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;
233 static char *arg_settings_filename
= NULL
;
235 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
257 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
258 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
259 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
260 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
262 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
264 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
265 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
266 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
267 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
269 static int handle_arg_console(const char *arg
) {
270 if (streq(arg
, "help")) {
279 if (streq(arg
, "interactive"))
280 arg_console_mode
= CONSOLE_INTERACTIVE
;
281 else if (streq(arg
, "read-only"))
282 arg_console_mode
= CONSOLE_READ_ONLY
;
283 else if (streq(arg
, "passive"))
284 arg_console_mode
= CONSOLE_PASSIVE
;
285 else if (streq(arg
, "pipe")) {
286 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
287 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
288 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
289 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
290 "Proceeding anyway.");
292 arg_console_mode
= CONSOLE_PIPE
;
293 } else if (streq(arg
, "autopipe")) {
294 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
295 arg_console_mode
= CONSOLE_INTERACTIVE
;
297 arg_console_mode
= CONSOLE_PIPE
;
299 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
301 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
305 static int help(void) {
306 _cleanup_free_
char *link
= NULL
;
309 pager_open(arg_pager_flags
);
311 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
315 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
316 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
317 " -h --help Show this help\n"
318 " --version Print version string\n"
319 " -q --quiet Do not show status information\n"
320 " --no-pager Do not pipe output into a pager\n"
321 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
323 " -D --directory=PATH Root directory for the container\n"
324 " --template=PATH Initialize root directory from template directory,\n"
326 " -x --ephemeral Run container with snapshot of root directory, and\n"
327 " remove it after exit\n"
328 " -i --image=PATH Root file system disk image (or device node) for\n"
330 " --oci-bundle=PATH OCI bundle directory\n"
331 " --read-only Mount the root directory read-only\n"
332 " --volatile[=MODE] Run the system in volatile mode\n"
333 " --root-hash=HASH Specify verity root hash for root disk image\n"
334 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
335 " as a DER encoded PKCS7, either as a path to a file\n"
336 " or as an ASCII base64 encoded string prefixed by\n"
338 " --verity-data=PATH Specify hash device for verity\n"
339 " --pivot-root=PATH[:PATH]\n"
340 " Pivot root to given directory in the container\n\n"
341 "%3$sExecution:%4$s\n"
342 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
343 " -b --boot Boot up full system (i.e. invoke init)\n"
344 " --chdir=PATH Set working directory in the container\n"
345 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
346 " -u --user=USER Run the command under specified user or UID\n"
347 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
348 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
349 " --suppress-sync=BOOLEAN\n"
350 " Suppress any form of disk data synchronization\n\n"
351 "%3$sSystem Identity:%4$s\n"
352 " -M --machine=NAME Set the machine name for the container\n"
353 " --hostname=NAME Override the hostname for the container\n"
354 " --uuid=UUID Set a specific machine UUID for the container\n\n"
355 "%3$sProperties:%4$s\n"
356 " -S --slice=SLICE Place the container in the specified slice\n"
357 " --property=NAME=VALUE Set scope unit property\n"
358 " --register=BOOLEAN Register container as machine\n"
359 " --keep-unit Do not register a scope for the machine, reuse\n"
360 " the service unit nspawn is running in\n\n"
361 "%3$sUser Namespacing:%4$s\n"
362 " --private-users=no Run without user namespacing\n"
363 " --private-users=yes|pick|identity\n"
364 " Run within user namespace, autoselect UID/GID range\n"
365 " --private-users=UIDBASE[:NUIDS]\n"
366 " Similar, but with user configured UID/GID range\n"
367 " --private-users-ownership=MODE\n"
368 " Adjust ('chown') or map ('map') OS tree ownership\n"
369 " to private UID/GID range\n"
370 " -U Equivalent to --private-users=pick and\n"
371 " --private-users-ownership=auto\n\n"
372 "%3$sNetworking:%4$s\n"
373 " --private-network Disable network in container\n"
374 " --network-interface=INTERFACE\n"
375 " Assign an existing network interface to the\n"
377 " --network-macvlan=INTERFACE\n"
378 " Create a macvlan network interface based on an\n"
379 " existing network interface to the container\n"
380 " --network-ipvlan=INTERFACE\n"
381 " Create an ipvlan network interface based on an\n"
382 " existing network interface to the container\n"
383 " -n --network-veth Add a virtual Ethernet connection between host\n"
385 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
386 " Add an additional virtual Ethernet link between\n"
387 " host and container\n"
388 " --network-bridge=INTERFACE\n"
389 " Add a virtual Ethernet connection to the container\n"
390 " and attach it to an existing bridge on the host\n"
391 " --network-zone=NAME Similar, but attach the new interface to an\n"
392 " an automatically managed bridge interface\n"
393 " --network-namespace-path=PATH\n"
394 " Set network namespace to the one represented by\n"
395 " the specified kernel namespace file node\n"
396 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
397 " Expose a container IP port on the host\n\n"
398 "%3$sSecurity:%4$s\n"
399 " --capability=CAP In addition to the default, retain specified\n"
401 " --drop-capability=CAP Drop the specified capability from the default set\n"
402 " --ambient-capability=CAP\n"
403 " Sets the specified capability for the started\n"
404 " process. Not useful if booting a machine.\n"
405 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
406 " --system-call-filter=LIST|~LIST\n"
407 " Permit/prohibit specific system calls\n"
408 " -Z --selinux-context=SECLABEL\n"
409 " Set the SELinux security context to be used by\n"
410 " processes in the container\n"
411 " -L --selinux-apifs-context=SECLABEL\n"
412 " Set the SELinux security context to be used by\n"
413 " API/tmpfs file systems in the container\n\n"
414 "%3$sResources:%4$s\n"
415 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
416 " --oom-score-adjust=VALUE\n"
417 " Adjust the OOM score value for the payload\n"
418 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
419 " --personality=ARCH Pick personality for this container\n\n"
420 "%3$sIntegration:%4$s\n"
421 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
422 " --timezone=MODE Select mode of /etc/localtime initialization\n"
423 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
424 " host, try-guest, try-host\n"
425 " -j Equivalent to --link-journal=try-guest\n\n"
427 " --bind=PATH[:PATH[:OPTIONS]]\n"
428 " Bind mount a file or directory from the host into\n"
430 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
431 " Similar, but creates a read-only bind mount\n"
432 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
434 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
435 " --overlay=PATH[:PATH...]:PATH\n"
436 " Create an overlay mount from the host to \n"
438 " --overlay-ro=PATH[:PATH...]:PATH\n"
439 " Similar, but creates a read-only overlay mount\n"
440 " --bind-user=NAME Bind user from host to container\n\n"
441 "%3$sInput/Output:%4$s\n"
442 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
443 " set up for the container.\n"
444 " -P --pipe Equivalent to --console=pipe\n\n"
445 "%3$sCredentials:%4$s\n"
446 " --set-credential=ID:VALUE\n"
447 " Pass a credential with literal value to container.\n"
448 " --load-credential=ID:PATH\n"
449 " Load credential to pass to container from file or\n"
450 " AF_UNIX stream socket.\n"
451 "\nSee the %2$s for details.\n",
452 program_invocation_short_name
,
462 static int custom_mount_check_all(void) {
465 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
466 CustomMount
*m
= &arg_custom_mounts
[i
];
468 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
469 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
470 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
471 "--private-users-ownership=own may not be combined with custom root mounts.");
472 if (arg_uid_shift
== UID_INVALID
)
473 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
474 "--private-users with automatic UID shift may not be combined with custom root mounts.");
481 static int detect_unified_cgroup_hierarchy_from_environment(void) {
482 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
485 /* Allow the user to control whether the unified hierarchy is used */
489 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
490 var
= "UNIFIED_CGROUP_HIERARCHY";
495 r
= parse_boolean(e
);
497 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
499 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
501 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
507 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
510 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
511 * in the image actually supports. */
512 r
= cg_all_unified();
514 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
516 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
517 * routine only detects 231, so we'll have a false negative here for 230. */
518 r
= systemd_installation_has_version(directory
, "230");
520 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
522 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
524 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
525 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
526 /* Mixed cgroup hierarchy support was added in 233 */
527 r
= systemd_installation_has_version(directory
, "233");
529 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
531 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
533 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
535 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
537 log_debug("Using %s hierarchy for container.",
538 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
539 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
544 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
549 _cleanup_free_
char *t
= NULL
;
551 r
= extract_first_word(&spec
, &t
, ",", 0);
553 return log_error_errno(r
, "Failed to parse capability %s.", t
);
557 if (streq(t
, "help")) {
558 for (int i
= 0; i
< capability_list_length(); i
++) {
561 name
= capability_to_name(i
);
572 r
= capability_from_name(t
);
574 return log_error_errno(r
, "Failed to parse capability %s.", t
);
581 return 1; /* continue */
584 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
587 r
= getenv_bool(name
);
591 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
593 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
594 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
598 static int parse_mount_settings_env(void) {
602 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
603 if (r
< 0 && r
!= -ENXIO
)
604 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
606 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
608 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
609 if (streq_ptr(e
, "network"))
610 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
613 r
= parse_boolean(e
);
615 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
617 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
618 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
624 static int parse_environment(void) {
628 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
631 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
634 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
637 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
641 r
= parse_mount_settings_env();
645 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
646 * even if it is supported. If not supported, it has no effect. */
647 if (!cg_ns_supported())
648 arg_use_cgns
= false;
650 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
653 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
657 arg_use_cgns
= r
> 0;
658 arg_settings_mask
|= SETTING_USE_CGNS
;
662 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
664 arg_container_service_name
= e
;
666 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
668 arg_suppress_sync
= r
;
669 else if (r
!= -ENXIO
)
670 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
672 return detect_unified_cgroup_hierarchy_from_environment();
675 static int parse_argv(int argc
, char *argv
[]) {
682 ARG_AMBIENT_CAPABILITY
,
694 ARG_NETWORK_INTERFACE
,
699 ARG_NETWORK_VETH_EXTRA
,
700 ARG_NETWORK_NAMESPACE_PATH
,
710 ARG_PRIVATE_USERS_CHOWN
,
711 ARG_PRIVATE_USERS_OWNERSHIP
,
716 ARG_SYSTEM_CALL_FILTER
,
719 ARG_NO_NEW_PRIVILEGES
,
720 ARG_OOM_SCORE_ADJUST
,
734 static const struct option options
[] = {
735 { "help", no_argument
, NULL
, 'h' },
736 { "version", no_argument
, NULL
, ARG_VERSION
},
737 { "directory", required_argument
, NULL
, 'D' },
738 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
739 { "ephemeral", no_argument
, NULL
, 'x' },
740 { "user", required_argument
, NULL
, 'u' },
741 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
742 { "as-pid2", no_argument
, NULL
, 'a' },
743 { "boot", no_argument
, NULL
, 'b' },
744 { "uuid", required_argument
, NULL
, ARG_UUID
},
745 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
746 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
747 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
748 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
749 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
750 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
751 { "bind", required_argument
, NULL
, ARG_BIND
},
752 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
753 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
754 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
755 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
756 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
757 { "machine", required_argument
, NULL
, 'M' },
758 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
759 { "slice", required_argument
, NULL
, 'S' },
760 { "setenv", required_argument
, NULL
, 'E' },
761 { "selinux-context", required_argument
, NULL
, 'Z' },
762 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
763 { "quiet", no_argument
, NULL
, 'q' },
764 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
765 { "register", required_argument
, NULL
, ARG_REGISTER
},
766 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
767 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
768 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
769 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
770 { "network-veth", no_argument
, NULL
, 'n' },
771 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
772 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
773 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
774 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
775 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
776 { "image", required_argument
, NULL
, 'i' },
777 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
778 { "port", required_argument
, NULL
, 'p' },
779 { "property", required_argument
, NULL
, ARG_PROPERTY
},
780 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
781 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
782 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
783 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
784 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
785 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
786 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
787 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
788 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
789 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
790 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
791 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
792 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
793 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
794 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
795 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
796 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
797 { "console", required_argument
, NULL
, ARG_CONSOLE
},
798 { "pipe", no_argument
, NULL
, ARG_PIPE
},
799 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
800 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
801 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
802 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
803 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
804 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
809 uint64_t plus
= 0, minus
= 0;
810 bool mask_all_settings
= false, mask_no_settings
= false;
815 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
825 r
= parse_path_argument(optarg
, false, &arg_directory
);
829 arg_settings_mask
|= SETTING_DIRECTORY
;
833 r
= parse_path_argument(optarg
, false, &arg_template
);
837 arg_settings_mask
|= SETTING_DIRECTORY
;
841 r
= parse_path_argument(optarg
, false, &arg_image
);
845 arg_settings_mask
|= SETTING_DIRECTORY
;
849 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
856 arg_ephemeral
= true;
857 arg_settings_mask
|= SETTING_EPHEMERAL
;
861 r
= free_and_strdup(&arg_user
, optarg
);
865 arg_settings_mask
|= SETTING_USER
;
868 case ARG_NETWORK_ZONE
: {
871 j
= strjoin("vz-", optarg
);
875 if (!ifname_valid(j
)) {
876 log_error("Network zone name not valid: %s", j
);
881 free_and_replace(arg_network_zone
, j
);
883 arg_network_veth
= true;
884 arg_private_network
= true;
885 arg_settings_mask
|= SETTING_NETWORK
;
889 case ARG_NETWORK_BRIDGE
:
891 if (!ifname_valid(optarg
))
892 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
893 "Bridge interface name not valid: %s", optarg
);
895 r
= free_and_strdup(&arg_network_bridge
, optarg
);
901 arg_network_veth
= true;
902 arg_private_network
= true;
903 arg_settings_mask
|= SETTING_NETWORK
;
906 case ARG_NETWORK_VETH_EXTRA
:
907 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
909 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
911 arg_private_network
= true;
912 arg_settings_mask
|= SETTING_NETWORK
;
915 case ARG_NETWORK_INTERFACE
:
916 if (!ifname_valid(optarg
))
917 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
918 "Network interface name not valid: %s", optarg
);
920 r
= test_network_interface_initialized(optarg
);
924 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
927 arg_private_network
= true;
928 arg_settings_mask
|= SETTING_NETWORK
;
931 case ARG_NETWORK_MACVLAN
:
933 if (!ifname_valid(optarg
))
934 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
935 "MACVLAN network interface name not valid: %s", optarg
);
937 r
= test_network_interface_initialized(optarg
);
941 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
944 arg_private_network
= true;
945 arg_settings_mask
|= SETTING_NETWORK
;
948 case ARG_NETWORK_IPVLAN
:
950 if (!ifname_valid(optarg
))
951 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
952 "IPVLAN network interface name not valid: %s", optarg
);
954 r
= test_network_interface_initialized(optarg
);
958 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
962 case ARG_PRIVATE_NETWORK
:
963 arg_private_network
= true;
964 arg_settings_mask
|= SETTING_NETWORK
;
967 case ARG_NETWORK_NAMESPACE_PATH
:
968 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
972 arg_settings_mask
|= SETTING_NETWORK
;
976 if (arg_start_mode
== START_PID2
)
977 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
978 "--boot and --as-pid2 may not be combined.");
980 arg_start_mode
= START_BOOT
;
981 arg_settings_mask
|= SETTING_START_MODE
;
985 if (arg_start_mode
== START_BOOT
)
986 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
987 "--boot and --as-pid2 may not be combined.");
989 arg_start_mode
= START_PID2
;
990 arg_settings_mask
|= SETTING_START_MODE
;
994 r
= sd_id128_from_string(optarg
, &arg_uuid
);
996 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
998 if (sd_id128_is_null(arg_uuid
))
999 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1000 "Machine UUID may not be all zeroes.");
1002 arg_settings_mask
|= SETTING_MACHINE_ID
;
1006 _cleanup_free_
char *mangled
= NULL
;
1008 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1012 free_and_replace(arg_slice
, mangled
);
1013 arg_settings_mask
|= SETTING_SLICE
;
1018 if (isempty(optarg
))
1019 arg_machine
= mfree(arg_machine
);
1021 if (!hostname_is_valid(optarg
, 0))
1022 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1023 "Invalid machine name: %s", optarg
);
1025 r
= free_and_strdup(&arg_machine
, optarg
);
1032 if (isempty(optarg
))
1033 arg_hostname
= mfree(arg_hostname
);
1035 if (!hostname_is_valid(optarg
, 0))
1036 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1037 "Invalid hostname: %s", optarg
);
1039 r
= free_and_strdup(&arg_hostname
, optarg
);
1044 arg_settings_mask
|= SETTING_HOSTNAME
;
1048 arg_selinux_context
= optarg
;
1052 arg_selinux_apifs_context
= optarg
;
1056 arg_read_only
= true;
1057 arg_settings_mask
|= SETTING_READ_ONLY
;
1060 case ARG_AMBIENT_CAPABILITY
: {
1062 r
= parse_capability_spec(optarg
, &m
);
1065 arg_caps_ambient
|= m
;
1066 arg_settings_mask
|= SETTING_CAPABILITY
;
1069 case ARG_CAPABILITY
:
1070 case ARG_DROP_CAPABILITY
: {
1072 r
= parse_capability_spec(optarg
, &m
);
1076 if (c
== ARG_CAPABILITY
)
1080 arg_settings_mask
|= SETTING_CAPABILITY
;
1083 case ARG_NO_NEW_PRIVILEGES
:
1084 r
= parse_boolean(optarg
);
1086 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1088 arg_no_new_privileges
= r
;
1089 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1093 arg_link_journal
= LINK_GUEST
;
1094 arg_link_journal_try
= true;
1095 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1098 case ARG_LINK_JOURNAL
:
1099 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1101 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1103 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1108 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1110 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1112 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1116 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1118 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1120 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1124 case ARG_OVERLAY_RO
:
1125 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1126 if (r
== -EADDRNOTAVAIL
)
1127 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1129 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1131 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1134 case ARG_INACCESSIBLE
:
1135 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1137 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1139 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1143 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1145 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1147 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1154 case ARG_SHARE_SYSTEM
:
1155 /* We don't officially support this anymore, except for compat reasons. People should use the
1156 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1157 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1158 arg_clone_ns_flags
= 0;
1162 r
= parse_boolean(optarg
);
1164 log_error("Failed to parse --register= argument: %s", optarg
);
1172 arg_keep_unit
= true;
1175 case ARG_PERSONALITY
:
1177 arg_personality
= personality_from_string(optarg
);
1178 if (arg_personality
== PERSONALITY_INVALID
)
1179 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1180 "Unknown or unsupported personality '%s'.", optarg
);
1182 arg_settings_mask
|= SETTING_PERSONALITY
;
1188 arg_volatile_mode
= VOLATILE_YES
;
1189 else if (streq(optarg
, "help")) {
1190 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1195 m
= volatile_mode_from_string(optarg
);
1197 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1198 "Failed to parse --volatile= argument: %s", optarg
);
1200 arg_volatile_mode
= m
;
1203 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1207 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1209 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1211 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1213 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1217 if (strv_extend(&arg_property
, optarg
) < 0)
1222 case ARG_PRIVATE_USERS
: {
1227 else if (!in_charset(optarg
, DIGITS
))
1228 /* do *not* parse numbers as booleans */
1229 boolean
= parse_boolean(optarg
);
1234 /* no: User namespacing off */
1235 arg_userns_mode
= USER_NAMESPACE_NO
;
1236 arg_uid_shift
= UID_INVALID
;
1237 arg_uid_range
= UINT32_C(0x10000);
1238 } else if (boolean
> 0) {
1239 /* yes: User namespacing on, UID range is read from root dir */
1240 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1241 arg_uid_shift
= UID_INVALID
;
1242 arg_uid_range
= UINT32_C(0x10000);
1243 } else if (streq(optarg
, "pick")) {
1244 /* pick: User namespacing on, UID range is picked randomly */
1245 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1246 * implied by USER_NAMESPACE_PICK
1248 arg_uid_shift
= UID_INVALID
;
1249 arg_uid_range
= UINT32_C(0x10000);
1251 } else if (streq(optarg
, "identity")) {
1252 /* identitiy: User namespaces on, UID range is map the 0…0xFFFF range to
1253 * itself, i.e. we don't actually map anything, but do take benefit of
1254 * isolation of capability sets. */
1255 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1257 arg_uid_range
= UINT32_C(0x10000);
1259 _cleanup_free_
char *buffer
= NULL
;
1260 const char *range
, *shift
;
1262 /* anything else: User namespacing on, UID range is explicitly configured */
1264 range
= strchr(optarg
, ':');
1266 buffer
= strndup(optarg
, range
- optarg
);
1272 r
= safe_atou32(range
, &arg_uid_range
);
1274 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1278 r
= parse_uid(shift
, &arg_uid_shift
);
1280 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1282 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1284 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1285 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1288 arg_settings_mask
|= SETTING_USERNS
;
1293 if (userns_supported()) {
1294 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1295 * implied by USER_NAMESPACE_PICK
1297 arg_uid_shift
= UID_INVALID
;
1298 arg_uid_range
= UINT32_C(0x10000);
1300 arg_settings_mask
|= SETTING_USERNS
;
1305 case ARG_PRIVATE_USERS_CHOWN
:
1306 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1308 arg_settings_mask
|= SETTING_USERNS
;
1311 case ARG_PRIVATE_USERS_OWNERSHIP
:
1312 if (streq(optarg
, "help")) {
1313 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1317 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1318 if (arg_userns_ownership
< 0)
1319 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1321 arg_settings_mask
|= SETTING_USERNS
;
1324 case ARG_KILL_SIGNAL
:
1325 if (streq(optarg
, "help")) {
1326 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1330 arg_kill_signal
= signal_from_string(optarg
);
1331 if (arg_kill_signal
< 0)
1332 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1334 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1339 /* no → do not read files
1340 * yes → read files, do not override cmdline, trust only subset
1341 * override → read files, override cmdline, trust only subset
1342 * trusted → read files, do not override cmdline, trust all
1345 r
= parse_boolean(optarg
);
1347 if (streq(optarg
, "trusted")) {
1348 mask_all_settings
= false;
1349 mask_no_settings
= false;
1350 arg_settings_trusted
= true;
1352 } else if (streq(optarg
, "override")) {
1353 mask_all_settings
= false;
1354 mask_no_settings
= true;
1355 arg_settings_trusted
= -1;
1357 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1360 mask_all_settings
= false;
1361 mask_no_settings
= false;
1362 arg_settings_trusted
= -1;
1365 mask_all_settings
= true;
1366 mask_no_settings
= false;
1367 arg_settings_trusted
= false;
1373 if (!path_is_absolute(optarg
))
1374 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1375 "Working directory %s is not an absolute path.", optarg
);
1377 r
= free_and_strdup(&arg_chdir
, optarg
);
1381 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1384 case ARG_PIVOT_ROOT
:
1385 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1387 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1389 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1392 case ARG_NOTIFY_READY
:
1393 r
= parse_boolean(optarg
);
1395 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1396 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1397 arg_notify_ready
= r
;
1398 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1401 case ARG_ROOT_HASH
: {
1402 _cleanup_free_
void *k
= NULL
;
1405 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1407 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1408 if (l
< sizeof(sd_id128_t
))
1409 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1411 free_and_replace(arg_verity_settings
.root_hash
, k
);
1412 arg_verity_settings
.root_hash_size
= l
;
1416 case ARG_ROOT_HASH_SIG
: {
1421 if ((value
= startswith(optarg
, "base64:"))) {
1422 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1424 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1427 r
= read_full_file(optarg
, (char**) &p
, &l
);
1429 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1432 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1433 arg_verity_settings
.root_hash_sig_size
= l
;
1437 case ARG_VERITY_DATA
:
1438 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1443 case ARG_SYSTEM_CALL_FILTER
: {
1447 negative
= optarg
[0] == '~';
1448 items
= negative
? optarg
+ 1 : optarg
;
1451 _cleanup_free_
char *word
= NULL
;
1453 r
= extract_first_word(&items
, &word
, NULL
, 0);
1459 return log_error_errno(r
, "Failed to parse system call filter: %m");
1462 r
= strv_extend(&arg_syscall_deny_list
, word
);
1464 r
= strv_extend(&arg_syscall_allow_list
, word
);
1469 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1475 _cleanup_free_
char *name
= NULL
;
1478 if (streq(optarg
, "help")) {
1479 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1483 eq
= strchr(optarg
, '=');
1485 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1486 "--rlimit= expects an '=' assignment.");
1488 name
= strndup(optarg
, eq
- optarg
);
1492 rl
= rlimit_from_string_harder(name
);
1494 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1496 if (!arg_rlimit
[rl
]) {
1497 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1498 if (!arg_rlimit
[rl
])
1502 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1504 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1506 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1510 case ARG_OOM_SCORE_ADJUST
:
1511 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1513 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1515 arg_oom_score_adjust_set
= true;
1516 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1519 case ARG_CPU_AFFINITY
: {
1522 r
= parse_cpu_set(optarg
, &cpuset
);
1524 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1526 cpu_set_reset(&arg_cpu_set
);
1527 arg_cpu_set
= cpuset
;
1528 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1532 case ARG_RESOLV_CONF
:
1533 if (streq(optarg
, "help")) {
1534 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1538 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1539 if (arg_resolv_conf
< 0)
1540 return log_error_errno(arg_resolv_conf
,
1541 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1543 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1547 if (streq(optarg
, "help")) {
1548 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1552 arg_timezone
= timezone_mode_from_string(optarg
);
1553 if (arg_timezone
< 0)
1554 return log_error_errno(arg_timezone
,
1555 "Failed to parse /etc/localtime mode: %s", optarg
);
1557 arg_settings_mask
|= SETTING_TIMEZONE
;
1561 r
= handle_arg_console(optarg
);
1568 r
= handle_arg_console("pipe");
1574 arg_pager_flags
|= PAGER_DISABLE
;
1577 case ARG_SET_CREDENTIAL
: {
1578 _cleanup_free_
char *word
= NULL
, *data
= NULL
;
1579 const char *p
= optarg
;
1583 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1587 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1589 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1591 if (!credential_name_valid(word
))
1592 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1594 for (size_t i
= 0; i
< arg_n_credentials
; i
++)
1595 if (streq(arg_credentials
[i
].id
, word
))
1596 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1598 l
= cunescape(p
, UNESCAPE_ACCEPT_NUL
, &data
);
1600 return log_error_errno(l
, "Failed to unescape credential data: %s", p
);
1602 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1606 a
[arg_n_credentials
++] = (Credential
) {
1607 .id
= TAKE_PTR(word
),
1608 .data
= TAKE_PTR(data
),
1612 arg_credentials
= a
;
1614 arg_settings_mask
|= SETTING_CREDENTIALS
;
1618 case ARG_LOAD_CREDENTIAL
: {
1619 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
1620 _cleanup_(erase_and_freep
) char *data
= NULL
;
1621 _cleanup_free_
char *word
= NULL
, *j
= NULL
;
1622 const char *p
= optarg
;
1626 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1630 return log_error_errno(r
, "Failed to parse --load-credential= parameter: %m");
1632 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --load-credential=: %s", optarg
);
1634 if (!credential_name_valid(word
))
1635 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1637 for (i
= 0; i
< arg_n_credentials
; i
++)
1638 if (streq(arg_credentials
[i
].id
, word
))
1639 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1641 if (path_is_absolute(p
))
1642 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
1646 r
= get_credentials_dir(&e
);
1648 return log_error_errno(r
, "Credential not available (no credentials passed at all): %s", word
);
1650 j
= path_join(e
, p
);
1655 r
= read_full_file_full(AT_FDCWD
, j
?: p
, UINT64_MAX
, SIZE_MAX
,
1660 return log_error_errno(r
, "Failed to read credential '%s': %m", j
?: p
);
1662 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1666 a
[arg_n_credentials
++] = (Credential
) {
1667 .id
= TAKE_PTR(word
),
1668 .data
= TAKE_PTR(data
),
1672 arg_credentials
= a
;
1674 arg_settings_mask
|= SETTING_CREDENTIALS
;
1679 if (!valid_user_group_name(optarg
, 0))
1680 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1682 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1685 arg_settings_mask
|= SETTING_BIND_USER
;
1688 case ARG_SUPPRESS_SYNC
:
1689 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1693 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1700 assert_not_reached();
1703 if (argc
> optind
) {
1704 strv_free(arg_parameters
);
1705 arg_parameters
= strv_copy(argv
+ optind
);
1706 if (!arg_parameters
)
1709 arg_settings_mask
|= SETTING_START_MODE
;
1712 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1713 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1714 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1715 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1717 arg_directory
= TAKE_PTR(arg_template
);
1719 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1721 /* Make sure to parse environment before we reset the settings mask below */
1722 r
= parse_environment();
1726 /* Load all settings from .nspawn files */
1727 if (mask_no_settings
)
1728 arg_settings_mask
= 0;
1730 /* Don't load any settings from .nspawn files */
1731 if (mask_all_settings
)
1732 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1737 static int verify_arguments(void) {
1740 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1741 /* If we are running the stub init in the container, we don't need to look at what the init
1742 * in the container supports, because we are not using it. Let's immediately pick the right
1743 * setting based on the host system configuration.
1745 * We only do this, if the user didn't use an environment variable to override the detection.
1748 r
= cg_all_unified();
1750 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1752 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1753 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1754 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1756 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1759 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1760 arg_mount_settings
|= MOUNT_USE_USERNS
;
1762 if (arg_private_network
)
1763 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1765 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1766 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1767 arg_register
= false;
1768 if (arg_start_mode
!= START_PID1
)
1769 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1772 if (arg_userns_ownership
< 0)
1773 arg_userns_ownership
=
1774 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1775 USER_NAMESPACE_OWNERSHIP_OFF
;
1777 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1778 arg_kill_signal
= SIGRTMIN
+3;
1780 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 */
1781 arg_read_only
= true;
1783 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1784 arg_read_only
= true;
1786 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1787 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1788 * The latter is not technically a user session, but we don't need to labour the point. */
1789 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1791 if (arg_directory
&& arg_image
)
1792 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1794 if (arg_template
&& arg_image
)
1795 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1797 if (arg_template
&& !(arg_directory
|| arg_machine
))
1798 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1800 if (arg_ephemeral
&& arg_template
)
1801 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1803 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1804 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1806 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1807 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1809 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1810 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1811 "--read-only and --private-users-ownership=chown may not be combined.");
1813 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1814 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1815 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1816 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1817 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1819 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1820 * we need to error out, to avoid conflicts between different network options. */
1821 if (arg_network_namespace_path
&&
1822 (arg_network_interfaces
|| arg_network_macvlan
||
1823 arg_network_ipvlan
|| arg_network_veth_extra
||
1824 arg_network_bridge
|| arg_network_zone
||
1826 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1828 if (arg_network_bridge
&& arg_network_zone
)
1829 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1830 "--network-bridge= and --network-zone= may not be combined.");
1832 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1833 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1835 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1836 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1838 if (arg_expose_ports
&& !arg_private_network
)
1839 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1841 if (arg_caps_ambient
) {
1842 if (arg_caps_ambient
== UINT64_MAX
)
1843 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1845 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1846 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1848 if (arg_start_mode
== START_BOOT
)
1849 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1852 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1853 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1855 /* Drop duplicate --bind-user= entries */
1856 strv_uniq(arg_bind_user
);
1858 r
= custom_mount_check_all();
1865 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1868 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1871 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1874 if (uid
!= UID_INVALID
) {
1875 uid
+= arg_uid_shift
;
1877 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1881 if (gid
!= GID_INVALID
) {
1882 gid
+= (gid_t
) arg_uid_shift
;
1884 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1888 return RET_NERRNO(lchown(p
, uid
, gid
));
1891 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1895 q
= prefix_roota(root
, path
);
1896 r
= RET_NERRNO(mkdir(q
, mode
));
1902 return userns_lchown(q
, uid
, gid
);
1905 static const char *timezone_from_path(const char *path
) {
1906 return PATH_STARTSWITH_SET(
1908 "../usr/share/zoneinfo/",
1909 "/usr/share/zoneinfo/");
1912 static bool etc_writable(void) {
1913 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1916 static int setup_timezone(const char *dest
) {
1917 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1918 const char *where
, *check
;
1924 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1925 r
= readlink_malloc("/etc/localtime", &p
);
1926 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1927 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1928 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1929 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1931 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1932 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1936 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1939 } else if (arg_timezone
== TIMEZONE_AUTO
)
1940 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1946 if (m
== TIMEZONE_OFF
)
1949 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1951 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1955 where
= strjoina(etc
, "/localtime");
1959 case TIMEZONE_DELETE
:
1960 if (unlink(where
) < 0)
1961 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1965 case TIMEZONE_SYMLINK
: {
1966 _cleanup_free_
char *q
= NULL
;
1967 const char *z
, *what
;
1969 z
= timezone_from_path(p
);
1971 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1975 r
= readlink_malloc(where
, &q
);
1976 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1977 return 0; /* Already pointing to the right place? Then do nothing .. */
1979 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1980 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1982 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1984 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1985 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1986 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1990 what
= strjoina("../usr/share/zoneinfo/", z
);
1991 if (symlink(what
, where
) < 0) {
1992 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1993 errno
, "Failed to correct timezone of container, ignoring: %m");
2003 case TIMEZONE_BIND
: {
2004 _cleanup_free_
char *resolved
= NULL
;
2007 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2009 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
2013 if (found
== 0) /* missing? */
2014 (void) touch(resolved
);
2016 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
2018 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2024 /* If mounting failed, try to copy */
2025 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2027 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2028 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
2035 assert_not_reached();
2038 /* Fix permissions of the symlink or file copy we just created */
2039 r
= userns_lchown(where
, 0, 0);
2041 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
2046 static int have_resolv_conf(const char *path
) {
2049 if (access(path
, F_OK
) < 0) {
2050 if (errno
== ENOENT
)
2053 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
2059 static int resolved_listening(void) {
2060 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2061 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
2062 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
2065 /* Check if resolved is listening */
2067 r
= sd_bus_open_system(&bus
);
2069 return log_debug_errno(r
, "Failed to open system bus: %m");
2071 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2073 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2077 r
= sd_bus_get_property_string(bus
,
2078 "org.freedesktop.resolve1",
2079 "/org/freedesktop/resolve1",
2080 "org.freedesktop.resolve1.Manager",
2083 &dns_stub_listener_mode
);
2085 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2087 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2090 static int setup_resolv_conf(const char *dest
) {
2091 _cleanup_free_
char *etc
= NULL
;
2092 const char *where
, *what
;
2098 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2099 if (arg_private_network
)
2100 m
= RESOLV_CONF_OFF
;
2101 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2102 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2103 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2104 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2106 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2109 m
= arg_resolv_conf
;
2111 if (m
== RESOLV_CONF_OFF
)
2114 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2116 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2120 where
= strjoina(etc
, "/resolv.conf");
2122 if (m
== RESOLV_CONF_DELETE
) {
2123 if (unlink(where
) < 0)
2124 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2129 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2130 what
= PRIVATE_STATIC_RESOLV_CONF
;
2131 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2132 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2133 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2134 what
= PRIVATE_STUB_RESOLV_CONF
;
2136 what
= "/etc/resolv.conf";
2138 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2139 _cleanup_free_
char *resolved
= NULL
;
2142 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2144 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2148 if (found
== 0) /* missing? */
2149 (void) touch(resolved
);
2151 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2153 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2155 /* If that didn't work, let's copy the file */
2158 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2159 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2161 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
2163 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2164 * resolved or something similar runs inside and the symlink points there.
2166 * If the disk image is read-only, there's also no point in complaining.
2168 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2169 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2170 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2174 r
= userns_lchown(where
, 0, 0);
2176 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2181 static int setup_boot_id(void) {
2182 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2183 _cleanup_free_
char *path
= NULL
;
2184 sd_id128_t rnd
= SD_ID128_NULL
;
2188 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2190 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2192 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2194 r
= sd_id128_randomize(&rnd
);
2196 return log_error_errno(r
, "Failed to generate random boot id: %m");
2198 r
= id128_write(path
, ID128_UUID
, rnd
, false);
2200 return log_error_errno(r
, "Failed to write boot id: %m");
2202 from
= TAKE_PTR(path
);
2203 to
= "/proc/sys/kernel/random/boot_id";
2205 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2209 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2212 static int copy_devnodes(const char *dest
) {
2213 static const char devnodes
[] =
2227 BLOCK_WITH_UMASK(0000);
2229 /* Create /dev/net, so that we can create /dev/net/tun in it */
2230 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2231 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2233 NULSTR_FOREACH(d
, devnodes
) {
2234 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2237 from
= path_join("/dev/", d
);
2241 to
= path_join(dest
, from
);
2245 if (stat(from
, &st
) < 0) {
2247 if (errno
!= ENOENT
)
2248 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2250 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2251 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2252 "%s is not a char or block device, cannot copy.", from
);
2254 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2256 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2257 /* Explicitly warn the user when /dev is already populated. */
2258 if (errno
== EEXIST
)
2259 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
);
2261 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2263 /* Some systems abusively restrict mknod but allow bind mounts. */
2266 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2267 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2269 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2272 r
= userns_lchown(to
, 0, 0);
2274 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2276 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2280 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2282 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2284 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2287 prefixed
= path_join(dest
, sl
);
2291 t
= path_join("..", d
);
2295 if (symlink(t
, prefixed
) < 0)
2296 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2303 static int make_extra_nodes(const char *dest
) {
2307 BLOCK_WITH_UMASK(0000);
2309 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2310 _cleanup_free_
char *path
= NULL
;
2311 DeviceNode
*n
= arg_extra_nodes
+ i
;
2313 path
= path_join(dest
, n
->path
);
2317 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2318 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2320 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2322 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2328 static int setup_pts(const char *dest
) {
2329 _cleanup_free_
char *options
= NULL
;
2334 if (arg_selinux_apifs_context
)
2335 (void) asprintf(&options
,
2336 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2337 arg_uid_shift
+ TTY_GID
,
2338 arg_selinux_apifs_context
);
2341 (void) asprintf(&options
,
2342 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2343 arg_uid_shift
+ TTY_GID
);
2348 /* Mount /dev/pts itself */
2349 p
= prefix_roota(dest
, "/dev/pts");
2350 r
= RET_NERRNO(mkdir(p
, 0755));
2352 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2354 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2357 r
= userns_lchown(p
, 0, 0);
2359 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2361 /* Create /dev/ptmx symlink */
2362 p
= prefix_roota(dest
, "/dev/ptmx");
2363 if (symlink("pts/ptmx", p
) < 0)
2364 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2365 r
= userns_lchown(p
, 0, 0);
2367 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2369 /* And fix /dev/pts/ptmx ownership */
2370 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2371 r
= userns_lchown(p
, 0, 0);
2373 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2378 static int setup_stdio_as_dev_console(void) {
2379 _cleanup_close_
int terminal
= -1;
2382 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2383 * explicitly, if we are configured to. */
2384 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2386 return log_error_errno(terminal
, "Failed to open console: %m");
2388 /* Make sure we can continue logging to the original stderr, even if
2389 * stderr points elsewhere now */
2390 r
= log_dup_console();
2392 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2394 /* invalidates 'terminal' on success and failure */
2395 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2398 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2403 static int setup_dev_console(const char *console
) {
2404 _cleanup_free_
char *p
= NULL
;
2407 /* Create /dev/console symlink */
2408 r
= path_make_relative("/dev", console
, &p
);
2410 return log_error_errno(r
, "Failed to create relative path: %m");
2412 if (symlink(p
, "/dev/console") < 0)
2413 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2418 static int setup_keyring(void) {
2419 key_serial_t keyring
;
2421 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2422 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2423 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2424 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2425 * into the container. */
2427 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2428 if (keyring
== -1) {
2429 if (errno
== ENOSYS
)
2430 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2431 else if (ERRNO_IS_PRIVILEGE(errno
))
2432 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2434 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2440 static int setup_credentials(const char *root
) {
2444 if (arg_n_credentials
<= 0)
2447 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2449 return log_error_errno(r
, "Failed to create /run/host: %m");
2451 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2453 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2455 q
= prefix_roota(root
, "/run/host/credentials");
2456 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2460 for (size_t i
= 0; i
< arg_n_credentials
; i
++) {
2461 _cleanup_free_
char *j
= NULL
;
2462 _cleanup_close_
int fd
= -1;
2464 j
= path_join(q
, arg_credentials
[i
].id
);
2468 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2470 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2472 r
= loop_write(fd
, arg_credentials
[i
].data
, arg_credentials
[i
].size
, /* do_poll= */ false);
2474 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2476 if (fchmod(fd
, 0400) < 0)
2477 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2479 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2480 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2481 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2485 if (chmod(q
, 0500) < 0)
2486 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2488 r
= userns_lchown(q
, 0, 0);
2492 /* Make both mount and superblock read-only now */
2493 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2497 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2500 static int setup_kmsg(int kmsg_socket
) {
2501 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2502 _cleanup_free_
char *fifo
= NULL
;
2503 _cleanup_close_
int fd
= -1;
2506 assert(kmsg_socket
>= 0);
2508 BLOCK_WITH_UMASK(0000);
2510 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2511 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2512 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2513 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2515 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2517 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2519 if (mkfifo(fifo
, 0600) < 0)
2520 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2522 from
= TAKE_PTR(fifo
);
2524 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2528 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2530 return log_error_errno(errno
, "Failed to open fifo: %m");
2532 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2533 r
= send_one_fd(kmsg_socket
, fd
, 0);
2535 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2541 union in_addr_union address4
;
2542 union in_addr_union address6
;
2543 struct FirewallContext
*fw_ctx
;
2546 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2547 struct ExposeArgs
*args
= ASSERT_PTR(userdata
);
2552 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2553 (void) expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2557 static int setup_hostname(void) {
2560 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2563 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2565 return log_error_errno(r
, "Failed to set hostname: %m");
2570 static int setup_journal(const char *directory
) {
2571 _cleanup_free_
char *d
= NULL
;
2577 /* Don't link journals in ephemeral mode */
2581 if (arg_link_journal
== LINK_NO
)
2584 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2586 r
= sd_id128_get_machine(&this_id
);
2588 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2590 if (sd_id128_equal(arg_uuid
, this_id
)) {
2591 log_full(try ? LOG_WARNING
: LOG_ERR
,
2592 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2598 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2599 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2601 bool ignore
= r
== -EROFS
&& try;
2602 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2603 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2604 return ignore
? 0 : r
;
2608 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2609 q
= prefix_roota(directory
, p
);
2611 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2615 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2616 "%s: already a mount point, refusing to use for journal", p
);
2619 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2623 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2624 "%s: already a mount point, refusing to use for journal", q
);
2627 r
= readlink_and_make_absolute(p
, &d
);
2629 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2632 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2634 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2639 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2640 } else if (r
== -EINVAL
) {
2642 if (arg_link_journal
== LINK_GUEST
&&
2645 if (errno
== ENOTDIR
) {
2646 log_error("%s already exists and is neither a symlink nor a directory", p
);
2649 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2651 } else if (r
!= -ENOENT
)
2652 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2654 if (arg_link_journal
== LINK_GUEST
) {
2656 if (symlink(q
, p
) < 0) {
2658 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2661 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2664 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2666 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2670 if (arg_link_journal
== LINK_HOST
) {
2671 /* don't create parents here — if the host doesn't have
2672 * permanent journal set up, don't force it here */
2674 r
= RET_NERRNO(mkdir(p
, 0755));
2675 if (r
< 0 && r
!= -EEXIST
) {
2677 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2680 return log_error_errno(r
, "Failed to create %s: %m", p
);
2683 } else if (access(p
, F_OK
) < 0)
2686 if (dir_is_empty(q
, /* ignore_hidden_or_backup= */ false) == 0)
2687 log_warning("%s is not empty, proceeding anyway.", q
);
2689 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2691 return log_error_errno(r
, "Failed to create %s: %m", q
);
2693 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2695 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2700 static int drop_capabilities(uid_t uid
) {
2701 CapabilityQuintet q
;
2703 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2704 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2705 * arg_caps_retain. */
2707 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2708 q
= arg_full_capabilities
;
2710 if (q
.bounding
== UINT64_MAX
)
2711 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2713 if (q
.effective
== UINT64_MAX
)
2714 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2716 if (q
.inheritable
== UINT64_MAX
)
2717 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2719 if (q
.permitted
== UINT64_MAX
)
2720 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2722 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2723 q
.ambient
= arg_caps_ambient
;
2725 if (capability_quintet_mangle(&q
))
2726 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2729 q
= (CapabilityQuintet
) {
2730 .bounding
= arg_caps_retain
,
2731 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2732 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2733 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2734 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2737 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2738 * in order to maintain the same behavior as systemd < 242. */
2739 if (capability_quintet_mangle(&q
))
2740 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2741 "Some capabilities will not be set because they are not in the current bounding set.");
2745 return capability_quintet_enforce(&q
);
2748 static int reset_audit_loginuid(void) {
2749 _cleanup_free_
char *p
= NULL
;
2752 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2755 r
= read_one_line_file("/proc/self/loginuid", &p
);
2759 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2761 /* Already reset? */
2762 if (streq(p
, "4294967295"))
2765 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2768 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2769 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2770 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2771 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2772 "using systemd-nspawn. Sleeping for 5s... (%m)");
2780 static int setup_propagate(const char *root
) {
2784 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2785 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2786 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2787 (void) mkdir_p(p
, 0600);
2789 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2791 return log_error_errno(r
, "Failed to create /run/host: %m");
2793 r
= userns_mkdir(root
, "/run/host/incoming", 0600, 0, 0);
2795 return log_error_errno(r
, "Failed to create /run/host/incoming: %m");
2797 q
= prefix_roota(root
, "/run/host/incoming");
2798 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2802 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2806 /* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
2807 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2810 static int setup_machine_id(const char *directory
) {
2811 const char *etc_machine_id
;
2815 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2816 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2817 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2818 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2819 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2820 * container behaves nicely). */
2822 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2824 r
= id128_read(etc_machine_id
, ID128_PLAIN_OR_UNINIT
, &id
);
2826 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2827 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2829 if (sd_id128_is_null(arg_uuid
)) {
2830 r
= sd_id128_randomize(&arg_uuid
);
2832 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2835 if (sd_id128_is_null(id
))
2836 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2837 "Machine ID in container image is zero, refusing.");
2845 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2850 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2853 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2854 if (r
== -EOPNOTSUPP
)
2855 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.");
2857 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2859 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2861 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2863 log_debug("Patched directory tree to match UID/GID range.");
2870 * < 0 : wait_for_terminate() failed to get the state of the
2871 * container, the container was terminated by a signal, or
2872 * failed for an unknown reason. No change is made to the
2873 * container argument.
2874 * > 0 : The program executed in the container terminated with an
2875 * error. The exit code of the program executed in the
2876 * container is returned. The container argument has been set
2877 * to CONTAINER_TERMINATED.
2878 * 0 : The container is being rebooted, has been shut down or exited
2879 * successfully. The container argument has been set to either
2880 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2882 * That is, success is indicated by a return value of zero, and an
2883 * error is indicated by a non-zero value.
2885 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2889 r
= wait_for_terminate(pid
, &status
);
2891 return log_warning_errno(r
, "Failed to wait for container: %m");
2893 switch (status
.si_code
) {
2896 if (status
.si_status
== 0)
2897 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2899 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2901 *container
= CONTAINER_TERMINATED
;
2902 return status
.si_status
;
2905 if (status
.si_status
== SIGINT
) {
2906 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2907 *container
= CONTAINER_TERMINATED
;
2910 } else if (status
.si_status
== SIGHUP
) {
2911 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2912 *container
= CONTAINER_REBOOTED
;
2918 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2919 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2922 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2923 "Container %s failed due to unknown reason.", arg_machine
);
2927 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2930 pid
= PTR_TO_PID(userdata
);
2932 if (kill(pid
, arg_kill_signal
) >= 0) {
2933 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2934 sd_event_source_set_userdata(s
, NULL
);
2939 sd_event_exit(sd_event_source_get_event(s
), 0);
2943 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2949 pid
= PTR_TO_PID(userdata
);
2954 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2955 return log_error_errno(errno
, "Failed to waitid(): %m");
2956 if (si
.si_pid
== 0) /* No pending children. */
2958 if (si
.si_pid
== pid
) {
2959 /* The main process we care for has exited. Return from
2960 * signal handler but leave the zombie. */
2961 sd_event_exit(sd_event_source_get_event(s
), 0);
2965 /* Reap all other children. */
2966 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2972 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2977 pid
= PTR_TO_PID(userdata
);
2979 if (arg_kill_signal
> 0) {
2980 log_info("Container termination requested. Attempting to halt container.");
2981 (void) kill(pid
, arg_kill_signal
);
2983 log_info("Container termination requested. Exiting.");
2984 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2990 static int determine_names(void) {
2993 if (arg_template
&& !arg_directory
&& arg_machine
) {
2995 /* If --template= was specified then we should not
2996 * search for a machine, but instead create a new one
2997 * in /var/lib/machine. */
2999 arg_directory
= path_join("/var/lib/machines", arg_machine
);
3004 if (!arg_image
&& !arg_directory
) {
3006 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3008 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3010 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3012 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3014 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3015 r
= free_and_strdup(&arg_image
, i
->path
);
3017 r
= free_and_strdup(&arg_directory
, i
->path
);
3022 arg_read_only
= arg_read_only
|| i
->read_only
;
3024 r
= safe_getcwd(&arg_directory
);
3026 return log_error_errno(r
, "Failed to determine current directory: %m");
3029 if (!arg_directory
&& !arg_image
)
3030 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3034 if (arg_directory
&& path_equal(arg_directory
, "/"))
3035 arg_machine
= gethostname_malloc();
3036 else if (arg_image
) {
3039 arg_machine
= strdup(basename(arg_image
));
3041 /* Truncate suffix if there is one */
3042 e
= endswith(arg_machine
, ".raw");
3046 arg_machine
= strdup(basename(arg_directory
));
3050 hostname_cleanup(arg_machine
);
3051 if (!hostname_is_valid(arg_machine
, 0))
3052 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3054 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3055 * to match fixed config file names. */
3056 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3057 if (!arg_settings_filename
)
3060 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3061 * instances at once without manually having to specify -M each time. */
3063 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3066 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3067 if (!arg_settings_filename
)
3074 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
3083 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
3085 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3087 return free_and_replace(*p
, chased
);
3090 static int determine_uid_shift(const char *directory
) {
3092 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3097 if (arg_uid_shift
== UID_INVALID
) {
3100 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3102 if (stat(directory
, &st
) < 0)
3103 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3105 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3107 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3108 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3109 "UID and GID base of %s don't match.", directory
);
3111 arg_uid_range
= UINT32_C(0x10000);
3113 if (arg_uid_shift
!= 0) {
3114 /* If the image is shifted already, then we'll fall back to classic chowning, for
3115 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3117 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3118 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3119 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3120 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3121 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3122 "UID base of %s is not zero, UID mapping not supported.", directory
);
3126 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3127 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3132 static unsigned long effective_clone_ns_flags(void) {
3133 unsigned long flags
= arg_clone_ns_flags
;
3135 if (arg_private_network
)
3136 flags
|= CLONE_NEWNET
;
3138 flags
|= CLONE_NEWCGROUP
;
3139 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3140 flags
|= CLONE_NEWUSER
;
3145 static int patch_sysctl(void) {
3147 /* This table is inspired by runc's sysctl() function */
3148 static const struct {
3151 unsigned long clone_flags
;
3153 { "kernel.hostname", false, CLONE_NEWUTS
},
3154 { "kernel.domainname", false, CLONE_NEWUTS
},
3155 { "kernel.msgmax", false, CLONE_NEWIPC
},
3156 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3157 { "kernel.msgmni", false, CLONE_NEWIPC
},
3158 { "kernel.sem", false, CLONE_NEWIPC
},
3159 { "kernel.shmall", false, CLONE_NEWIPC
},
3160 { "kernel.shmmax", false, CLONE_NEWIPC
},
3161 { "kernel.shmmni", false, CLONE_NEWIPC
},
3162 { "fs.mqueue.", true, CLONE_NEWIPC
},
3163 { "net.", true, CLONE_NEWNET
},
3166 unsigned long flags
;
3169 flags
= effective_clone_ns_flags();
3171 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3175 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3177 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3180 if (safe_sysctl
[i
].prefix
)
3181 good
= startswith(*k
, safe_sysctl
[i
].key
);
3183 good
= streq(*k
, safe_sysctl
[i
].key
);
3190 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3192 r
= sysctl_write(*k
, *v
);
3194 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3200 static int inner_child(
3202 const char *directory
,
3206 int master_pty_socket
,
3208 char **os_release_pairs
) {
3210 _cleanup_free_
char *home
= NULL
;
3213 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3214 NULL
, /* container */
3219 NULL
, /* container_uuid */
3220 NULL
, /* LISTEN_FDS */
3221 NULL
, /* LISTEN_PID */
3222 NULL
, /* NOTIFY_SOCKET */
3223 NULL
, /* CREDENTIALS_DIRECTORY */
3227 const char *exec_target
;
3228 _cleanup_strv_free_
char **env_use
= NULL
;
3229 int r
, which_failed
;
3231 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3232 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3233 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3234 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3235 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3236 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3239 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3240 * unshare(). See below. */
3244 assert(kmsg_socket
>= 0);
3246 log_debug("Inner child is initializing.");
3248 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3249 /* Tell the parent, that it now can write the UID map. */
3250 (void) barrier_place(barrier
); /* #1 */
3252 /* Wait until the parent wrote the UID map */
3253 if (!barrier_place_and_sync(barrier
)) /* #2 */
3254 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3256 /* Become the new root user inside our namespace */
3257 r
= reset_uid_gid();
3259 return log_error_errno(r
, "Couldn't become new root: %m");
3261 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3262 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3263 * propagation, but simply create new peer groups for all our mounts). */
3264 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3270 arg_mount_settings
| MOUNT_IN_USERNS
,
3272 arg_selinux_apifs_context
);
3276 if (!arg_network_namespace_path
&& arg_private_network
) {
3277 r
= unshare(CLONE_NEWNET
);
3279 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3281 /* Tell the parent that it can setup network interfaces. */
3282 (void) barrier_place(barrier
); /* #3 */
3285 r
= mount_sysfs(NULL
, arg_mount_settings
);
3289 /* Wait until we are cgroup-ified, so that we
3290 * can mount the right cgroup path writable */
3291 if (!barrier_place_and_sync(barrier
)) /* #4 */
3292 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3293 "Parent died too early");
3296 r
= unshare(CLONE_NEWCGROUP
);
3298 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3301 arg_unified_cgroup_hierarchy
,
3302 arg_userns_mode
!= USER_NAMESPACE_NO
,
3305 arg_selinux_apifs_context
,
3308 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3312 r
= setup_boot_id();
3316 r
= setup_kmsg(kmsg_socket
);
3319 kmsg_socket
= safe_close(kmsg_socket
);
3324 arg_n_custom_mounts
,
3327 arg_selinux_apifs_context
,
3328 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3333 return log_error_errno(errno
, "setsid() failed: %m");
3335 if (arg_private_network
)
3336 (void) loopback_setup();
3338 if (arg_expose_ports
) {
3339 r
= expose_port_send_rtnl(rtnl_socket
);
3342 rtnl_socket
= safe_close(rtnl_socket
);
3345 if (arg_console_mode
!= CONSOLE_PIPE
) {
3346 _cleanup_close_
int master
= -1;
3347 _cleanup_free_
char *console
= NULL
;
3349 /* Allocate a pty and make it available as /dev/console. */
3350 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3352 return log_error_errno(master
, "Failed to allocate a pty: %m");
3354 r
= setup_dev_console(console
);
3356 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3358 r
= send_one_fd(master_pty_socket
, master
, 0);
3360 return log_error_errno(r
, "Failed to send master fd: %m");
3361 master_pty_socket
= safe_close(master_pty_socket
);
3363 r
= setup_stdio_as_dev_console();
3372 if (arg_oom_score_adjust_set
) {
3373 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3375 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3378 if (arg_cpu_set
.set
)
3379 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3380 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3382 (void) setup_hostname();
3384 if (arg_personality
!= PERSONALITY_INVALID
) {
3385 r
= safe_personality(arg_personality
);
3387 return log_error_errno(r
, "personality() failed: %m");
3388 } else if (secondary
) {
3389 r
= safe_personality(PER_LINUX32
);
3391 return log_error_errno(r
, "personality() failed: %m");
3394 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3396 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3401 if (is_seccomp_available()) {
3403 r
= seccomp_load(arg_seccomp
);
3404 if (ERRNO_IS_SECCOMP_FATAL(r
))
3405 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3407 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3412 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3417 if (arg_suppress_sync
) {
3419 r
= seccomp_suppress_sync();
3421 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3423 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3428 if (arg_selinux_context
)
3429 if (setexeccon(arg_selinux_context
) < 0)
3430 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3433 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3434 * if we need to later on. */
3435 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3436 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3438 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3439 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3441 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3445 r
= drop_capabilities(getuid());
3447 return log_error_errno(r
, "Dropping capabilities failed: %m");
3449 if (arg_no_new_privileges
)
3450 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3451 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3453 /* LXC sets container=lxc, so follow the scheme here */
3454 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3456 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3460 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3461 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3464 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3465 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3466 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3469 assert(!sd_id128_is_null(arg_uuid
));
3471 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3474 if (fdset_size(fds
) > 0) {
3475 r
= fdset_cloexec(fds
, false);
3477 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3479 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3480 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3483 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3486 if (arg_n_credentials
> 0) {
3487 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3493 if (arg_start_mode
!= START_BOOT
) {
3494 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3500 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3504 /* Let the parent know that we are ready and
3505 * wait until the parent is ready with the
3507 if (!barrier_place_and_sync(barrier
)) /* #5 */
3508 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3511 if (chdir(arg_chdir
) < 0)
3512 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3514 if (arg_start_mode
== START_PID2
) {
3515 r
= stub_pid1(arg_uuid
);
3520 if (arg_console_mode
!= CONSOLE_PIPE
) {
3521 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3522 * are configured for that. Acquire it as controlling tty. */
3523 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3524 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3527 log_debug("Inner child completed, invoking payload.");
3529 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3530 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3531 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3533 log_set_open_when_needed(true);
3535 (void) fdset_close_others(fds
);
3537 if (arg_start_mode
== START_BOOT
) {
3541 /* Automatically search for the init system */
3543 m
= strv_length(arg_parameters
);
3544 a
= newa(char*, m
+ 2);
3545 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3548 FOREACH_STRING(init
,
3549 "/usr/lib/systemd/systemd",
3550 "/lib/systemd/systemd",
3552 a
[0] = (char*) init
;
3553 execve(a
[0], a
, env_use
);
3556 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3557 } else if (!strv_isempty(arg_parameters
)) {
3558 const char *dollar_path
;
3560 exec_target
= arg_parameters
[0];
3562 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3564 dollar_path
= strv_env_get(env_use
, "PATH");
3566 if (setenv("PATH", dollar_path
, 1) < 0)
3567 return log_error_errno(errno
, "Failed to update $PATH: %m");
3570 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3573 /* If we cannot change the directory, we'll end up in /, that is expected. */
3574 (void) chdir(home
?: "/root");
3576 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3577 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3578 execle("/bin/bash", "-bash", NULL
, env_use
);
3579 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3580 execle("/bin/sh", "-sh", NULL
, env_use
);
3582 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3585 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3588 static int setup_notify_child(void) {
3589 _cleanup_close_
int fd
= -1;
3590 static const union sockaddr_union sa
= {
3591 .un
.sun_family
= AF_UNIX
,
3592 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3596 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3598 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3600 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3601 (void) sockaddr_un_unlink(&sa
.un
);
3603 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3605 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3607 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3609 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3611 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3613 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3618 static int outer_child(
3620 const char *directory
,
3621 DissectedImage
*dissected_image
,
3628 int uid_shift_socket
,
3629 int master_pty_socket
,
3630 int unified_cgroup_hierarchy_socket
,
3634 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3635 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3636 _cleanup_close_
int fd
= -1;
3643 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3644 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3645 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3646 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3647 * forked off it, and it exits. */
3651 assert(pid_socket
>= 0);
3652 assert(uuid_socket
>= 0);
3653 assert(notify_socket
>= 0);
3654 assert(master_pty_socket
>= 0);
3655 assert(kmsg_socket
>= 0);
3657 log_debug("Outer child is initializing.");
3659 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3661 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3663 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3664 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3666 r
= reset_audit_loginuid();
3670 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3671 * mounts to the real root. */
3672 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3676 if (dissected_image
) {
3677 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3678 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3679 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3680 * right place right away. This makes sure ESP partitions and userns are compatible. */
3682 r
= dissected_image_mount_and_warn(
3687 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3688 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3689 DISSECT_IMAGE_USR_NO_ROOT
|
3690 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3691 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3696 r
= determine_uid_shift(directory
);
3700 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3701 /* Let the parent know which UID shift we read from the image */
3702 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3704 return log_error_errno(errno
, "Failed to send UID shift: %m");
3705 if (l
!= sizeof(arg_uid_shift
))
3706 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3707 "Short write while sending UID shift.");
3709 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3710 /* When we are supposed to pick the UID shift, the parent will check now whether the
3711 * UID shift we just read from the image is available. If yes, it will send the UID
3712 * shift back to us, if not it will pick a different one, and send it back to us. */
3714 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3716 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3717 if (l
!= sizeof(arg_uid_shift
))
3718 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3719 "Short read while receiving UID shift.");
3722 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3723 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3726 if (path_equal(directory
, "/")) {
3727 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3728 * place, so that we can make changes to its mount structure (for example, to implement
3729 * --volatile=) without this interfering with our ability to access files such as
3730 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3731 * (instead of a temporary directory, since we are living in our own mount namspace here
3732 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3733 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3735 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3739 directory
= "/run/systemd/nspawn-root";
3742 r
= setup_pivot_root(
3745 arg_pivot_root_old
);
3749 r
= setup_volatile_mode(
3753 arg_selinux_apifs_context
);
3757 r
= bind_user_prepare(
3762 &arg_custom_mounts
, &arg_n_custom_mounts
,
3763 &bind_user_context
);
3767 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3768 /* Send the user maps we determined to the parent, so that it installs it in our user
3769 * namespace UID map table */
3771 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3773 bind_user_context
->data
[i
].payload_user
->uid
,
3774 bind_user_context
->data
[i
].host_user
->uid
,
3775 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3776 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3779 l
= send(uid_shift_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3781 return log_error_errno(errno
, "Failed to send user UID map: %m");
3782 if (l
!= sizeof(map
))
3783 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3784 "Short write while sending user UID map.");
3791 arg_n_custom_mounts
,
3794 arg_selinux_apifs_context
,
3799 /* Make sure we always have a mount that we can move to root later on. */
3800 r
= make_mount_point(directory
);
3804 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3805 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3806 arg_uid_shift
!= 0) {
3808 r
= remount_idmap(directory
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3809 if (r
== -EINVAL
|| ERRNO_IS_NOT_SUPPORTED(r
)) {
3810 /* This might fail because the kernel or file system doesn't support idmapping. We
3811 * can't really distinguish this nicely, nor do we have any guarantees about the
3812 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3813 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3814 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3815 "ID mapped mounts are apparently not available, sorry.");
3817 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3818 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3820 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3822 log_debug("ID mapped mounts available, making use of them.");
3827 if (dissected_image
) {
3828 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3829 r
= dissected_image_mount(
3834 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3835 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3836 DISSECT_IMAGE_USR_NO_ROOT
|
3837 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3838 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3840 return log_error_errno(r
, "File system check for image failed: %m");
3842 return log_error_errno(r
, "Failed to mount image file system: %m");
3845 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3846 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3848 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3852 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3854 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3855 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3856 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3857 "Short write while sending cgroup mode.");
3859 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3862 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3863 * mounts available in systemd services inside the container that create a new mount namespace. See
3864 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3865 * will inherit the shared propagation mode.
3867 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3868 * directory mount to root later on.
3869 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3871 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3875 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3879 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3883 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3884 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3885 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3887 return log_error_errno(r
, "Failed to make tree read-only: %m");
3890 r
= mount_all(directory
,
3893 arg_selinux_apifs_context
);
3897 r
= copy_devnodes(directory
);
3901 r
= make_extra_nodes(directory
);
3905 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3907 p
= prefix_roota(directory
, "/run/host");
3908 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3910 r
= setup_pts(directory
);
3914 r
= setup_propagate(directory
);
3918 r
= setup_keyring();
3922 r
= setup_credentials(directory
);
3926 r
= bind_user_setup(bind_user_context
, directory
);
3933 arg_n_custom_mounts
,
3936 arg_selinux_apifs_context
,
3937 MOUNT_NON_ROOT_ONLY
);
3941 r
= setup_timezone(directory
);
3945 r
= setup_resolv_conf(directory
);
3949 r
= setup_machine_id(directory
);
3953 r
= setup_journal(directory
);
3957 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3958 p
= prefix_roota(directory
, "/run/host/container-manager");
3959 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
);
3961 /* The same stuff as the $container_uuid env var */
3962 p
= prefix_roota(directory
, "/run/host/container-uuid");
3963 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
3965 if (!arg_use_cgns
) {
3968 arg_unified_cgroup_hierarchy
,
3969 arg_userns_mode
!= USER_NAMESPACE_NO
,
3972 arg_selinux_apifs_context
,
3978 r
= mount_move_root(directory
);
3980 return log_error_errno(r
, "Failed to move root directory: %m");
3982 fd
= setup_notify_child();
3986 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3987 arg_clone_ns_flags
|
3988 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3990 return log_error_errno(errno
, "Failed to fork inner child: %m");
3992 pid_socket
= safe_close(pid_socket
);
3993 uuid_socket
= safe_close(uuid_socket
);
3994 notify_socket
= safe_close(notify_socket
);
3995 uid_shift_socket
= safe_close(uid_shift_socket
);
3997 /* The inner child has all namespaces that are requested, so that we all are owned by the
3998 * user if user namespaces are turned on. */
4000 if (arg_network_namespace_path
) {
4001 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
4003 return log_error_errno(r
, "Failed to join network namespace: %m");
4006 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
, os_release_pairs
);
4008 _exit(EXIT_FAILURE
);
4010 _exit(EXIT_SUCCESS
);
4013 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4015 return log_error_errno(errno
, "Failed to send PID: %m");
4016 if (l
!= sizeof(pid
))
4017 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4018 "Short write while sending PID.");
4020 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4022 return log_error_errno(errno
, "Failed to send machine ID: %m");
4023 if (l
!= sizeof(arg_uuid
))
4024 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4025 "Short write while sending machine ID.");
4027 l
= send_one_fd(notify_socket
, fd
, 0);
4029 return log_error_errno(l
, "Failed to send notify fd: %m");
4031 pid_socket
= safe_close(pid_socket
);
4032 uuid_socket
= safe_close(uuid_socket
);
4033 notify_socket
= safe_close(notify_socket
);
4034 master_pty_socket
= safe_close(master_pty_socket
);
4035 kmsg_socket
= safe_close(kmsg_socket
);
4036 rtnl_socket
= safe_close(rtnl_socket
);
4037 netns_fd
= safe_close(netns_fd
);
4042 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4043 bool tried_hashed
= false;
4044 unsigned n_tries
= 100;
4049 assert(ret_lock_file
);
4050 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4051 assert(arg_uid_range
== 0x10000U
);
4055 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4058 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4059 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4064 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4066 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4069 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4070 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4071 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4076 /* Make some superficial checks whether the range is currently known in the user database */
4077 if (getpwuid(candidate
))
4079 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4081 if (getgrgid(candidate
))
4083 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4086 *ret_lock_file
= lf
;
4087 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4092 if (arg_machine
&& !tried_hashed
) {
4093 /* Try to hash the base from the container name */
4095 static const uint8_t hash_key
[] = {
4096 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4097 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4100 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4102 tried_hashed
= true;
4104 random_bytes(&candidate
, sizeof(candidate
));
4106 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4107 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4111 static int add_one_uid_map(
4113 uid_t container_uid
,
4117 return strextendf(p
,
4118 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4119 container_uid
, host_uid
, range
);
4122 static int make_uid_map_string(
4123 const uid_t bind_user_uid
[],
4124 size_t n_bind_user_uid
,
4128 _cleanup_free_
char *s
= NULL
;
4129 uid_t previous_uid
= 0;
4132 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4133 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4136 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4137 * quadruplet, consisting of host and container UID + GID. */
4139 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4140 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4141 host_uid
= bind_user_uid
[i
*4+offset
+1];
4143 assert(previous_uid
<= payload_uid
);
4144 assert(payload_uid
< arg_uid_range
);
4146 /* Add a range to close the gap to previous entry */
4147 if (payload_uid
> previous_uid
) {
4148 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4153 /* Map this specific user */
4154 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4158 previous_uid
= payload_uid
+ 1;
4161 /* And add a range to close the gap to finish the range */
4162 if (arg_uid_range
> previous_uid
) {
4163 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4174 static int setup_uid_map(
4176 const uid_t bind_user_uid
[],
4177 size_t n_bind_user_uid
) {
4179 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4180 _cleanup_free_
char *s
= NULL
;
4185 /* Build the UID map string */
4186 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4189 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4190 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4192 return log_error_errno(r
, "Failed to write UID map: %m");
4194 /* And now build the GID map string */
4196 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4199 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4200 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4202 return log_error_errno(r
, "Failed to write GID map: %m");
4207 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4208 char buf
[NOTIFY_BUFFER_MAX
+1];
4210 struct iovec iovec
= {
4212 .iov_len
= sizeof(buf
)-1,
4214 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4215 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4216 struct msghdr msghdr
= {
4219 .msg_control
= &control
,
4220 .msg_controllen
= sizeof(control
),
4222 struct ucred
*ucred
;
4224 pid_t inner_child_pid
;
4225 _cleanup_strv_free_
char **tags
= NULL
;
4230 inner_child_pid
= PTR_TO_PID(userdata
);
4232 if (revents
!= EPOLLIN
) {
4233 log_warning("Got unexpected poll event for notify fd.");
4237 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4239 if (ERRNO_IS_TRANSIENT(n
))
4242 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4245 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4248 cmsg_close_all(&msghdr
);
4250 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4251 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4252 log_debug("Received notify message without valid credentials. Ignoring.");
4256 if ((size_t) n
>= sizeof(buf
)) {
4257 log_warning("Received notify message exceeded maximum size. Ignoring.");
4262 tags
= strv_split(buf
, "\n\r");
4266 if (strv_contains(tags
, "READY=1")) {
4267 r
= sd_notify(false, "READY=1\n");
4269 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4272 p
= strv_find_startswith(tags
, "STATUS=");
4274 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4279 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4282 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4284 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4286 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4291 static int merge_settings(Settings
*settings
, const char *path
) {
4297 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4298 * that this steals the fields of the Settings* structure, and hence modifies it. */
4300 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4301 settings
->start_mode
>= 0) {
4302 arg_start_mode
= settings
->start_mode
;
4303 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4306 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4307 settings
->ephemeral
>= 0)
4308 arg_ephemeral
= settings
->ephemeral
;
4310 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4313 if (!arg_settings_trusted
)
4314 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4316 free_and_replace(arg_directory
, settings
->root
);
4319 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4320 settings
->pivot_root_new
) {
4321 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4322 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4325 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4326 settings
->working_directory
)
4327 free_and_replace(arg_chdir
, settings
->working_directory
);
4329 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4330 settings
->environment
)
4331 strv_free_and_replace(arg_setenv
, settings
->environment
);
4333 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4336 free_and_replace(arg_user
, settings
->user
);
4338 if (uid_is_valid(settings
->uid
))
4339 arg_uid
= settings
->uid
;
4340 if (gid_is_valid(settings
->gid
))
4341 arg_gid
= settings
->gid
;
4342 if (settings
->n_supplementary_gids
> 0) {
4343 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4344 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4348 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4349 uint64_t plus
, minus
;
4350 uint64_t network_minus
= 0;
4353 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4354 * Settings structure */
4356 plus
= settings
->capability
;
4357 minus
= settings
->drop_capability
;
4359 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4360 settings_network_configured(settings
)) {
4361 if (settings_private_network(settings
))
4362 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4364 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4367 if (!arg_settings_trusted
&& plus
!= 0) {
4368 if (settings
->capability
!= 0)
4369 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4371 arg_caps_retain
&= ~network_minus
;
4372 arg_caps_retain
|= plus
;
4375 arg_caps_retain
&= ~minus
;
4377 /* Copy the full capabilities over too */
4378 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4379 if (!arg_settings_trusted
)
4380 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4382 arg_full_capabilities
= settings
->full_capabilities
;
4385 ambient
= settings
->ambient_capability
;
4386 if (!arg_settings_trusted
&& ambient
!= 0)
4387 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4389 arg_caps_ambient
|= ambient
;
4392 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4393 settings
->kill_signal
> 0)
4394 arg_kill_signal
= settings
->kill_signal
;
4396 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4397 settings
->personality
!= PERSONALITY_INVALID
)
4398 arg_personality
= settings
->personality
;
4400 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4401 !sd_id128_is_null(settings
->machine_id
)) {
4403 if (!arg_settings_trusted
)
4404 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4406 arg_uuid
= settings
->machine_id
;
4409 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4410 settings
->read_only
>= 0)
4411 arg_read_only
= settings
->read_only
;
4413 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4414 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4415 arg_volatile_mode
= settings
->volatile_mode
;
4417 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4418 settings
->n_custom_mounts
> 0) {
4420 if (!arg_settings_trusted
)
4421 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4423 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4424 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4425 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4426 settings
->n_custom_mounts
= 0;
4430 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4431 settings_network_configured(settings
)) {
4433 if (!arg_settings_trusted
)
4434 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4436 arg_network_veth
= settings_network_veth(settings
);
4437 arg_private_network
= settings_private_network(settings
);
4439 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4440 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4441 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4442 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4444 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4445 free_and_replace(arg_network_zone
, settings
->network_zone
);
4447 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4451 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4452 settings
->expose_ports
) {
4454 if (!arg_settings_trusted
)
4455 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4457 expose_port_free_all(arg_expose_ports
);
4458 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4462 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4463 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4465 if (!arg_settings_trusted
)
4466 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4468 arg_userns_mode
= settings
->userns_mode
;
4469 arg_uid_shift
= settings
->uid_shift
;
4470 arg_uid_range
= settings
->uid_range
;
4471 arg_userns_ownership
= settings
->userns_ownership
;
4475 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4476 !strv_isempty(settings
->bind_user
))
4477 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4479 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4480 settings
->notify_ready
>= 0)
4481 arg_notify_ready
= settings
->notify_ready
;
4483 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4485 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4486 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4487 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4489 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4490 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4495 if (settings
->seccomp
) {
4496 if (!arg_settings_trusted
)
4497 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4499 seccomp_release(arg_seccomp
);
4500 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4506 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4507 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4510 if (!settings
->rlimit
[rl
])
4513 if (!arg_settings_trusted
) {
4514 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4518 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4521 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4523 free_and_replace(arg_hostname
, settings
->hostname
);
4525 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4526 settings
->no_new_privileges
>= 0)
4527 arg_no_new_privileges
= settings
->no_new_privileges
;
4529 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4530 settings
->oom_score_adjust_set
) {
4532 if (!arg_settings_trusted
)
4533 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4535 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4536 arg_oom_score_adjust_set
= true;
4540 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4541 settings
->cpu_set
.set
) {
4543 if (!arg_settings_trusted
)
4544 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4546 cpu_set_reset(&arg_cpu_set
);
4547 arg_cpu_set
= settings
->cpu_set
;
4548 settings
->cpu_set
= (CPUSet
) {};
4552 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4553 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4554 arg_resolv_conf
= settings
->resolv_conf
;
4556 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4557 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4559 if (!arg_settings_trusted
)
4560 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4562 arg_link_journal
= settings
->link_journal
;
4563 arg_link_journal_try
= settings
->link_journal_try
;
4567 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4568 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4569 arg_timezone
= settings
->timezone
;
4571 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4574 if (!arg_settings_trusted
)
4575 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4577 free_and_replace(arg_slice
, settings
->slice
);
4580 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4581 settings
->use_cgns
>= 0) {
4583 if (!arg_settings_trusted
)
4584 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4586 arg_use_cgns
= settings
->use_cgns
;
4589 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4590 settings
->clone_ns_flags
!= ULONG_MAX
) {
4592 if (!arg_settings_trusted
)
4593 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4595 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4598 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4599 settings
->console_mode
>= 0) {
4601 if (!arg_settings_trusted
)
4602 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4604 arg_console_mode
= settings
->console_mode
;
4607 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4608 settings
->suppress_sync
>= 0)
4609 arg_suppress_sync
= settings
->suppress_sync
;
4611 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4612 * don't consult arg_settings_mask for them. */
4614 sd_bus_message_unref(arg_property_message
);
4615 arg_property_message
= TAKE_PTR(settings
->properties
);
4617 arg_console_width
= settings
->console_width
;
4618 arg_console_height
= settings
->console_height
;
4620 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4621 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4622 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4627 static int load_settings(void) {
4628 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4629 _cleanup_fclose_
FILE *f
= NULL
;
4630 _cleanup_free_
char *p
= NULL
;
4636 /* If all settings are masked, there's no point in looking for
4637 * the settings file */
4638 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4641 /* We first look in the admin's directories in /etc and /run */
4642 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4643 _cleanup_free_
char *j
= NULL
;
4645 j
= path_join(i
, arg_settings_filename
);
4653 /* By default, we trust configuration from /etc and /run */
4654 if (arg_settings_trusted
< 0)
4655 arg_settings_trusted
= true;
4660 if (errno
!= ENOENT
)
4661 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4665 /* After that, let's look for a file next to the
4666 * actual image we shall boot. */
4669 p
= file_in_same_dir(arg_image
, arg_settings_filename
);
4672 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4673 p
= file_in_same_dir(arg_directory
, arg_settings_filename
);
4680 if (!f
&& errno
!= ENOENT
)
4681 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4683 /* By default, we do not trust configuration from /var/lib/machines */
4684 if (arg_settings_trusted
< 0)
4685 arg_settings_trusted
= false;
4692 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4694 r
= settings_load(f
, p
, &settings
);
4698 return merge_settings(settings
, p
);
4701 static int load_oci_bundle(void) {
4702 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4705 if (!arg_oci_bundle
)
4708 /* By default let's trust OCI bundles */
4709 if (arg_settings_trusted
< 0)
4710 arg_settings_trusted
= true;
4712 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4716 return merge_settings(settings
, arg_oci_bundle
);
4719 static int run_container(
4720 DissectedImage
*dissected_image
,
4723 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4724 struct ExposeArgs
*expose_args
,
4725 int *master
, pid_t
*pid
, int *ret
) {
4727 static const struct sigaction sa
= {
4728 .sa_handler
= nop_signal_handler
,
4729 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4732 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4733 _cleanup_close_
int etc_passwd_lock
= -1;
4734 _cleanup_close_pair_
int
4735 kmsg_socket_pair
[2] = { -1, -1 },
4736 rtnl_socket_pair
[2] = { -1, -1 },
4737 pid_socket_pair
[2] = { -1, -1 },
4738 uuid_socket_pair
[2] = { -1, -1 },
4739 notify_socket_pair
[2] = { -1, -1 },
4740 uid_shift_socket_pair
[2] = { -1, -1 },
4741 master_pty_socket_pair
[2] = { -1, -1 },
4742 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4744 _cleanup_close_
int notify_socket
= -1;
4745 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4746 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4747 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4748 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4749 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4750 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4751 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4752 size_t n_bind_user_uid
= 0;
4753 ContainerStatus container_status
= 0;
4757 _cleanup_close_
int child_netns_fd
= -1;
4759 assert_se(sigemptyset(&mask_chld
) == 0);
4760 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4762 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4763 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4764 * check with getpwuid() if the specific user already exists. Note that /etc might be
4765 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4766 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4767 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4770 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4771 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4772 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4775 r
= barrier_create(&barrier
);
4777 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4779 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4780 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4782 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4783 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4785 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4786 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4788 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4789 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4791 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4792 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4794 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4795 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4797 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4798 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4799 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4801 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4802 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4803 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4805 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4806 * parent's blocking calls and give it a chance to call wait() and terminate. */
4807 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4809 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4811 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4813 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4815 if (arg_network_namespace_path
) {
4816 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4817 if (child_netns_fd
< 0)
4818 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4820 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4822 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
);
4824 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4826 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4827 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4830 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4832 return log_error_errno(errno
, "clone() failed%s: %m",
4834 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4837 /* The outer child only has a file system namespace. */
4838 barrier_set_role(&barrier
, BARRIER_CHILD
);
4840 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4841 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4842 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4843 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4844 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4845 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4846 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4847 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4849 (void) reset_all_signal_handlers();
4850 (void) reset_signal_mask();
4852 r
= outer_child(&barrier
,
4857 uuid_socket_pair
[1],
4858 notify_socket_pair
[1],
4859 kmsg_socket_pair
[1],
4860 rtnl_socket_pair
[1],
4861 uid_shift_socket_pair
[1],
4862 master_pty_socket_pair
[1],
4863 unified_cgroup_hierarchy_socket_pair
[1],
4867 _exit(EXIT_FAILURE
);
4869 _exit(EXIT_SUCCESS
);
4872 barrier_set_role(&barrier
, BARRIER_PARENT
);
4876 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4877 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4878 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4879 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4880 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4881 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4882 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4883 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4885 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4886 /* The child just let us know the UID shift it might have read from the image. */
4887 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4889 return log_error_errno(errno
, "Failed to read UID shift: %m");
4890 if (l
!= sizeof arg_uid_shift
)
4891 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4893 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4894 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4895 * image, but if that's already in use, pick a new one, and report back to the child,
4896 * which one we now picked. */
4898 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4900 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4902 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4904 return log_error_errno(errno
, "Failed to send UID shift: %m");
4905 if (l
!= sizeof arg_uid_shift
)
4906 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4909 n_bind_user_uid
= strv_length(arg_bind_user
);
4910 if (n_bind_user_uid
> 0) {
4911 /* Right after the UID shift, we'll receive the list of UID mappings for the
4912 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4914 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4918 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4919 l
= recv(uid_shift_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4921 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4922 if (l
!= sizeof(uid_t
)*4)
4923 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4924 SYNTHETIC_ERRNO(EIO
),
4925 "Short read while reading bind user UID pairs.");
4930 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4931 /* The child let us know the support cgroup mode it might have read from the image. */
4932 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4934 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4935 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4936 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
4937 l
, l
== 0 ? " The child is most likely dead." : "");
4940 /* Wait for the outer child. */
4941 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4944 if (r
!= EXIT_SUCCESS
)
4947 /* And now retrieve the PID of the inner child. */
4948 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4950 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4951 if (l
!= sizeof *pid
)
4952 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4954 /* We also retrieve container UUID in case it was generated by outer child */
4955 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4957 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4958 if (l
!= sizeof(arg_uuid
))
4959 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4961 /* We also retrieve the socket used for notifications generated by outer child */
4962 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4963 if (notify_socket
< 0)
4964 return log_error_errno(notify_socket
,
4965 "Failed to receive notification socket from the outer child: %m");
4967 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4969 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4970 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4971 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4973 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4977 (void) barrier_place(&barrier
); /* #2 */
4980 if (arg_private_network
) {
4981 if (!arg_network_namespace_path
) {
4982 /* Wait until the child has unshared its network namespace. */
4983 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4984 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4987 if (child_netns_fd
< 0) {
4988 /* Make sure we have an open file descriptor to the child's network
4989 * namespace so it stays alive even if the child exits. */
4990 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4992 return log_error_errno(r
, "Failed to open child network namespace: %m");
4995 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4999 if (arg_network_veth
) {
5000 r
= setup_veth(arg_machine
, *pid
, veth_name
,
5001 arg_network_bridge
|| arg_network_zone
);
5007 if (arg_network_bridge
) {
5008 /* Add the interface to a bridge */
5009 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
5014 } else if (arg_network_zone
) {
5015 /* Add the interface to a bridge, possibly creating it */
5016 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5024 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5028 /* We created the primary and extra veth links now; let's remember this, so that we know to
5029 remove them later on. Note that we don't bother with removing veth links that were created
5030 here when their setup failed half-way, because in that case the kernel should be able to
5031 remove them on its own, since they cannot be referenced by anything yet. */
5032 *veth_created
= true;
5034 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5038 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5043 if (arg_register
|| !arg_keep_unit
) {
5044 r
= sd_bus_default_system(&bus
);
5046 return log_error_errno(r
, "Failed to open system bus: %m");
5048 r
= sd_bus_set_close_on_exit(bus
, false);
5050 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5053 if (!arg_keep_unit
) {
5054 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5055 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5056 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5058 r
= sd_bus_match_signal_async(
5061 "org.freedesktop.systemd1",
5063 "org.freedesktop.systemd1.Scope",
5065 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5067 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5071 r
= register_machine(
5079 arg_custom_mounts
, arg_n_custom_mounts
,
5082 arg_property_message
,
5084 arg_container_service_name
);
5088 } else if (!arg_keep_unit
) {
5094 arg_custom_mounts
, arg_n_custom_mounts
,
5097 arg_property_message
);
5101 } else if (arg_slice
|| arg_property
)
5102 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5104 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5108 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5112 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5116 /* Notify the child that the parent is ready with all
5117 * its setup (including cgroup-ification), and that
5118 * the child can now hand over control to the code to
5119 * run inside the container. */
5120 (void) barrier_place(&barrier
); /* #4 */
5122 /* Block SIGCHLD here, before notifying child.
5123 * process_pty() will handle it with the other signals. */
5124 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5126 /* Reset signal to default */
5127 r
= default_signals(SIGCHLD
);
5129 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5131 r
= sd_event_new(&event
);
5133 return log_error_errno(r
, "Failed to get default event source: %m");
5135 (void) sd_event_set_watchdog(event
, true);
5138 r
= sd_bus_attach_event(bus
, event
, 0);
5140 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5143 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5147 /* Let the child know that we are ready and wait that the child is completely ready now. */
5148 if (!barrier_place_and_sync(&barrier
)) /* #5 */
5149 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5151 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5152 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5153 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5155 (void) sd_notifyf(false,
5156 "STATUS=Container running.\n"
5157 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5158 if (!arg_notify_ready
) {
5159 r
= sd_notify(false, "READY=1\n");
5161 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5164 if (arg_kill_signal
> 0) {
5165 /* Try to kill the init system on SIGINT or SIGTERM */
5166 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5167 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5169 /* Immediately exit */
5170 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5171 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5174 /* Exit when the child exits */
5175 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5177 if (arg_expose_ports
) {
5178 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5182 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5183 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5186 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
5188 if (arg_console_mode
!= CONSOLE_PIPE
) {
5189 _cleanup_close_
int fd
= -1;
5190 PTYForwardFlags flags
= 0;
5192 /* Retrieve the master pty allocated by inner child */
5193 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
5195 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5197 switch (arg_console_mode
) {
5199 case CONSOLE_READ_ONLY
:
5200 flags
|= PTY_FORWARD_READ_ONLY
;
5204 case CONSOLE_INTERACTIVE
:
5205 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5207 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5209 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5211 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5212 (void) pty_forward_set_width_height(forward
,
5214 arg_console_height
);
5218 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5221 *master
= TAKE_FD(fd
);
5224 r
= sd_event_loop(event
);
5226 return log_error_errno(r
, "Failed to run event loop: %m");
5231 (void) pty_forward_get_last_char(forward
, &last_char
);
5232 forward
= pty_forward_free(forward
);
5234 if (!arg_quiet
&& last_char
!= '\n')
5238 /* Kill if it is not dead yet anyway */
5239 if (!arg_register
&& !arg_keep_unit
&& bus
)
5240 terminate_scope(bus
, arg_machine
);
5242 /* Normally redundant, but better safe than sorry */
5243 (void) kill(*pid
, SIGKILL
);
5245 if (arg_private_network
) {
5246 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5247 * to avoid having to move the parent to the child network namespace. */
5248 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
5253 _cleanup_close_
int parent_netns_fd
= -1;
5255 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5257 log_error_errno(r
, "Failed to open parent network namespace: %m");
5258 _exit(EXIT_FAILURE
);
5261 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5263 log_error_errno(r
, "Failed to enter child network namespace: %m");
5264 _exit(EXIT_FAILURE
);
5267 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5269 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5271 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5275 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5277 /* Tell machined that we are gone. */
5279 (void) unregister_machine(bus
, arg_machine
);
5282 /* We failed to wait for the container, or the container exited abnormally. */
5284 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5285 /* r > 0 → The container exited with a non-zero status.
5286 * As a special case, we need to replace 133 with a different value,
5287 * because 133 is special-cased in the service file to reboot the container.
5288 * otherwise → The container exited with zero status and a reboot was not requested.
5290 if (r
== EXIT_FORCE_RESTART
)
5291 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5293 return 0; /* finito */
5296 /* CONTAINER_REBOOTED, loop again */
5298 if (arg_keep_unit
) {
5299 /* Special handling if we are running as a service: instead of simply
5300 * restarting the machine we want to restart the entire service, so let's
5301 * inform systemd about this with the special exit code 133. The service
5302 * file uses RestartForceExitStatus=133 so that this results in a full
5303 * nspawn restart. This is necessary since we might have cgroup parameters
5304 * set we want to have flushed out. */
5305 *ret
= EXIT_FORCE_RESTART
;
5306 return 0; /* finito */
5309 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5310 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5312 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5313 *veth_created
= false;
5314 return 1; /* loop again */
5317 static int initialize_rlimits(void) {
5318 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5319 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5320 * container execution environments. */
5322 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5323 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5324 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5325 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5326 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5327 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5328 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5329 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5330 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5331 [RLIMIT_NICE
] = { 0, 0 },
5332 [RLIMIT_NOFILE
] = { 1024, 4096 },
5333 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5334 [RLIMIT_RTPRIO
] = { 0, 0 },
5335 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5336 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5338 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5339 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5340 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5341 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5342 * that PID 1 changes a number of other resource limits during early initialization which is why we
5343 * don't read the other limits from PID 1 but prefer the static table above. */
5348 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5349 /* Let's only fill in what the user hasn't explicitly configured anyway */
5350 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5351 const struct rlimit
*v
;
5352 struct rlimit buffer
;
5354 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5355 /* For these two let's read the limits off PID 1. See above for an explanation. */
5357 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
5358 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5361 } else if (rl
== RLIMIT_NOFILE
) {
5362 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5363 * userspace. Given that nspawn containers are often run without our PID 1,
5364 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5365 * so that container userspace gets similar resources as host userspace
5367 buffer
= kernel_defaults
[rl
];
5368 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5371 v
= kernel_defaults
+ rl
;
5373 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5374 if (!arg_rlimit
[rl
])
5378 if (DEBUG_LOGGING
) {
5379 _cleanup_free_
char *k
= NULL
;
5381 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5382 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5389 static int cant_be_in_netns(void) {
5390 char udev_path
[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t
)];
5391 _cleanup_free_
char *udev_ns
= NULL
, *our_ns
= NULL
;
5392 _cleanup_close_
int fd
= -1;
5396 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5397 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5400 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5403 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5405 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5407 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5409 if (r
== -ENOENT
|| ERRNO_IS_DISCONNECT(r
))
5410 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5411 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5413 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5416 r
= getpeercred(fd
, &ucred
);
5418 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5420 xsprintf(udev_path
, "/proc/" PID_FMT
"/ns/net", ucred
.pid
);
5421 r
= readlink_malloc(udev_path
, &udev_ns
);
5423 return log_error_errno(r
, "Failed to read network namespace of udev: %m");
5425 r
= readlink_malloc("/proc/self/ns/net", &our_ns
);
5427 return log_error_errno(r
, "Failed to read our own network namespace: %m");
5429 if (!streq(our_ns
, udev_ns
))
5430 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5431 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5435 static int run(int argc
, char *argv
[]) {
5436 bool secondary
= false, remove_directory
= false, remove_image
= false,
5437 veth_created
= false, remove_tmprootdir
= false;
5438 _cleanup_close_
int master
= -1;
5439 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5440 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5441 char veth_name
[IFNAMSIZ
] = "";
5442 struct ExposeArgs expose_args
= {};
5443 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5444 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5445 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5446 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5447 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5450 log_parse_environment();
5453 r
= parse_argv(argc
, argv
);
5457 if (geteuid() != 0) {
5458 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5459 argc
>= 2 ? "Need to be root." :
5460 "Need to be root (and some arguments are usually required).\nHint: try --help");
5464 r
= cant_be_in_netns();
5468 r
= initialize_rlimits();
5472 r
= load_oci_bundle();
5476 r
= determine_names();
5480 r
= load_settings();
5486 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5490 r
= verify_arguments();
5494 /* Reapply environment settings. */
5495 (void) detect_unified_cgroup_hierarchy_from_environment();
5497 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5498 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5499 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5500 (void) ignore_signals(SIGPIPE
);
5502 n_fd_passed
= sd_listen_fds(false);
5503 if (n_fd_passed
> 0) {
5504 r
= fdset_new_listen_fds(&fds
, false);
5506 log_error_errno(r
, "Failed to collect file descriptors: %m");
5511 /* The "default" umask. This is appropriate for most file and directory
5512 * operations performed by nspawn, and is the umask that will be used for
5513 * the child. Functions like copy_devnodes() change the umask temporarily. */
5516 if (arg_directory
) {
5519 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5520 * /var from the host will propagate into container dynamically (because bad things happen if
5521 * two systems write to the same /var). Let's allow it for the special cases where /var is
5522 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5523 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5524 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5529 if (arg_ephemeral
) {
5530 _cleanup_free_
char *np
= NULL
;
5532 r
= chase_symlinks_and_update(&arg_directory
, 0);
5536 /* If the specified path is a mount point we generate the new snapshot immediately
5537 * inside it under a random name. However if the specified is not a mount point we
5538 * create the new snapshot in the parent directory, just next to it. */
5539 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5541 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5545 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5547 r
= tempfn_random(arg_directory
, "machine.", &np
);
5549 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5553 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5554 * only owned by us and no one else. */
5555 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5557 log_error_errno(r
, "Failed to lock %s: %m", np
);
5562 BLOCK_SIGNALS(SIGINT
);
5563 r
= btrfs_subvol_snapshot(arg_directory
, np
,
5564 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5565 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5566 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5567 BTRFS_SNAPSHOT_RECURSIVE
|
5568 BTRFS_SNAPSHOT_QUOTA
|
5569 BTRFS_SNAPSHOT_SIGINT
);
5572 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5576 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5580 free_and_replace(arg_directory
, np
);
5581 remove_directory
= true;
5583 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5587 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5589 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5593 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5598 r
= chase_symlinks_and_update(&arg_template
, 0);
5603 BLOCK_SIGNALS(SIGINT
);
5604 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5605 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5606 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5607 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5608 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5609 BTRFS_SNAPSHOT_RECURSIVE
|
5610 BTRFS_SNAPSHOT_QUOTA
|
5611 BTRFS_SNAPSHOT_SIGINT
);
5614 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5615 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5616 else if (r
== -EINTR
) {
5617 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5620 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5623 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5624 "Populated %s from template %s.", arg_directory
, arg_template
);
5628 if (arg_start_mode
== START_BOOT
) {
5629 _cleanup_free_
char *b
= NULL
;
5632 if (arg_pivot_root_new
) {
5633 b
= path_join(arg_directory
, arg_pivot_root_new
);
5641 if (path_is_os_tree(p
) <= 0) {
5642 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5643 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5647 _cleanup_free_
char *p
= NULL
;
5649 if (arg_pivot_root_new
)
5650 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5652 p
= path_join(arg_directory
, "/usr/");
5656 if (laccess(p
, F_OK
) < 0) {
5657 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5658 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5664 DissectImageFlags dissect_image_flags
=
5665 DISSECT_IMAGE_GENERIC_ROOT
|
5666 DISSECT_IMAGE_REQUIRE_ROOT
|
5667 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5668 DISSECT_IMAGE_USR_NO_ROOT
;
5670 assert(!arg_template
);
5672 r
= chase_symlinks_and_update(&arg_image
, 0);
5676 if (arg_ephemeral
) {
5677 _cleanup_free_
char *np
= NULL
;
5679 r
= tempfn_random(arg_image
, "machine.", &np
);
5681 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5685 /* Always take an exclusive lock on our own ephemeral copy. */
5686 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5688 r
= log_error_errno(r
, "Failed to create image lock: %m");
5693 BLOCK_SIGNALS(SIGINT
);
5694 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
);
5697 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5701 r
= log_error_errno(r
, "Failed to copy image file: %m");
5705 free_and_replace(arg_image
, np
);
5706 remove_image
= true;
5708 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5710 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5714 r
= log_error_errno(r
, "Failed to create image lock: %m");
5718 r
= verity_settings_load(
5719 &arg_verity_settings
,
5720 arg_image
, NULL
, NULL
);
5722 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5726 if (arg_verity_settings
.data_path
)
5727 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5730 if (!mkdtemp(tmprootdir
)) {
5731 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5735 remove_tmprootdir
= true;
5737 arg_directory
= strdup(tmprootdir
);
5738 if (!arg_directory
) {
5743 r
= loop_device_make_by_path(
5745 arg_read_only
? O_RDONLY
: O_RDWR
,
5746 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5750 log_error_errno(r
, "Failed to set up loopback block device: %m");
5754 r
= dissect_loop_device_and_warn(
5756 &arg_verity_settings
,
5758 dissect_image_flags
,
5761 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5762 log_notice("Note that the disk image needs to\n"
5763 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5764 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5765 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5766 " d) or contain a file system without a partition table\n"
5767 "in order to be bootable with systemd-nspawn.");
5773 r
= dissected_image_load_verity_sig_partition(
5776 &arg_verity_settings
);
5780 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5781 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5782 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5784 r
= dissected_image_decrypt_interactively(
5787 &arg_verity_settings
,
5792 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5793 if (remove_image
&& unlink(arg_image
) >= 0)
5794 remove_image
= false;
5797 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5801 if (arg_console_mode
< 0)
5803 isatty(STDIN_FILENO
) > 0 &&
5804 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5806 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5810 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5811 arg_machine
, arg_image
?: arg_directory
);
5813 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5815 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5816 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5820 if (arg_expose_ports
) {
5821 r
= fw_ctx_new(&fw_ctx
);
5823 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5826 expose_args
.fw_ctx
= fw_ctx
;
5829 r
= run_container(dissected_image
,
5832 veth_name
, &veth_created
,
5833 &expose_args
, &master
,
5840 (void) sd_notify(false,
5841 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5842 "STOPPING=1\nSTATUS=Terminating...");
5845 (void) kill(pid
, SIGKILL
);
5847 /* Try to flush whatever is still queued in the pty */
5849 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5850 master
= safe_close(master
);
5854 (void) wait_for_terminate(pid
, NULL
);
5858 if (remove_directory
&& arg_directory
) {
5861 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5863 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5866 if (remove_image
&& arg_image
) {
5867 if (unlink(arg_image
) < 0)
5868 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5871 if (remove_tmprootdir
) {
5872 if (rmdir(tmprootdir
) < 0)
5873 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5879 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5880 (void) rm_rf(p
, REMOVE_ROOT
);
5883 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5884 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5887 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5888 (void) remove_bridge(arg_network_zone
);
5890 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5891 expose_port_free_all(arg_expose_ports
);
5892 rlimit_free_all(arg_rlimit
);
5893 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5894 credential_free_all(arg_credentials
, arg_n_credentials
);
5902 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);