1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
8 #include <linux/loop.h>
10 #include <selinux/selinux.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.h>
23 #include "sd-daemon.h"
26 #include "alloc-util.h"
28 #include "base-filesystem.h"
29 #include "blkid-util.h"
30 #include "btrfs-util.h"
31 #include "bus-error.h"
34 #include "capability-util.h"
35 #include "cgroup-util.h"
37 #include "cpu-set-util.h"
38 #include "creds-util.h"
39 #include "dev-setup.h"
40 #include "discover-image.h"
41 #include "dissect-image.h"
47 #include "format-util.h"
50 #include "hexdecoct.h"
51 #include "hostname-setup.h"
52 #include "hostname-util.h"
53 #include "id128-util.h"
56 #include "loop-util.h"
57 #include "loopback-setup.h"
59 #include "main-func.h"
60 #include "missing_sched.h"
62 #include "mount-util.h"
63 #include "mountpoint-util.h"
64 #include "namespace-util.h"
65 #include "netlink-util.h"
66 #include "nspawn-bind-user.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-creds.h"
69 #include "nspawn-def.h"
70 #include "nspawn-expose-ports.h"
71 #include "nspawn-mount.h"
72 #include "nspawn-network.h"
73 #include "nspawn-oci.h"
74 #include "nspawn-patch-uid.h"
75 #include "nspawn-register.h"
76 #include "nspawn-seccomp.h"
77 #include "nspawn-settings.h"
78 #include "nspawn-setuid.h"
79 #include "nspawn-stub-pid1.h"
81 #include "nulstr-util.h"
84 #include "parse-argument.h"
85 #include "parse-util.h"
86 #include "path-util.h"
87 #include "pretty-print.h"
88 #include "process-util.h"
90 #include "random-util.h"
91 #include "raw-clone.h"
92 #include "resolve-util.h"
93 #include "rlimit-util.h"
96 #include "seccomp-util.h"
98 #include "selinux-util.h"
99 #include "signal-util.h"
100 #include "socket-util.h"
101 #include "stat-util.h"
102 #include "stdio-util.h"
103 #include "string-table.h"
104 #include "string-util.h"
106 #include "sysctl-util.h"
107 #include "terminal-util.h"
108 #include "tmpfile-util.h"
109 #include "umask-util.h"
110 #include "unit-name.h"
111 #include "user-util.h"
114 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
115 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
117 #define EXIT_FORCE_RESTART 133
119 typedef enum ContainerStatus
{
120 CONTAINER_TERMINATED
,
124 static char *arg_directory
= NULL
;
125 static char *arg_template
= NULL
;
126 static char *arg_chdir
= NULL
;
127 static char *arg_pivot_root_new
= NULL
;
128 static char *arg_pivot_root_old
= NULL
;
129 static char *arg_user
= NULL
;
130 static uid_t arg_uid
= UID_INVALID
;
131 static gid_t arg_gid
= GID_INVALID
;
132 static gid_t
* arg_supplementary_gids
= NULL
;
133 static size_t arg_n_supplementary_gids
= 0;
134 static sd_id128_t arg_uuid
= {};
135 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
136 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
137 static const char *arg_selinux_context
= NULL
;
138 static const char *arg_selinux_apifs_context
= NULL
;
139 static char *arg_slice
= NULL
;
140 static bool arg_private_network
= false;
141 static bool arg_read_only
= false;
142 static StartMode arg_start_mode
= START_PID1
;
143 static bool arg_ephemeral
= false;
144 static LinkJournal arg_link_journal
= LINK_AUTO
;
145 static bool arg_link_journal_try
= false;
146 static uint64_t arg_caps_retain
=
147 (1ULL << CAP_AUDIT_CONTROL
) |
148 (1ULL << CAP_AUDIT_WRITE
) |
149 (1ULL << CAP_CHOWN
) |
150 (1ULL << CAP_DAC_OVERRIDE
) |
151 (1ULL << CAP_DAC_READ_SEARCH
) |
152 (1ULL << CAP_FOWNER
) |
153 (1ULL << CAP_FSETID
) |
154 (1ULL << CAP_IPC_OWNER
) |
156 (1ULL << CAP_LEASE
) |
157 (1ULL << CAP_LINUX_IMMUTABLE
) |
158 (1ULL << CAP_MKNOD
) |
159 (1ULL << CAP_NET_BIND_SERVICE
) |
160 (1ULL << CAP_NET_BROADCAST
) |
161 (1ULL << CAP_NET_RAW
) |
162 (1ULL << CAP_SETFCAP
) |
163 (1ULL << CAP_SETGID
) |
164 (1ULL << CAP_SETPCAP
) |
165 (1ULL << CAP_SETUID
) |
166 (1ULL << CAP_SYS_ADMIN
) |
167 (1ULL << CAP_SYS_BOOT
) |
168 (1ULL << CAP_SYS_CHROOT
) |
169 (1ULL << CAP_SYS_NICE
) |
170 (1ULL << CAP_SYS_PTRACE
) |
171 (1ULL << CAP_SYS_RESOURCE
) |
172 (1ULL << CAP_SYS_TTY_CONFIG
);
173 static uint64_t arg_caps_ambient
= 0;
174 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
175 static CustomMount
*arg_custom_mounts
= NULL
;
176 static size_t arg_n_custom_mounts
= 0;
177 static char **arg_setenv
= NULL
;
178 static bool arg_quiet
= false;
179 static bool arg_register
= true;
180 static bool arg_keep_unit
= false;
181 static char **arg_network_interfaces
= NULL
;
182 static char **arg_network_macvlan
= NULL
;
183 static char **arg_network_ipvlan
= NULL
;
184 static bool arg_network_veth
= false;
185 static char **arg_network_veth_extra
= NULL
;
186 static char *arg_network_bridge
= NULL
;
187 static char *arg_network_zone
= NULL
;
188 static char *arg_network_namespace_path
= NULL
;
189 static PagerFlags arg_pager_flags
= 0;
190 static unsigned long arg_personality
= PERSONALITY_INVALID
;
191 static char *arg_image
= NULL
;
192 static char *arg_oci_bundle
= NULL
;
193 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
194 static ExposePort
*arg_expose_ports
= NULL
;
195 static char **arg_property
= NULL
;
196 static sd_bus_message
*arg_property_message
= NULL
;
197 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
198 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
199 static UserNamespaceOwnership arg_userns_ownership
= _USER_NAMESPACE_OWNERSHIP_INVALID
;
200 static int arg_kill_signal
= 0;
201 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
202 static SettingsMask arg_settings_mask
= 0;
203 static int arg_settings_trusted
= -1;
204 static char **arg_parameters
= NULL
;
205 static const char *arg_container_service_name
= "systemd-nspawn";
206 static bool arg_notify_ready
= false;
207 static bool arg_use_cgns
= true;
208 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
209 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
210 static VeritySettings arg_verity_settings
= VERITY_SETTINGS_DEFAULT
;
211 static char **arg_syscall_allow_list
= NULL
;
212 static char **arg_syscall_deny_list
= NULL
;
214 static scmp_filter_ctx arg_seccomp
= NULL
;
216 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
217 static bool arg_no_new_privileges
= false;
218 static int arg_oom_score_adjust
= 0;
219 static bool arg_oom_score_adjust_set
= false;
220 static CPUSet arg_cpu_set
= {};
221 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
222 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
223 static unsigned arg_console_width
= UINT_MAX
, arg_console_height
= UINT_MAX
;
224 static DeviceNode
* arg_extra_nodes
= NULL
;
225 static size_t arg_n_extra_nodes
= 0;
226 static char **arg_sysctl
= NULL
;
227 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
228 static Credential
*arg_credentials
= NULL
;
229 static size_t arg_n_credentials
= 0;
230 static char **arg_bind_user
= NULL
;
232 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
233 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
254 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings
, verity_settings_done
);
256 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
259 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
261 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
262 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
263 STATIC_DESTRUCTOR_REGISTER(arg_bind_user
, strv_freep
);
265 static int handle_arg_console(const char *arg
) {
266 if (streq(arg
, "help")) {
275 if (streq(arg
, "interactive"))
276 arg_console_mode
= CONSOLE_INTERACTIVE
;
277 else if (streq(arg
, "read-only"))
278 arg_console_mode
= CONSOLE_READ_ONLY
;
279 else if (streq(arg
, "passive"))
280 arg_console_mode
= CONSOLE_PASSIVE
;
281 else if (streq(arg
, "pipe")) {
282 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
283 log_full(arg_quiet
? LOG_DEBUG
: LOG_NOTICE
,
284 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
285 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
286 "Proceeding anyway.");
288 arg_console_mode
= CONSOLE_PIPE
;
289 } else if (streq(arg
, "autopipe")) {
290 if (isatty(STDIN_FILENO
) > 0 && isatty(STDOUT_FILENO
) > 0)
291 arg_console_mode
= CONSOLE_INTERACTIVE
;
293 arg_console_mode
= CONSOLE_PIPE
;
295 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
297 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
301 static int help(void) {
302 _cleanup_free_
char *link
= NULL
;
305 (void) pager_open(arg_pager_flags
);
307 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
311 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
312 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
313 " -h --help Show this help\n"
314 " --version Print version string\n"
315 " -q --quiet Do not show status information\n"
316 " --no-pager Do not pipe output into a pager\n"
317 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
319 " -D --directory=PATH Root directory for the container\n"
320 " --template=PATH Initialize root directory from template directory,\n"
322 " -x --ephemeral Run container with snapshot of root directory, and\n"
323 " remove it after exit\n"
324 " -i --image=PATH Root file system disk image (or device node) for\n"
326 " --oci-bundle=PATH OCI bundle directory\n"
327 " --read-only Mount the root directory read-only\n"
328 " --volatile[=MODE] Run the system in volatile mode\n"
329 " --root-hash=HASH Specify verity root hash for root disk image\n"
330 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
331 " as a DER encoded PKCS7, either as a path to a file\n"
332 " or as an ASCII base64 encoded string prefixed by\n"
334 " --verity-data=PATH Specify hash device for verity\n"
335 " --pivot-root=PATH[:PATH]\n"
336 " Pivot root to given directory in the container\n\n"
337 "%3$sExecution:%4$s\n"
338 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
339 " -b --boot Boot up full system (i.e. invoke init)\n"
340 " --chdir=PATH Set working directory in the container\n"
341 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
342 " -u --user=USER Run the command under specified user or UID\n"
343 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
344 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
345 "%3$sSystem Identity:%4$s\n"
346 " -M --machine=NAME Set the machine name for the container\n"
347 " --hostname=NAME Override the hostname for the container\n"
348 " --uuid=UUID Set a specific machine UUID for the container\n\n"
349 "%3$sProperties:%4$s\n"
350 " -S --slice=SLICE Place the container in the specified slice\n"
351 " --property=NAME=VALUE Set scope unit property\n"
352 " --register=BOOLEAN Register container as machine\n"
353 " --keep-unit Do not register a scope for the machine, reuse\n"
354 " the service unit nspawn is running in\n\n"
355 "%3$sUser Namespacing:%4$s\n"
356 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
357 " --private-users[=UIDBASE[:NUIDS]]\n"
358 " Similar, but with user configured UID/GID range\n"
359 " --private-users-ownership=MODE\n"
360 " Adjust ('chown') or map ('map') OS tree ownership\n"
361 " to private UID/GID range\n\n"
362 "%3$sNetworking:%4$s\n"
363 " --private-network Disable network in container\n"
364 " --network-interface=INTERFACE\n"
365 " Assign an existing network interface to the\n"
367 " --network-macvlan=INTERFACE\n"
368 " Create a macvlan network interface based on an\n"
369 " existing network interface to the container\n"
370 " --network-ipvlan=INTERFACE\n"
371 " Create an ipvlan network interface based on an\n"
372 " existing network interface to the container\n"
373 " -n --network-veth Add a virtual Ethernet connection between host\n"
375 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
376 " Add an additional virtual Ethernet link between\n"
377 " host and container\n"
378 " --network-bridge=INTERFACE\n"
379 " Add a virtual Ethernet connection to the container\n"
380 " and attach it to an existing bridge on the host\n"
381 " --network-zone=NAME Similar, but attach the new interface to an\n"
382 " an automatically managed bridge interface\n"
383 " --network-namespace-path=PATH\n"
384 " Set network namespace to the one represented by\n"
385 " the specified kernel namespace file node\n"
386 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
387 " Expose a container IP port on the host\n\n"
388 "%3$sSecurity:%4$s\n"
389 " --capability=CAP In addition to the default, retain specified\n"
391 " --drop-capability=CAP Drop the specified capability from the default set\n"
392 " --ambient-capability=CAP\n"
393 " Sets the specified capability for the started\n"
394 " process. Not useful if booting a machine.\n"
395 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
396 " --system-call-filter=LIST|~LIST\n"
397 " Permit/prohibit specific system calls\n"
398 " -Z --selinux-context=SECLABEL\n"
399 " Set the SELinux security context to be used by\n"
400 " processes in the container\n"
401 " -L --selinux-apifs-context=SECLABEL\n"
402 " Set the SELinux security context to be used by\n"
403 " API/tmpfs file systems in the container\n\n"
404 "%3$sResources:%4$s\n"
405 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
406 " --oom-score-adjust=VALUE\n"
407 " Adjust the OOM score value for the payload\n"
408 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
409 " --personality=ARCH Pick personality for this container\n\n"
410 "%3$sIntegration:%4$s\n"
411 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
412 " --timezone=MODE Select mode of /etc/localtime initialization\n"
413 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
414 " host, try-guest, try-host\n"
415 " -j Equivalent to --link-journal=try-guest\n\n"
417 " --bind=PATH[:PATH[:OPTIONS]]\n"
418 " Bind mount a file or directory from the host into\n"
420 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
421 " Similar, but creates a read-only bind mount\n"
422 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
424 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
425 " --overlay=PATH[:PATH...]:PATH\n"
426 " Create an overlay mount from the host to \n"
428 " --overlay-ro=PATH[:PATH...]:PATH\n"
429 " Similar, but creates a read-only overlay mount\n"
430 " --bind-user=NAME Bind user from host to container\n\n"
431 "%3$sInput/Output:%4$s\n"
432 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
433 " set up for the container.\n"
434 " -P --pipe Equivalent to --console=pipe\n\n"
435 "%3$sCredentials:%4$s\n"
436 " --set-credential=ID:VALUE\n"
437 " Pass a credential with literal value to container.\n"
438 " --load-credential=ID:PATH\n"
439 " Load credential to pass to container from file or\n"
440 " AF_UNIX stream socket.\n"
441 "\nSee the %2$s for details.\n",
442 program_invocation_short_name
,
452 static int custom_mount_check_all(void) {
455 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
456 CustomMount
*m
= &arg_custom_mounts
[i
];
458 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
459 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_OFF
)
460 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
461 "--private-users-ownership=own may not be combined with custom root mounts.");
462 if (arg_uid_shift
== UID_INVALID
)
463 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
464 "--private-users with automatic UID shift may not be combined with custom root mounts.");
471 static int detect_unified_cgroup_hierarchy_from_environment(void) {
472 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
475 /* Allow the user to control whether the unified hierarchy is used */
479 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
480 var
= "UNIFIED_CGROUP_HIERARCHY";
485 r
= parse_boolean(e
);
487 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
489 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
491 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
497 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
500 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
501 * in the image actually supports. */
502 r
= cg_all_unified();
504 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
506 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
507 * routine only detects 231, so we'll have a false negative here for 230. */
508 r
= systemd_installation_has_version(directory
, 230);
510 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
512 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
514 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
515 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
516 /* Mixed cgroup hierarchy support was added in 233 */
517 r
= systemd_installation_has_version(directory
, 233);
519 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
521 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
523 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
525 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
527 log_debug("Using %s hierarchy for container.",
528 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
529 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
534 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
539 _cleanup_free_
char *t
= NULL
;
541 r
= extract_first_word(&spec
, &t
, ",", 0);
543 return log_error_errno(r
, "Failed to parse capability %s.", t
);
547 if (streq(t
, "help")) {
548 for (int i
= 0; i
< capability_list_length(); i
++) {
551 name
= capability_to_name(i
);
562 r
= capability_from_name(t
);
564 return log_error_errno(r
, "Failed to parse capability %s.", t
);
571 return 1; /* continue */
574 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
577 r
= getenv_bool(name
);
581 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
583 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
584 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
588 static int parse_mount_settings_env(void) {
592 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
593 if (r
< 0 && r
!= -ENXIO
)
594 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
596 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
598 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
599 if (streq_ptr(e
, "network"))
600 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
603 r
= parse_boolean(e
);
605 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
607 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
608 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
614 static int parse_environment(void) {
618 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
621 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
624 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
627 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
631 r
= parse_mount_settings_env();
635 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
636 * even if it is supported. If not supported, it has no effect. */
637 if (!cg_ns_supported())
638 arg_use_cgns
= false;
640 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
643 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
647 arg_use_cgns
= r
> 0;
648 arg_settings_mask
|= SETTING_USE_CGNS
;
652 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
654 arg_container_service_name
= e
;
656 return detect_unified_cgroup_hierarchy_from_environment();
659 static int parse_argv(int argc
, char *argv
[]) {
666 ARG_AMBIENT_CAPABILITY
,
678 ARG_NETWORK_INTERFACE
,
683 ARG_NETWORK_VETH_EXTRA
,
684 ARG_NETWORK_NAMESPACE_PATH
,
694 ARG_PRIVATE_USERS_CHOWN
,
695 ARG_PRIVATE_USERS_OWNERSHIP
,
700 ARG_SYSTEM_CALL_FILTER
,
703 ARG_NO_NEW_PRIVILEGES
,
704 ARG_OOM_SCORE_ADJUST
,
717 static const struct option options
[] = {
718 { "help", no_argument
, NULL
, 'h' },
719 { "version", no_argument
, NULL
, ARG_VERSION
},
720 { "directory", required_argument
, NULL
, 'D' },
721 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
722 { "ephemeral", no_argument
, NULL
, 'x' },
723 { "user", required_argument
, NULL
, 'u' },
724 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
725 { "as-pid2", no_argument
, NULL
, 'a' },
726 { "boot", no_argument
, NULL
, 'b' },
727 { "uuid", required_argument
, NULL
, ARG_UUID
},
728 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
729 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
730 { "ambient-capability", required_argument
, NULL
, ARG_AMBIENT_CAPABILITY
},
731 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
732 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
733 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
734 { "bind", required_argument
, NULL
, ARG_BIND
},
735 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
736 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
737 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
738 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
739 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
740 { "machine", required_argument
, NULL
, 'M' },
741 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
742 { "slice", required_argument
, NULL
, 'S' },
743 { "setenv", required_argument
, NULL
, 'E' },
744 { "selinux-context", required_argument
, NULL
, 'Z' },
745 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
746 { "quiet", no_argument
, NULL
, 'q' },
747 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
748 { "register", required_argument
, NULL
, ARG_REGISTER
},
749 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
750 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
751 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
752 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
753 { "network-veth", no_argument
, NULL
, 'n' },
754 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
755 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
756 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
757 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
758 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
759 { "image", required_argument
, NULL
, 'i' },
760 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
761 { "port", required_argument
, NULL
, 'p' },
762 { "property", required_argument
, NULL
, ARG_PROPERTY
},
763 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
764 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
}, /* obsolete */
765 { "private-users-ownership",required_argument
, NULL
, ARG_PRIVATE_USERS_OWNERSHIP
},
766 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
767 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
768 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
769 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
770 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
771 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
772 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
773 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
774 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
775 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
776 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
777 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
778 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
779 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
780 { "console", required_argument
, NULL
, ARG_CONSOLE
},
781 { "pipe", no_argument
, NULL
, ARG_PIPE
},
782 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
783 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
784 { "set-credential", required_argument
, NULL
, ARG_SET_CREDENTIAL
},
785 { "load-credential", required_argument
, NULL
, ARG_LOAD_CREDENTIAL
},
786 { "bind-user", required_argument
, NULL
, ARG_BIND_USER
},
791 uint64_t plus
= 0, minus
= 0;
792 bool mask_all_settings
= false, mask_no_settings
= false;
797 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
807 r
= parse_path_argument(optarg
, false, &arg_directory
);
811 arg_settings_mask
|= SETTING_DIRECTORY
;
815 r
= parse_path_argument(optarg
, false, &arg_template
);
819 arg_settings_mask
|= SETTING_DIRECTORY
;
823 r
= parse_path_argument(optarg
, false, &arg_image
);
827 arg_settings_mask
|= SETTING_DIRECTORY
;
831 r
= parse_path_argument(optarg
, false, &arg_oci_bundle
);
838 arg_ephemeral
= true;
839 arg_settings_mask
|= SETTING_EPHEMERAL
;
843 r
= free_and_strdup(&arg_user
, optarg
);
847 arg_settings_mask
|= SETTING_USER
;
850 case ARG_NETWORK_ZONE
: {
853 j
= strjoin("vz-", optarg
);
857 if (!ifname_valid(j
)) {
858 log_error("Network zone name not valid: %s", j
);
863 free_and_replace(arg_network_zone
, j
);
865 arg_network_veth
= true;
866 arg_private_network
= true;
867 arg_settings_mask
|= SETTING_NETWORK
;
871 case ARG_NETWORK_BRIDGE
:
873 if (!ifname_valid(optarg
))
874 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
875 "Bridge interface name not valid: %s", optarg
);
877 r
= free_and_strdup(&arg_network_bridge
, optarg
);
883 arg_network_veth
= true;
884 arg_private_network
= true;
885 arg_settings_mask
|= SETTING_NETWORK
;
888 case ARG_NETWORK_VETH_EXTRA
:
889 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
891 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
893 arg_private_network
= true;
894 arg_settings_mask
|= SETTING_NETWORK
;
897 case ARG_NETWORK_INTERFACE
:
898 if (!ifname_valid(optarg
))
899 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
900 "Network interface name not valid: %s", optarg
);
902 r
= test_network_interface_initialized(optarg
);
906 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
909 arg_private_network
= true;
910 arg_settings_mask
|= SETTING_NETWORK
;
913 case ARG_NETWORK_MACVLAN
:
915 if (!ifname_valid(optarg
))
916 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
917 "MACVLAN network interface name not valid: %s", optarg
);
919 r
= test_network_interface_initialized(optarg
);
923 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
926 arg_private_network
= true;
927 arg_settings_mask
|= SETTING_NETWORK
;
930 case ARG_NETWORK_IPVLAN
:
932 if (!ifname_valid(optarg
))
933 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
934 "IPVLAN network interface name not valid: %s", optarg
);
936 r
= test_network_interface_initialized(optarg
);
940 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
944 case ARG_PRIVATE_NETWORK
:
945 arg_private_network
= true;
946 arg_settings_mask
|= SETTING_NETWORK
;
949 case ARG_NETWORK_NAMESPACE_PATH
:
950 r
= parse_path_argument(optarg
, false, &arg_network_namespace_path
);
954 arg_settings_mask
|= SETTING_NETWORK
;
958 if (arg_start_mode
== START_PID2
)
959 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
960 "--boot and --as-pid2 may not be combined.");
962 arg_start_mode
= START_BOOT
;
963 arg_settings_mask
|= SETTING_START_MODE
;
967 if (arg_start_mode
== START_BOOT
)
968 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
969 "--boot and --as-pid2 may not be combined.");
971 arg_start_mode
= START_PID2
;
972 arg_settings_mask
|= SETTING_START_MODE
;
976 r
= sd_id128_from_string(optarg
, &arg_uuid
);
978 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
980 if (sd_id128_is_null(arg_uuid
))
981 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
982 "Machine UUID may not be all zeroes.");
984 arg_settings_mask
|= SETTING_MACHINE_ID
;
988 _cleanup_free_
char *mangled
= NULL
;
990 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
994 free_and_replace(arg_slice
, mangled
);
995 arg_settings_mask
|= SETTING_SLICE
;
1000 if (isempty(optarg
))
1001 arg_machine
= mfree(arg_machine
);
1003 if (!hostname_is_valid(optarg
, 0))
1004 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1005 "Invalid machine name: %s", optarg
);
1007 r
= free_and_strdup(&arg_machine
, optarg
);
1014 if (isempty(optarg
))
1015 arg_hostname
= mfree(arg_hostname
);
1017 if (!hostname_is_valid(optarg
, 0))
1018 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1019 "Invalid hostname: %s", optarg
);
1021 r
= free_and_strdup(&arg_hostname
, optarg
);
1026 arg_settings_mask
|= SETTING_HOSTNAME
;
1030 arg_selinux_context
= optarg
;
1034 arg_selinux_apifs_context
= optarg
;
1038 arg_read_only
= true;
1039 arg_settings_mask
|= SETTING_READ_ONLY
;
1042 case ARG_AMBIENT_CAPABILITY
: {
1044 r
= parse_capability_spec(optarg
, &m
);
1047 arg_caps_ambient
|= m
;
1048 arg_settings_mask
|= SETTING_CAPABILITY
;
1051 case ARG_CAPABILITY
:
1052 case ARG_DROP_CAPABILITY
: {
1054 r
= parse_capability_spec(optarg
, &m
);
1058 if (c
== ARG_CAPABILITY
)
1062 arg_settings_mask
|= SETTING_CAPABILITY
;
1065 case ARG_NO_NEW_PRIVILEGES
:
1066 r
= parse_boolean(optarg
);
1068 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1070 arg_no_new_privileges
= r
;
1071 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1075 arg_link_journal
= LINK_GUEST
;
1076 arg_link_journal_try
= true;
1077 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1080 case ARG_LINK_JOURNAL
:
1081 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1083 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1085 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1090 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1092 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1094 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1098 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1100 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1102 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1106 case ARG_OVERLAY_RO
:
1107 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1108 if (r
== -EADDRNOTAVAIL
)
1109 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1111 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1113 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1116 case ARG_INACCESSIBLE
:
1117 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1119 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1121 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1125 r
= strv_env_replace_strdup_passthrough(&arg_setenv
, optarg
);
1127 return log_error_errno(r
, "Cannot assign environment variable %s: %m", optarg
);
1129 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1136 case ARG_SHARE_SYSTEM
:
1137 /* We don't officially support this anymore, except for compat reasons. People should use the
1138 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1139 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1140 arg_clone_ns_flags
= 0;
1144 r
= parse_boolean(optarg
);
1146 log_error("Failed to parse --register= argument: %s", optarg
);
1154 arg_keep_unit
= true;
1157 case ARG_PERSONALITY
:
1159 arg_personality
= personality_from_string(optarg
);
1160 if (arg_personality
== PERSONALITY_INVALID
)
1161 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1162 "Unknown or unsupported personality '%s'.", optarg
);
1164 arg_settings_mask
|= SETTING_PERSONALITY
;
1170 arg_volatile_mode
= VOLATILE_YES
;
1171 else if (streq(optarg
, "help")) {
1172 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1177 m
= volatile_mode_from_string(optarg
);
1179 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1180 "Failed to parse --volatile= argument: %s", optarg
);
1182 arg_volatile_mode
= m
;
1185 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1189 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1191 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1193 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1195 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1199 if (strv_extend(&arg_property
, optarg
) < 0)
1204 case ARG_PRIVATE_USERS
: {
1209 else if (!in_charset(optarg
, DIGITS
))
1210 /* do *not* parse numbers as booleans */
1211 boolean
= parse_boolean(optarg
);
1216 /* no: User namespacing off */
1217 arg_userns_mode
= USER_NAMESPACE_NO
;
1218 arg_uid_shift
= UID_INVALID
;
1219 arg_uid_range
= UINT32_C(0x10000);
1220 } else if (boolean
> 0) {
1221 /* yes: User namespacing on, UID range is read from root dir */
1222 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1223 arg_uid_shift
= UID_INVALID
;
1224 arg_uid_range
= UINT32_C(0x10000);
1225 } else if (streq(optarg
, "pick")) {
1226 /* pick: User namespacing on, UID range is picked randomly */
1227 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1228 * implied by USER_NAMESPACE_PICK
1230 arg_uid_shift
= UID_INVALID
;
1231 arg_uid_range
= UINT32_C(0x10000);
1233 } else if (streq(optarg
, "identity")) {
1234 /* identitiy: User namespaces on, UID range is map the 0…0xFFFF range to
1235 * itself, i.e. we don't actually map anything, but do take benefit of
1236 * isolation of capability sets. */
1237 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1239 arg_uid_range
= UINT32_C(0x10000);
1241 _cleanup_free_
char *buffer
= NULL
;
1242 const char *range
, *shift
;
1244 /* anything else: User namespacing on, UID range is explicitly configured */
1246 range
= strchr(optarg
, ':');
1248 buffer
= strndup(optarg
, range
- optarg
);
1254 r
= safe_atou32(range
, &arg_uid_range
);
1256 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1260 r
= parse_uid(shift
, &arg_uid_shift
);
1262 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1264 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1266 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
1267 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID range cannot be empty or go beyond " UID_FMT
".", UID_INVALID
);
1270 arg_settings_mask
|= SETTING_USERNS
;
1275 if (userns_supported()) {
1276 arg_userns_mode
= USER_NAMESPACE_PICK
; /* Note that arg_userns_ownership is
1277 * implied by USER_NAMESPACE_PICK
1279 arg_uid_shift
= UID_INVALID
;
1280 arg_uid_range
= UINT32_C(0x10000);
1282 arg_settings_mask
|= SETTING_USERNS
;
1287 case ARG_PRIVATE_USERS_CHOWN
:
1288 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
1290 arg_settings_mask
|= SETTING_USERNS
;
1293 case ARG_PRIVATE_USERS_OWNERSHIP
:
1294 if (streq(optarg
, "help")) {
1295 DUMP_STRING_TABLE(user_namespace_ownership
, UserNamespaceOwnership
, _USER_NAMESPACE_OWNERSHIP_MAX
);
1299 arg_userns_ownership
= user_namespace_ownership_from_string(optarg
);
1300 if (arg_userns_ownership
< 0)
1301 return log_error_errno(arg_userns_ownership
, "Cannot parse --user-namespace-ownership= value: %s", optarg
);
1303 arg_settings_mask
|= SETTING_USERNS
;
1306 case ARG_KILL_SIGNAL
:
1307 if (streq(optarg
, "help")) {
1308 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1312 arg_kill_signal
= signal_from_string(optarg
);
1313 if (arg_kill_signal
< 0)
1314 return log_error_errno(arg_kill_signal
, "Cannot parse signal: %s", optarg
);
1316 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1321 /* no → do not read files
1322 * yes → read files, do not override cmdline, trust only subset
1323 * override → read files, override cmdline, trust only subset
1324 * trusted → read files, do not override cmdline, trust all
1327 r
= parse_boolean(optarg
);
1329 if (streq(optarg
, "trusted")) {
1330 mask_all_settings
= false;
1331 mask_no_settings
= false;
1332 arg_settings_trusted
= true;
1334 } else if (streq(optarg
, "override")) {
1335 mask_all_settings
= false;
1336 mask_no_settings
= true;
1337 arg_settings_trusted
= -1;
1339 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1342 mask_all_settings
= false;
1343 mask_no_settings
= false;
1344 arg_settings_trusted
= -1;
1347 mask_all_settings
= true;
1348 mask_no_settings
= false;
1349 arg_settings_trusted
= false;
1355 if (!path_is_absolute(optarg
))
1356 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1357 "Working directory %s is not an absolute path.", optarg
);
1359 r
= free_and_strdup(&arg_chdir
, optarg
);
1363 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1366 case ARG_PIVOT_ROOT
:
1367 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1369 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1371 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1374 case ARG_NOTIFY_READY
:
1375 r
= parse_boolean(optarg
);
1377 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1378 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1379 arg_notify_ready
= r
;
1380 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1383 case ARG_ROOT_HASH
: {
1384 _cleanup_free_
void *k
= NULL
;
1387 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1389 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1390 if (l
< sizeof(sd_id128_t
))
1391 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1393 free_and_replace(arg_verity_settings
.root_hash
, k
);
1394 arg_verity_settings
.root_hash_size
= l
;
1398 case ARG_ROOT_HASH_SIG
: {
1403 if ((value
= startswith(optarg
, "base64:"))) {
1404 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1406 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1409 r
= read_full_file(optarg
, (char**) &p
, &l
);
1411 return log_error_errno(r
, "Failed parse root hash signature file '%s': %m", optarg
);
1414 free_and_replace(arg_verity_settings
.root_hash_sig
, p
);
1415 arg_verity_settings
.root_hash_sig_size
= l
;
1419 case ARG_VERITY_DATA
:
1420 r
= parse_path_argument(optarg
, false, &arg_verity_settings
.data_path
);
1425 case ARG_SYSTEM_CALL_FILTER
: {
1429 negative
= optarg
[0] == '~';
1430 items
= negative
? optarg
+ 1 : optarg
;
1433 _cleanup_free_
char *word
= NULL
;
1435 r
= extract_first_word(&items
, &word
, NULL
, 0);
1441 return log_error_errno(r
, "Failed to parse system call filter: %m");
1444 r
= strv_extend(&arg_syscall_deny_list
, word
);
1446 r
= strv_extend(&arg_syscall_allow_list
, word
);
1451 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1457 _cleanup_free_
char *name
= NULL
;
1460 if (streq(optarg
, "help")) {
1461 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1465 eq
= strchr(optarg
, '=');
1467 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1468 "--rlimit= expects an '=' assignment.");
1470 name
= strndup(optarg
, eq
- optarg
);
1474 rl
= rlimit_from_string_harder(name
);
1476 return log_error_errno(rl
, "Unknown resource limit: %s", name
);
1478 if (!arg_rlimit
[rl
]) {
1479 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1480 if (!arg_rlimit
[rl
])
1484 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1486 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1488 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1492 case ARG_OOM_SCORE_ADJUST
:
1493 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1495 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1497 arg_oom_score_adjust_set
= true;
1498 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1501 case ARG_CPU_AFFINITY
: {
1504 r
= parse_cpu_set(optarg
, &cpuset
);
1506 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1508 cpu_set_reset(&arg_cpu_set
);
1509 arg_cpu_set
= cpuset
;
1510 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1514 case ARG_RESOLV_CONF
:
1515 if (streq(optarg
, "help")) {
1516 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1520 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1521 if (arg_resolv_conf
< 0)
1522 return log_error_errno(arg_resolv_conf
,
1523 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1525 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1529 if (streq(optarg
, "help")) {
1530 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1534 arg_timezone
= timezone_mode_from_string(optarg
);
1535 if (arg_timezone
< 0)
1536 return log_error_errno(arg_timezone
,
1537 "Failed to parse /etc/localtime mode: %s", optarg
);
1539 arg_settings_mask
|= SETTING_TIMEZONE
;
1543 r
= handle_arg_console(optarg
);
1550 r
= handle_arg_console("pipe");
1556 arg_pager_flags
|= PAGER_DISABLE
;
1559 case ARG_SET_CREDENTIAL
: {
1560 _cleanup_free_
char *word
= NULL
, *data
= NULL
;
1561 const char *p
= optarg
;
1565 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1569 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1571 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1573 if (!credential_name_valid(word
))
1574 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1576 for (size_t i
= 0; i
< arg_n_credentials
; i
++)
1577 if (streq(arg_credentials
[i
].id
, word
))
1578 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1580 l
= cunescape(p
, UNESCAPE_ACCEPT_NUL
, &data
);
1582 return log_error_errno(l
, "Failed to unescape credential data: %s", p
);
1584 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1588 a
[arg_n_credentials
++] = (Credential
) {
1589 .id
= TAKE_PTR(word
),
1590 .data
= TAKE_PTR(data
),
1594 arg_credentials
= a
;
1596 arg_settings_mask
|= SETTING_CREDENTIALS
;
1600 case ARG_LOAD_CREDENTIAL
: {
1601 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
1602 _cleanup_(erase_and_freep
) char *data
= NULL
;
1603 _cleanup_free_
char *word
= NULL
, *j
= NULL
;
1604 const char *p
= optarg
;
1608 r
= extract_first_word(&p
, &word
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1612 return log_error_errno(r
, "Failed to parse --set-credential= parameter: %m");
1614 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Missing value for --set-credential=: %s", optarg
);
1616 if (!credential_name_valid(word
))
1617 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential name is not valid: %s", word
);
1619 for (i
= 0; i
< arg_n_credentials
; i
++)
1620 if (streq(arg_credentials
[i
].id
, word
))
1621 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Duplicate credential '%s', refusing.", word
);
1623 if (path_is_absolute(p
))
1624 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
1628 r
= get_credentials_dir(&e
);
1630 return log_error_errno(r
, "Credential not available (no credentials passed at all): %s", word
);
1632 j
= path_join(e
, p
);
1637 r
= read_full_file_full(AT_FDCWD
, j
?: p
, UINT64_MAX
, SIZE_MAX
,
1642 return log_error_errno(r
, "Failed to read credential '%s': %m", j
?: p
);
1644 a
= reallocarray(arg_credentials
, arg_n_credentials
+ 1, sizeof(Credential
));
1648 a
[arg_n_credentials
++] = (Credential
) {
1649 .id
= TAKE_PTR(word
),
1650 .data
= TAKE_PTR(data
),
1654 arg_credentials
= a
;
1656 arg_settings_mask
|= SETTING_CREDENTIALS
;
1661 if (!valid_user_group_name(optarg
, 0))
1662 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid user name to bind: %s", optarg
);
1664 if (strv_extend(&arg_bind_user
, optarg
) < 0)
1667 arg_settings_mask
|= SETTING_BIND_USER
;
1674 assert_not_reached();
1677 if (argc
> optind
) {
1678 strv_free(arg_parameters
);
1679 arg_parameters
= strv_copy(argv
+ optind
);
1680 if (!arg_parameters
)
1683 arg_settings_mask
|= SETTING_START_MODE
;
1686 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1687 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1688 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1689 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1691 arg_directory
= TAKE_PTR(arg_template
);
1693 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1695 /* Make sure to parse environment before we reset the settings mask below */
1696 r
= parse_environment();
1700 /* Load all settings from .nspawn files */
1701 if (mask_no_settings
)
1702 arg_settings_mask
= 0;
1704 /* Don't load any settings from .nspawn files */
1705 if (mask_all_settings
)
1706 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1711 static int verify_arguments(void) {
1714 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1715 /* If we are running the stub init in the container, we don't need to look at what the init
1716 * in the container supports, because we are not using it. Let's immediately pick the right
1717 * setting based on the host system configuration.
1719 * We only do this, if the user didn't use an environment variable to override the detection.
1722 r
= cg_all_unified();
1724 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1726 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1727 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1728 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1730 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1733 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1734 arg_mount_settings
|= MOUNT_USE_USERNS
;
1736 if (arg_private_network
)
1737 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1739 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1740 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1741 arg_register
= false;
1742 if (arg_start_mode
!= START_PID1
)
1743 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1746 if (arg_userns_ownership
< 0)
1747 arg_userns_ownership
=
1748 arg_userns_mode
== USER_NAMESPACE_PICK
? USER_NAMESPACE_OWNERSHIP_AUTO
:
1749 USER_NAMESPACE_OWNERSHIP_OFF
;
1751 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1752 arg_kill_signal
= SIGRTMIN
+3;
1754 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 */
1755 arg_read_only
= true;
1757 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1758 arg_read_only
= true;
1760 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1761 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1762 * The latter is not technically a user session, but we don't need to labour the point. */
1763 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1765 if (arg_directory
&& arg_image
)
1766 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1768 if (arg_template
&& arg_image
)
1769 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1771 if (arg_template
&& !(arg_directory
|| arg_machine
))
1772 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1774 if (arg_ephemeral
&& arg_template
)
1775 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1777 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1778 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1780 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1781 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1783 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_read_only
)
1784 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1785 "--read-only and --private-users-ownership=chown may not be combined.");
1787 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1788 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1789 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1790 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_CHOWN
&& arg_volatile_mode
!= VOLATILE_NO
)
1791 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-ownership=chown may not be combined.");
1793 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1794 * we need to error out, to avoid conflicts between different network options. */
1795 if (arg_network_namespace_path
&&
1796 (arg_network_interfaces
|| arg_network_macvlan
||
1797 arg_network_ipvlan
|| arg_network_veth_extra
||
1798 arg_network_bridge
|| arg_network_zone
||
1800 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1802 if (arg_network_bridge
&& arg_network_zone
)
1803 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1804 "--network-bridge= and --network-zone= may not be combined.");
1806 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1807 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1809 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1810 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1812 if (arg_expose_ports
&& !arg_private_network
)
1813 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1815 if (arg_caps_ambient
) {
1816 if (arg_caps_ambient
== UINT64_MAX
)
1817 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= does not support the value all.");
1819 if ((arg_caps_ambient
& arg_caps_retain
) != arg_caps_ambient
)
1820 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not fully covered by Capability= setting.");
1822 if (arg_start_mode
== START_BOOT
)
1823 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "AmbientCapability= setting is not useful for boot mode.");
1826 if (arg_userns_mode
== USER_NAMESPACE_NO
&& !strv_isempty(arg_bind_user
))
1827 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--bind-user= requires --private-users");
1829 /* Drop duplicate --bind-user= entries */
1830 strv_uniq(arg_bind_user
);
1832 r
= custom_mount_check_all();
1839 int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1842 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1845 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1848 if (uid
!= UID_INVALID
) {
1849 uid
+= arg_uid_shift
;
1851 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1855 if (gid
!= GID_INVALID
) {
1856 gid
+= (gid_t
) arg_uid_shift
;
1858 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1862 if (lchown(p
, uid
, gid
) < 0)
1868 int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1872 q
= prefix_roota(root
, path
);
1873 r
= mkdir_errno_wrapper(q
, mode
);
1879 return userns_lchown(q
, uid
, gid
);
1882 static const char *timezone_from_path(const char *path
) {
1883 return PATH_STARTSWITH_SET(
1885 "../usr/share/zoneinfo/",
1886 "/usr/share/zoneinfo/");
1889 static bool etc_writable(void) {
1890 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1893 static int setup_timezone(const char *dest
) {
1894 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1895 const char *where
, *check
;
1901 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1902 r
= readlink_malloc("/etc/localtime", &p
);
1903 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1904 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1905 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1906 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1908 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1909 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1913 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1916 } else if (arg_timezone
== TIMEZONE_AUTO
)
1917 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1923 if (m
== TIMEZONE_OFF
)
1926 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1928 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1932 where
= strjoina(etc
, "/localtime");
1936 case TIMEZONE_DELETE
:
1937 if (unlink(where
) < 0)
1938 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1942 case TIMEZONE_SYMLINK
: {
1943 _cleanup_free_
char *q
= NULL
;
1944 const char *z
, *what
;
1946 z
= timezone_from_path(p
);
1948 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1952 r
= readlink_malloc(where
, &q
);
1953 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1954 return 0; /* Already pointing to the right place? Then do nothing .. */
1956 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1957 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1959 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1961 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1962 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1963 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1967 what
= strjoina("../usr/share/zoneinfo/", z
);
1968 if (symlink(what
, where
) < 0) {
1969 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1970 errno
, "Failed to correct timezone of container, ignoring: %m");
1980 case TIMEZONE_BIND
: {
1981 _cleanup_free_
char *resolved
= NULL
;
1984 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1986 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1990 if (found
== 0) /* missing? */
1991 (void) touch(resolved
);
1993 r
= mount_nofollow_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1995 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2001 /* If mounting failed, try to copy */
2002 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2004 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2005 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
2012 assert_not_reached();
2015 /* Fix permissions of the symlink or file copy we just created */
2016 r
= userns_lchown(where
, 0, 0);
2018 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
2023 static int have_resolv_conf(const char *path
) {
2026 if (access(path
, F_OK
) < 0) {
2027 if (errno
== ENOENT
)
2030 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
2036 static int resolved_listening(void) {
2037 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2038 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
2039 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
2042 /* Check if resolved is listening */
2044 r
= sd_bus_open_system(&bus
);
2046 return log_debug_errno(r
, "Failed to open system bus: %m");
2048 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
2050 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2054 r
= sd_bus_get_property_string(bus
,
2055 "org.freedesktop.resolve1",
2056 "/org/freedesktop/resolve1",
2057 "org.freedesktop.resolve1.Manager",
2060 &dns_stub_listener_mode
);
2062 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
2064 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
2067 static int setup_resolv_conf(const char *dest
) {
2068 _cleanup_free_
char *etc
= NULL
;
2069 const char *where
, *what
;
2075 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
2076 if (arg_private_network
)
2077 m
= RESOLV_CONF_OFF
;
2078 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
2079 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
2080 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2081 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
2083 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
2086 m
= arg_resolv_conf
;
2088 if (m
== RESOLV_CONF_OFF
)
2091 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
2093 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
2097 where
= strjoina(etc
, "/resolv.conf");
2099 if (m
== RESOLV_CONF_DELETE
) {
2100 if (unlink(where
) < 0)
2101 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
2106 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
2107 what
= PRIVATE_STATIC_RESOLV_CONF
;
2108 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
2109 what
= PRIVATE_UPLINK_RESOLV_CONF
;
2110 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
2111 what
= PRIVATE_STUB_RESOLV_CONF
;
2113 what
= "/etc/resolv.conf";
2115 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
2116 _cleanup_free_
char *resolved
= NULL
;
2119 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
2121 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2125 if (found
== 0) /* missing? */
2126 (void) touch(resolved
);
2128 r
= mount_nofollow_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
2130 return mount_nofollow_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
2132 /* If that didn't work, let's copy the file */
2135 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
2136 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
2138 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
2140 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2141 * resolved or something similar runs inside and the symlink points there.
2143 * If the disk image is read-only, there's also no point in complaining.
2145 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
2146 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2147 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
2151 r
= userns_lchown(where
, 0, 0);
2153 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
2158 static int setup_boot_id(void) {
2159 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2160 _cleanup_free_
char *path
= NULL
;
2161 sd_id128_t rnd
= SD_ID128_NULL
;
2165 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2167 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
2169 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
2171 r
= sd_id128_randomize(&rnd
);
2173 return log_error_errno(r
, "Failed to generate random boot id: %m");
2175 r
= id128_write(path
, ID128_UUID
, rnd
, false);
2177 return log_error_errno(r
, "Failed to write boot id: %m");
2179 from
= TAKE_PTR(path
);
2180 to
= "/proc/sys/kernel/random/boot_id";
2182 r
= mount_nofollow_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2186 return mount_nofollow_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2189 static int copy_devnodes(const char *dest
) {
2190 static const char devnodes
[] =
2199 _cleanup_umask_ mode_t u
;
2207 /* Create /dev/net, so that we can create /dev/net/tun in it */
2208 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2209 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2211 NULSTR_FOREACH(d
, devnodes
) {
2212 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2215 from
= path_join("/dev/", d
);
2219 to
= path_join(dest
, from
);
2223 if (stat(from
, &st
) < 0) {
2225 if (errno
!= ENOENT
)
2226 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2228 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2229 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2230 "%s is not a char or block device, cannot copy.", from
);
2232 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2234 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2235 /* Explicitly warn the user when /dev is already populated. */
2236 if (errno
== EEXIST
)
2237 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
);
2239 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2241 /* Some systems abusively restrict mknod but allow bind mounts. */
2244 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2245 r
= mount_nofollow_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2247 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2250 r
= userns_lchown(to
, 0, 0);
2252 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2254 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2258 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2260 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2262 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2265 prefixed
= path_join(dest
, sl
);
2269 t
= path_join("..", d
);
2273 if (symlink(t
, prefixed
) < 0)
2274 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2281 static int make_extra_nodes(const char *dest
) {
2282 _cleanup_umask_ mode_t u
;
2288 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2289 _cleanup_free_
char *path
= NULL
;
2290 DeviceNode
*n
= arg_extra_nodes
+ i
;
2292 path
= path_join(dest
, n
->path
);
2296 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2297 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2299 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2301 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2307 static int setup_pts(const char *dest
) {
2308 _cleanup_free_
char *options
= NULL
;
2313 if (arg_selinux_apifs_context
)
2314 (void) asprintf(&options
,
2315 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2316 arg_uid_shift
+ TTY_GID
,
2317 arg_selinux_apifs_context
);
2320 (void) asprintf(&options
,
2321 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2322 arg_uid_shift
+ TTY_GID
);
2327 /* Mount /dev/pts itself */
2328 p
= prefix_roota(dest
, "/dev/pts");
2329 r
= mkdir_errno_wrapper(p
, 0755);
2331 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2333 r
= mount_nofollow_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2336 r
= userns_lchown(p
, 0, 0);
2338 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2340 /* Create /dev/ptmx symlink */
2341 p
= prefix_roota(dest
, "/dev/ptmx");
2342 if (symlink("pts/ptmx", p
) < 0)
2343 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2344 r
= userns_lchown(p
, 0, 0);
2346 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2348 /* And fix /dev/pts/ptmx ownership */
2349 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2350 r
= userns_lchown(p
, 0, 0);
2352 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2357 static int setup_stdio_as_dev_console(void) {
2358 _cleanup_close_
int terminal
= -1;
2361 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2362 * explicitly, if we are configured to. */
2363 terminal
= open_terminal("/dev/console", O_RDWR
|O_NOCTTY
);
2365 return log_error_errno(terminal
, "Failed to open console: %m");
2367 /* Make sure we can continue logging to the original stderr, even if
2368 * stderr points elsewhere now */
2369 r
= log_dup_console();
2371 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2373 /* invalidates 'terminal' on success and failure */
2374 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2377 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2382 static int setup_dev_console(const char *console
) {
2383 _cleanup_free_
char *p
= NULL
;
2386 /* Create /dev/console symlink */
2387 r
= path_make_relative("/dev", console
, &p
);
2389 return log_error_errno(r
, "Failed to create relative path: %m");
2391 if (symlink(p
, "/dev/console") < 0)
2392 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2397 static int setup_keyring(void) {
2398 key_serial_t keyring
;
2400 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2401 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2402 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2403 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2404 * into the container. */
2406 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2407 if (keyring
== -1) {
2408 if (errno
== ENOSYS
)
2409 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2410 else if (ERRNO_IS_PRIVILEGE(errno
))
2411 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2413 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2419 static int setup_credentials(const char *root
) {
2423 if (arg_n_credentials
<= 0)
2426 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2428 return log_error_errno(r
, "Failed to create /run/host: %m");
2430 r
= userns_mkdir(root
, "/run/host/credentials", 0700, 0, 0);
2432 return log_error_errno(r
, "Failed to create /run/host/credentials: %m");
2434 q
= prefix_roota(root
, "/run/host/credentials");
2435 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, "ramfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0700");
2439 for (size_t i
= 0; i
< arg_n_credentials
; i
++) {
2440 _cleanup_free_
char *j
= NULL
;
2441 _cleanup_close_
int fd
= -1;
2443 j
= path_join(q
, arg_credentials
[i
].id
);
2447 fd
= open(j
, O_CREAT
|O_EXCL
|O_WRONLY
|O_CLOEXEC
|O_NOFOLLOW
, 0600);
2449 return log_error_errno(errno
, "Failed to create credential file %s: %m", j
);
2451 r
= loop_write(fd
, arg_credentials
[i
].data
, arg_credentials
[i
].size
, /* do_poll= */ false);
2453 return log_error_errno(r
, "Failed to write credential to file %s: %m", j
);
2455 if (fchmod(fd
, 0400) < 0)
2456 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", j
);
2458 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2459 if (fchown(fd
, arg_uid_shift
, arg_uid_shift
) < 0)
2460 return log_error_errno(errno
, "Failed to adjust ownership of %s: %m", j
);
2464 if (chmod(q
, 0500) < 0)
2465 return log_error_errno(errno
, "Failed to adjust access mode of %s: %m", q
);
2467 r
= userns_lchown(q
, 0, 0);
2471 /* Make both mount and superblock read-only now */
2472 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2476 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, "mode=0500");
2479 static int setup_kmsg(int kmsg_socket
) {
2480 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2481 _cleanup_free_
char *fifo
= NULL
;
2482 _cleanup_close_
int fd
= -1;
2483 _cleanup_umask_ mode_t u
;
2486 assert(kmsg_socket
>= 0);
2490 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2491 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2492 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2493 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2495 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2497 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2499 if (mkfifo(fifo
, 0600) < 0)
2500 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2502 from
= TAKE_PTR(fifo
);
2504 r
= mount_nofollow_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2508 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2510 return log_error_errno(errno
, "Failed to open fifo: %m");
2512 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2513 r
= send_one_fd(kmsg_socket
, fd
, 0);
2515 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2521 union in_addr_union address4
;
2522 union in_addr_union address6
;
2523 struct FirewallContext
*fw_ctx
;
2526 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2527 struct ExposeArgs
*args
= userdata
;
2533 expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET
, &args
->address4
);
2534 expose_port_execute(rtnl
, &args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &args
->address6
);
2538 static int setup_hostname(void) {
2541 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2544 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2546 return log_error_errno(r
, "Failed to set hostname: %m");
2551 static int setup_journal(const char *directory
) {
2552 _cleanup_free_
char *d
= NULL
;
2553 const char *dirname
, *p
, *q
;
2558 /* Don't link journals in ephemeral mode */
2562 if (arg_link_journal
== LINK_NO
)
2565 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2567 r
= sd_id128_get_machine(&this_id
);
2569 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2571 if (sd_id128_equal(arg_uuid
, this_id
)) {
2572 log_full(try ? LOG_WARNING
: LOG_ERR
,
2573 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid
));
2579 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2580 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2582 bool ignore
= r
== -EROFS
&& try;
2583 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2584 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2585 return ignore
? 0 : r
;
2589 p
= strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid
));
2590 q
= prefix_roota(directory
, p
);
2592 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2596 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2597 "%s: already a mount point, refusing to use for journal", p
);
2600 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2604 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2605 "%s: already a mount point, refusing to use for journal", q
);
2608 r
= readlink_and_make_absolute(p
, &d
);
2610 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2613 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2615 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2620 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2621 } else if (r
== -EINVAL
) {
2623 if (arg_link_journal
== LINK_GUEST
&&
2626 if (errno
== ENOTDIR
) {
2627 log_error("%s already exists and is neither a symlink nor a directory", p
);
2630 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2632 } else if (r
!= -ENOENT
)
2633 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2635 if (arg_link_journal
== LINK_GUEST
) {
2637 if (symlink(q
, p
) < 0) {
2639 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2642 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2645 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2647 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2651 if (arg_link_journal
== LINK_HOST
) {
2652 /* don't create parents here — if the host doesn't have
2653 * permanent journal set up, don't force it here */
2655 r
= mkdir_errno_wrapper(p
, 0755);
2656 if (r
< 0 && r
!= -EEXIST
) {
2658 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2661 return log_error_errno(r
, "Failed to create %s: %m", p
);
2664 } else if (access(p
, F_OK
) < 0)
2667 if (dir_is_empty(q
) == 0)
2668 log_warning("%s is not empty, proceeding anyway.", q
);
2670 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2672 return log_error_errno(r
, "Failed to create %s: %m", q
);
2674 r
= mount_nofollow_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2676 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2681 static int drop_capabilities(uid_t uid
) {
2682 CapabilityQuintet q
;
2684 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2685 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2686 * arg_caps_retain. */
2688 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2689 q
= arg_full_capabilities
;
2691 if (q
.bounding
== UINT64_MAX
)
2692 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2694 if (q
.effective
== UINT64_MAX
)
2695 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2697 if (q
.inheritable
== UINT64_MAX
)
2698 q
.inheritable
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2700 if (q
.permitted
== UINT64_MAX
)
2701 q
.permitted
= uid
== 0 ? q
.bounding
: arg_caps_ambient
;
2703 if (q
.ambient
== UINT64_MAX
&& ambient_capabilities_supported())
2704 q
.ambient
= arg_caps_ambient
;
2706 if (capability_quintet_mangle(&q
))
2707 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2710 q
= (CapabilityQuintet
) {
2711 .bounding
= arg_caps_retain
,
2712 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2713 .inheritable
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2714 .permitted
= uid
== 0 ? arg_caps_retain
: arg_caps_ambient
,
2715 .ambient
= ambient_capabilities_supported() ? arg_caps_ambient
: UINT64_MAX
,
2718 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2719 * in order to maintain the same behavior as systemd < 242. */
2720 if (capability_quintet_mangle(&q
))
2721 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2722 "Some capabilities will not be set because they are not in the current bounding set.");
2726 return capability_quintet_enforce(&q
);
2729 static int reset_audit_loginuid(void) {
2730 _cleanup_free_
char *p
= NULL
;
2733 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2736 r
= read_one_line_file("/proc/self/loginuid", &p
);
2740 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2742 /* Already reset? */
2743 if (streq(p
, "4294967295"))
2746 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2749 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2750 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2751 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2752 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2753 "using systemd-nspawn. Sleeping for 5s... (%m)");
2761 static int setup_propagate(const char *root
) {
2765 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2766 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2767 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2768 (void) mkdir_p(p
, 0600);
2770 r
= userns_mkdir(root
, "/run/host", 0755, 0, 0);
2772 return log_error_errno(r
, "Failed to create /run/host: %m");
2774 r
= userns_mkdir(root
, "/run/host/incoming", 0600, 0, 0);
2776 return log_error_errno(r
, "Failed to create /run/host/incoming: %m");
2778 q
= prefix_roota(root
, "/run/host/incoming");
2779 r
= mount_nofollow_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2783 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2787 /* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
2788 return mount_nofollow_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2791 static int setup_machine_id(const char *directory
) {
2792 const char *etc_machine_id
;
2796 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2797 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2798 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2799 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2800 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2801 * container behaves nicely). */
2803 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2805 r
= id128_read(etc_machine_id
, ID128_PLAIN_OR_UNINIT
, &id
);
2807 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2808 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2810 if (sd_id128_is_null(arg_uuid
)) {
2811 r
= sd_id128_randomize(&arg_uuid
);
2813 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2816 if (sd_id128_is_null(id
))
2817 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2818 "Machine ID in container image is zero, refusing.");
2826 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2831 if (arg_userns_mode
== USER_NAMESPACE_NO
|| arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_CHOWN
)
2834 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2835 if (r
== -EOPNOTSUPP
)
2836 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.");
2838 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2840 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2842 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2844 log_debug("Patched directory tree to match UID/GID range.");
2851 * < 0 : wait_for_terminate() failed to get the state of the
2852 * container, the container was terminated by a signal, or
2853 * failed for an unknown reason. No change is made to the
2854 * container argument.
2855 * > 0 : The program executed in the container terminated with an
2856 * error. The exit code of the program executed in the
2857 * container is returned. The container argument has been set
2858 * to CONTAINER_TERMINATED.
2859 * 0 : The container is being rebooted, has been shut down or exited
2860 * successfully. The container argument has been set to either
2861 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2863 * That is, success is indicated by a return value of zero, and an
2864 * error is indicated by a non-zero value.
2866 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2870 r
= wait_for_terminate(pid
, &status
);
2872 return log_warning_errno(r
, "Failed to wait for container: %m");
2874 switch (status
.si_code
) {
2877 if (status
.si_status
== 0)
2878 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2880 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2882 *container
= CONTAINER_TERMINATED
;
2883 return status
.si_status
;
2886 if (status
.si_status
== SIGINT
) {
2887 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2888 *container
= CONTAINER_TERMINATED
;
2891 } else if (status
.si_status
== SIGHUP
) {
2892 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2893 *container
= CONTAINER_REBOOTED
;
2899 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2900 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2903 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2904 "Container %s failed due to unknown reason.", arg_machine
);
2908 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2911 pid
= PTR_TO_PID(userdata
);
2913 if (kill(pid
, arg_kill_signal
) >= 0) {
2914 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2915 sd_event_source_set_userdata(s
, NULL
);
2920 sd_event_exit(sd_event_source_get_event(s
), 0);
2924 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2930 pid
= PTR_TO_PID(userdata
);
2935 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2936 return log_error_errno(errno
, "Failed to waitid(): %m");
2937 if (si
.si_pid
== 0) /* No pending children. */
2939 if (si
.si_pid
== pid
) {
2940 /* The main process we care for has exited. Return from
2941 * signal handler but leave the zombie. */
2942 sd_event_exit(sd_event_source_get_event(s
), 0);
2946 /* Reap all other children. */
2947 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2953 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2958 pid
= PTR_TO_PID(userdata
);
2960 if (arg_kill_signal
> 0) {
2961 log_info("Container termination requested. Attempting to halt container.");
2962 (void) kill(pid
, arg_kill_signal
);
2964 log_info("Container termination requested. Exiting.");
2965 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2971 static int determine_names(void) {
2974 if (arg_template
&& !arg_directory
&& arg_machine
) {
2976 /* If --template= was specified then we should not
2977 * search for a machine, but instead create a new one
2978 * in /var/lib/machine. */
2980 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2985 if (!arg_image
&& !arg_directory
) {
2987 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2989 r
= image_find(IMAGE_MACHINE
, arg_machine
, NULL
, &i
);
2991 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2993 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2995 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2996 r
= free_and_strdup(&arg_image
, i
->path
);
2998 r
= free_and_strdup(&arg_directory
, i
->path
);
3003 arg_read_only
= arg_read_only
|| i
->read_only
;
3005 r
= safe_getcwd(&arg_directory
);
3007 return log_error_errno(r
, "Failed to determine current directory: %m");
3010 if (!arg_directory
&& !arg_image
)
3011 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
3015 if (arg_directory
&& path_equal(arg_directory
, "/"))
3016 arg_machine
= gethostname_malloc();
3017 else if (arg_image
) {
3020 arg_machine
= strdup(basename(arg_image
));
3022 /* Truncate suffix if there is one */
3023 e
= endswith(arg_machine
, ".raw");
3027 arg_machine
= strdup(basename(arg_directory
));
3031 hostname_cleanup(arg_machine
);
3032 if (!hostname_is_valid(arg_machine
, 0))
3033 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
3035 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3036 * instances at once without manually having to specify -M each time. */
3038 if (strextendf(&arg_machine
, "-%016" PRIx64
, random_u64()) < 0)
3045 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
3054 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
3056 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
3058 return free_and_replace(*p
, chased
);
3061 static int determine_uid_shift(const char *directory
) {
3063 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
3068 if (arg_uid_shift
== UID_INVALID
) {
3071 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3073 if (stat(directory
, &st
) < 0)
3074 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
3076 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
3078 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
3079 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3080 "UID and GID base of %s don't match.", directory
);
3082 arg_uid_range
= UINT32_C(0x10000);
3084 if (arg_uid_shift
!= 0) {
3085 /* If the image is shifted already, then we'll fall back to classic chowning, for
3086 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3088 if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_AUTO
) {
3089 log_debug("UID base of %s is non-zero, not using UID mapping.", directory
);
3090 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3091 } else if (arg_userns_ownership
== USER_NAMESPACE_OWNERSHIP_MAP
)
3092 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
3093 "UID base of %s is not zero, UID mapping not supported.", directory
);
3097 if (!userns_shift_range_valid(arg_uid_shift
, arg_uid_range
))
3098 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "UID base too high for UID range.");
3103 static unsigned long effective_clone_ns_flags(void) {
3104 unsigned long flags
= arg_clone_ns_flags
;
3106 if (arg_private_network
)
3107 flags
|= CLONE_NEWNET
;
3109 flags
|= CLONE_NEWCGROUP
;
3110 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3111 flags
|= CLONE_NEWUSER
;
3116 static int patch_sysctl(void) {
3118 /* This table is inspired by runc's sysctl() function */
3119 static const struct {
3122 unsigned long clone_flags
;
3124 { "kernel.hostname", false, CLONE_NEWUTS
},
3125 { "kernel.domainname", false, CLONE_NEWUTS
},
3126 { "kernel.msgmax", false, CLONE_NEWIPC
},
3127 { "kernel.msgmnb", false, CLONE_NEWIPC
},
3128 { "kernel.msgmni", false, CLONE_NEWIPC
},
3129 { "kernel.sem", false, CLONE_NEWIPC
},
3130 { "kernel.shmall", false, CLONE_NEWIPC
},
3131 { "kernel.shmmax", false, CLONE_NEWIPC
},
3132 { "kernel.shmmni", false, CLONE_NEWIPC
},
3133 { "fs.mqueue.", true, CLONE_NEWIPC
},
3134 { "net.", true, CLONE_NEWNET
},
3137 unsigned long flags
;
3141 flags
= effective_clone_ns_flags();
3143 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
3147 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
3149 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
3152 if (safe_sysctl
[i
].prefix
)
3153 good
= startswith(*k
, safe_sysctl
[i
].key
);
3155 good
= streq(*k
, safe_sysctl
[i
].key
);
3162 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
3164 r
= sysctl_write(*k
, *v
);
3166 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
3172 static int inner_child(
3174 const char *directory
,
3178 int master_pty_socket
,
3180 char **os_release_pairs
) {
3182 _cleanup_free_
char *home
= NULL
;
3185 (char*) "PATH=" DEFAULT_PATH_COMPAT
,
3186 NULL
, /* container */
3191 NULL
, /* container_uuid */
3192 NULL
, /* LISTEN_FDS */
3193 NULL
, /* LISTEN_PID */
3194 NULL
, /* NOTIFY_SOCKET */
3195 NULL
, /* CREDENTIALS_DIRECTORY */
3198 const char *exec_target
;
3199 _cleanup_strv_free_
char **env_use
= NULL
;
3200 int r
, which_failed
;
3202 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3203 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3204 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3205 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3206 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3207 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3210 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3211 * unshare(). See below. */
3215 assert(kmsg_socket
>= 0);
3217 log_debug("Inner child is initializing.");
3219 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3220 /* Tell the parent, that it now can write the UID map. */
3221 (void) barrier_place(barrier
); /* #1 */
3223 /* Wait until the parent wrote the UID map */
3224 if (!barrier_place_and_sync(barrier
)) /* #2 */
3225 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3227 /* Become the new root user inside our namespace */
3228 r
= reset_uid_gid();
3230 return log_error_errno(r
, "Couldn't become new root: %m");
3232 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3233 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3234 * propagation, but simply create new peer groups for all our mounts). */
3235 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SHARED
|MS_REC
, NULL
);
3241 arg_mount_settings
| MOUNT_IN_USERNS
,
3243 arg_selinux_apifs_context
);
3247 if (!arg_network_namespace_path
&& arg_private_network
) {
3248 r
= unshare(CLONE_NEWNET
);
3250 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3252 /* Tell the parent that it can setup network interfaces. */
3253 (void) barrier_place(barrier
); /* #3 */
3256 r
= mount_sysfs(NULL
, arg_mount_settings
);
3260 /* Wait until we are cgroup-ified, so that we
3261 * can mount the right cgroup path writable */
3262 if (!barrier_place_and_sync(barrier
)) /* #4 */
3263 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3264 "Parent died too early");
3267 r
= unshare(CLONE_NEWCGROUP
);
3269 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3272 arg_unified_cgroup_hierarchy
,
3273 arg_userns_mode
!= USER_NAMESPACE_NO
,
3276 arg_selinux_apifs_context
,
3279 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3283 r
= setup_boot_id();
3287 r
= setup_kmsg(kmsg_socket
);
3290 kmsg_socket
= safe_close(kmsg_socket
);
3295 arg_n_custom_mounts
,
3297 arg_selinux_apifs_context
,
3298 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3303 return log_error_errno(errno
, "setsid() failed: %m");
3305 if (arg_private_network
)
3306 (void) loopback_setup();
3308 if (arg_expose_ports
) {
3309 r
= expose_port_send_rtnl(rtnl_socket
);
3312 rtnl_socket
= safe_close(rtnl_socket
);
3315 if (arg_console_mode
!= CONSOLE_PIPE
) {
3316 _cleanup_close_
int master
= -1;
3317 _cleanup_free_
char *console
= NULL
;
3319 /* Allocate a pty and make it available as /dev/console. */
3320 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3322 return log_error_errno(master
, "Failed to allocate a pty: %m");
3324 r
= setup_dev_console(console
);
3326 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3328 r
= send_one_fd(master_pty_socket
, master
, 0);
3330 return log_error_errno(r
, "Failed to send master fd: %m");
3331 master_pty_socket
= safe_close(master_pty_socket
);
3333 r
= setup_stdio_as_dev_console();
3342 if (arg_oom_score_adjust_set
) {
3343 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3345 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3348 if (arg_cpu_set
.set
)
3349 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3350 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3352 (void) setup_hostname();
3354 if (arg_personality
!= PERSONALITY_INVALID
) {
3355 r
= safe_personality(arg_personality
);
3357 return log_error_errno(r
, "personality() failed: %m");
3358 } else if (secondary
) {
3359 r
= safe_personality(PER_LINUX32
);
3361 return log_error_errno(r
, "personality() failed: %m");
3364 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3366 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3371 if (is_seccomp_available()) {
3373 r
= seccomp_load(arg_seccomp
);
3374 if (ERRNO_IS_SECCOMP_FATAL(r
))
3375 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3377 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3382 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3388 if (arg_selinux_context
)
3389 if (setexeccon(arg_selinux_context
) < 0)
3390 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3393 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3394 * if we need to later on. */
3395 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3396 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3398 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3399 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
, arg_console_mode
!= CONSOLE_PIPE
);
3401 r
= change_uid_gid(arg_user
, arg_console_mode
!= CONSOLE_PIPE
, &home
);
3405 r
= drop_capabilities(getuid());
3407 return log_error_errno(r
, "Dropping capabilities failed: %m");
3409 if (arg_no_new_privileges
)
3410 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3411 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3413 /* LXC sets container=lxc, so follow the scheme here */
3414 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3416 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3420 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3421 if (asprintf(envp
+ n_env
++, "HOME=%s", home
?: "/root") < 0)
3424 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3425 if (asprintf(envp
+ n_env
++, "USER=%s", arg_user
?: "root") < 0 ||
3426 asprintf(envp
+ n_env
++, "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3429 assert(!sd_id128_is_null(arg_uuid
));
3431 if (asprintf(envp
+ n_env
++, "container_uuid=%s", ID128_TO_UUID_STRING(arg_uuid
)) < 0)
3434 if (fdset_size(fds
) > 0) {
3435 r
= fdset_cloexec(fds
, false);
3437 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3439 if ((asprintf(envp
+ n_env
++, "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3440 (asprintf(envp
+ n_env
++, "LISTEN_PID=1") < 0))
3443 if (asprintf(envp
+ n_env
++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3446 if (arg_n_credentials
> 0) {
3447 envp
[n_env
] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3453 env_use
= strv_env_merge(envp
, os_release_pairs
, arg_setenv
);
3457 /* Let the parent know that we are ready and
3458 * wait until the parent is ready with the
3460 if (!barrier_place_and_sync(barrier
)) /* #5 */
3461 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Parent died too early");
3464 if (chdir(arg_chdir
) < 0)
3465 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3467 if (arg_start_mode
== START_PID2
) {
3468 r
= stub_pid1(arg_uuid
);
3473 if (arg_console_mode
!= CONSOLE_PIPE
) {
3474 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3475 * are configured for that. Acquire it as controlling tty. */
3476 if (ioctl(STDIN_FILENO
, TIOCSCTTY
) < 0)
3477 return log_error_errno(errno
, "Failed to acquire controlling TTY: %m");
3480 log_debug("Inner child completed, invoking payload.");
3482 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3483 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3484 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3486 log_set_open_when_needed(true);
3488 (void) fdset_close_others(fds
);
3490 if (arg_start_mode
== START_BOOT
) {
3494 /* Automatically search for the init system */
3496 m
= strv_length(arg_parameters
);
3497 a
= newa(char*, m
+ 2);
3498 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3501 a
[0] = (char*) "/usr/lib/systemd/systemd";
3502 execve(a
[0], a
, env_use
);
3504 a
[0] = (char*) "/lib/systemd/systemd";
3505 execve(a
[0], a
, env_use
);
3507 a
[0] = (char*) "/sbin/init";
3508 execve(a
[0], a
, env_use
);
3510 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3511 } else if (!strv_isempty(arg_parameters
)) {
3512 const char *dollar_path
;
3514 exec_target
= arg_parameters
[0];
3516 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3518 dollar_path
= strv_env_get(env_use
, "PATH");
3520 if (setenv("PATH", dollar_path
, 1) < 0)
3521 return log_error_errno(errno
, "Failed to update $PATH: %m");
3524 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3527 /* If we cannot change the directory, we'll end up in /, that is expected. */
3528 (void) chdir(home
?: "/root");
3530 execle("/bin/bash", "-bash", NULL
, env_use
);
3531 execle("/bin/sh", "-sh", NULL
, env_use
);
3533 exec_target
= "/bin/bash, /bin/sh";
3536 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3539 static int setup_notify_child(void) {
3540 _cleanup_close_
int fd
= -1;
3541 union sockaddr_union sa
= {
3542 .un
.sun_family
= AF_UNIX
,
3543 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3547 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3549 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3551 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3552 (void) sockaddr_un_unlink(&sa
.un
);
3554 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3556 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3558 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3560 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3562 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3564 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3569 static int outer_child(
3571 const char *directory
,
3572 DissectedImage
*dissected_image
,
3579 int uid_shift_socket
,
3580 int master_pty_socket
,
3581 int unified_cgroup_hierarchy_socket
,
3585 _cleanup_(bind_user_context_freep
) BindUserContext
*bind_user_context
= NULL
;
3586 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3587 _cleanup_close_
int fd
= -1;
3594 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3595 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3596 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3597 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3601 assert(pid_socket
>= 0);
3602 assert(uuid_socket
>= 0);
3603 assert(notify_socket
>= 0);
3604 assert(master_pty_socket
>= 0);
3605 assert(kmsg_socket
>= 0);
3607 log_debug("Outer child is initializing.");
3609 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3611 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3613 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3614 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3616 r
= reset_audit_loginuid();
3620 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3621 * mounts to the real root. */
3622 r
= mount_follow_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3626 if (dissected_image
) {
3627 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3628 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3629 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3630 * makes sure ESP partitions and userns are compatible. */
3632 r
= dissected_image_mount_and_warn(
3637 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3638 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3639 DISSECT_IMAGE_USR_NO_ROOT
|
3640 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3641 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3646 r
= determine_uid_shift(directory
);
3650 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3651 /* Let the parent know which UID shift we read from the image */
3652 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3654 return log_error_errno(errno
, "Failed to send UID shift: %m");
3655 if (l
!= sizeof(arg_uid_shift
))
3656 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3657 "Short write while sending UID shift.");
3659 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3660 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3661 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3662 * not it will pick a different one, and send it back to us. */
3664 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3666 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3667 if (l
!= sizeof(arg_uid_shift
))
3668 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3669 "Short read while receiving UID shift.");
3672 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3673 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3676 if (path_equal(directory
, "/")) {
3677 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3678 * place, so that we can make changes to its mount structure (for example, to implement
3679 * --volatile=) without this interfering with our ability to access files such as
3680 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3681 * (instead of a temporary directory, since we are living in our own mount namspace here
3682 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3683 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3685 r
= mount_nofollow_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3689 directory
= "/run/systemd/nspawn-root";
3692 if (arg_userns_mode
!= USER_NAMESPACE_NO
&&
3693 IN_SET(arg_userns_ownership
, USER_NAMESPACE_OWNERSHIP_MAP
, USER_NAMESPACE_OWNERSHIP_AUTO
) &&
3694 arg_uid_shift
!= 0) {
3695 r
= make_mount_point(directory
);
3699 r
= remount_idmap(directory
, arg_uid_shift
, arg_uid_range
);
3700 if (r
== -EINVAL
|| ERRNO_IS_NOT_SUPPORTED(r
)) {
3701 /* This might fail because the kernel or file system doesn't support idmapping. We
3702 * can't really distinguish this nicely, nor do we have any guarantees about the
3703 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3704 if (arg_userns_ownership
!= USER_NAMESPACE_OWNERSHIP_AUTO
)
3705 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
3706 "ID mapped mounts are apparently not available, sorry.");
3708 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3709 arg_userns_ownership
= USER_NAMESPACE_OWNERSHIP_CHOWN
;
3711 return log_error_errno(r
, "Failed to set up ID mapped mounts: %m");
3713 log_debug("ID mapped mounts available, making use of them.");
3718 r
= setup_pivot_root(
3721 arg_pivot_root_old
);
3725 r
= setup_volatile_mode(
3729 arg_selinux_apifs_context
);
3733 r
= bind_user_prepare(
3738 &arg_custom_mounts
, &arg_n_custom_mounts
,
3739 &bind_user_context
);
3743 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& bind_user_context
) {
3744 /* Send the user maps we determined to the parent, so that it installs it in our user namespace UID map table */
3746 for (size_t i
= 0; i
< bind_user_context
->n_data
; i
++) {
3748 bind_user_context
->data
[i
].payload_user
->uid
,
3749 bind_user_context
->data
[i
].host_user
->uid
,
3750 (uid_t
) bind_user_context
->data
[i
].payload_group
->gid
,
3751 (uid_t
) bind_user_context
->data
[i
].host_group
->gid
,
3754 l
= send(uid_shift_socket
, map
, sizeof(map
), MSG_NOSIGNAL
);
3756 return log_error_errno(errno
, "Failed to send user UID map: %m");
3757 if (l
!= sizeof(map
))
3758 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3759 "Short write while sending user UID map.");
3766 arg_n_custom_mounts
,
3768 arg_selinux_apifs_context
,
3773 /* Make sure we always have a mount that we can move to root later on. */
3774 r
= make_mount_point(directory
);
3778 if (dissected_image
) {
3779 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3780 r
= dissected_image_mount(
3785 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|
3786 DISSECT_IMAGE_DISCARD_ON_LOOP
|
3787 DISSECT_IMAGE_USR_NO_ROOT
|
3788 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
|DISSECT_IMAGE_GROWFS
)|
3789 (idmap
? DISSECT_IMAGE_MOUNT_IDMAPPED
: 0));
3791 return log_error_errno(r
, "File system check for image failed: %m");
3793 return log_error_errno(r
, "Failed to mount image file system: %m");
3796 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3797 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3799 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3803 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3805 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3806 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3807 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3808 "Short write while sending cgroup mode.");
3810 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3813 /* Mark everything as shared so our mounts get propagated down. This is
3814 * required to make new bind mounts available in systemd services
3815 * inside the container that create a new mount namespace.
3816 * See https://github.com/systemd/systemd/issues/3860
3817 * Further submounts (such as /dev) done after this will inherit the
3818 * shared propagation mode.
3820 * IMPORTANT: Do not overmount the root directory anymore from now on to
3821 * enable moving the root directory mount to root later on.
3822 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3824 r
= mount_nofollow_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3828 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3832 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3836 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3837 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3838 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3840 return log_error_errno(r
, "Failed to make tree read-only: %m");
3843 r
= mount_all(directory
,
3846 arg_selinux_apifs_context
);
3850 r
= copy_devnodes(directory
);
3854 r
= make_extra_nodes(directory
);
3858 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3860 p
= prefix_roota(directory
, "/run/host");
3861 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3863 r
= setup_pts(directory
);
3867 r
= setup_propagate(directory
);
3871 r
= setup_keyring();
3875 r
= setup_credentials(directory
);
3879 r
= bind_user_setup(bind_user_context
, directory
);
3886 arg_n_custom_mounts
,
3888 arg_selinux_apifs_context
,
3889 MOUNT_NON_ROOT_ONLY
);
3893 r
= setup_timezone(directory
);
3897 r
= setup_resolv_conf(directory
);
3901 r
= setup_machine_id(directory
);
3905 r
= setup_journal(directory
);
3909 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3910 p
= prefix_roota(directory
, "/run/host/container-manager");
3911 (void) write_string_file(p
, arg_container_service_name
, WRITE_STRING_FILE_CREATE
);
3913 /* The same stuff as the $container_uuid env var */
3914 p
= prefix_roota(directory
, "/run/host/container-uuid");
3915 (void) write_string_filef(p
, WRITE_STRING_FILE_CREATE
, SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(arg_uuid
));
3917 if (!arg_use_cgns
) {
3920 arg_unified_cgroup_hierarchy
,
3921 arg_userns_mode
!= USER_NAMESPACE_NO
,
3924 arg_selinux_apifs_context
,
3930 r
= mount_move_root(directory
);
3932 return log_error_errno(r
, "Failed to move root directory: %m");
3934 fd
= setup_notify_child();
3938 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3939 arg_clone_ns_flags
|
3940 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3942 return log_error_errno(errno
, "Failed to fork inner child: %m");
3944 pid_socket
= safe_close(pid_socket
);
3945 uuid_socket
= safe_close(uuid_socket
);
3946 notify_socket
= safe_close(notify_socket
);
3947 uid_shift_socket
= safe_close(uid_shift_socket
);
3949 /* The inner child has all namespaces that are requested, so that we all are owned by the
3950 * user if user namespaces are turned on. */
3952 if (arg_network_namespace_path
) {
3953 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3955 return log_error_errno(r
, "Failed to join network namespace: %m");
3958 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
, os_release_pairs
);
3960 _exit(EXIT_FAILURE
);
3962 _exit(EXIT_SUCCESS
);
3965 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3967 return log_error_errno(errno
, "Failed to send PID: %m");
3968 if (l
!= sizeof(pid
))
3969 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3970 "Short write while sending PID.");
3972 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3974 return log_error_errno(errno
, "Failed to send machine ID: %m");
3975 if (l
!= sizeof(arg_uuid
))
3976 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3977 "Short write while sending machine ID.");
3979 l
= send_one_fd(notify_socket
, fd
, 0);
3981 return log_error_errno(l
, "Failed to send notify fd: %m");
3983 pid_socket
= safe_close(pid_socket
);
3984 uuid_socket
= safe_close(uuid_socket
);
3985 notify_socket
= safe_close(notify_socket
);
3986 master_pty_socket
= safe_close(master_pty_socket
);
3987 kmsg_socket
= safe_close(kmsg_socket
);
3988 rtnl_socket
= safe_close(rtnl_socket
);
3989 netns_fd
= safe_close(netns_fd
);
3994 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3995 bool tried_hashed
= false;
3996 unsigned n_tries
= 100;
4001 assert(ret_lock_file
);
4002 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
4003 assert(arg_uid_range
== 0x10000U
);
4007 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4010 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
4011 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
4016 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
4018 if ((candidate
& UINT32_C(0xFFFF)) != 0)
4021 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
4022 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
4023 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
4028 /* Make some superficial checks whether the range is currently known in the user database */
4029 if (getpwuid(candidate
))
4031 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
4033 if (getgrgid(candidate
))
4035 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
4038 *ret_lock_file
= lf
;
4039 lf
= (struct LockFile
) LOCK_FILE_INIT
;
4044 if (arg_machine
&& !tried_hashed
) {
4045 /* Try to hash the base from the container name */
4047 static const uint8_t hash_key
[] = {
4048 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4049 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4052 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
4054 tried_hashed
= true;
4056 random_bytes(&candidate
, sizeof(candidate
));
4058 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
4059 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
4063 static int add_one_uid_map(
4065 uid_t container_uid
,
4069 return strextendf(p
,
4070 UID_FMT
" " UID_FMT
" " UID_FMT
"\n",
4071 container_uid
, host_uid
, range
);
4074 static int make_uid_map_string(
4075 const uid_t bind_user_uid
[],
4076 size_t n_bind_user_uid
,
4080 _cleanup_free_
char *s
= NULL
;
4081 uid_t previous_uid
= 0;
4084 assert(n_bind_user_uid
== 0 || bind_user_uid
);
4085 assert(IN_SET(offset
, 0, 2)); /* used to switch between UID and GID map */
4088 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4089 * quadruplet, consisting of host and container UID + GID. */
4091 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4092 uid_t payload_uid
= bind_user_uid
[i
*2+offset
],
4093 host_uid
= bind_user_uid
[i
*2+offset
+1];
4095 assert(previous_uid
<= payload_uid
);
4096 assert(payload_uid
< arg_uid_range
);
4098 /* Add a range to close the gap to previous entry */
4099 if (payload_uid
> previous_uid
) {
4100 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, payload_uid
- previous_uid
);
4105 /* Map this specific user */
4106 r
= add_one_uid_map(&s
, payload_uid
, host_uid
, 1);
4110 previous_uid
= payload_uid
+ 1;
4113 /* And add a range to close the gap to finish the range */
4114 if (arg_uid_range
> previous_uid
) {
4115 r
= add_one_uid_map(&s
, previous_uid
, arg_uid_shift
+ previous_uid
, arg_uid_range
- previous_uid
);
4126 static int setup_uid_map(
4128 const uid_t bind_user_uid
[],
4129 size_t n_bind_user_uid
) {
4131 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1];
4132 _cleanup_free_
char *s
= NULL
;
4137 /* Build the UID map string */
4138 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 0, &s
) < 0) /* offset=0 contains the UID pair */
4141 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
4142 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4144 return log_error_errno(r
, "Failed to write UID map: %m");
4146 /* And now build the GID map string */
4148 if (make_uid_map_string(bind_user_uid
, n_bind_user_uid
, 2, &s
) < 0) /* offset=2 contains the GID pair */
4151 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
4152 r
= write_string_file(uid_map
, s
, WRITE_STRING_FILE_DISABLE_BUFFER
);
4154 return log_error_errno(r
, "Failed to write GID map: %m");
4159 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4160 char buf
[NOTIFY_BUFFER_MAX
+1];
4162 struct iovec iovec
= {
4164 .iov_len
= sizeof(buf
)-1,
4166 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
4167 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
4168 struct msghdr msghdr
= {
4171 .msg_control
= &control
,
4172 .msg_controllen
= sizeof(control
),
4174 struct ucred
*ucred
;
4176 pid_t inner_child_pid
;
4177 _cleanup_strv_free_
char **tags
= NULL
;
4181 inner_child_pid
= PTR_TO_PID(userdata
);
4183 if (revents
!= EPOLLIN
) {
4184 log_warning("Got unexpected poll event for notify fd.");
4188 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
4189 if (IN_SET(n
, -EAGAIN
, -EINTR
))
4192 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4196 return log_warning_errno(n
, "Couldn't read notification socket: %m");
4198 cmsg_close_all(&msghdr
);
4200 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
4201 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
4202 log_debug("Received notify message without valid credentials. Ignoring.");
4206 if ((size_t) n
>= sizeof(buf
)) {
4207 log_warning("Received notify message exceeded maximum size. Ignoring.");
4212 tags
= strv_split(buf
, "\n\r");
4216 if (strv_find(tags
, "READY=1"))
4217 (void) sd_notifyf(false, "READY=1\n");
4219 p
= strv_find_startswith(tags
, "STATUS=");
4221 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
4226 static int setup_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
4229 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
4231 return log_error_errno(r
, "Failed to allocate notify event source: %m");
4233 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
4238 static int merge_settings(Settings
*settings
, const char *path
) {
4244 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4245 * that this steals the fields of the Settings* structure, and hence modifies it. */
4247 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
4248 settings
->start_mode
>= 0) {
4249 arg_start_mode
= settings
->start_mode
;
4250 strv_free_and_replace(arg_parameters
, settings
->parameters
);
4253 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
4254 arg_ephemeral
= settings
->ephemeral
;
4256 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
4259 if (!arg_settings_trusted
)
4260 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
4262 free_and_replace(arg_directory
, settings
->root
);
4265 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
4266 settings
->pivot_root_new
) {
4267 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
4268 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
4271 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
4272 settings
->working_directory
)
4273 free_and_replace(arg_chdir
, settings
->working_directory
);
4275 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
4276 settings
->environment
)
4277 strv_free_and_replace(arg_setenv
, settings
->environment
);
4279 if ((arg_settings_mask
& SETTING_USER
) == 0) {
4282 free_and_replace(arg_user
, settings
->user
);
4284 if (uid_is_valid(settings
->uid
))
4285 arg_uid
= settings
->uid
;
4286 if (gid_is_valid(settings
->gid
))
4287 arg_gid
= settings
->gid
;
4288 if (settings
->n_supplementary_gids
> 0) {
4289 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
4290 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
4294 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
4295 uint64_t plus
, minus
;
4296 uint64_t network_minus
= 0;
4299 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4300 * Settings structure */
4302 plus
= settings
->capability
;
4303 minus
= settings
->drop_capability
;
4305 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
4306 if (settings_private_network(settings
))
4307 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4309 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
4312 if (!arg_settings_trusted
&& plus
!= 0) {
4313 if (settings
->capability
!= 0)
4314 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
4316 arg_caps_retain
&= ~network_minus
;
4317 arg_caps_retain
|= plus
;
4320 arg_caps_retain
&= ~minus
;
4322 /* Copy the full capabilities over too */
4323 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
4324 if (!arg_settings_trusted
)
4325 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
4327 arg_full_capabilities
= settings
->full_capabilities
;
4330 ambient
= settings
->ambient_capability
;
4331 if (!arg_settings_trusted
&& ambient
!= 0)
4332 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path
);
4334 arg_caps_ambient
|= ambient
;
4337 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
4338 settings
->kill_signal
> 0)
4339 arg_kill_signal
= settings
->kill_signal
;
4341 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
4342 settings
->personality
!= PERSONALITY_INVALID
)
4343 arg_personality
= settings
->personality
;
4345 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
4346 !sd_id128_is_null(settings
->machine_id
)) {
4348 if (!arg_settings_trusted
)
4349 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
4351 arg_uuid
= settings
->machine_id
;
4354 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
4355 settings
->read_only
>= 0)
4356 arg_read_only
= settings
->read_only
;
4358 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
4359 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
4360 arg_volatile_mode
= settings
->volatile_mode
;
4362 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
4363 settings
->n_custom_mounts
> 0) {
4365 if (!arg_settings_trusted
)
4366 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
4368 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4369 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
4370 arg_n_custom_mounts
= settings
->n_custom_mounts
;
4371 settings
->n_custom_mounts
= 0;
4375 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
4376 (settings
->private_network
>= 0 ||
4377 settings
->network_veth
>= 0 ||
4378 settings
->network_bridge
||
4379 settings
->network_zone
||
4380 settings
->network_interfaces
||
4381 settings
->network_macvlan
||
4382 settings
->network_ipvlan
||
4383 settings
->network_veth_extra
||
4384 settings
->network_namespace_path
)) {
4386 if (!arg_settings_trusted
)
4387 log_warning("Ignoring network settings, file %s is not trusted.", path
);
4389 arg_network_veth
= settings_network_veth(settings
);
4390 arg_private_network
= settings_private_network(settings
);
4392 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
4393 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
4394 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
4395 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
4397 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
4398 free_and_replace(arg_network_zone
, settings
->network_zone
);
4400 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
4404 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
4405 settings
->expose_ports
) {
4407 if (!arg_settings_trusted
)
4408 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
4410 expose_port_free_all(arg_expose_ports
);
4411 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
4415 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
4416 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
4418 if (!arg_settings_trusted
)
4419 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
4421 arg_userns_mode
= settings
->userns_mode
;
4422 arg_uid_shift
= settings
->uid_shift
;
4423 arg_uid_range
= settings
->uid_range
;
4424 arg_userns_ownership
= settings
->userns_ownership
;
4428 if ((arg_settings_mask
& SETTING_BIND_USER
) == 0)
4429 strv_free_and_replace(arg_bind_user
, settings
->bind_user
);
4431 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
4432 arg_notify_ready
= settings
->notify_ready
;
4434 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4436 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4437 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4439 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4440 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4444 if (!arg_settings_trusted
&& settings
->seccomp
)
4445 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4447 seccomp_release(arg_seccomp
);
4448 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4453 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4454 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4457 if (!settings
->rlimit
[rl
])
4460 if (!arg_settings_trusted
) {
4461 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4465 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4468 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4470 free_and_replace(arg_hostname
, settings
->hostname
);
4472 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4473 settings
->no_new_privileges
>= 0)
4474 arg_no_new_privileges
= settings
->no_new_privileges
;
4476 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4477 settings
->oom_score_adjust_set
) {
4479 if (!arg_settings_trusted
)
4480 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4482 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4483 arg_oom_score_adjust_set
= true;
4487 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4488 settings
->cpu_set
.set
) {
4490 if (!arg_settings_trusted
)
4491 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4493 cpu_set_reset(&arg_cpu_set
);
4494 arg_cpu_set
= settings
->cpu_set
;
4495 settings
->cpu_set
= (CPUSet
) {};
4499 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4500 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4501 arg_resolv_conf
= settings
->resolv_conf
;
4503 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4504 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4506 if (!arg_settings_trusted
)
4507 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4509 arg_link_journal
= settings
->link_journal
;
4510 arg_link_journal_try
= settings
->link_journal_try
;
4514 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4515 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4516 arg_timezone
= settings
->timezone
;
4518 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4521 if (!arg_settings_trusted
)
4522 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4524 free_and_replace(arg_slice
, settings
->slice
);
4527 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4528 settings
->use_cgns
>= 0) {
4530 if (!arg_settings_trusted
)
4531 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4533 arg_use_cgns
= settings
->use_cgns
;
4536 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4537 settings
->clone_ns_flags
!= ULONG_MAX
) {
4539 if (!arg_settings_trusted
)
4540 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4542 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4545 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4546 settings
->console_mode
>= 0) {
4548 if (!arg_settings_trusted
)
4549 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4551 arg_console_mode
= settings
->console_mode
;
4554 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4555 * don't consult arg_settings_mask for them. */
4557 sd_bus_message_unref(arg_property_message
);
4558 arg_property_message
= TAKE_PTR(settings
->properties
);
4560 arg_console_width
= settings
->console_width
;
4561 arg_console_height
= settings
->console_height
;
4563 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4564 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4565 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4570 static int load_settings(void) {
4571 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4572 _cleanup_fclose_
FILE *f
= NULL
;
4573 _cleanup_free_
char *p
= NULL
;
4580 /* If all settings are masked, there's no point in looking for
4581 * the settings file */
4582 if (FLAGS_SET(arg_settings_mask
, _SETTINGS_MASK_ALL
))
4585 fn
= strjoina(arg_machine
, ".nspawn");
4587 /* We first look in the admin's directories in /etc and /run */
4588 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4589 _cleanup_free_
char *j
= NULL
;
4591 j
= path_join(i
, fn
);
4599 /* By default, we trust configuration from /etc and /run */
4600 if (arg_settings_trusted
< 0)
4601 arg_settings_trusted
= true;
4606 if (errno
!= ENOENT
)
4607 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4611 /* After that, let's look for a file next to the
4612 * actual image we shall boot. */
4615 p
= file_in_same_dir(arg_image
, fn
);
4618 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4619 p
= file_in_same_dir(arg_directory
, fn
);
4626 if (!f
&& errno
!= ENOENT
)
4627 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4629 /* By default, we do not trust configuration from /var/lib/machines */
4630 if (arg_settings_trusted
< 0)
4631 arg_settings_trusted
= false;
4638 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4640 r
= settings_load(f
, p
, &settings
);
4644 return merge_settings(settings
, p
);
4647 static int load_oci_bundle(void) {
4648 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4651 if (!arg_oci_bundle
)
4654 /* By default let's trust OCI bundles */
4655 if (arg_settings_trusted
< 0)
4656 arg_settings_trusted
= true;
4658 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4662 return merge_settings(settings
, arg_oci_bundle
);
4665 static int run_container(
4666 DissectedImage
*dissected_image
,
4669 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4670 struct ExposeArgs
*expose_args
,
4671 int *master
, pid_t
*pid
, int *ret
) {
4673 static const struct sigaction sa
= {
4674 .sa_handler
= nop_signal_handler
,
4675 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4678 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4679 _cleanup_close_
int etc_passwd_lock
= -1;
4680 _cleanup_close_pair_
int
4681 kmsg_socket_pair
[2] = { -1, -1 },
4682 rtnl_socket_pair
[2] = { -1, -1 },
4683 pid_socket_pair
[2] = { -1, -1 },
4684 uuid_socket_pair
[2] = { -1, -1 },
4685 notify_socket_pair
[2] = { -1, -1 },
4686 uid_shift_socket_pair
[2] = { -1, -1 },
4687 master_pty_socket_pair
[2] = { -1, -1 },
4688 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4690 _cleanup_close_
int notify_socket
= -1;
4691 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4692 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4693 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4694 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4695 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4696 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4697 _cleanup_free_ uid_t
*bind_user_uid
= NULL
;
4698 size_t n_bind_user_uid
= 0;
4699 ContainerStatus container_status
= 0;
4703 _cleanup_close_
int child_netns_fd
= -1;
4705 assert_se(sigemptyset(&mask_chld
) == 0);
4706 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4708 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4709 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4710 * check with getpwuid() if the specific user already exists. Note that /etc might be
4711 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4712 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4713 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4716 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4717 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4718 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4721 r
= barrier_create(&barrier
);
4723 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4725 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4726 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4728 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4729 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4731 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4732 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4734 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4735 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4737 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4738 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4740 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4741 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4743 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4744 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4745 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4747 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4748 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4749 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4751 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4752 * parent's blocking calls and give it a chance to call wait() and terminate. */
4753 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4755 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4757 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4759 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4761 if (arg_network_namespace_path
) {
4762 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4763 if (child_netns_fd
< 0)
4764 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4766 r
= fd_is_ns(child_netns_fd
, CLONE_NEWNET
);
4768 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
);
4770 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4772 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4773 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4776 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4778 return log_error_errno(errno
, "clone() failed%s: %m",
4780 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4783 /* The outer child only has a file system namespace. */
4784 barrier_set_role(&barrier
, BARRIER_CHILD
);
4786 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4787 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4788 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4789 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4790 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4791 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4792 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4793 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4795 (void) reset_all_signal_handlers();
4796 (void) reset_signal_mask();
4798 r
= outer_child(&barrier
,
4803 uuid_socket_pair
[1],
4804 notify_socket_pair
[1],
4805 kmsg_socket_pair
[1],
4806 rtnl_socket_pair
[1],
4807 uid_shift_socket_pair
[1],
4808 master_pty_socket_pair
[1],
4809 unified_cgroup_hierarchy_socket_pair
[1],
4813 _exit(EXIT_FAILURE
);
4815 _exit(EXIT_SUCCESS
);
4818 barrier_set_role(&barrier
, BARRIER_PARENT
);
4822 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4823 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4824 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4825 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4826 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4827 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4828 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4829 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4831 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4832 /* The child just let us know the UID shift it might have read from the image. */
4833 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4835 return log_error_errno(errno
, "Failed to read UID shift: %m");
4836 if (l
!= sizeof arg_uid_shift
)
4837 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4839 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4840 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4841 * image, but if that's already in use, pick a new one, and report back to the child,
4842 * which one we now picked. */
4844 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4846 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4848 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4850 return log_error_errno(errno
, "Failed to send UID shift: %m");
4851 if (l
!= sizeof arg_uid_shift
)
4852 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4855 n_bind_user_uid
= strv_length(arg_bind_user
);
4856 if (n_bind_user_uid
> 0) {
4857 /* Right after the UID shift, we'll receive the list of UID mappings for the
4858 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4860 bind_user_uid
= new(uid_t
, n_bind_user_uid
*4);
4864 for (size_t i
= 0; i
< n_bind_user_uid
; i
++) {
4865 l
= recv(uid_shift_socket_pair
[0], bind_user_uid
+ i
*4, sizeof(uid_t
)*4, 0);
4867 return log_error_errno(errno
, "Failed to read user UID map pair: %m");
4868 if (l
!= sizeof(uid_t
)*4)
4869 return log_full_errno(l
== 0 ? LOG_DEBUG
: LOG_WARNING
,
4870 SYNTHETIC_ERRNO(EIO
),
4871 "Short read while reading bind user UID pairs.");
4876 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4877 /* The child let us know the support cgroup mode it might have read from the image. */
4878 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4880 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4881 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4882 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4883 l
, l
== 0 ? " The child is most likely dead." : "");
4886 /* Wait for the outer child. */
4887 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4890 if (r
!= EXIT_SUCCESS
)
4893 /* And now retrieve the PID of the inner child. */
4894 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4896 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4897 if (l
!= sizeof *pid
)
4898 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4900 /* We also retrieve container UUID in case it was generated by outer child */
4901 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4903 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4904 if (l
!= sizeof(arg_uuid
))
4905 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4907 /* We also retrieve the socket used for notifications generated by outer child */
4908 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4909 if (notify_socket
< 0)
4910 return log_error_errno(notify_socket
,
4911 "Failed to receive notification socket from the outer child: %m");
4913 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4915 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4916 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4917 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4919 r
= setup_uid_map(*pid
, bind_user_uid
, n_bind_user_uid
);
4923 (void) barrier_place(&barrier
); /* #2 */
4926 if (arg_private_network
) {
4927 if (!arg_network_namespace_path
) {
4928 /* Wait until the child has unshared its network namespace. */
4929 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4930 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4933 if (child_netns_fd
< 0) {
4934 /* Make sure we have an open file descriptor to the child's network
4935 * namespace so it stays alive even if the child exits. */
4936 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4938 return log_error_errno(r
, "Failed to open child network namespace: %m");
4941 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4945 if (arg_network_veth
) {
4946 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4947 arg_network_bridge
|| arg_network_zone
);
4953 if (arg_network_bridge
) {
4954 /* Add the interface to a bridge */
4955 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4960 } else if (arg_network_zone
) {
4961 /* Add the interface to a bridge, possibly creating it */
4962 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4970 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4974 /* We created the primary and extra veth links now; let's remember this, so that we know to
4975 remove them later on. Note that we don't bother with removing veth links that were created
4976 here when their setup failed half-way, because in that case the kernel should be able to
4977 remove them on its own, since they cannot be referenced by anything yet. */
4978 *veth_created
= true;
4980 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4984 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4989 if (arg_register
|| !arg_keep_unit
) {
4990 r
= sd_bus_default_system(&bus
);
4992 return log_error_errno(r
, "Failed to open system bus: %m");
4994 r
= sd_bus_set_close_on_exit(bus
, false);
4996 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4999 if (!arg_keep_unit
) {
5000 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5001 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5002 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5004 r
= sd_bus_match_signal_async(
5007 "org.freedesktop.systemd1",
5009 "org.freedesktop.systemd1.Scope",
5011 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
5013 return log_error_errno(r
, "Failed to request RequestStop match: %m");
5017 r
= register_machine(
5025 arg_custom_mounts
, arg_n_custom_mounts
,
5028 arg_property_message
,
5030 arg_container_service_name
);
5034 } else if (!arg_keep_unit
) {
5040 arg_custom_mounts
, arg_n_custom_mounts
,
5043 arg_property_message
);
5047 } else if (arg_slice
|| arg_property
)
5048 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5050 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
5054 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5058 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
5062 /* Notify the child that the parent is ready with all
5063 * its setup (including cgroup-ification), and that
5064 * the child can now hand over control to the code to
5065 * run inside the container. */
5066 (void) barrier_place(&barrier
); /* #4 */
5068 /* Block SIGCHLD here, before notifying child.
5069 * process_pty() will handle it with the other signals. */
5070 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
5072 /* Reset signal to default */
5073 r
= default_signals(SIGCHLD
);
5075 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
5077 r
= sd_event_new(&event
);
5079 return log_error_errno(r
, "Failed to get default event source: %m");
5081 (void) sd_event_set_watchdog(event
, true);
5084 r
= sd_bus_attach_event(bus
, event
, 0);
5086 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
5089 r
= setup_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
5093 /* Let the child know that we are ready and wait that the child is completely ready now. */
5094 if (!barrier_place_and_sync(&barrier
)) /* #5 */
5095 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
5097 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5098 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5099 etc_passwd_lock
= safe_close(etc_passwd_lock
);
5101 (void) sd_notifyf(false,
5102 "STATUS=Container running.\n"
5103 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
5104 if (!arg_notify_ready
)
5105 (void) sd_notify(false, "READY=1\n");
5107 if (arg_kill_signal
> 0) {
5108 /* Try to kill the init system on SIGINT or SIGTERM */
5109 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5110 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
5112 /* Immediately exit */
5113 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
5114 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
5117 /* Exit when the child exits */
5118 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
5120 if (arg_expose_ports
) {
5121 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, expose_args
, &rtnl
);
5125 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5126 (void) expose_port_execute(rtnl
, &expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5129 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
5131 if (arg_console_mode
!= CONSOLE_PIPE
) {
5132 _cleanup_close_
int fd
= -1;
5133 PTYForwardFlags flags
= 0;
5135 /* Retrieve the master pty allocated by inner child */
5136 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
5138 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
5140 switch (arg_console_mode
) {
5142 case CONSOLE_READ_ONLY
:
5143 flags
|= PTY_FORWARD_READ_ONLY
;
5147 case CONSOLE_INTERACTIVE
:
5148 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
5150 r
= pty_forward_new(event
, fd
, flags
, &forward
);
5152 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
5154 if (arg_console_width
!= UINT_MAX
|| arg_console_height
!= UINT_MAX
)
5155 (void) pty_forward_set_width_height(forward
,
5157 arg_console_height
);
5161 assert(arg_console_mode
== CONSOLE_PASSIVE
);
5164 *master
= TAKE_FD(fd
);
5167 r
= sd_event_loop(event
);
5169 return log_error_errno(r
, "Failed to run event loop: %m");
5174 (void) pty_forward_get_last_char(forward
, &last_char
);
5175 forward
= pty_forward_free(forward
);
5177 if (!arg_quiet
&& last_char
!= '\n')
5181 /* Kill if it is not dead yet anyway */
5182 if (!arg_register
&& !arg_keep_unit
&& bus
)
5183 terminate_scope(bus
, arg_machine
);
5185 /* Normally redundant, but better safe than sorry */
5186 (void) kill(*pid
, SIGKILL
);
5188 if (arg_private_network
) {
5189 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5190 * to avoid having to move the parent to the child network namespace. */
5191 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
5196 _cleanup_close_
int parent_netns_fd
= -1;
5198 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
5200 log_error_errno(r
, "Failed to open parent network namespace: %m");
5201 _exit(EXIT_FAILURE
);
5204 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
5206 log_error_errno(r
, "Failed to enter child network namespace: %m");
5207 _exit(EXIT_FAILURE
);
5210 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
5212 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
5214 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
5218 r
= wait_for_container(*pid
, &container_status
);
5221 /* Tell machined that we are gone. */
5223 (void) unregister_machine(bus
, arg_machine
);
5226 /* We failed to wait for the container, or the container exited abnormally. */
5228 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
5229 /* r > 0 → The container exited with a non-zero status.
5230 * As a special case, we need to replace 133 with a different value,
5231 * because 133 is special-cased in the service file to reboot the container.
5232 * otherwise → The container exited with zero status and a reboot was not requested.
5234 if (r
== EXIT_FORCE_RESTART
)
5235 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
5237 return 0; /* finito */
5240 /* CONTAINER_REBOOTED, loop again */
5242 if (arg_keep_unit
) {
5243 /* Special handling if we are running as a service: instead of simply
5244 * restarting the machine we want to restart the entire service, so let's
5245 * inform systemd about this with the special exit code 133. The service
5246 * file uses RestartForceExitStatus=133 so that this results in a full
5247 * nspawn restart. This is necessary since we might have cgroup parameters
5248 * set we want to have flushed out. */
5249 *ret
= EXIT_FORCE_RESTART
;
5250 return 0; /* finito */
5253 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
->address4
);
5254 expose_port_flush(&expose_args
->fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
->address6
);
5256 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5257 *veth_created
= false;
5258 return 1; /* loop again */
5261 static int initialize_rlimits(void) {
5262 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
5263 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5264 * container execution environments. */
5266 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
5267 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5268 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
5269 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5270 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5271 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5272 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5273 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
5274 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
5275 [RLIMIT_NICE
] = { 0, 0 },
5276 [RLIMIT_NOFILE
] = { 1024, 4096 },
5277 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5278 [RLIMIT_RTPRIO
] = { 0, 0 },
5279 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
5280 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
5282 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5283 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5284 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5285 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5286 * that PID 1 changes a number of other resource limits during early initialization which is why we
5287 * don't read the other limits from PID 1 but prefer the static table above. */
5292 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
5293 /* Let's only fill in what the user hasn't explicitly configured anyway */
5294 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
5295 const struct rlimit
*v
;
5296 struct rlimit buffer
;
5298 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
5299 /* For these two let's read the limits off PID 1. See above for an explanation. */
5301 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
5302 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
5306 v
= kernel_defaults
+ rl
;
5308 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
5309 if (!arg_rlimit
[rl
])
5313 if (DEBUG_LOGGING
) {
5314 _cleanup_free_
char *k
= NULL
;
5316 (void) rlimit_format(arg_rlimit
[rl
], &k
);
5317 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
5324 static int cant_be_in_netns(void) {
5325 union sockaddr_union sa
= {
5327 .sun_family
= AF_UNIX
,
5328 .sun_path
= "/run/udev/control",
5331 char udev_path
[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t
)];
5332 _cleanup_free_
char *udev_ns
= NULL
, *our_ns
= NULL
;
5333 _cleanup_close_
int fd
= -1;
5337 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5338 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5341 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5344 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
5346 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
5348 if (connect(fd
, &sa
.un
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
5350 if (errno
== ENOENT
|| ERRNO_IS_DISCONNECT(errno
))
5351 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5352 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5354 return log_error_errno(errno
, "Failed to connect socket to udev control socket: %m");
5357 r
= getpeercred(fd
, &ucred
);
5359 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
5361 xsprintf(udev_path
, "/proc/" PID_FMT
"/ns/net", ucred
.pid
);
5362 r
= readlink_malloc(udev_path
, &udev_ns
);
5364 return log_error_errno(r
, "Failed to read network namespace of udev: %m");
5366 r
= readlink_malloc("/proc/self/ns/net", &our_ns
);
5368 return log_error_errno(r
, "Failed to read our own network namespace: %m");
5370 if (!streq(our_ns
, udev_ns
))
5371 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
5372 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5376 static int run(int argc
, char *argv
[]) {
5377 bool secondary
= false, remove_directory
= false, remove_image
= false,
5378 veth_created
= false, remove_tmprootdir
= false;
5379 _cleanup_close_
int master
= -1;
5380 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
5381 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
5382 char veth_name
[IFNAMSIZ
] = "";
5383 struct ExposeArgs expose_args
= {};
5384 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
5385 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
5386 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
5387 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
5388 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
5389 _cleanup_(fw_ctx_freep
) FirewallContext
*fw_ctx
= NULL
;
5392 log_parse_environment();
5395 r
= parse_argv(argc
, argv
);
5399 if (geteuid() != 0) {
5400 r
= log_warning_errno(SYNTHETIC_ERRNO(EPERM
),
5401 argc
>= 2 ? "Need to be root." :
5402 "Need to be root (and some arguments are usually required).\nHint: try --help");
5406 r
= cant_be_in_netns();
5410 r
= initialize_rlimits();
5414 r
= load_oci_bundle();
5418 r
= determine_names();
5422 r
= load_settings();
5428 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5432 r
= verify_arguments();
5436 /* Reapply environment settings. */
5437 (void) detect_unified_cgroup_hierarchy_from_environment();
5439 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5440 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5441 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5442 (void) ignore_signals(SIGPIPE
);
5444 n_fd_passed
= sd_listen_fds(false);
5445 if (n_fd_passed
> 0) {
5446 r
= fdset_new_listen_fds(&fds
, false);
5448 log_error_errno(r
, "Failed to collect file descriptors: %m");
5453 /* The "default" umask. This is appropriate for most file and directory
5454 * operations performed by nspawn, and is the umask that will be used for
5455 * the child. Functions like copy_devnodes() change the umask temporarily. */
5458 if (arg_directory
) {
5461 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5462 * /var from the host will propagate into container dynamically (because bad things happen if
5463 * two systems write to the same /var). Let's allow it for the special cases where /var is
5464 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5465 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5466 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5471 if (arg_ephemeral
) {
5472 _cleanup_free_
char *np
= NULL
;
5474 r
= chase_symlinks_and_update(&arg_directory
, 0);
5478 /* If the specified path is a mount point we generate the new snapshot immediately
5479 * inside it under a random name. However if the specified is not a mount point we
5480 * create the new snapshot in the parent directory, just next to it. */
5481 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5483 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5487 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5489 r
= tempfn_random(arg_directory
, "machine.", &np
);
5491 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5495 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5496 * only owned by us and no one else. */
5497 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5499 log_error_errno(r
, "Failed to lock %s: %m", np
);
5504 BLOCK_SIGNALS(SIGINT
);
5505 r
= btrfs_subvol_snapshot(arg_directory
, np
,
5506 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5507 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5508 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5509 BTRFS_SNAPSHOT_RECURSIVE
|
5510 BTRFS_SNAPSHOT_QUOTA
|
5511 BTRFS_SNAPSHOT_SIGINT
);
5514 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5518 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5522 free_and_replace(arg_directory
, np
);
5523 remove_directory
= true;
5525 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5529 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5531 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5535 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5540 r
= chase_symlinks_and_update(&arg_template
, 0);
5545 BLOCK_SIGNALS(SIGINT
);
5546 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5547 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5548 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5549 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5550 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5551 BTRFS_SNAPSHOT_RECURSIVE
|
5552 BTRFS_SNAPSHOT_QUOTA
|
5553 BTRFS_SNAPSHOT_SIGINT
);
5556 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5557 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5558 else if (r
== -EINTR
) {
5559 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5562 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5565 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5566 "Populated %s from template %s.", arg_directory
, arg_template
);
5570 if (arg_start_mode
== START_BOOT
) {
5573 if (arg_pivot_root_new
)
5574 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5578 if (path_is_os_tree(p
) <= 0) {
5579 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5586 if (arg_pivot_root_new
)
5587 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5591 q
= strjoina(p
, "/usr/");
5593 if (laccess(q
, F_OK
) < 0) {
5594 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
5601 DissectImageFlags dissect_image_flags
=
5602 DISSECT_IMAGE_GENERIC_ROOT
|
5603 DISSECT_IMAGE_REQUIRE_ROOT
|
5604 DISSECT_IMAGE_RELAX_VAR_CHECK
|
5605 DISSECT_IMAGE_USR_NO_ROOT
;
5607 assert(!arg_template
);
5609 r
= chase_symlinks_and_update(&arg_image
, 0);
5613 if (arg_ephemeral
) {
5614 _cleanup_free_
char *np
= NULL
;
5616 r
= tempfn_random(arg_image
, "machine.", &np
);
5618 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5622 /* Always take an exclusive lock on our own ephemeral copy. */
5623 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5625 r
= log_error_errno(r
, "Failed to create image lock: %m");
5630 BLOCK_SIGNALS(SIGINT
);
5631 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
);
5634 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5638 r
= log_error_errno(r
, "Failed to copy image file: %m");
5642 free_and_replace(arg_image
, np
);
5643 remove_image
= true;
5645 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5647 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5651 r
= log_error_errno(r
, "Failed to create image lock: %m");
5655 r
= verity_settings_load(
5656 &arg_verity_settings
,
5657 arg_image
, NULL
, NULL
);
5659 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5663 if (arg_verity_settings
.data_path
)
5664 dissect_image_flags
|= DISSECT_IMAGE_NO_PARTITION_TABLE
;
5667 if (!mkdtemp(tmprootdir
)) {
5668 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5672 remove_tmprootdir
= true;
5674 arg_directory
= strdup(tmprootdir
);
5675 if (!arg_directory
) {
5680 r
= loop_device_make_by_path(
5682 arg_read_only
? O_RDONLY
: O_RDWR
,
5683 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
5686 log_error_errno(r
, "Failed to set up loopback block device: %m");
5690 r
= dissect_image_and_warn(
5693 &arg_verity_settings
,
5695 loop
->uevent_seqnum_not_before
,
5696 loop
->timestamp_not_before
,
5697 dissect_image_flags
,
5700 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5701 log_notice("Note that the disk image needs to\n"
5702 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5703 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5704 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5705 " d) or contain a file system without a partition table\n"
5706 "in order to be bootable with systemd-nspawn.");
5712 if (!arg_verity_settings
.root_hash
&& dissected_image
->can_verity
)
5713 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5715 r
= dissected_image_decrypt_interactively(
5718 &arg_verity_settings
,
5724 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5725 if (remove_image
&& unlink(arg_image
) >= 0)
5726 remove_image
= false;
5729 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5733 if (arg_console_mode
< 0)
5735 isatty(STDIN_FILENO
) > 0 &&
5736 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5738 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5742 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5743 arg_machine
, arg_image
?: arg_directory
);
5745 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5747 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5748 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5752 if (arg_expose_ports
) {
5753 r
= fw_ctx_new(&fw_ctx
);
5755 log_error_errno(r
, "Cannot expose configured ports, firewall initialization failed: %m");
5758 expose_args
.fw_ctx
= fw_ctx
;
5761 r
= run_container(dissected_image
,
5764 veth_name
, &veth_created
,
5765 &expose_args
, &master
,
5772 (void) sd_notify(false,
5773 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5774 "STOPPING=1\nSTATUS=Terminating...");
5777 (void) kill(pid
, SIGKILL
);
5779 /* Try to flush whatever is still queued in the pty */
5781 (void) copy_bytes(master
, STDOUT_FILENO
, UINT64_MAX
, 0);
5782 master
= safe_close(master
);
5786 (void) wait_for_terminate(pid
, NULL
);
5790 if (remove_directory
&& arg_directory
) {
5793 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5795 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5798 if (remove_image
&& arg_image
) {
5799 if (unlink(arg_image
) < 0)
5800 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5803 if (remove_tmprootdir
) {
5804 if (rmdir(tmprootdir
) < 0)
5805 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5811 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5812 (void) rm_rf(p
, REMOVE_ROOT
);
5815 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET
, &expose_args
.address4
);
5816 expose_port_flush(&fw_ctx
, arg_expose_ports
, AF_INET6
, &expose_args
.address6
);
5819 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5820 (void) remove_bridge(arg_network_zone
);
5822 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5823 expose_port_free_all(arg_expose_ports
);
5824 rlimit_free_all(arg_rlimit
);
5825 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5826 credential_free_all(arg_credentials
, arg_n_credentials
);
5834 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);