!mkdtemp(tmprootdir
)) {
5718 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5722 remove_tmprootdir
= true;
5724 arg_directory
= strdup(tmprootdir
);
5725 if (!arg_directory
) {
5730 r
= loop_device_make_by_path(
5732 arg_read_only
? O_RDONLY
: O_RDWR
,
5733 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5736 log_error_errno(r
, "Failed to set up loopback block device: %m");
5740 /* Take a LOCK_SH lock on the device, so that udevd doesn't issue BLKRRPART in our back */
5741 r
= loop_device_flock(loop
, LOCK_SH
);
5743 log_error_errno(r
, "Failed to take lock on loopback block device: %m");
5747 r
= dissect_image_and_warn(
5750 &arg_verity_settings
,
5753 loop
->uevent_seqnum_not_before
,
5754 loop
->timestamp_not_before
,
5755 dissect_image_flags
,
5758 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5759 log_notice("Note that the disk image needs to\n"
5760 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5761 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5762 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5763 " d) or contain a file system without a partition table\n"
5764 "in order to be bootable with systemd-nspawn.");
5770 r
= dissected_image_load_verity_sig_partition(
5773 &arg_verity_settings
);
5777 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5778 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5779 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5781 r
= dissected_image_decrypt_interactively(
5784 &arg_verity_settings
,
5790 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5791 if (remove_image
&& unlink(arg_image
) >= 0)
5792 remove_image
= false;
5795 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5799 if (arg_console_mode
< 0)
5801 isatty(STDIN_FILENO
) > 0 &&
5802 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5804 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5808 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5809 arg_machine
, arg_image
?: arg_directory
);
5811 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5813 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5814 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5818 if (arg_expose_ports
) {
5819 r
= fw_ctx_new(&fw_ctx
);
5821 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5824 expose_args
.fw_ctx
= fw_ctx
;
5827 r
= run_container(dissected_image
,
5830 veth_name
, &veth_created
,
5831 &expose_args
, &master
,
5838 (void) sd_notify(false,
5839 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5840 "STOPPING=1\nSTATUS=Terminating...");
5843 (void) kill(pid
, SIGKILL
);
5845 /* Try to flush whatever is still queued in the pty */
5847 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5848 master
= safe_close(master
);
5852 (void) wait_for_terminate(pid
, NULL
);
5856 if (remove_directory
&& arg_directory
) {
5859 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5861 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5864 if (remove_image
&& arg_image
) {
5865 if (unlink(arg_image
) < 0)
5866 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5869 if (remove_tmprootdir
) {
5870 if (rmdir(tmprootdir
) < 0)
5871 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5877 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5878 (void) rm_rf(p
, REMOVE_ROOT
);
5881 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5882 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5885 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5886 (void) remove_bridge(arg_network_zone
);
5888 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5889 expose_port_free_all(arg_expose_ports
);
5890 rlimit_free_all(arg_rlimit
);
5891 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5892 credential_free_all(arg_credentials
, arg_n_credentials
);
5900 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);