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"
32 #include "bus-error.h"
35 #include "capability-util.h"
36 #include "cgroup-util.h"
37 #include "chase-symlinks.h"
39 #include "cpu-set-util.h"
40 #include "creds-util.h"
41 #include "dev-setup.h"
42 #include "discover-image.h"
43 #include "dissect-image.h"
49 #include "format-util.h"
52 #include "hexdecoct.h"
53 #include "hostname-setup.h"
54 #include "hostname-util.h"
55 #include "id128-util.h"
58 #include "loop-util.h"
59 #include "loopback-setup.h"
61 #include "main-func.h"
62 #include "missing_sched.h"
64 #include "mount-util.h"
65 #include "mountpoint-util.h"
66 #include "namespace-util.h"
67 #include "netlink-util.h"
68 #include "nspawn-bind-user.h"
69 #include "nspawn-cgroup.h"
70 #include "nspawn-creds.h"
71 #include "nspawn-def.h"
72 #include "nspawn-expose-ports.h"
73 #include "nspawn-mount.h"
74 #include "nspawn-network.h"
75 #include "nspawn-oci.h"
76 #include "nspawn-patch-uid.h"
77 #include "nspawn-register.h"
78 #include "nspawn-seccomp.h"
79 #include "nspawn-settings.h"
80 #include "nspawn-setuid.h"
81 #include "nspawn-stub-pid1.h"
82 #include "nspawn-util.h"
84 #include "nulstr-util.h"
87 #include "parse-argument.h"
88 #include "parse-util.h"
89 #include "pretty-print.h"
90 #include "process-util.h"
92 #include "random-util.h"
93 #include "raw-clone.h"
94 #include "resolve-util.h"
95 #include "rlimit-util.h"
98 #include "seccomp-util.h"
100 #include "selinux-util.h"
101 #include "signal-util.h"
102 #include "socket-util.h"
103 #include "stat-util.h"
104 #include "stdio-util.h"
105 #include "string-table.h"
106 #include "string-util.h"
108 #include "sysctl-util.h"
109 #include "terminal-util.h"
110 #include "tmpfile-util.h"
111 #include "umask-util.h"
112 #include "unit-name.h"
113 #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"
117 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
119 #define EXIT_FORCE_RESTART 133
121 typedef enum ContainerStatus
{
122 CONTAINER_TERMINATED
,
126 static char *arg_directory
= NULL
;
127 static char *arg_template
= NULL
;
128 static char *arg_chdir
= NULL
;
129 static char *arg_pivot_root_new
= NULL
;
130 static char *arg_pivot_root_old
= NULL
;
131 static char *arg_user
= NULL
;
132 static uid_t arg_uid
= UID_INVALID
;
133 static gid_t arg_gid
= GID_INVALID
;
134 static gid_t
* arg_supplementary_gids
= NULL
;
135 static size_t arg_n_supplementary_gids
= 0;
136 static sd_id128_t arg_uuid
= {};
137 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
138 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
139 static const char *arg_selinux_context
= NULL
;
140 static const char *arg_selinux_apifs_context
= NULL
;
141 static char *arg_slice
= NULL
;
142 static bool arg_private_network
= false;
143 static bool arg_read_only
= false;
144 static StartMode arg_start_mode
= START_PID1
;
145 static bool arg_ephemeral
= false;
146 static LinkJournal arg_link_journal
= LINK_AUTO
;
147 static bool arg_link_journal_try
= false;
148 static uint64_t arg_caps_retain
=
149 (1ULL << CAP_AUDIT_CONTROL
) |
150 (1ULL << CAP_AUDIT_WRITE
) |
151 (1ULL << CAP_CHOWN
) |
152 (1ULL << CAP_DAC_OVERRIDE
) |
153 (1ULL << CAP_DAC_READ_SEARCH
) |
154 (1ULL << CAP_FOWNER
) |
155 (1ULL << CAP_FSETID
) |
156 (1ULL << CAP_IPC_OWNER
) |
158 (1ULL << CAP_LEASE
) |
159 (1ULL << CAP_LINUX_IMMUTABLE
) |
160 (1ULL << CAP_MKNOD
) |
161 (1ULL << CAP_NET_BIND_SERVICE
) |
162 (1ULL << CAP_NET_BROADCAST
) |
163 (1ULL << CAP_NET_RAW
) |
164 (1ULL << CAP_SETFCAP
) |
165 (1ULL << CAP_SETGID
) |
166 (1ULL << CAP_SETPCAP
) |
167 (1ULL << CAP_SETUID
) |
168 (1ULL << CAP_SYS_ADMIN
) |
169 (1ULL << CAP_SYS_BOOT
) |
170 (1ULL << CAP_SYS_CHROOT
) |
171 (1ULL << CAP_SYS_NICE
) |
172 (1ULL << CAP_SYS_PTRACE
) |
173 (1ULL << CAP_SYS_RESOURCE
) |
174 (1ULL << CAP_SYS_TTY_CONFIG
);
175 static uint64_t arg_caps_ambient
= 0;
176 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
177 static CustomMount
*arg_custom_mounts
= NULL
;
178 static size_t arg_n_custom_mounts
= 0;
179 static char **arg_setenv
= NULL
;
180 static bool arg_quiet
= false;
181 static bool arg_register
= true;
182 static bool arg_keep_unit
= false;
183 static char **arg_network_interfaces
= NULL
;
184 static char **arg_network_macvlan
= NULL
;
185 static char **arg_network_ipvlan
= NULL
;
186 static bool arg_network_veth
= false;
187 static char **arg_network_veth_extra
= NULL
;
188 static char *arg_network_bridge
= NULL
;
189 static char *arg_network_zone
= NULL
;
190 static char *arg_network_namespace_path
= NULL
;
191 static PagerFlags arg_pager_flags
= 0;
192 static unsigned long arg_personality
= PERSONALITY_INVALID
;
193 static char *arg_image
= NULL
;
194 static char *arg_oci_bundle
= NULL
;
195 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
196 static ExposePort
*arg_expose_ports
= NULL
;
197 static char **arg_property
= NULL
;
198 static sd_bus_message
*arg_property_message
= NULL
;
199 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
200 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
201 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
202 static int arg_kill_signal
= 0;
203 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
204 static SettingsMask arg_settings_mask
= 0;
205 static int arg_settings_trusted
= -1;
206 static char **arg_parameters
= NULL
;
207 static const char *arg_container_service_name
= "systemd-nspawn";
208 static bool arg_notify_ready
= false;
209 static bool arg_use_cgns
= true;
210 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
211 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
212 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
213 static char **arg_syscall_allow_list
= NULL
;
214 static char **arg_syscall_deny_list
= NULL
;
216 static scmp_filter_ctx arg_seccomp
= NULL
;
218 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
219 static bool arg_no_new_privileges
= false;
220 static int arg_oom_score_adjust
= 0;
221 static bool arg_oom_score_adjust_set
= false;
222 static CPUSet arg_cpu_set
= {};
223 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
224 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
225 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
226 static DeviceNode
* arg_extra_nodes
= NULL
;
227 static size_t arg_n_extra_nodes
= 0;
228 static char **arg_sysctl
= NULL
;
229 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
230 static Credential
*arg_credentials
= NULL
;
231 static size_t arg_n_credentials
= 0;
232 static char **arg_bind_user
= NULL
;
233 static bool arg_suppress_sync
= false;
234 static char *arg_settings_filename
= NULL
;
236 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
254 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
256 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
258 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
260 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
265 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
266 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
267 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
268 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename
, freep
);
270 static int handle_arg_console(const char *arg
) {
271 if (streq(arg
, "help")) {
280 if (streq(arg
, "interactive"))
281 arg_console_mode
= CONSOLE_INTERACTIVE
;
282 else if (streq(arg
, "read-only"))
283 arg_console_mode
= CONSOLE_READ_ONLY
;
284 else if (streq(arg
, "passive"))
285 arg_console_mode
= CONSOLE_PASSIVE
;
286 else if (streq(arg
, "pipe")) {
287 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
288 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
289 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
290 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
291 "Proceeding anyway.");
293 arg_console_mode
= CONSOLE_PIPE
;
294 } else if (streq(arg
, "autopipe")) {
295 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
296 arg_console_mode
= CONSOLE_INTERACTIVE
;
298 arg_console_mode
= CONSOLE_PIPE
;
300 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
302 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
306 static int help(void) {
307 _cleanup_free_
char *link
= NULL
;
310 pager_open(arg_pager_flags
);
312 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
316 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
317 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
318 " -h --help Show this help\n"
319 " --version Print version string\n"
320 " -q --quiet Do not show status information\n"
321 " --no-pager Do not pipe output into a pager\n"
322 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
324 " -D --directory=PATH Root directory for the container\n"
325 " --template=PATH Initialize root directory from template directory,\n"
327 " -x --ephemeral Run container with snapshot of root directory, and\n"
328 " remove it after exit\n"
329 " -i --image=PATH Root file system disk image (or device node) for\n"
331 " --oci-bundle=PATH OCI bundle directory\n"
332 " --read-only Mount the root directory read-only\n"
333 " --volatile[=MODE] Run the system in volatile mode\n"
334 " --root-hash=HASH Specify verity root hash for root disk image\n"
335 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
336 " as a DER encoded PKCS7, either as a path to a file\n"
337 " or as an ASCII base64 encoded string prefixed by\n"
339 " --verity-data=PATH Specify hash device for verity\n"
340 " --pivot-root=PATH[:PATH]\n"
341 " Pivot root to given directory in the container\n\n"
342 "%3$sExecution:%4$s\n"
343 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
344 " -b --boot Boot up full system (i.e. invoke init)\n"
345 " --chdir=PATH Set working directory in the container\n"
346 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
347 " -u --user=USER Run the command under specified user or UID\n"
348 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
349 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
350 " --suppress-sync=BOOLEAN\n"
351 " Suppress any form of disk data synchronization\n\n"
352 "%3$sSystem Identity:%4$s\n"
353 " -M --machine=NAME Set the machine name for the container\n"
354 " --hostname=NAME Override the hostname for the container\n"
355 " --uuid=UUID Set a specific machine UUID for the container\n\n"
356 "%3$sProperties:%4$s\n"
357 " -S --slice=SLICE Place the container in the specified slice\n"
358 " --property=NAME=VALUE Set scope unit property\n"
359 " --register=BOOLEAN Register container as machine\n"
360 " --keep-unit Do not register a scope for the machine, reuse\n"
361 " the service unit nspawn is running in\n\n"
362 "%3$sUser Namespacing:%4$s\n"
363 " --private-users=no Run without user namespacing\n"
364 " --private-users=yes|pick|identity\n"
365 " Run within user namespace, autoselect UID/GID range\n"
366 " --private-users=UIDBASE[:NUIDS]\n"
367 " Similar, but with user configured UID/GID range\n"
368 " --private-users-ownership=MODE\n"
369 " Adjust ('chown') or map ('map') OS tree ownership\n"
370 " to private UID/GID range\n"
371 " -U Equivalent to --private-users=pick and\n"
372 " --private-users-ownership=auto\n\n"
373 "%3$sNetworking:%4$s\n"
374 " --private-network Disable network in container\n"
375 " --network-interface=INTERFACE\n"
376 " Assign an existing network interface to the\n"
378 " --network-macvlan=INTERFACE\n"
379 " Create a macvlan network interface based on an\n"
380 " existing network interface to the container\n"
381 " --network-ipvlan=INTERFACE\n"
382 " Create an ipvlan network interface based on an\n"
383 " existing network interface to the container\n"
384 " -n --network-veth Add a virtual Ethernet connection between host\n"
386 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
387 " Add an additional virtual Ethernet link between\n"
388 " host and container\n"
389 " --network-bridge=INTERFACE\n"
390 " Add a virtual Ethernet connection to the container\n"
391 " and attach it to an existing bridge on the host\n"
392 " --network-zone=NAME Similar, but attach the new interface to an\n"
393 " an automatically managed bridge interface\n"
394 " --network-namespace-path=PATH\n"
395 " Set network namespace to the one represented by\n"
396 " the specified kernel namespace file node\n"
397 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
398 " Expose a container IP port on the host\n\n"
399 "%3$sSecurity:%4$s\n"
400 " --capability=CAP In addition to the default, retain specified\n"
402 " --drop-capability=CAP Drop the specified capability from the default set\n"
403 " --ambient-capability=CAP\n"
404 " Sets the specified capability for the started\n"
405 " process. Not useful if booting a machine.\n"
406 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
407 " --system-call-filter=LIST|~LIST\n"
408 " Permit/prohibit specific system calls\n"
409 " -Z --selinux-context=SECLABEL\n"
410 " Set the SELinux security context to be used by\n"
411 " processes in the container\n"
412 " -L --selinux-apifs-context=SECLABEL\n"
413 " Set the SELinux security context to be used by\n"
414 " API/tmpfs file systems in the container\n\n"
415 "%3$sResources:%4$s\n"
416 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
417 " --oom-score-adjust=VALUE\n"
418 " Adjust the OOM score value for the payload\n"
419 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
420 " --personality=ARCH Pick personality for this container\n\n"
421 "%3$sIntegration:%4$s\n"
422 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
423 " --timezone=MODE Select mode of /etc/localtime initialization\n"
424 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
425 " host, try-guest, try-host\n"
426 " -j Equivalent to --link-journal=try-guest\n\n"
428 " --bind=PATH[:PATH[:OPTIONS]]\n"
429 " Bind mount a file or directory from the host into\n"
431 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
432 " Similar, but creates a read-only bind mount\n"
433 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
435 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
436 " --overlay=PATH[:PATH...]:PATH\n"
437 " Create an overlay mount from the host to \n"
439 " --overlay-ro=PATH[:PATH...]:PATH\n"
440 " Similar, but creates a read-only overlay mount\n"
441 " --bind-user=NAME Bind user from host to container\n\n"
442 "%3$sInput/Output:%4$s\n"
443 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
444 " set up for the container.\n"
445 " -P --pipe Equivalent to --console=pipe\n\n"
446 "%3$sCredentials:%4$s\n"
447 " --set-credential=ID:VALUE\n"
448 " Pass a credential with literal value to container.\n"
449 " --load-credential=ID:PATH\n"
450 " Load credential to pass to container from file or\n"
451 " AF_UNIX stream socket.\n"
452 "\nSee the %2$s for details.\n",
453 program_invocation_short_name
,
463 static int custom_mount_check_all(void) {
466 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
467 CustomMount
*m
= &arg_custom_mounts
[i
];
469 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
470 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
471 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
472 "--private-users-ownership=own may not be combined with custom root mounts.");
473 if (arg_uid_shift
== UID_INVALID
)
474 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
475 "--private-users with automatic UID shift may not be combined with custom root mounts.");
482 static int detect_unified_cgroup_hierarchy_from_environment(void) {
483 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
486 /* Allow the user to control whether the unified hierarchy is used */
490 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
491 var
= "UNIFIED_CGROUP_HIERARCHY";
496 r
= parse_boolean(e
);
498 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
500 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
502 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
508 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
511 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
512 * in the image actually supports. */
513 r
= cg_all_unified();
515 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
517 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
518 * routine only detects 231, so we'll have a false negative here for 230. */
519 r
= systemd_installation_has_version(directory
, "230");
521 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
523 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
525 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
526 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
527 /* Mixed cgroup hierarchy support was added in 233 */
528 r
= systemd_installation_has_version(directory
, "233");
530 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
532 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
534 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
536 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
538 log_debug("Using %s hierarchy for container.",
539 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
540 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
545 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
550 _cleanup_free_
char *t
= NULL
;
552 r
= extract_first_word(&spec
, &t
, ",", 0);
554 return log_error_errno(r
, "Failed to parse capability %s.", t
);
558 if (streq(t
, "help")) {
559 for (int i
= 0; i
< capability_list_length(); i
++) {
562 name
= capability_to_name(i
);
573 r
= capability_from_name(t
);
575 return log_error_errno(r
, "Failed to parse capability %s.", t
);
582 return 1; /* continue */
585 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
588 r
= getenv_bool(name
);
592 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
594 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
595 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
599 static int parse_mount_settings_env(void) {
603 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
604 if (r
< 0 && r
!= -ENXIO
)
605 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
607 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
609 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
610 if (streq_ptr(e
, "network"))
611 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
614 r
= parse_boolean(e
);
616 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
618 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
619 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
625 static int parse_environment(void) {
629 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
632 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
635 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
638 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
642 r
= parse_mount_settings_env();
646 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
647 * even if it is supported. If not supported, it has no effect. */
648 if (!cg_ns_supported())
649 arg_use_cgns
= false;
651 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
654 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
658 arg_use_cgns
= r
> 0;
659 arg_settings_mask
|= SETTING_USE_CGNS
;
663 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
665 arg_container_service_name
= e
;
667 r
= getenv_bool("SYSTEMD_SUPPRESS_SYNC");
669 arg_suppress_sync
= r
;
670 else if (r
!= -ENXIO
)
671 log_debug_errno(r
, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
673 return detect_unified_cgroup_hierarchy_from_environment();
676 static int parse_argv(int argc
, char *argv
[]) {
683 ARG_AMBIENT_CAPABILITY
,
695 ARG_NETWORK_INTERFACE
,
700 ARG_NETWORK_VETH_EXTRA
,
701 ARG_NETWORK_NAMESPACE_PATH
,
711 ARG_PRIVATE_USERS_CHOWN
,
712 ARG_PRIVATE_USERS_OWNERSHIP
,
717 ARG_SYSTEM_CALL_FILTER
,
720 ARG_NO_NEW_PRIVILEGES
,
721 ARG_OOM_SCORE_ADJUST
,
735 static const struct option options
[] = {
736 { "help", no_argument
, NULL
, 'h' },
737 { "version", no_argument
, NULL
, ARG_VERSION
},
738 { "directory", required_argument
, NULL
, 'D' },
739 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
740 { "ephemeral", no_argument
, NULL
, 'x' },
741 { "user", required_argument
, NULL
, 'u' },
742 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
743 { "as-pid2", no_argument
, NULL
, 'a' },
744 { "boot", no_argument
, NULL
, 'b' },
745 { "uuid", required_argument
, NULL
, ARG_UUID
},
746 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
747 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
748 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
749 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
750 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
751 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
752 { "bind", required_argument
, NULL
, ARG_BIND
},
753 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
754 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
755 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
756 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
757 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
758 { "machine", required_argument
, NULL
, 'M' },
759 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
760 { "slice", required_argument
, NULL
, 'S' },
761 { "setenv", required_argument
, NULL
, 'E' },
762 { "selinux-context", required_argument
, NULL
, 'Z' },
763 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
764 { "quiet", no_argument
, NULL
, 'q' },
765 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
766 { "register", required_argument
, NULL
, ARG_REGISTER
},
767 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
768 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
769 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
770 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
771 { "network-veth", no_argument
, NULL
, 'n' },
772 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
773 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
774 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
775 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
776 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
777 { "image", required_argument
, NULL
, 'i' },
778 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
779 { "port", required_argument
, NULL
, 'p' },
780 { "property", required_argument
, NULL
, ARG_PROPERTY
},
781 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
782 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
783 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
784 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
785 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
786 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
787 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
788 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
789 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
790 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
791 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
792 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
793 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
794 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
795 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
796 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
797 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
798 { "console", required_argument
, NULL
, ARG_CONSOLE
},
799 { "pipe", no_argument
, NULL
, ARG_PIPE
},
800 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
801 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
802 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
803 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
804 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
805 { "suppress-sync", required_argument
, NULL
, ARG_SUPPRESS_SYNC
},
810 uint64_t plus
= 0, minus
= 0;
811 bool mask_all_settings
= false, mask_no_settings
= false;
816 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
826 r
= parse_path_argument(optarg
, false, &arg_directory
);
830 arg_settings_mask
|= SETTING_DIRECTORY
;
834 r
= parse_path_argument(optarg
, false, &arg_template
);
838 arg_settings_mask
|= SETTING_DIRECTORY
;
842 r
= parse_path_argument(optarg
, false, &arg_image
);
846 arg_settings_mask
|= SETTING_DIRECTORY
;
850 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
857 arg_ephemeral
= true;
858 arg_settings_mask
|= SETTING_EPHEMERAL
;
862 r
= free_and_strdup(&arg_user
, optarg
);
866 arg_settings_mask
|= SETTING_USER
;
869 case ARG_NETWORK_ZONE
: {
872 j
= strjoin("vz-", optarg
);
876 if (!ifname_valid(j
)) {
877 log_error("Network zone name not valid: %s", j
);
882 free_and_replace(arg_network_zone
, j
);
884 arg_network_veth
= true;
885 arg_private_network
= true;
886 arg_settings_mask
|= SETTING_NETWORK
;
890 case ARG_NETWORK_BRIDGE
:
892 if (!ifname_valid(optarg
))
893 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
894 "Bridge interface name not valid: %s", optarg
);
896 r
= free_and_strdup(&arg_network_bridge
, optarg
);
902 arg_network_veth
= true;
903 arg_private_network
= true;
904 arg_settings_mask
|= SETTING_NETWORK
;
907 case ARG_NETWORK_VETH_EXTRA
:
908 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
910 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
912 arg_private_network
= true;
913 arg_settings_mask
|= SETTING_NETWORK
;
916 case ARG_NETWORK_INTERFACE
:
917 if (!ifname_valid(optarg
))
918 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
919 "Network interface name not valid: %s", optarg
);
921 r
= test_network_interface_initialized(optarg
);
925 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
928 arg_private_network
= true;
929 arg_settings_mask
|= SETTING_NETWORK
;
932 case ARG_NETWORK_MACVLAN
:
934 if (!ifname_valid(optarg
))
935 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
936 "MACVLAN network interface name not valid: %s", optarg
);
938 r
= test_network_interface_initialized(optarg
);
942 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
945 arg_private_network
= true;
946 arg_settings_mask
|= SETTING_NETWORK
;
949 case ARG_NETWORK_IPVLAN
:
951 if (!ifname_valid(optarg
))
952 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
953 "IPVLAN network interface name not valid: %s", optarg
);
955 r
= test_network_interface_initialized(optarg
);
959 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
963 case ARG_PRIVATE_NETWORK
:
964 arg_private_network
= true;
965 arg_settings_mask
|= SETTING_NETWORK
;
968 case ARG_NETWORK_NAMESPACE_PATH
:
969 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
973 arg_settings_mask
|= SETTING_NETWORK
;
977 if (arg_start_mode
== START_PID2
)
978 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
979 "--boot and --as-pid2 may not be combined.");
981 arg_start_mode
= START_BOOT
;
982 arg_settings_mask
|= SETTING_START_MODE
;
986 if (arg_start_mode
== START_BOOT
)
987 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
988 "--boot and --as-pid2 may not be combined.");
990 arg_start_mode
= START_PID2
;
991 arg_settings_mask
|= SETTING_START_MODE
;
995 r
= sd_id128_from_string(optarg
, &arg_uuid
);
997 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
999 if (sd_id128_is_null(arg_uuid
))
1000 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1001 "Machine UUID may not be all zeroes.");
1003 arg_settings_mask
|= SETTING_MACHINE_ID
;
1007 _cleanup_free_
char *mangled
= NULL
;
1009 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
1013 free_and_replace(arg_slice
, mangled
);
1014 arg_settings_mask
|= SETTING_SLICE
;
1019 if (isempty(optarg
))
1020 arg_machine
= mfree(arg_machine
);
1022 if (!hostname_is_valid(optarg
, 0))
1023 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1024 "Invalid machine name: %s", optarg
);
1026 r
= free_and_strdup(&arg_machine
, optarg
);
1033 if (isempty(optarg
))
1034 arg_hostname
= mfree(arg_hostname
);
1036 if (!hostname_is_valid(optarg
, 0))
1037 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1038 "Invalid hostname: %s", optarg
);
1040 r
= free_and_strdup(&arg_hostname
, optarg
);
1045 arg_settings_mask
|= SETTING_HOSTNAME
;
1049 arg_selinux_context
= optarg
;
1053 arg_selinux_apifs_context
= optarg
;
1057 arg_read_only
= true;
1058 arg_settings_mask
|= SETTING_READ_ONLY
;
1061 case ARG_AMBIENT_CAPABILITY
: {
1063 r
= parse_capability_spec(optarg
, &m
);
1066 arg_caps_ambient
|= m
;
1067 arg_settings_mask
|= SETTING_CAPABILITY
;
1070 case ARG_CAPABILITY
:
1071 case ARG_DROP_CAPABILITY
: {
1073 r
= parse_capability_spec(optarg
, &m
);
1077 if (c
== ARG_CAPABILITY
)
1081 arg_settings_mask
|= SETTING_CAPABILITY
;
1084 case ARG_NO_NEW_PRIVILEGES
:
1085 r
= parse_boolean(optarg
);
1087 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1089 arg_no_new_privileges
= r
;
1090 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1094 arg_link_journal
= LINK_GUEST
;
1095 arg_link_journal_try
= true;
1096 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1099 case ARG_LINK_JOURNAL
:
1100 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1102 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1104 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1109 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1111 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1113 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1117 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1119 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1121 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1125 case ARG_OVERLAY_RO
:
1126 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1127 if (r
== -EADDRNOTAVAIL
)
1128 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1130 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1132 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1135 case ARG_INACCESSIBLE
:
1136 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1138 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1140 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1144 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1146 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1148 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1155 case ARG_SHARE_SYSTEM
:
1156 /* We don't officially support this anymore, except for compat reasons. People should use the
1157 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1158 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1159 arg_clone_ns_flags
= 0;
1163 r
= parse_boolean(optarg
);
1165 log_error("Failed to parse --register= argument: %s", optarg
);
1173 arg_keep_unit
= true;
1176 case ARG_PERSONALITY
:
1178 arg_personality
= personality_from_string(optarg
);
1179 if (arg_personality
== PERSONALITY_INVALID
)
1180 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1181 "Unknown or unsupported personality '%s'.", optarg
);
1183 arg_settings_mask
|= SETTING_PERSONALITY
;
1189 arg_volatile_mode
= VOLATILE_YES
;
1190 else if (streq(optarg
, "help")) {
1191 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1196 m
= volatile_mode_from_string(optarg
);
1198 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1199 "Failed to parse --volatile= argument: %s", optarg
);
1201 arg_volatile_mode
= m
;
1204 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1208 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1210 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1212 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1214 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1218 if (strv_extend(&arg_property
, optarg
) < 0)
1223 case ARG_PRIVATE_USERS
: {
1228 else if (!in_charset(optarg
, DIGITS
))
1229 /* do *not* parse numbers as booleans */
1230 boolean
= parse_boolean(optarg
);
1235 /* no: User namespacing off */
1236 arg_userns_mode
= USER_NAMESPACE_NO
;
1237 arg_uid_shift
= UID_INVALID
;
1238 arg_uid_range
= UINT32_C(0x10000);
1239 } else if (boolean
> 0) {
1240 /* yes: User namespacing on, UID range is read from root dir */
1241 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1242 arg_uid_shift
= UID_INVALID
;
1243 arg_uid_range
= UINT32_C(0x10000);
1244 } else if (streq(optarg
, "pick")) {
1245 /* pick: User namespacing on, UID range is picked randomly */
1246 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1247 * implied by USER_NAMESPACE_PICK
1249 arg_uid_shift
= UID_INVALID
;
1250 arg_uid_range
= UINT32_C(0x10000);
1252 } else if (streq(optarg
, "identity")) {
1253 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1254 * itself, i.e. we don't actually map anything, but do take benefit of
1255 * isolation of capability sets. */
1256 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1258 arg_uid_range
= UINT32_C(0x10000);
1260 _cleanup_free_
char *buffer
= NULL
;
1261 const char *range
, *shift
;
1263 /* anything else: User namespacing on, UID range is explicitly configured */
1265 range
= strchr(optarg
, ':');
1267 buffer
= strndup(optarg
, range
- optarg
);
1273 r
= safe_atou32(range
, &arg_uid_range
);
1275 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1279 r
= parse_uid(shift
, &arg_uid_shift
);
1281 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1283 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1285 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1286 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1289 arg_settings_mask
|= SETTING_USERNS
;
1294 if (userns_supported()) {
1295 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1296 * implied by USER_NAMESPACE_PICK
1298 arg_uid_shift
= UID_INVALID
;
1299 arg_uid_range
= UINT32_C(0x10000);
1301 arg_settings_mask
|= SETTING_USERNS
;
1306 case ARG_PRIVATE_USERS_CHOWN
:
1307 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1309 arg_settings_mask
|= SETTING_USERNS
;
1312 case ARG_PRIVATE_USERS_OWNERSHIP
:
1313 if (streq(optarg
, "help")) {
1314 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1318 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1319 if (arg_userns_ownership
< 0)
1320 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1322 arg_settings_mask
|= SETTING_USERNS
;
1325 case ARG_KILL_SIGNAL
:
1326 if (streq(optarg
, "help")) {
1327 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1331 arg_kill_signal
= signal_from_string(optarg
);
1332 if (arg_kill_signal
< 0)
1333 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1335 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1340 /* no → do not read files
1341 * yes → read files, do not override cmdline, trust only subset
1342 * override → read files, override cmdline, trust only subset
1343 * trusted → read files, do not override cmdline, trust all
1346 r
= parse_boolean(optarg
);
1348 if (streq(optarg
, "trusted")) {
1349 mask_all_settings
= false;
1350 mask_no_settings
= false;
1351 arg_settings_trusted
= true;
1353 } else if (streq(optarg
, "override")) {
1354 mask_all_settings
= false;
1355 mask_no_settings
= true;
1356 arg_settings_trusted
= -1;
1358 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1361 mask_all_settings
= false;
1362 mask_no_settings
= false;
1363 arg_settings_trusted
= -1;
1366 mask_all_settings
= true;
1367 mask_no_settings
= false;
1368 arg_settings_trusted
= false;
1374 if (!path_is_absolute(optarg
))
1375 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1376 "Working directory %s is not an absolute path.", optarg
);
1378 r
= free_and_strdup(&arg_chdir
, optarg
);
1382 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1385 case ARG_PIVOT_ROOT
:
1386 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1388 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1390 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1393 case ARG_NOTIFY_READY
:
1394 r
= parse_boolean(optarg
);
1396 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1397 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1398 arg_notify_ready
= r
;
1399 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1402 case ARG_ROOT_HASH
: {
1403 _cleanup_free_
void *k
= NULL
;
1406 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1408 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1409 if (l
< sizeof(sd_id128_t
))
1410 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1412 free_and_replace(arg_verity_settings
.root_hash
, k
);
1413 arg_verity_settings
.root_hash_size
= l
;
1417 case ARG_ROOT_HASH_SIG
: {
1422 if ((value
= startswith(optarg
, "base64:"))) {
1423 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1425 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1428 r
= read_full_file(optarg
, (char**) &p
, &l
);
1430 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1433 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1434 arg_verity_settings
.root_hash_sig_size
= l
;
1438 case ARG_VERITY_DATA
:
1439 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1444 case ARG_SYSTEM_CALL_FILTER
: {
1448 negative
= optarg
[0] == '~';
1449 items
= negative
? optarg
+ 1 : optarg
;
1452 _cleanup_free_
char *word
= NULL
;
1454 r
= extract_first_word(&items
, &word
, NULL
, 0);
1460 return log_error_errno(r
, "Failed to parse system call filter: %m");
1463 r
= strv_extend(&arg_syscall_deny_list
, word
);
1465 r
= strv_extend(&arg_syscall_allow_list
, word
);
1470 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1476 _cleanup_free_
char *name
= NULL
;
1479 if (streq(optarg
, "help")) {
1480 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1484 eq
= strchr(optarg
, '=');
1486 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1487 "--rlimit= expects an '=' assignment.");
1489 name
= strndup(optarg
, eq
- optarg
);
1493 rl
= rlimit_from_string_harder(name
);
1495 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1497 if (!arg_rlimit
[rl
]) {
1498 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1499 if (!arg_rlimit
[rl
])
1503 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1505 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1507 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1511 case ARG_OOM_SCORE_ADJUST
:
1512 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1514 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1516 arg_oom_score_adjust_set
= true;
1517 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1520 case ARG_CPU_AFFINITY
: {
1523 r
= parse_cpu_set(optarg
, &cpuset
);
1525 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1527 cpu_set_reset(&arg_cpu_set
);
1528 arg_cpu_set
= cpuset
;
1529 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1533 case ARG_RESOLV_CONF
:
1534 if (streq(optarg
, "help")) {
1535 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1539 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1540 if (arg_resolv_conf
< 0)
1541 return log_error_errno(arg_resolv_conf
,
1542 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1544 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1548 if (streq(optarg
, "help")) {
1549 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1553 arg_timezone
= timezone_mode_from_string(optarg
);
1554 if (arg_timezone
< 0)
1555 return log_error_errno(arg_timezone
,
1556 "Failed to parse /etc/localtime mode: %s", optarg
);
1558 arg_settings_mask
|= SETTING_TIMEZONE
;
1562 r
= handle_arg_console(optarg
);
1569 r
= handle_arg_console("pipe");
1575 arg_pager_flags
|= PAGER_DISABLE
;
1578 case ARG_SET_CREDENTIAL
: {
1579 _cleanup_free_
char *word
= NULL
, *data
= NULL
;
1580 const char *p
= optarg
;
1584 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1588 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1590 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1592 if (!credential_name_valid(word
))
1593 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1595 for (size_t i
= 0; i
< arg_n_credentials
; i
++)
1596 if (streq(arg_credentials
[i
].id
, word
))
1597 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1599 l
= cunescape(p
, UNESCAPE_ACCEPT_NUL
, &data
);
1601 return log_error_errno(l
, "Failed to unescape credential data: %s", p
);
1603 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1607 a
[arg_n_credentials
++] = (Credential
) {
1608 .id
= TAKE_PTR(word
),
1609 .data
= TAKE_PTR(data
),
1613 arg_credentials
= a
;
1615 arg_settings_mask
|= SETTING_CREDENTIALS
;
1619 case ARG_LOAD_CREDENTIAL
: {
1620 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
1621 _cleanup_(erase_and_freep
) char *data
= NULL
;
1622 _cleanup_free_
char *word
= NULL
, *j
= NULL
;
1623 const char *p
= optarg
;
1627 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1631 return log_error_errno(r
, "Failed to parse --load-credential= parameter: %m");
1633 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --load-credential=: %s", optarg
);
1635 if (!credential_name_valid(word
))
1636 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1638 for (i
= 0; i
< arg_n_credentials
; i
++)
1639 if (streq(arg_credentials
[i
].id
, word
))
1640 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1642 if (path_is_absolute(p
))
1643 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
1647 r
= get_credentials_dir(&e
);
1649 return log_error_errno(r
, "Credential not available (no credentials passed at all): %s", word
);
1651 j
= path_join(e
, p
);
1656 r
= read_full_file_full(AT_FDCWD
, j
?: p
, UINT64_MAX
, SIZE_MAX
,
1661 return log_error_errno(r
, "Failed to read credential '%s': %m", j
?: p
);
1663 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1667 a
[arg_n_credentials
++] = (Credential
) {
1668 .id
= TAKE_PTR(word
),
1669 .data
= TAKE_PTR(data
),
1673 arg_credentials
= a
;
1675 arg_settings_mask
|= SETTING_CREDENTIALS
;
1680 if (!valid_user_group_name(optarg
, 0))
1681 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1683 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1686 arg_settings_mask
|= SETTING_BIND_USER
;
1689 case ARG_SUPPRESS_SYNC
:
1690 r
= parse_boolean_argument("--suppress-sync=", optarg
, &arg_suppress_sync
);
1694 arg_settings_mask
|= SETTING_SUPPRESS_SYNC
;
1701 assert_not_reached();
1704 if (argc
> optind
) {
1705 strv_free(arg_parameters
);
1706 arg_parameters
= strv_copy(argv
+ optind
);
1707 if (!arg_parameters
)
1710 arg_settings_mask
|= SETTING_START_MODE
;
1713 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1714 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1715 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1716 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1718 arg_directory
= TAKE_PTR(arg_template
);
1720 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1722 /* Make sure to parse environment before we reset the settings mask below */
1723 r
= parse_environment();
1727 /* Load all settings from .nspawn files */
1728 if (mask_no_settings
)
1729 arg_settings_mask
= 0;
1731 /* Don't load any settings from .nspawn files */
1732 if (mask_all_settings
)
1733 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1738 static int verify_arguments(void) {
1741 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1742 /* If we are running the stub init in the container, we don't need to look at what the init
1743 * in the container supports, because we are not using it. Let's immediately pick the right
1744 * setting based on the host system configuration.
1746 * We only do this, if the user didn't use an environment variable to override the detection.
1749 r
= cg_all_unified();
1751 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1753 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1754 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1755 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1757 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1760 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1761 arg_mount_settings
|= MOUNT_USE_USERNS
;
1763 if (arg_private_network
)
1764 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1766 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1767 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1768 arg_register
= false;
1769 if (arg_start_mode
!= START_PID1
)
1770 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1773 if (arg_userns_ownership
< 0)
1774 arg_userns_ownership
=
1775 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1776 USER_NAMESPACE_OWNERSHIP_OFF
;
1778 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1779 arg_kill_signal
= SIGRTMIN
+3;
1781 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 */
1782 arg_read_only
= true;
1784 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1785 arg_read_only
= true;
1787 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1788 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1789 * The latter is not technically a user session, but we don't need to labour the point. */
1790 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1792 if (arg_directory
&& arg_image
)
1793 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1795 if (arg_template
&& arg_image
)
1796 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1798 if (arg_template
&& !(arg_directory
|| arg_machine
))
1799 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1801 if (arg_ephemeral
&& arg_template
)
1802 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1804 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1805 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1807 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1808 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1810 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1811 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1812 "--read-only and --private-users-ownership=chown may not be combined.");
1814 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1815 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1816 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1817 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1818 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1820 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1821 * we need to error out, to avoid conflicts between different network options. */
1822 if (arg_network_namespace_path
&&
1823 (arg_network_interfaces
|| arg_network_macvlan
||
1824 arg_network_ipvlan
|| arg_network_veth_extra
||
1825 arg_network_bridge
|| arg_network_zone
||
1827 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1829 if (arg_network_bridge
&& arg_network_zone
)
1830 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1831 "--network-bridge= and --network-zone= may not be combined.");
1833 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1834 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1836 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1837 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1839 if (arg_expose_ports
&& !arg_private_network
)
1840 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1842 if (arg_caps_ambient
) {
1843 if (arg_caps_ambient
== UINT64_MAX
)
1844 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1846 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1847 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1849 if (arg_start_mode
== START_BOOT
)
1850 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1853 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1854 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1856 /* Drop duplicate --bind-user= entries */
1857 strv_uniq(arg_bind_user
);
1859 r
= custom_mount_check_all();
1866 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1869 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1872 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1875 if (uid
!= UID_INVALID
) {
1876 uid
+= arg_uid_shift
;
1878 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1882 if (gid
!= GID_INVALID
) {
1883 gid
+= (gid_t
) arg_uid_shift
;
1885 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1889 return RET_NERRNO(lchown(p
, uid
, gid
));
1892 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1896 q
= prefix_roota(root
, path
);
1897 r
= RET_NERRNO(mkdir(q
, mode
));
1903 return userns_lchown(q
, uid
, gid
);
1906 static const char *timezone_from_path(const char *path
) {
1907 return PATH_STARTSWITH_SET(
1909 "../usr/share/zoneinfo/",
1910 "/usr/share/zoneinfo/");
1913 static bool etc_writable(void) {
1914 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1917 static int setup_timezone(const char *dest
) {
1918 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1919 const char *where
, *check
;
1925 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1926 r
= readlink_malloc("/etc/localtime", &p
);
1927 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1928 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1929 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1930 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1932 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1933 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1937 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1940 } else if (arg_timezone
== TIMEZONE_AUTO
)
1941 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1947 if (m
== TIMEZONE_OFF
)
1950 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1952 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1956 where
= strjoina(etc
, "/localtime");
1960 case TIMEZONE_DELETE
:
1961 if (unlink(where
) < 0)
1962 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1966 case TIMEZONE_SYMLINK
: {
1967 _cleanup_free_
char *q
= NULL
;
1968 const char *z
, *what
;
1970 z
= timezone_from_path(p
);
1972 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1976 r
= readlink_malloc(where
, &q
);
1977 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1978 return 0; /* Already pointing to the right place? Then do nothing .. */
1980 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1981 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1983 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1985 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1986 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1987 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1991 what
= strjoina("../usr/share/zoneinfo/", z
);
1992 if (symlink(what
, where
) < 0) {
1993 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1994 errno
, "Failed to correct timezone of container, ignoring: %m");
2004 case TIMEZONE_BIND
: {
2005 _cleanup_free_
char *resolved
= NULL
;
2008 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2010 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
2014 if (found
== 0) /* missing? */
2015 (void) touch(resolved
);
2017 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
2019 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2025 /* If mounting failed, try to copy */
2026 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2028 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2029 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
2036 assert_not_reached();
2039 /* Fix permissions of the symlink or file copy we just created */
2040 r
= userns_lchown(where
, 0, 0);
2042 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
2047 static int have_resolv_conf(const char *path
) {
2050 if (access(path
, F_OK
) < 0) {
2051 if (errno
== ENOENT
)
2054 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
2060 static int resolved_listening(void) {
2061 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2062 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
2063 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
2066 /* Check if resolved is listening */
2068 r
= sd_bus_open_system(&bus
);
2070 return log_debug_errno(r
, "Failed to open system bus: %m");
2072 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2074 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2078 r
= sd_bus_get_property_string(bus
,
2079 "org.freedesktop.resolve1",
2080 "/org/freedesktop/resolve1",
2081 "org.freedesktop.resolve1.Manager",
2084 &dns_stub_listener_mode
);
2086 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2088 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2091 static int setup_resolv_conf(const char *dest
) {
2092 _cleanup_free_
char *etc
= NULL
;
2093 const char *where
, *what
;
2099 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2100 if (arg_private_network
)
2101 m
= RESOLV_CONF_OFF
;
2102 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2103 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2104 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2105 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2107 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2110 m
= arg_resolv_conf
;
2112 if (m
== RESOLV_CONF_OFF
)
2115 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2117 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2121 where
= strjoina(etc
, "/resolv.conf");
2123 if (m
== RESOLV_CONF_DELETE
) {
2124 if (unlink(where
) < 0)
2125 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2130 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2131 what
= PRIVATE_STATIC_RESOLV_CONF
;
2132 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2133 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2134 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2135 what
= PRIVATE_STUB_RESOLV_CONF
;
2137 what
= "/etc/resolv.conf";
2139 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2140 _cleanup_free_
char *resolved
= NULL
;
2143 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2145 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2149 if (found
== 0) /* missing? */
2150 (void) touch(resolved
);
2152 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2154 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2156 /* If that didn't work, let's copy the file */
2159 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2160 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2162 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
2164 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2165 * resolved or something similar runs inside and the symlink points there.
2167 * If the disk image is read-only, there's also no point in complaining.
2169 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2170 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2171 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2175 r
= userns_lchown(where
, 0, 0);
2177 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2182 static int setup_boot_id(void) {
2183 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2184 _cleanup_free_
char *path
= NULL
;
2185 sd_id128_t rnd
= SD_ID128_NULL
;
2189 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2191 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2193 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2195 r
= sd_id128_randomize(&rnd
);
2197 return log_error_errno(r
, "Failed to generate random boot id: %m");
2199 r
= id128_write(path
, ID128_FORMAT_UUID
, rnd
);
2201 return log_error_errno(r
, "Failed to write boot id: %m");
2203 from
= TAKE_PTR(path
);
2204 to
= "/proc/sys/kernel/random/boot_id";
2206 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2210 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2213 static int copy_devnodes(const char *dest
) {
2214 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 mount_tunnel_dig(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
, NSPAWN_MOUNT_TUNNEL
, 0600, 0, 0);
2795 return log_error_errno(r
, "Failed to create "NSPAWN_MOUNT_TUNNEL
": %m");
2797 q
= prefix_roota(root
, NSPAWN_MOUNT_TUNNEL
);
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
);
2809 static int mount_tunnel_open(void) {
2812 r
= mount_follow_verbose(LOG_ERR
, NULL
, NSPAWN_MOUNT_TUNNEL
, NULL
, MS_SLAVE
, NULL
);
2819 static int setup_machine_id(const char *directory
) {
2820 const char *etc_machine_id
;
2824 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2825 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2826 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2827 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2828 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2829 * container behaves nicely). */
2831 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2833 r
= id128_read(etc_machine_id
, ID128_FORMAT_PLAIN
, &id
);
2835 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
, -ENOPKG
)) /* If the file is missing, empty, or uninitialized, we don't mind */
2836 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2838 if (sd_id128_is_null(arg_uuid
)) {
2839 r
= sd_id128_randomize(&arg_uuid
);
2841 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2844 if (sd_id128_is_null(id
))
2845 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2846 "Machine ID in container image is zero, refusing.");
2854 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2859 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2862 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2863 if (r
== -EOPNOTSUPP
)
2864 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.");
2866 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2868 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2870 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2872 log_debug("Patched directory tree to match UID/GID range.");
2879 * < 0 : wait_for_terminate() failed to get the state of the
2880 * container, the container was terminated by a signal, or
2881 * failed for an unknown reason. No change is made to the
2882 * container argument.
2883 * > 0 : The program executed in the container terminated with an
2884 * error. The exit code of the program executed in the
2885 * container is returned. The container argument has been set
2886 * to CONTAINER_TERMINATED.
2887 * 0 : The container is being rebooted, has been shut down or exited
2888 * successfully. The container argument has been set to either
2889 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2891 * That is, success is indicated by a return value of zero, and an
2892 * error is indicated by a non-zero value.
2894 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2898 r
= wait_for_terminate(pid
, &status
);
2900 return log_warning_errno(r
, "Failed to wait for container: %m");
2902 switch (status
.si_code
) {
2905 if (status
.si_status
== 0)
2906 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2908 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2910 *container
= CONTAINER_TERMINATED
;
2911 return status
.si_status
;
2914 if (status
.si_status
== SIGINT
) {
2915 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2916 *container
= CONTAINER_TERMINATED
;
2919 } else if (status
.si_status
== SIGHUP
) {
2920 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2921 *container
= CONTAINER_REBOOTED
;
2927 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2928 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2931 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2932 "Container %s failed due to unknown reason.", arg_machine
);
2936 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2939 pid
= PTR_TO_PID(userdata
);
2941 if (kill(pid
, arg_kill_signal
) >= 0) {
2942 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2943 sd_event_source_set_userdata(s
, NULL
);
2948 sd_event_exit(sd_event_source_get_event(s
), 0);
2952 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2958 pid
= PTR_TO_PID(userdata
);
2963 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2964 return log_error_errno(errno
, "Failed to waitid(): %m");
2965 if (si
.si_pid
== 0) /* No pending children. */
2967 if (si
.si_pid
== pid
) {
2968 /* The main process we care for has exited. Return from
2969 * signal handler but leave the zombie. */
2970 sd_event_exit(sd_event_source_get_event(s
), 0);
2974 /* Reap all other children. */
2975 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2981 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2986 pid
= PTR_TO_PID(userdata
);
2988 if (arg_kill_signal
> 0) {
2989 log_info("Container termination requested. Attempting to halt container.");
2990 (void) kill(pid
, arg_kill_signal
);
2992 log_info("Container termination requested. Exiting.");
2993 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2999 static int determine_names(void) {
3002 if (arg_template
&& !arg_directory
&& arg_machine
) {
3004 /* If --template= was specified then we should not
3005 * search for a machine, but instead create a new one
3006 * in /var/lib/machine. */
3008 arg_directory
= path_join("/var/lib/machines", arg_machine
);
3013 if (!arg_image
&& !arg_directory
) {
3015 _cleanup_(image_unrefp
) Image
*i
= NULL
;
3017 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
3019 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
3021 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
3023 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
3024 r
= free_and_strdup(&arg_image
, i
->path
);
3026 r
= free_and_strdup(&arg_directory
, i
->path
);
3031 arg_read_only
= arg_read_only
|| i
->read_only
;
3033 r
= safe_getcwd(&arg_directory
);
3035 return log_error_errno(r
, "Failed to determine current directory: %m");
3038 if (!arg_directory
&& !arg_image
)
3039 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3043 if (arg_directory
&& path_equal(arg_directory
, "/"))
3044 arg_machine
= gethostname_malloc();
3045 else if (arg_image
) {
3048 arg_machine
= strdup(basename(arg_image
));
3050 /* Truncate suffix if there is one */
3051 e
= endswith(arg_machine
, ".raw");
3055 arg_machine
= strdup(basename(arg_directory
));
3059 hostname_cleanup(arg_machine
);
3060 if (!hostname_is_valid(arg_machine
, 0))
3061 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3063 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3064 * to match fixed config file names. */
3065 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3066 if (!arg_settings_filename
)
3069 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3070 * instances at once without manually having to specify -M each time. */
3072 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3075 arg_settings_filename
= strjoin(arg_machine
, ".nspawn");
3076 if (!arg_settings_filename
)
3083 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
3092 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
3094 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3096 return free_and_replace(*p
, chased
);
3099 static int determine_uid_shift(const char *directory
) {
3101 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3106 if (arg_uid_shift
== UID_INVALID
) {
3109 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3111 if (stat(directory
, &st
) < 0)
3112 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3114 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3116 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3117 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3118 "UID and GID base of %s don't match.", directory
);
3120 arg_uid_range
= UINT32_C(0x10000);
3122 if (arg_uid_shift
!= 0) {
3123 /* If the image is shifted already, then we'll fall back to classic chowning, for
3124 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3126 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3127 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3128 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3129 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3130 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3131 "UID base of %s is not zero, UID mapping not supported.", directory
);
3135 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3136 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3141 static unsigned long effective_clone_ns_flags(void) {
3142 unsigned long flags
= arg_clone_ns_flags
;
3144 if (arg_private_network
)
3145 flags
|= CLONE_NEWNET
;
3147 flags
|= CLONE_NEWCGROUP
;
3148 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3149 flags
|= CLONE_NEWUSER
;
3154 static int patch_sysctl(void) {
3156 /* This table is inspired by runc's sysctl() function */
3157 static const struct {
3160 unsigned long clone_flags
;
3162 { "kernel.hostname", false, CLONE_NEWUTS
},
3163 { "kernel.domainname", false, CLONE_NEWUTS
},
3164 { "kernel.msgmax", false, CLONE_NEWIPC
},
3165 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3166 { "kernel.msgmni", false, CLONE_NEWIPC
},
3167 { "kernel.sem", false, CLONE_NEWIPC
},
3168 { "kernel.shmall", false, CLONE_NEWIPC
},
3169 { "kernel.shmmax", false, CLONE_NEWIPC
},
3170 { "kernel.shmmni", false, CLONE_NEWIPC
},
3171 { "fs.mqueue.", true, CLONE_NEWIPC
},
3172 { "net.", true, CLONE_NEWNET
},
3175 unsigned long flags
;
3178 flags
= effective_clone_ns_flags();
3180 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3184 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3186 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3189 if (safe_sysctl
[i
].prefix
)
3190 good
= startswith(*k
, safe_sysctl
[i
].key
);
3192 good
= streq(*k
, safe_sysctl
[i
].key
);
3199 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3201 r
= sysctl_write(*k
, *v
);
3203 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3209 static int inner_child(
3211 const char *directory
,
3215 int master_pty_socket
,
3217 char **os_release_pairs
) {
3219 _cleanup_free_
char *home
= NULL
;
3222 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3223 NULL
, /* container */
3228 NULL
, /* container_uuid */
3229 NULL
, /* LISTEN_FDS */
3230 NULL
, /* LISTEN_PID */
3231 NULL
, /* NOTIFY_SOCKET */
3232 NULL
, /* CREDENTIALS_DIRECTORY */
3236 const char *exec_target
;
3237 _cleanup_strv_free_
char **env_use
= NULL
;
3238 int r
, which_failed
;
3240 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3241 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3242 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3243 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3244 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3245 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3248 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3249 * unshare(). See below. */
3253 assert(kmsg_socket
>= 0);
3255 log_debug("Inner child is initializing.");
3257 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3258 /* Tell the parent, that it now can write the UID map. */
3259 (void) barrier_place(barrier
); /* #1 */
3261 /* Wait until the parent wrote the UID map */
3262 if (!barrier_place_and_sync(barrier
)) /* #2 */
3263 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3265 /* Become the new root user inside our namespace */
3266 r
= reset_uid_gid();
3268 return log_error_errno(r
, "Couldn't become new root: %m");
3270 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3271 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3272 * propagation, but simply create new peer groups for all our mounts). */
3273 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3279 arg_mount_settings
| MOUNT_IN_USERNS
,
3281 arg_selinux_apifs_context
);
3285 if (!arg_network_namespace_path
&& arg_private_network
) {
3286 r
= unshare(CLONE_NEWNET
);
3288 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3290 /* Tell the parent that it can setup network interfaces. */
3291 (void) barrier_place(barrier
); /* #3 */
3294 r
= mount_sysfs(NULL
, arg_mount_settings
);
3298 /* Wait until we are cgroup-ified, so that we
3299 * can mount the right cgroup path writable */
3300 if (!barrier_place_and_sync(barrier
)) /* #4 */
3301 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3302 "Parent died too early");
3305 r
= unshare(CLONE_NEWCGROUP
);
3307 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3310 arg_unified_cgroup_hierarchy
,
3311 arg_userns_mode
!= USER_NAMESPACE_NO
,
3314 arg_selinux_apifs_context
,
3317 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3321 r
= setup_boot_id();
3325 r
= setup_kmsg(kmsg_socket
);
3328 kmsg_socket
= safe_close(kmsg_socket
);
3333 arg_n_custom_mounts
,
3336 arg_selinux_apifs_context
,
3337 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3342 return log_error_errno(errno
, "setsid() failed: %m");
3344 if (arg_private_network
)
3345 (void) loopback_setup();
3347 if (arg_expose_ports
) {
3348 r
= expose_port_send_rtnl(rtnl_socket
);
3351 rtnl_socket
= safe_close(rtnl_socket
);
3354 if (arg_console_mode
!= CONSOLE_PIPE
) {
3355 _cleanup_close_
int master
= -1;
3356 _cleanup_free_
char *console
= NULL
;
3358 /* Allocate a pty and make it available as /dev/console. */
3359 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3361 return log_error_errno(master
, "Failed to allocate a pty: %m");
3363 r
= setup_dev_console(console
);
3365 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3367 r
= send_one_fd(master_pty_socket
, master
, 0);
3369 return log_error_errno(r
, "Failed to send master fd: %m");
3370 master_pty_socket
= safe_close(master_pty_socket
);
3372 r
= setup_stdio_as_dev_console();
3381 if (arg_oom_score_adjust_set
) {
3382 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3384 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3387 if (arg_cpu_set
.set
)
3388 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3389 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3391 (void) setup_hostname();
3393 if (arg_personality
!= PERSONALITY_INVALID
) {
3394 r
= safe_personality(arg_personality
);
3396 return log_error_errno(r
, "personality() failed: %m");
3397 } else if (secondary
) {
3398 r
= safe_personality(PER_LINUX32
);
3400 return log_error_errno(r
, "personality() failed: %m");
3403 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3405 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3410 if (is_seccomp_available()) {
3412 r
= seccomp_load(arg_seccomp
);
3413 if (ERRNO_IS_SECCOMP_FATAL(r
))
3414 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3416 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3421 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3426 if (arg_suppress_sync
) {
3428 r
= seccomp_suppress_sync();
3430 log_debug_errno(r
, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3432 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3437 if (arg_selinux_context
)
3438 if (setexeccon(arg_selinux_context
) < 0)
3439 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3442 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3443 * if we need to later on. */
3444 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3445 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3447 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3448 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3450 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3454 r
= drop_capabilities(getuid());
3456 return log_error_errno(r
, "Dropping capabilities failed: %m");
3458 if (arg_no_new_privileges
)
3459 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3460 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3462 /* LXC sets container=lxc, so follow the scheme here */
3463 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3465 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3469 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3470 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3473 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3474 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3475 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3478 assert(!sd_id128_is_null(arg_uuid
));
3480 if (asprintf(envp
+ n_env
++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3483 if (fdset_size(fds
) > 0) {
3484 r
= fdset_cloexec(fds
, false);
3486 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3488 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3489 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3492 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3495 if (arg_n_credentials
> 0) {
3496 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3502 if (arg_start_mode
!= START_BOOT
) {
3503 envp
[n_env
] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE
);
3509 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3513 /* Let the parent know that we are ready and
3514 * wait until the parent is ready with the
3516 if (!barrier_place_and_sync(barrier
)) /* #5 */
3517 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3520 if (chdir(arg_chdir
) < 0)
3521 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3523 if (arg_start_mode
== START_PID2
) {
3524 r
= stub_pid1(arg_uuid
);
3529 if (arg_console_mode
!= CONSOLE_PIPE
) {
3530 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3531 * are configured for that. Acquire it as controlling tty. */
3532 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3533 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3536 log_debug("Inner child completed, invoking payload.");
3538 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3539 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3540 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3542 log_set_open_when_needed(true);
3544 (void) fdset_close_others(fds
);
3546 if (arg_start_mode
== START_BOOT
) {
3550 /* Automatically search for the init system */
3552 m
= strv_length(arg_parameters
);
3553 a
= newa(char*, m
+ 2);
3554 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3557 FOREACH_STRING(init
,
3558 "/usr/lib/systemd/systemd",
3559 "/lib/systemd/systemd",
3561 a
[0] = (char*) init
;
3562 execve(a
[0], a
, env_use
);
3565 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3566 } else if (!strv_isempty(arg_parameters
)) {
3567 const char *dollar_path
;
3569 exec_target
= arg_parameters
[0];
3571 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3573 dollar_path
= strv_env_get(env_use
, "PATH");
3575 if (setenv("PATH", dollar_path
, 1) < 0)
3576 return log_error_errno(errno
, "Failed to update $PATH: %m");
3579 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3582 /* If we cannot change the directory, we'll end up in /, that is expected. */
3583 (void) chdir(home
?: "/root");
3585 execle(DEFAULT_USER_SHELL
, "-" DEFAULT_USER_SHELL_NAME
, NULL
, env_use
);
3586 if (!streq(DEFAULT_USER_SHELL
, "/bin/bash"))
3587 execle("/bin/bash", "-bash", NULL
, env_use
);
3588 if (!streq(DEFAULT_USER_SHELL
, "/bin/sh"))
3589 execle("/bin/sh", "-sh", NULL
, env_use
);
3591 exec_target
= DEFAULT_USER_SHELL
", /bin/bash, /bin/sh";
3594 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3597 static int setup_notify_child(void) {
3598 _cleanup_close_
int fd
= -1;
3599 static const union sockaddr_union sa
= {
3600 .un
.sun_family
= AF_UNIX
,
3601 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3605 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3607 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3609 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3610 (void) sockaddr_un_unlink(&sa
.un
);
3612 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3614 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3616 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3618 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3620 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3622 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3627 static int outer_child(
3629 const char *directory
,
3630 DissectedImage
*dissected_image
,
3635 int master_pty_socket
,
3636 int unified_cgroup_hierarchy_socket
,
3640 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3641 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3642 _cleanup_close_
int fd
= -1, mntns_fd
= -EBADF
;
3649 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3650 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3651 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3652 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3653 * forked off it, and it exits. */
3657 assert(fd_socket
>= 0);
3658 assert(master_pty_socket
>= 0);
3659 assert(kmsg_socket
>= 0);
3661 log_debug("Outer child is initializing.");
3663 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3665 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3667 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3668 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3670 r
= reset_audit_loginuid();
3674 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3675 * mounts to the real root. */
3676 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3680 if (dissected_image
) {
3681 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3682 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3683 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3684 * right place right away. This makes sure ESP partitions and userns are compatible. */
3686 r
= dissected_image_mount_and_warn(
3691 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3692 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3693 DISSECT_IMAGE_USR_NO_ROOT
|
3694 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3695 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3700 r
= determine_uid_shift(directory
);
3704 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3705 r
= namespace_open(0, NULL
, &mntns_fd
, NULL
, NULL
, NULL
);
3707 return log_error_errno(r
, "Failed to pin outer mount namespace: %m");
3709 l
= send_one_fd(fd_socket
, mntns_fd
, 0);
3711 return log_error_errno(l
, "Failed to send outer mount namespace fd: %m");
3712 mntns_fd
= safe_close(mntns_fd
);
3714 /* Let the parent know which UID shift we read from the image */
3715 l
= send(fd_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3717 return log_error_errno(errno
, "Failed to send UID shift: %m");
3718 if (l
!= sizeof(arg_uid_shift
))
3719 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3720 "Short write while sending UID shift.");
3722 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3723 /* When we are supposed to pick the UID shift, the parent will check now whether the
3724 * UID shift we just read from the image is available. If yes, it will send the UID
3725 * shift back to us, if not it will pick a different one, and send it back to us. */
3727 l
= recv(fd_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3729 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3730 if (l
!= sizeof(arg_uid_shift
))
3731 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3732 "Short read while receiving UID shift.");
3735 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3736 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3739 if (path_equal(directory
, "/")) {
3740 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3741 * place, so that we can make changes to its mount structure (for example, to implement
3742 * --volatile=) without this interfering with our ability to access files such as
3743 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3744 * (instead of a temporary directory, since we are living in our own mount namespace here
3745 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3746 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3748 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3752 directory
= "/run/systemd/nspawn-root";
3755 r
= setup_pivot_root(
3758 arg_pivot_root_old
);
3762 r
= setup_volatile_mode(
3766 arg_selinux_apifs_context
);
3770 r
= bind_user_prepare(
3775 &arg_custom_mounts
, &arg_n_custom_mounts
,
3776 &bind_user_context
);
3780 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3781 /* Send the user maps we determined to the parent, so that it installs it in our user
3782 * namespace UID map table */
3784 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3786 bind_user_context
->data
[i
].payload_user
->uid
,
3787 bind_user_context
->data
[i
].host_user
->uid
,
3788 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3789 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3792 l
= send(fd_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3794 return log_error_errno(errno
, "Failed to send user UID map: %m");
3795 if (l
!= sizeof(map
))
3796 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3797 "Short write while sending user UID map.");
3804 arg_n_custom_mounts
,
3807 arg_selinux_apifs_context
,
3812 /* Make sure we always have a mount that we can move to root later on. */
3813 r
= make_mount_point(directory
);
3817 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3818 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3819 arg_uid_shift
!= 0) {
3821 r
= remount_idmap(directory
, arg_uid_shift
, arg_uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
3822 if (r
== -EINVAL
|| ERRNO_IS_NOT_SUPPORTED(r
)) {
3823 /* This might fail because the kernel or file system doesn't support idmapping. We
3824 * can't really distinguish this nicely, nor do we have any guarantees about the
3825 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3826 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3827 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3828 "ID mapped mounts are apparently not available, sorry.");
3830 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3831 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3833 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3835 log_debug("ID mapped mounts available, making use of them.");
3840 if (dissected_image
) {
3841 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3842 r
= dissected_image_mount(
3847 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3848 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3849 DISSECT_IMAGE_USR_NO_ROOT
|
3850 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3851 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3853 return log_error_errno(r
, "File system check for image failed: %m");
3855 return log_error_errno(r
, "Failed to mount image file system: %m");
3858 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3859 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3861 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3865 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3867 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3868 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3869 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3870 "Short write while sending cgroup mode.");
3872 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
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
= mount_tunnel_dig(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 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3979 * mounts available in systemd services inside the container that create a new mount namespace. See
3980 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3981 * will inherit the shared propagation mode.
3983 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3984 * directory mount to root later on.
3985 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3987 r
= mount_switch_root(directory
, MOUNT_ATTR_PROPAGATION_SHARED
);
3989 return log_error_errno(r
, "Failed to move root directory: %m");
3991 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
3992 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
3994 r
= mount_tunnel_open();
3998 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3999 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4000 * requires that a fully visible instance is already present in the target mount
4001 * namespace. Mount one here so the inner child can mount its own instances. Later
4002 * we umount the temporary instances created here before we actually exec the
4003 * payload. Since the rootfs is shared the umount will propagate into the container.
4004 * Note, the inner child wouldn't be able to unmount the instances on its own since
4005 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4007 r
= pin_fully_visible_fs();
4012 fd
= setup_notify_child();
4016 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
4017 arg_clone_ns_flags
|
4018 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
4020 return log_error_errno(errno
, "Failed to fork inner child: %m");
4022 fd_socket
= safe_close(fd_socket
);
4024 /* The inner child has all namespaces that are requested, so that we all are owned by the
4025 * user if user namespaces are turned on. */
4027 if (arg_network_namespace_path
) {
4028 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
4030 return log_error_errno(r
, "Failed to join network namespace: %m");
4033 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
, os_release_pairs
);
4035 _exit(EXIT_FAILURE
);
4037 _exit(EXIT_SUCCESS
);
4040 l
= send(fd_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
4042 return log_error_errno(errno
, "Failed to send PID: %m");
4043 if (l
!= sizeof(pid
))
4044 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4045 "Short write while sending PID.");
4047 l
= send(fd_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
4049 return log_error_errno(errno
, "Failed to send machine ID: %m");
4050 if (l
!= sizeof(arg_uuid
))
4051 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
4052 "Short write while sending machine ID.");
4054 l
= send_one_fd(fd_socket
, fd
, 0);
4056 return log_error_errno(l
, "Failed to send notify fd: %m");
4058 fd_socket
= safe_close(fd_socket
);
4059 master_pty_socket
= safe_close(master_pty_socket
);
4060 kmsg_socket
= safe_close(kmsg_socket
);
4061 rtnl_socket
= safe_close(rtnl_socket
);
4062 netns_fd
= safe_close(netns_fd
);
4067 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
4068 bool tried_hashed
= false;
4069 unsigned n_tries
= 100;
4074 assert(ret_lock_file
);
4075 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4076 assert(arg_uid_range
== 0x10000U
);
4080 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4083 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4084 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4089 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4091 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4094 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4095 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4096 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4101 /* Make some superficial checks whether the range is currently known in the user database */
4102 if (getpwuid(candidate
))
4104 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4106 if (getgrgid(candidate
))
4108 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4111 *ret_lock_file
= lf
;
4112 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4117 if (arg_machine
&& !tried_hashed
) {
4118 /* Try to hash the base from the container name */
4120 static const uint8_t hash_key
[] = {
4121 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4122 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4125 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4127 tried_hashed
= true;
4129 random_bytes(&candidate
, sizeof(candidate
));
4131 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4132 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4136 static int add_one_uid_map(
4138 uid_t container_uid
,
4142 return strextendf(p
,
4143 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4144 container_uid
, host_uid
, range
);
4147 static int make_uid_map_string(
4148 const uid_t bind_user_uid
[],
4149 size_t n_bind_user_uid
,
4153 _cleanup_free_
char *s
= NULL
;
4154 uid_t previous_uid
= 0;
4157 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4158 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4161 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4162 * quadruplet, consisting of host and container UID + GID. */
4164 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4165 uid_t payload_uid
= bind_user_uid
[i
*4+offset
],
4166 host_uid
= bind_user_uid
[i
*4+offset
+1];
4168 assert(previous_uid
<= payload_uid
);
4169 assert(payload_uid
< arg_uid_range
);
4171 /* Add a range to close the gap to previous entry */
4172 if (payload_uid
> previous_uid
) {
4173 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4178 /* Map this specific user */
4179 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4183 previous_uid
= payload_uid
+ 1;
4186 /* And add a range to close the gap to finish the range */
4187 if (arg_uid_range
> previous_uid
) {
4188 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4199 static int setup_uid_map(
4201 const uid_t bind_user_uid
[],
4202 size_t n_bind_user_uid
) {
4204 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4205 _cleanup_free_
char *s
= NULL
;
4210 /* Build the UID map string */
4211 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4214 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4215 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4217 return log_error_errno(r
, "Failed to write UID map: %m");
4219 /* And now build the GID map string */
4221 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4224 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4225 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4227 return log_error_errno(r
, "Failed to write GID map: %m");
4232 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4233 char buf
[NOTIFY_BUFFER_MAX
+1];
4235 struct iovec iovec
= {
4237 .iov_len
= sizeof(buf
)-1,
4239 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4240 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4241 struct msghdr msghdr
= {
4244 .msg_control
= &control
,
4245 .msg_controllen
= sizeof(control
),
4247 struct ucred
*ucred
;
4249 pid_t inner_child_pid
;
4250 _cleanup_strv_free_
char **tags
= NULL
;
4255 inner_child_pid
= PTR_TO_PID(userdata
);
4257 if (revents
!= EPOLLIN
) {
4258 log_warning("Got unexpected poll event for notify fd.");
4262 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4264 if (ERRNO_IS_TRANSIENT(n
))
4267 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4270 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4273 cmsg_close_all(&msghdr
);
4275 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4276 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4277 log_debug("Received notify message without valid credentials. Ignoring.");
4281 if ((size_t) n
>= sizeof(buf
)) {
4282 log_warning("Received notify message exceeded maximum size. Ignoring.");
4287 tags
= strv_split(buf
, "\n\r");
4291 if (strv_contains(tags
, "READY=1")) {
4292 r
= sd_notify(false, "READY=1\n");
4294 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
4297 p
= strv_find_startswith(tags
, "STATUS=");
4299 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4304 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4307 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4309 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4311 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4316 static int merge_settings(Settings
*settings
, const char *path
) {
4322 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4323 * that this steals the fields of the Settings* structure, and hence modifies it. */
4325 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4326 settings
->start_mode
>= 0) {
4327 arg_start_mode
= settings
->start_mode
;
4328 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4331 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0 &&
4332 settings
->ephemeral
>= 0)
4333 arg_ephemeral
= settings
->ephemeral
;
4335 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4338 if (!arg_settings_trusted
)
4339 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4341 free_and_replace(arg_directory
, settings
->root
);
4344 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4345 settings
->pivot_root_new
) {
4346 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4347 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4350 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4351 settings
->working_directory
)
4352 free_and_replace(arg_chdir
, settings
->working_directory
);
4354 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4355 settings
->environment
)
4356 strv_free_and_replace(arg_setenv
, settings
->environment
);
4358 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4361 free_and_replace(arg_user
, settings
->user
);
4363 if (uid_is_valid(settings
->uid
))
4364 arg_uid
= settings
->uid
;
4365 if (gid_is_valid(settings
->gid
))
4366 arg_gid
= settings
->gid
;
4367 if (settings
->n_supplementary_gids
> 0) {
4368 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4369 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4373 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4374 uint64_t plus
, minus
;
4375 uint64_t network_minus
= 0;
4378 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4379 * Settings structure */
4381 plus
= settings
->capability
;
4382 minus
= settings
->drop_capability
;
4384 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4385 settings_network_configured(settings
)) {
4386 if (settings_private_network(settings
))
4387 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4389 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4392 if (!arg_settings_trusted
&& plus
!= 0) {
4393 if (settings
->capability
!= 0)
4394 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4396 arg_caps_retain
&= ~network_minus
;
4397 arg_caps_retain
|= plus
;
4400 arg_caps_retain
&= ~minus
;
4402 /* Copy the full capabilities over too */
4403 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4404 if (!arg_settings_trusted
)
4405 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4407 arg_full_capabilities
= settings
->full_capabilities
;
4410 ambient
= settings
->ambient_capability
;
4411 if (!arg_settings_trusted
&& ambient
!= 0)
4412 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4414 arg_caps_ambient
|= ambient
;
4417 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4418 settings
->kill_signal
> 0)
4419 arg_kill_signal
= settings
->kill_signal
;
4421 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4422 settings
->personality
!= PERSONALITY_INVALID
)
4423 arg_personality
= settings
->personality
;
4425 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4426 !sd_id128_is_null(settings
->machine_id
)) {
4428 if (!arg_settings_trusted
)
4429 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4431 arg_uuid
= settings
->machine_id
;
4434 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4435 settings
->read_only
>= 0)
4436 arg_read_only
= settings
->read_only
;
4438 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4439 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4440 arg_volatile_mode
= settings
->volatile_mode
;
4442 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4443 settings
->n_custom_mounts
> 0) {
4445 if (!arg_settings_trusted
)
4446 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4448 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4449 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4450 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4451 settings
->n_custom_mounts
= 0;
4455 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4456 settings_network_configured(settings
)) {
4458 if (!arg_settings_trusted
)
4459 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4461 arg_network_veth
= settings_network_veth(settings
);
4462 arg_private_network
= settings_private_network(settings
);
4464 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4465 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4466 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4467 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4469 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4470 free_and_replace(arg_network_zone
, settings
->network_zone
);
4472 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4476 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4477 settings
->expose_ports
) {
4479 if (!arg_settings_trusted
)
4480 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4482 expose_port_free_all(arg_expose_ports
);
4483 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4487 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4488 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4490 if (!arg_settings_trusted
)
4491 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4493 arg_userns_mode
= settings
->userns_mode
;
4494 arg_uid_shift
= settings
->uid_shift
;
4495 arg_uid_range
= settings
->uid_range
;
4496 arg_userns_ownership
= settings
->userns_ownership
;
4500 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0 &&
4501 !strv_isempty(settings
->bind_user
))
4502 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4504 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0 &&
4505 settings
->notify_ready
>= 0)
4506 arg_notify_ready
= settings
->notify_ready
;
4508 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4510 if (!strv_isempty(settings
->syscall_allow_list
) || !strv_isempty(settings
->syscall_deny_list
)) {
4511 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4512 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4514 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4515 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4520 if (settings
->seccomp
) {
4521 if (!arg_settings_trusted
)
4522 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4524 seccomp_release(arg_seccomp
);
4525 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4531 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4532 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4535 if (!settings
->rlimit
[rl
])
4538 if (!arg_settings_trusted
) {
4539 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4543 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4546 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4548 free_and_replace(arg_hostname
, settings
->hostname
);
4550 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4551 settings
->no_new_privileges
>= 0)
4552 arg_no_new_privileges
= settings
->no_new_privileges
;
4554 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4555 settings
->oom_score_adjust_set
) {
4557 if (!arg_settings_trusted
)
4558 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4560 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4561 arg_oom_score_adjust_set
= true;
4565 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4566 settings
->cpu_set
.set
) {
4568 if (!arg_settings_trusted
)
4569 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4571 cpu_set_reset(&arg_cpu_set
);
4572 arg_cpu_set
= settings
->cpu_set
;
4573 settings
->cpu_set
= (CPUSet
) {};
4577 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4578 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4579 arg_resolv_conf
= settings
->resolv_conf
;
4581 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4582 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4584 if (!arg_settings_trusted
)
4585 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4587 arg_link_journal
= settings
->link_journal
;
4588 arg_link_journal_try
= settings
->link_journal_try
;
4592 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4593 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4594 arg_timezone
= settings
->timezone
;
4596 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4599 if (!arg_settings_trusted
)
4600 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4602 free_and_replace(arg_slice
, settings
->slice
);
4605 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4606 settings
->use_cgns
>= 0) {
4608 if (!arg_settings_trusted
)
4609 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4611 arg_use_cgns
= settings
->use_cgns
;
4614 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4615 settings
->clone_ns_flags
!= ULONG_MAX
) {
4617 if (!arg_settings_trusted
)
4618 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4620 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4623 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4624 settings
->console_mode
>= 0) {
4626 if (!arg_settings_trusted
)
4627 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4629 arg_console_mode
= settings
->console_mode
;
4632 if ((arg_settings_mask
& SETTING_SUPPRESS_SYNC
) == 0 &&
4633 settings
->suppress_sync
>= 0)
4634 arg_suppress_sync
= settings
->suppress_sync
;
4636 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4637 * don't consult arg_settings_mask for them. */
4639 sd_bus_message_unref(arg_property_message
);
4640 arg_property_message
= TAKE_PTR(settings
->properties
);
4642 arg_console_width
= settings
->console_width
;
4643 arg_console_height
= settings
->console_height
;
4645 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4646 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4647 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4652 static int load_settings(void) {
4653 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4654 _cleanup_fclose_
FILE *f
= NULL
;
4655 _cleanup_free_
char *p
= NULL
;
4661 /* If all settings are masked, there's no point in looking for
4662 * the settings file */
4663 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4666 /* We first look in the admin's directories in /etc and /run */
4667 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4668 _cleanup_free_
char *j
= NULL
;
4670 j
= path_join(i
, arg_settings_filename
);
4678 /* By default, we trust configuration from /etc and /run */
4679 if (arg_settings_trusted
< 0)
4680 arg_settings_trusted
= true;
4685 if (errno
!= ENOENT
)
4686 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4690 /* After that, let's look for a file next to the
4691 * actual image we shall boot. */
4694 p
= file_in_same_dir(arg_image
, arg_settings_filename
);
4697 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4698 p
= file_in_same_dir(arg_directory
, arg_settings_filename
);
4705 if (!f
&& errno
!= ENOENT
)
4706 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4708 /* By default, we do not trust configuration from /var/lib/machines */
4709 if (arg_settings_trusted
< 0)
4710 arg_settings_trusted
= false;
4717 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4719 r
= settings_load(f
, p
, &settings
);
4723 return merge_settings(settings
, p
);
4726 static int load_oci_bundle(void) {
4727 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4730 if (!arg_oci_bundle
)
4733 /* By default let's trust OCI bundles */
4734 if (arg_settings_trusted
< 0)
4735 arg_settings_trusted
= true;
4737 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4741 return merge_settings(settings
, arg_oci_bundle
);
4744 static int run_container(
4745 DissectedImage
*dissected_image
,
4748 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4749 struct ExposeArgs
*expose_args
,
4750 int *master
, pid_t
*pid
, int *ret
) {
4752 static const struct sigaction sa
= {
4753 .sa_handler
= nop_signal_handler
,
4754 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4757 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4758 _cleanup_close_
int etc_passwd_lock
= -1;
4759 _cleanup_close_pair_
int
4760 kmsg_socket_pair
[2] = { -1, -1 },
4761 rtnl_socket_pair
[2] = { -1, -1 },
4762 fd_socket_pair
[2] = { -EBADF
, -EBADF
},
4763 master_pty_socket_pair
[2] = { -1, -1 },
4764 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4766 _cleanup_close_
int notify_socket
= -1, mntns_fd
= -EBADF
;
4767 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4768 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4769 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4770 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4771 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4772 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4773 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4774 size_t n_bind_user_uid
= 0;
4775 ContainerStatus container_status
= 0;
4779 _cleanup_close_
int child_netns_fd
= -1;
4781 assert_se(sigemptyset(&mask_chld
) == 0);
4782 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4784 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4785 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4786 * check with getpwuid() if the specific user already exists. Note that /etc might be
4787 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4788 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4789 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4792 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4793 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4794 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4797 r
= barrier_create(&barrier
);
4799 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4801 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4802 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4804 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4805 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4807 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, fd_socket_pair
) < 0)
4808 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4810 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4811 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4813 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4814 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4815 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4817 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4818 * parent's blocking calls and give it a chance to call wait() and terminate. */
4819 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4821 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4823 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4825 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4827 if (arg_network_namespace_path
) {
4828 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4829 if (child_netns_fd
< 0)
4830 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4832 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4834 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
);
4836 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4838 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4839 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4842 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4844 return log_error_errno(errno
, "clone() failed%s: %m",
4846 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4849 /* The outer child only has a file system namespace. */
4850 barrier_set_role(&barrier
, BARRIER_CHILD
);
4852 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4853 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4854 fd_socket_pair
[0] = safe_close(fd_socket_pair
[0]);
4855 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4856 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4858 (void) reset_all_signal_handlers();
4859 (void) reset_signal_mask();
4861 r
= outer_child(&barrier
,
4866 kmsg_socket_pair
[1],
4867 rtnl_socket_pair
[1],
4868 master_pty_socket_pair
[1],
4869 unified_cgroup_hierarchy_socket_pair
[1],
4873 _exit(EXIT_FAILURE
);
4875 _exit(EXIT_SUCCESS
);
4878 barrier_set_role(&barrier
, BARRIER_PARENT
);
4882 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4883 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4884 fd_socket_pair
[1] = safe_close(fd_socket_pair
[1]);
4885 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4886 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4888 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4889 mntns_fd
= receive_one_fd(fd_socket_pair
[0], 0);
4891 return log_error_errno(mntns_fd
, "Failed to receive mount namespace fd from outer child: %m");
4893 /* The child just let us know the UID shift it might have read from the image. */
4894 l
= recv(fd_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4896 return log_error_errno(errno
, "Failed to read UID shift: %m");
4897 if (l
!= sizeof arg_uid_shift
)
4898 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4900 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4901 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4902 * image, but if that's already in use, pick a new one, and report back to the child,
4903 * which one we now picked. */
4905 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4907 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4909 l
= send(fd_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4911 return log_error_errno(errno
, "Failed to send UID shift: %m");
4912 if (l
!= sizeof arg_uid_shift
)
4913 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4916 n_bind_user_uid
= strv_length(arg_bind_user
);
4917 if (n_bind_user_uid
> 0) {
4918 /* Right after the UID shift, we'll receive the list of UID mappings for the
4919 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4921 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4925 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4926 l
= recv(fd_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4928 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4929 if (l
!= sizeof(uid_t
)*4)
4930 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4931 SYNTHETIC_ERRNO(EIO
),
4932 "Short read while reading bind user UID pairs.");
4937 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4938 /* The child let us know the support cgroup mode it might have read from the image. */
4939 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4941 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4942 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4943 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zi bytes).%s",
4944 l
, l
== 0 ? " The child is most likely dead." : "");
4947 /* Wait for the outer child. */
4948 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4951 if (r
!= EXIT_SUCCESS
)
4954 /* And now retrieve the PID of the inner child. */
4955 l
= recv(fd_socket_pair
[0], pid
, sizeof *pid
, 0);
4957 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4958 if (l
!= sizeof *pid
)
4959 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4961 /* We also retrieve container UUID in case it was generated by outer child */
4962 l
= recv(fd_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4964 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4965 if (l
!= sizeof(arg_uuid
))
4966 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4968 /* We also retrieve the socket used for notifications generated by outer child */
4969 notify_socket
= receive_one_fd(fd_socket_pair
[0], 0);
4970 if (notify_socket
< 0)
4971 return log_error_errno(notify_socket
,
4972 "Failed to receive notification socket from the outer child: %m");
4974 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4976 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4977 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4978 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4980 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4984 (void) barrier_place(&barrier
); /* #2 */
4987 if (arg_private_network
) {
4988 if (!arg_network_namespace_path
) {
4989 /* Wait until the child has unshared its network namespace. */
4990 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4991 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4994 if (child_netns_fd
< 0) {
4995 /* Make sure we have an open file descriptor to the child's network
4996 * namespace so it stays alive even if the child exits. */
4997 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4999 return log_error_errno(r
, "Failed to open child network namespace: %m");
5002 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
5006 if (arg_network_veth
) {
5007 r
= setup_veth(arg_machine
, *pid
, veth_name
,
5008 arg_network_bridge
|| arg_network_zone
);
5014 if (arg_network_bridge
) {
5015 /* Add the interface to a bridge */
5016 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
5021 } else if (arg_network_zone
) {
5022 /* Add the interface to a bridge, possibly creating it */
5023 r
= setup_bridge(veth_name
, arg_network_zone
, true);
5031 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
5035 /* We created the primary and extra veth links now; let's remember this, so that we know to
5036 remove them later on. Note that we don't bother with removing veth links that were created
5037 here when their setup failed half-way, because in that case the kernel should be able to
5038 remove them on its own, since they cannot be referenced by anything yet. */
5039 *veth_created
= true;
5041 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
5045 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
5050 if (arg_register
|| !arg_keep_unit
) {
5051 r
= sd_bus_default_system(&bus
);
5053 return log_error_errno(r
, "Failed to open system bus: %m");
5055 r
= sd_bus_set_close_on_exit(bus
, false);
5057 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
5060 if (!arg_keep_unit
) {
5061 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5062 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5063 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5065 r
= sd_bus_match_signal_async(
5068 "org.freedesktop.systemd1",
5070 "org.freedesktop.systemd1.Scope",
5072 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5074 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5078 r
= register_machine(
5086 arg_custom_mounts
, arg_n_custom_mounts
,
5089 arg_property_message
,
5091 arg_container_service_name
);
5095 } else if (!arg_keep_unit
) {
5101 arg_custom_mounts
, arg_n_custom_mounts
,
5104 arg_property_message
);
5108 } else if (arg_slice
|| arg_property
)
5109 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5111 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5115 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5119 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5123 /* Notify the child that the parent is ready with all
5124 * its setup (including cgroup-ification), and that
5125 * the child can now hand over control to the code to
5126 * run inside the container. */
5127 (void) barrier_place(&barrier
); /* #4 */
5129 /* Block SIGCHLD here, before notifying child.
5130 * process_pty() will handle it with the other signals. */
5131 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5133 /* Reset signal to default */
5134 r
= default_signals(SIGCHLD
);
5136 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5138 r
= sd_event_new(&event
);
5140 return log_error_errno(r
, "Failed to get default event source: %m");
5142 (void) sd_event_set_watchdog(event
, true);
5145 r
= sd_bus_attach_event(bus
, event
, 0);
5147 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5150 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5154 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
5155 r
= wipe_fully_visible_fs(mntns_fd
);
5158 mntns_fd
= safe_close(mntns_fd
);
5161 /* Let the child know that we are ready and wait that the child is completely ready now. */
5162 if (!barrier_place_and_sync(&barrier
)) /* #5 */
5163 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5165 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5166 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5167 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5169 (void) sd_notifyf(false,
5170 "STATUS=Container running.\n"
5171 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5172 if (!arg_notify_ready
) {
5173 r
= sd_notify(false, "READY=1\n");
5175 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
5178 if (arg_kill_signal
> 0) {
5179 /* Try to kill the init system on SIGINT or SIGTERM */
5180 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5181 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5183 /* Immediately exit */
5184 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5185 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5188 /* Exit when the child exits */
5189 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5191 if (arg_expose_ports
) {
5192 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5196 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5197 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5200 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
5202 if (arg_console_mode
!= CONSOLE_PIPE
) {
5203 _cleanup_close_
int fd
= -1;
5204 PTYForwardFlags flags
= 0;
5206 /* Retrieve the master pty allocated by inner child */
5207 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
5209 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5211 switch (arg_console_mode
) {
5213 case CONSOLE_READ_ONLY
:
5214 flags
|= PTY_FORWARD_READ_ONLY
;
5218 case CONSOLE_INTERACTIVE
:
5219 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5221 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5223 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5225 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5226 (void) pty_forward_set_width_height(forward
,
5228 arg_console_height
);
5232 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5235 *master
= TAKE_FD(fd
);
5238 r
= sd_event_loop(event
);
5240 return log_error_errno(r
, "Failed to run event loop: %m");
5245 (void) pty_forward_get_last_char(forward
, &last_char
);
5246 forward
= pty_forward_free(forward
);
5248 if (!arg_quiet
&& last_char
!= '\n')
5252 /* Kill if it is not dead yet anyway */
5253 if (!arg_register
&& !arg_keep_unit
&& bus
)
5254 terminate_scope(bus
, arg_machine
);
5256 /* Normally redundant, but better safe than sorry */
5257 (void) kill(*pid
, SIGKILL
);
5259 if (arg_private_network
) {
5260 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5261 * to avoid having to move the parent to the child network namespace. */
5262 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
5267 _cleanup_close_
int parent_netns_fd
= -1;
5269 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5271 log_error_errno(r
, "Failed to open parent network namespace: %m");
5272 _exit(EXIT_FAILURE
);
5275 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5277 log_error_errno(r
, "Failed to enter child network namespace: %m");
5278 _exit(EXIT_FAILURE
);
5281 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5283 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5285 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5289 r
= wait_for_container(TAKE_PID(*pid
), &container_status
);
5291 /* Tell machined that we are gone. */
5293 (void) unregister_machine(bus
, arg_machine
);
5296 /* We failed to wait for the container, or the container exited abnormally. */
5298 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5299 /* r > 0 → The container exited with a non-zero status.
5300 * As a special case, we need to replace 133 with a different value,
5301 * because 133 is special-cased in the service file to reboot the container.
5302 * otherwise → The container exited with zero status and a reboot was not requested.
5304 if (r
== EXIT_FORCE_RESTART
)
5305 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5307 return 0; /* finito */
5310 /* CONTAINER_REBOOTED, loop again */
5312 if (arg_keep_unit
) {
5313 /* Special handling if we are running as a service: instead of simply
5314 * restarting the machine we want to restart the entire service, so let's
5315 * inform systemd about this with the special exit code 133. The service
5316 * file uses RestartForceExitStatus=133 so that this results in a full
5317 * nspawn restart. This is necessary since we might have cgroup parameters
5318 * set we want to have flushed out. */
5319 *ret
= EXIT_FORCE_RESTART
;
5320 return 0; /* finito */
5323 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5324 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5326 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5327 *veth_created
= false;
5328 return 1; /* loop again */
5331 static int initialize_rlimits(void) {
5332 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5333 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5334 * container execution environments. */
5336 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5337 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5338 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5339 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5340 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5341 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5342 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5343 [RLIMIT_MEMLOCK
] = { DEFAULT_RLIMIT_MEMLOCK
, DEFAULT_RLIMIT_MEMLOCK
},
5344 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5345 [RLIMIT_NICE
] = { 0, 0 },
5346 [RLIMIT_NOFILE
] = { 1024, 4096 },
5347 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5348 [RLIMIT_RTPRIO
] = { 0, 0 },
5349 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5350 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5352 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5353 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5354 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5355 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5356 * that PID 1 changes a number of other resource limits during early initialization which is why we
5357 * don't read the other limits from PID 1 but prefer the static table above. */
5362 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5363 /* Let's only fill in what the user hasn't explicitly configured anyway */
5364 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5365 const struct rlimit
*v
;
5366 struct rlimit buffer
;
5368 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5369 /* For these two let's read the limits off PID 1. See above for an explanation. */
5371 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
5372 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5375 } else if (rl
== RLIMIT_NOFILE
) {
5376 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5377 * userspace. Given that nspawn containers are often run without our PID 1,
5378 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5379 * so that container userspace gets similar resources as host userspace
5381 buffer
= kernel_defaults
[rl
];
5382 buffer
.rlim_max
= MIN((rlim_t
) read_nr_open(), (rlim_t
) HIGH_RLIMIT_NOFILE
);
5385 v
= kernel_defaults
+ rl
;
5387 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5388 if (!arg_rlimit
[rl
])
5392 if (DEBUG_LOGGING
) {
5393 _cleanup_free_
char *k
= NULL
;
5395 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5396 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5403 static int cant_be_in_netns(void) {
5404 _cleanup_close_
int fd
= -1;
5408 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5409 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5412 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5415 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5417 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5419 r
= connect_unix_path(fd
, AT_FDCWD
, "/run/udev/control");
5421 if (r
== -ENOENT
|| ERRNO_IS_DISCONNECT(r
))
5422 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5423 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5425 return log_error_errno(r
, "Failed to connect socket to udev control socket: %m");
5428 r
= getpeercred(fd
, &ucred
);
5430 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5432 r
= in_same_namespace(ucred
.pid
, 0, NAMESPACE_NET
);
5434 return log_error_errno(r
, "Failed to determine network namespace of udev: %m");
5436 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5437 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5441 static int run(int argc
, char *argv
[]) {
5442 bool secondary
= false, remove_directory
= false, remove_image
= false,
5443 veth_created
= false, remove_tmprootdir
= false;
5444 _cleanup_close_
int master
= -1;
5445 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5446 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5447 char veth_name
[IFNAMSIZ
] = "";
5448 struct ExposeArgs expose_args
= {};
5449 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5450 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5451 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5452 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5453 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5456 log_parse_environment();
5459 r
= parse_argv(argc
, argv
);
5463 if (geteuid() != 0) {
5464 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5465 argc
>= 2 ? "Need to be root." :
5466 "Need to be root (and some arguments are usually required).\nHint: try --help");
5470 r
= cant_be_in_netns();
5474 r
= initialize_rlimits();
5478 r
= load_oci_bundle();
5482 r
= determine_names();
5486 r
= load_settings();
5492 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5496 r
= verify_arguments();
5500 /* Reapply environment settings. */
5501 (void) detect_unified_cgroup_hierarchy_from_environment();
5503 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5504 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5505 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5506 (void) ignore_signals(SIGPIPE
);
5508 n_fd_passed
= sd_listen_fds(false);
5509 if (n_fd_passed
> 0) {
5510 r
= fdset_new_listen_fds(&fds
, false);
5512 log_error_errno(r
, "Failed to collect file descriptors: %m");
5517 /* The "default" umask. This is appropriate for most file and directory
5518 * operations performed by nspawn, and is the umask that will be used for
5519 * the child. Functions like copy_devnodes() change the umask temporarily. */
5522 if (arg_directory
) {
5525 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5526 * /var from the host will propagate into container dynamically (because bad things happen if
5527 * two systems write to the same /var). Let's allow it for the special cases where /var is
5528 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5529 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5530 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5535 if (arg_ephemeral
) {
5536 _cleanup_free_
char *np
= NULL
;
5538 r
= chase_symlinks_and_update(&arg_directory
, 0);
5542 /* If the specified path is a mount point we generate the new snapshot immediately
5543 * inside it under a random name. However if the specified is not a mount point we
5544 * create the new snapshot in the parent directory, just next to it. */
5545 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5547 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5551 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5553 r
= tempfn_random(arg_directory
, "machine.", &np
);
5555 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5559 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5560 * only owned by us and no one else. */
5561 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5563 log_error_errno(r
, "Failed to lock %s: %m", np
);
5568 BLOCK_SIGNALS(SIGINT
);
5569 r
= btrfs_subvol_snapshot(arg_directory
, np
,
5570 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5571 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5572 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5573 BTRFS_SNAPSHOT_RECURSIVE
|
5574 BTRFS_SNAPSHOT_QUOTA
|
5575 BTRFS_SNAPSHOT_SIGINT
);
5578 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5582 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5586 free_and_replace(arg_directory
, np
);
5587 remove_directory
= true;
5589 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5593 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5595 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5599 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5604 r
= chase_symlinks_and_update(&arg_template
, 0);
5609 BLOCK_SIGNALS(SIGINT
);
5610 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5611 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5612 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5613 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5614 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5615 BTRFS_SNAPSHOT_RECURSIVE
|
5616 BTRFS_SNAPSHOT_QUOTA
|
5617 BTRFS_SNAPSHOT_SIGINT
);
5620 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5621 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5622 else if (r
== -EINTR
) {
5623 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5626 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5629 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5630 "Populated %s from template %s.", arg_directory
, arg_template
);
5634 if (arg_start_mode
== START_BOOT
) {
5635 _cleanup_free_
char *b
= NULL
;
5638 if (arg_pivot_root_new
) {
5639 b
= path_join(arg_directory
, arg_pivot_root_new
);
5647 if (path_is_os_tree(p
) <= 0) {
5648 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5649 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5653 _cleanup_free_
char *p
= NULL
;
5655 if (arg_pivot_root_new
)
5656 p
= path_join(arg_directory
, arg_pivot_root_new
, "/usr/");
5658 p
= path_join(arg_directory
, "/usr/");
5662 if (laccess(p
, F_OK
) < 0) {
5663 r
= log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
5664 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory
);
5670 DissectImageFlags dissect_image_flags
=
5671 DISSECT_IMAGE_GENERIC_ROOT
|
5672 DISSECT_IMAGE_REQUIRE_ROOT
|
5673 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5674 DISSECT_IMAGE_USR_NO_ROOT
|
5675 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
5676 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
5678 assert(!arg_template
);
5680 r
= chase_symlinks_and_update(&arg_image
, 0);
5684 if (arg_ephemeral
) {
5685 _cleanup_free_
char *np
= NULL
;
5687 r
= tempfn_random(arg_image
, "machine.", &np
);
5689 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5693 /* Always take an exclusive lock on our own ephemeral copy. */
5694 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5696 r
= log_error_errno(r
, "Failed to create image lock: %m");
5701 BLOCK_SIGNALS(SIGINT
);
5702 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
);
5705 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5709 r
= log_error_errno(r
, "Failed to copy image file: %m");
5713 free_and_replace(arg_image
, np
);
5714 remove_image
= true;
5716 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5718 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5722 r
= log_error_errno(r
, "Failed to create image lock: %m");
5726 r
= verity_settings_load(
5727 &arg_verity_settings
,
5728 arg_image
, NULL
, NULL
);
5730 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5734 if (arg_verity_settings
.data_path
)
5735 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5738 if (!mkdtemp(tmprootdir
)) {
5739 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5743 remove_tmprootdir
= true;
5745 arg_directory
= strdup(tmprootdir
);
5746 if (!arg_directory
) {
5751 r
= loop_device_make_by_path(
5753 arg_read_only
? O_RDONLY
: O_RDWR
,
5754 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5758 log_error_errno(r
, "Failed to set up loopback block device: %m");
5762 r
= dissect_loop_device_and_warn(
5764 &arg_verity_settings
,
5766 dissect_image_flags
,
5769 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5770 log_notice("Note that the disk image needs to\n"
5771 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5772 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5773 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5774 " d) or contain a file system without a partition table\n"
5775 "in order to be bootable with systemd-nspawn.");
5781 r
= dissected_image_load_verity_sig_partition(
5784 &arg_verity_settings
);
5788 if (dissected_image
->has_verity
&& !arg_verity_settings
.root_hash
&& !dissected_image
->has_verity_sig
)
5789 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5790 "root hash signature found! Proceeding without integrity checking.", arg_image
);
5792 r
= dissected_image_decrypt_interactively(
5795 &arg_verity_settings
,
5800 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5801 if (remove_image
&& unlink(arg_image
) >= 0)
5802 remove_image
= false;
5805 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5809 if (arg_console_mode
< 0)
5811 isatty(STDIN_FILENO
) > 0 &&
5812 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5814 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5818 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5819 arg_machine
, arg_image
?: arg_directory
);
5821 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5823 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5824 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5828 if (arg_expose_ports
) {
5829 r
= fw_ctx_new(&fw_ctx
);
5831 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5834 expose_args
.fw_ctx
= fw_ctx
;
5837 r
= run_container(dissected_image
,
5840 veth_name
, &veth_created
,
5841 &expose_args
, &master
,
5848 (void) sd_notify(false,
5849 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5850 "STOPPING=1\nSTATUS=Terminating...");
5853 (void) kill(pid
, SIGKILL
);
5855 /* Try to flush whatever is still queued in the pty */
5857 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5858 master
= safe_close(master
);
5862 (void) wait_for_terminate(pid
, NULL
);
5866 if (remove_directory
&& arg_directory
) {
5869 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5871 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5874 if (remove_image
&& arg_image
) {
5875 if (unlink(arg_image
) < 0)
5876 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5879 if (remove_tmprootdir
) {
5880 if (rmdir(tmprootdir
) < 0)
5881 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5887 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5888 (void) rm_rf(p
, REMOVE_ROOT
);
5891 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5892 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5895 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5896 (void) remove_bridge(arg_network_zone
);
5898 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5899 expose_port_free_all(arg_expose_ports
);
5900 rlimit_free_all(arg_rlimit
);
5901 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5902 credential_free_all(arg_credentials
, arg_n_credentials
);
5910 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);