1 /* SPDX-License-Identifier: LGPL-2.1+ */
8 #include <linux/loop.h>
10 #include <selinux/selinux.h>
14 #include <sys/personality.h>
15 #include <sys/prctl.h>
16 #include <sys/types.h>
21 #include "sd-daemon.h"
24 #include "alloc-util.h"
26 #include "base-filesystem.h"
27 #include "blkid-util.h"
28 #include "btrfs-util.h"
29 #include "bus-error.h"
32 #include "capability-util.h"
33 #include "cgroup-util.h"
35 #include "cpu-set-util.h"
36 #include "dev-setup.h"
37 #include "dissect-image.h"
42 #include "format-util.h"
45 #include "hexdecoct.h"
46 #include "hostname-util.h"
47 #include "id128-util.h"
49 #include "loop-util.h"
50 #include "loopback-setup.h"
51 #include "machine-image.h"
53 #include "main-func.h"
54 #include "missing_sched.h"
56 #include "mount-util.h"
57 #include "mountpoint-util.h"
58 #include "namespace-util.h"
59 #include "netlink-util.h"
60 #include "nspawn-cgroup.h"
61 #include "nspawn-def.h"
62 #include "nspawn-expose-ports.h"
63 #include "nspawn-mount.h"
64 #include "nspawn-network.h"
65 #include "nspawn-oci.h"
66 #include "nspawn-patch-uid.h"
67 #include "nspawn-register.h"
68 #include "nspawn-seccomp.h"
69 #include "nspawn-settings.h"
70 #include "nspawn-setuid.h"
71 #include "nspawn-stub-pid1.h"
72 #include "nulstr-util.h"
75 #include "parse-util.h"
76 #include "path-util.h"
77 #include "pretty-print.h"
78 #include "process-util.h"
80 #include "random-util.h"
81 #include "raw-clone.h"
82 #include "rlimit-util.h"
85 #include "seccomp-util.h"
87 #include "selinux-util.h"
88 #include "signal-util.h"
89 #include "socket-util.h"
90 #include "stat-util.h"
91 #include "stdio-util.h"
92 #include "string-table.h"
93 #include "string-util.h"
95 #include "sysctl-util.h"
96 #include "terminal-util.h"
97 #include "tmpfile-util.h"
98 #include "umask-util.h"
99 #include "unit-name.h"
100 #include "user-util.h"
104 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
106 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
109 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
110 * nspawn_notify_socket_path is relative to the container
111 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
112 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
114 #define EXIT_FORCE_RESTART 133
116 typedef enum ContainerStatus
{
117 CONTAINER_TERMINATED
,
121 static char *arg_directory
= NULL
;
122 static char *arg_template
= NULL
;
123 static char *arg_chdir
= NULL
;
124 static char *arg_pivot_root_new
= NULL
;
125 static char *arg_pivot_root_old
= NULL
;
126 static char *arg_user
= NULL
;
127 static uid_t arg_uid
= UID_INVALID
;
128 static gid_t arg_gid
= GID_INVALID
;
129 static gid_t
* arg_supplementary_gids
= NULL
;
130 static size_t arg_n_supplementary_gids
= 0;
131 static sd_id128_t arg_uuid
= {};
132 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
133 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
134 static const char *arg_selinux_context
= NULL
;
135 static const char *arg_selinux_apifs_context
= NULL
;
136 static char *arg_slice
= NULL
;
137 static bool arg_private_network
= false;
138 static bool arg_read_only
= false;
139 static StartMode arg_start_mode
= START_PID1
;
140 static bool arg_ephemeral
= false;
141 static LinkJournal arg_link_journal
= LINK_AUTO
;
142 static bool arg_link_journal_try
= false;
143 static uint64_t arg_caps_retain
=
144 (1ULL << CAP_AUDIT_CONTROL
) |
145 (1ULL << CAP_AUDIT_WRITE
) |
146 (1ULL << CAP_CHOWN
) |
147 (1ULL << CAP_DAC_OVERRIDE
) |
148 (1ULL << CAP_DAC_READ_SEARCH
) |
149 (1ULL << CAP_FOWNER
) |
150 (1ULL << CAP_FSETID
) |
151 (1ULL << CAP_IPC_OWNER
) |
153 (1ULL << CAP_LEASE
) |
154 (1ULL << CAP_LINUX_IMMUTABLE
) |
155 (1ULL << CAP_MKNOD
) |
156 (1ULL << CAP_NET_BIND_SERVICE
) |
157 (1ULL << CAP_NET_BROADCAST
) |
158 (1ULL << CAP_NET_RAW
) |
159 (1ULL << CAP_SETFCAP
) |
160 (1ULL << CAP_SETGID
) |
161 (1ULL << CAP_SETPCAP
) |
162 (1ULL << CAP_SETUID
) |
163 (1ULL << CAP_SYS_ADMIN
) |
164 (1ULL << CAP_SYS_BOOT
) |
165 (1ULL << CAP_SYS_CHROOT
) |
166 (1ULL << CAP_SYS_NICE
) |
167 (1ULL << CAP_SYS_PTRACE
) |
168 (1ULL << CAP_SYS_RESOURCE
) |
169 (1ULL << CAP_SYS_TTY_CONFIG
);
170 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
171 static CustomMount
*arg_custom_mounts
= NULL
;
172 static size_t arg_n_custom_mounts
= 0;
173 static char **arg_setenv
= NULL
;
174 static bool arg_quiet
= false;
175 static bool arg_register
= true;
176 static bool arg_keep_unit
= false;
177 static char **arg_network_interfaces
= NULL
;
178 static char **arg_network_macvlan
= NULL
;
179 static char **arg_network_ipvlan
= NULL
;
180 static bool arg_network_veth
= false;
181 static char **arg_network_veth_extra
= NULL
;
182 static char *arg_network_bridge
= NULL
;
183 static char *arg_network_zone
= NULL
;
184 static char *arg_network_namespace_path
= NULL
;
185 static PagerFlags arg_pager_flags
= 0;
186 static unsigned long arg_personality
= PERSONALITY_INVALID
;
187 static char *arg_image
= NULL
;
188 static char *arg_oci_bundle
= NULL
;
189 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
190 static ExposePort
*arg_expose_ports
= NULL
;
191 static char **arg_property
= NULL
;
192 static sd_bus_message
*arg_property_message
= NULL
;
193 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
194 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
195 static bool arg_userns_chown
= false;
196 static int arg_kill_signal
= 0;
197 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
198 static SettingsMask arg_settings_mask
= 0;
199 static int arg_settings_trusted
= -1;
200 static char **arg_parameters
= NULL
;
201 static const char *arg_container_service_name
= "systemd-nspawn";
202 static bool arg_notify_ready
= false;
203 static bool arg_use_cgns
= true;
204 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
205 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
206 static void *arg_root_hash
= NULL
;
207 static size_t arg_root_hash_size
= 0;
208 static char **arg_syscall_whitelist
= NULL
;
209 static char **arg_syscall_blacklist
= NULL
;
211 static scmp_filter_ctx arg_seccomp
= NULL
;
213 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
214 static bool arg_no_new_privileges
= false;
215 static int arg_oom_score_adjust
= 0;
216 static bool arg_oom_score_adjust_set
= false;
217 static CPUSet arg_cpu_set
= {};
218 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
219 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
220 static unsigned arg_console_width
= (unsigned) -1, arg_console_height
= (unsigned) -1;
221 static DeviceNode
* arg_extra_nodes
= NULL
;
222 static size_t arg_n_extra_nodes
= 0;
223 static char **arg_sysctl
= NULL
;
224 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
226 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
227 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
228 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
229 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
230 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
231 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
232 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
233 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
247 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
248 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_root_hash
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist
, strv_freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
256 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
258 static int handle_arg_console(const char *arg
) {
259 if (streq(arg
, "help")) {
267 if (streq(arg
, "interactive"))
268 arg_console_mode
= CONSOLE_INTERACTIVE
;
269 else if (streq(arg
, "read-only"))
270 arg_console_mode
= CONSOLE_READ_ONLY
;
271 else if (streq(arg
, "passive"))
272 arg_console_mode
= CONSOLE_PASSIVE
;
273 else if (streq(arg
, "pipe"))
274 arg_console_mode
= CONSOLE_PIPE
;
276 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
278 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
282 static int help(void) {
283 _cleanup_free_
char *link
= NULL
;
286 (void) pager_open(arg_pager_flags
);
288 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
292 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
293 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
294 " -h --help Show this help\n"
295 " --version Print version string\n"
296 " -q --quiet Do not show status information\n"
297 " --no-pager Do not pipe output into a pager\n"
298 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
300 " -D --directory=PATH Root directory for the container\n"
301 " --template=PATH Initialize root directory from template directory,\n"
303 " -x --ephemeral Run container with snapshot of root directory, and\n"
304 " remove it after exit\n"
305 " -i --image=PATH Root file system disk image (or device node) for\n"
307 " --oci-bundle=PATH OCI bundle directory\n"
308 " --read-only Mount the root directory read-only\n"
309 " --volatile[=MODE] Run the system in volatile mode\n"
310 " --root-hash=HASH Specify verity root hash for root disk image\n"
311 " --pivot-root=PATH[:PATH]\n"
312 " Pivot root to given directory in the container\n\n"
313 "%3$sExecution:%4$s\n"
314 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
315 " -b --boot Boot up full system (i.e. invoke init)\n"
316 " --chdir=PATH Set working directory in the container\n"
317 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
318 " -u --user=USER Run the command under specified user or UID\n"
319 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
320 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
321 "%3$sSystem Identity:%4$s\n"
322 " -M --machine=NAME Set the machine name for the container\n"
323 " --hostname=NAME Override the hostname for the container\n"
324 " --uuid=UUID Set a specific machine UUID for the container\n\n"
325 "%3$sProperties:%4$s\n"
326 " -S --slice=SLICE Place the container in the specified slice\n"
327 " --property=NAME=VALUE Set scope unit property\n"
328 " --register=BOOLEAN Register container as machine\n"
329 " --keep-unit Do not register a scope for the machine, reuse\n"
330 " the service unit nspawn is running in\n\n"
331 "%3$sUser Namespacing:%4$s\n"
332 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
333 " --private-users[=UIDBASE[:NUIDS]]\n"
334 " Similar, but with user configured UID/GID range\n"
335 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
336 "%3$sNetworking:%4$s\n"
337 " --private-network Disable network in container\n"
338 " --network-interface=INTERFACE\n"
339 " Assign an existing network interface to the\n"
341 " --network-macvlan=INTERFACE\n"
342 " Create a macvlan network interface based on an\n"
343 " existing network interface to the container\n"
344 " --network-ipvlan=INTERFACE\n"
345 " Create a ipvlan network interface based on an\n"
346 " existing network interface to the container\n"
347 " -n --network-veth Add a virtual Ethernet connection between host\n"
349 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
350 " Add an additional virtual Ethernet link between\n"
351 " host and container\n"
352 " --network-bridge=INTERFACE\n"
353 " Add a virtual Ethernet connection to the container\n"
354 " and attach it to an existing bridge on the host\n"
355 " --network-zone=NAME Similar, but attach the new interface to an\n"
356 " an automatically managed bridge interface\n"
357 " --network-namespace-path=PATH\n"
358 " Set network namespace to the one represented by\n"
359 " the specified kernel namespace file node\n"
360 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
361 " Expose a container IP port on the host\n\n"
362 "%3$sSecurity:%4$s\n"
363 " --capability=CAP In addition to the default, retain specified\n"
365 " --drop-capability=CAP Drop the specified capability from the default set\n"
366 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
367 " --system-call-filter=LIST|~LIST\n"
368 " Permit/prohibit specific system calls\n"
369 " -Z --selinux-context=SECLABEL\n"
370 " Set the SELinux security context to be used by\n"
371 " processes in the container\n"
372 " -L --selinux-apifs-context=SECLABEL\n"
373 " Set the SELinux security context to be used by\n"
374 " API/tmpfs file systems in the container\n\n"
375 "%3$sResources:%4$s\n"
376 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
377 " --oom-score-adjust=VALUE\n"
378 " Adjust the OOM score value for the payload\n"
379 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
380 " --personality=ARCH Pick personality for this container\n\n"
381 "%3$sIntegration:%4$s\n"
382 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
383 " --timezone=MODE Select mode of /etc/localtime initialization\n"
384 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
385 " host, try-guest, try-host\n"
386 " -j Equivalent to --link-journal=try-guest\n\n"
388 " --bind=PATH[:PATH[:OPTIONS]]\n"
389 " Bind mount a file or directory from the host into\n"
391 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
392 " Similar, but creates a read-only bind mount\n"
393 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
395 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
396 " --overlay=PATH[:PATH...]:PATH\n"
397 " Create an overlay mount from the host to \n"
399 " --overlay-ro=PATH[:PATH...]:PATH\n"
400 " Similar, but creates a read-only overlay mount\n\n"
401 "%3$sInput/Output:%4$s\n"
402 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
403 " set up for the container.\n"
404 " -P --pipe Equivalent to --console=pipe\n"
405 "\nSee the %2$s for details.\n"
406 , program_invocation_short_name
408 , ansi_underline(), ansi_normal()
409 , ansi_highlight(), ansi_normal()
415 static int custom_mount_check_all(void) {
418 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
419 CustomMount
*m
= &arg_custom_mounts
[i
];
421 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
422 if (arg_userns_chown
)
423 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
424 "--private-users-chown may not be combined with custom root mounts.");
425 else if (arg_uid_shift
== UID_INVALID
)
426 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
427 "--private-users with automatic UID shift may not be combined with custom root mounts.");
434 static int detect_unified_cgroup_hierarchy_from_environment(void) {
435 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
438 /* Allow the user to control whether the unified hierarchy is used */
442 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
443 var
= "UNIFIED_CGROUP_HIERARCHY";
448 r
= parse_boolean(e
);
450 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
452 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
454 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
460 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
463 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
464 * in the image actually supports. */
465 r
= cg_all_unified();
467 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
469 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
470 * routine only detects 231, so we'll have a false negative here for 230. */
471 r
= systemd_installation_has_version(directory
, 230);
473 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
475 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
477 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
478 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
479 /* Mixed cgroup hierarchy support was added in 233 */
480 r
= systemd_installation_has_version(directory
, 233);
482 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
484 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
486 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
488 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
490 log_debug("Using %s hierarchy for container.",
491 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
492 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
497 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
502 _cleanup_free_
char *t
= NULL
;
504 r
= extract_first_word(&spec
, &t
, ",", 0);
506 return log_error_errno(r
, "Failed to parse capability %s.", t
);
510 if (streq(t
, "help")) {
511 for (int i
= 0; i
< capability_list_length(); i
++) {
514 name
= capability_to_name(i
);
523 mask
= (uint64_t) -1;
525 r
= capability_from_name(t
);
527 return log_error_errno(r
, "Failed to parse capability %s.", t
);
534 return 1; /* continue */
537 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
540 r
= getenv_bool(name
);
544 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
546 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
547 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
551 static int parse_mount_settings_env(void) {
555 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
556 if (r
< 0 && r
!= -ENXIO
)
557 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
559 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
561 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
562 if (streq_ptr(e
, "network"))
563 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
566 r
= parse_boolean(e
);
568 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
570 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
571 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
577 static int parse_environment(void) {
581 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
584 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
587 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
590 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
594 r
= parse_mount_settings_env();
598 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
599 * even if it is supported. If not supported, it has no effect. */
600 if (!cg_ns_supported())
601 arg_use_cgns
= false;
603 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
606 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
610 arg_use_cgns
= r
> 0;
611 arg_settings_mask
|= SETTING_USE_CGNS
;
615 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
617 arg_container_service_name
= e
;
619 return detect_unified_cgroup_hierarchy_from_environment();
622 static int parse_argv(int argc
, char *argv
[]) {
640 ARG_NETWORK_INTERFACE
,
645 ARG_NETWORK_VETH_EXTRA
,
646 ARG_NETWORK_NAMESPACE_PATH
,
656 ARG_PRIVATE_USERS_CHOWN
,
659 ARG_SYSTEM_CALL_FILTER
,
662 ARG_NO_NEW_PRIVILEGES
,
663 ARG_OOM_SCORE_ADJUST
,
673 static const struct option options
[] = {
674 { "help", no_argument
, NULL
, 'h' },
675 { "version", no_argument
, NULL
, ARG_VERSION
},
676 { "directory", required_argument
, NULL
, 'D' },
677 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
678 { "ephemeral", no_argument
, NULL
, 'x' },
679 { "user", required_argument
, NULL
, 'u' },
680 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
681 { "as-pid2", no_argument
, NULL
, 'a' },
682 { "boot", no_argument
, NULL
, 'b' },
683 { "uuid", required_argument
, NULL
, ARG_UUID
},
684 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
685 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
686 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
687 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
688 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
689 { "bind", required_argument
, NULL
, ARG_BIND
},
690 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
691 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
692 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
693 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
694 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
695 { "machine", required_argument
, NULL
, 'M' },
696 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
697 { "slice", required_argument
, NULL
, 'S' },
698 { "setenv", required_argument
, NULL
, 'E' },
699 { "selinux-context", required_argument
, NULL
, 'Z' },
700 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
701 { "quiet", no_argument
, NULL
, 'q' },
702 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
703 { "register", required_argument
, NULL
, ARG_REGISTER
},
704 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
705 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
706 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
707 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
708 { "network-veth", no_argument
, NULL
, 'n' },
709 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
710 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
711 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
712 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
713 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
714 { "image", required_argument
, NULL
, 'i' },
715 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
716 { "port", required_argument
, NULL
, 'p' },
717 { "property", required_argument
, NULL
, ARG_PROPERTY
},
718 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
719 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
720 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
721 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
722 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
723 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
724 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
725 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
726 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
727 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
728 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
729 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
730 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
731 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
732 { "console", required_argument
, NULL
, ARG_CONSOLE
},
733 { "pipe", no_argument
, NULL
, ARG_PIPE
},
734 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
735 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
740 uint64_t plus
= 0, minus
= 0;
741 bool mask_all_settings
= false, mask_no_settings
= false;
746 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
756 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
760 arg_settings_mask
|= SETTING_DIRECTORY
;
764 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
768 arg_settings_mask
|= SETTING_DIRECTORY
;
772 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
776 arg_settings_mask
|= SETTING_DIRECTORY
;
780 r
= parse_path_argument_and_warn(optarg
, false, &arg_oci_bundle
);
787 arg_ephemeral
= true;
788 arg_settings_mask
|= SETTING_EPHEMERAL
;
792 r
= free_and_strdup(&arg_user
, optarg
);
796 arg_settings_mask
|= SETTING_USER
;
799 case ARG_NETWORK_ZONE
: {
802 j
= strjoin("vz-", optarg
);
806 if (!ifname_valid(j
)) {
807 log_error("Network zone name not valid: %s", j
);
812 free_and_replace(arg_network_zone
, j
);
814 arg_network_veth
= true;
815 arg_private_network
= true;
816 arg_settings_mask
|= SETTING_NETWORK
;
820 case ARG_NETWORK_BRIDGE
:
822 if (!ifname_valid(optarg
))
823 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
824 "Bridge interface name not valid: %s", optarg
);
826 r
= free_and_strdup(&arg_network_bridge
, optarg
);
832 arg_network_veth
= true;
833 arg_private_network
= true;
834 arg_settings_mask
|= SETTING_NETWORK
;
837 case ARG_NETWORK_VETH_EXTRA
:
838 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
840 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
842 arg_private_network
= true;
843 arg_settings_mask
|= SETTING_NETWORK
;
846 case ARG_NETWORK_INTERFACE
:
847 if (!ifname_valid(optarg
))
848 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
849 "Network interface name not valid: %s", optarg
);
851 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
854 arg_private_network
= true;
855 arg_settings_mask
|= SETTING_NETWORK
;
858 case ARG_NETWORK_MACVLAN
:
860 if (!ifname_valid(optarg
))
861 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
862 "MACVLAN network interface name not valid: %s", optarg
);
864 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
867 arg_private_network
= true;
868 arg_settings_mask
|= SETTING_NETWORK
;
871 case ARG_NETWORK_IPVLAN
:
873 if (!ifname_valid(optarg
))
874 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
875 "IPVLAN network interface name not valid: %s", optarg
);
877 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
881 case ARG_PRIVATE_NETWORK
:
882 arg_private_network
= true;
883 arg_settings_mask
|= SETTING_NETWORK
;
886 case ARG_NETWORK_NAMESPACE_PATH
:
887 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
891 arg_settings_mask
|= SETTING_NETWORK
;
895 if (arg_start_mode
== START_PID2
)
896 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
897 "--boot and --as-pid2 may not be combined.");
899 arg_start_mode
= START_BOOT
;
900 arg_settings_mask
|= SETTING_START_MODE
;
904 if (arg_start_mode
== START_BOOT
)
905 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
906 "--boot and --as-pid2 may not be combined.");
908 arg_start_mode
= START_PID2
;
909 arg_settings_mask
|= SETTING_START_MODE
;
913 r
= sd_id128_from_string(optarg
, &arg_uuid
);
915 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
917 if (sd_id128_is_null(arg_uuid
))
918 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
919 "Machine UUID may not be all zeroes.");
921 arg_settings_mask
|= SETTING_MACHINE_ID
;
925 _cleanup_free_
char *mangled
= NULL
;
927 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
931 free_and_replace(arg_slice
, mangled
);
932 arg_settings_mask
|= SETTING_SLICE
;
938 arg_machine
= mfree(arg_machine
);
940 if (!machine_name_is_valid(optarg
))
941 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
942 "Invalid machine name: %s", optarg
);
944 r
= free_and_strdup(&arg_machine
, optarg
);
952 arg_hostname
= mfree(arg_hostname
);
954 if (!hostname_is_valid(optarg
, false))
955 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
956 "Invalid hostname: %s", optarg
);
958 r
= free_and_strdup(&arg_hostname
, optarg
);
963 arg_settings_mask
|= SETTING_HOSTNAME
;
967 arg_selinux_context
= optarg
;
971 arg_selinux_apifs_context
= optarg
;
975 arg_read_only
= true;
976 arg_settings_mask
|= SETTING_READ_ONLY
;
980 case ARG_DROP_CAPABILITY
: {
982 r
= parse_capability_spec(optarg
, &m
);
986 if (c
== ARG_CAPABILITY
)
990 arg_settings_mask
|= SETTING_CAPABILITY
;
993 case ARG_NO_NEW_PRIVILEGES
:
994 r
= parse_boolean(optarg
);
996 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
998 arg_no_new_privileges
= r
;
999 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1003 arg_link_journal
= LINK_GUEST
;
1004 arg_link_journal_try
= true;
1005 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1008 case ARG_LINK_JOURNAL
:
1009 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1011 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1013 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1018 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1020 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1022 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1026 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1028 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1030 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1034 case ARG_OVERLAY_RO
:
1035 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1036 if (r
== -EADDRNOTAVAIL
)
1037 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1039 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1041 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1044 case ARG_INACCESSIBLE
:
1045 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1047 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1049 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1055 if (!env_assignment_is_valid(optarg
))
1056 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1057 "Environment variable assignment '%s' is not valid.", optarg
);
1059 n
= strv_env_set(arg_setenv
, optarg
);
1063 strv_free_and_replace(arg_setenv
, n
);
1064 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1072 case ARG_SHARE_SYSTEM
:
1073 /* We don't officially support this anymore, except for compat reasons. People should use the
1074 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1075 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1076 arg_clone_ns_flags
= 0;
1080 r
= parse_boolean(optarg
);
1082 log_error("Failed to parse --register= argument: %s", optarg
);
1090 arg_keep_unit
= true;
1093 case ARG_PERSONALITY
:
1095 arg_personality
= personality_from_string(optarg
);
1096 if (arg_personality
== PERSONALITY_INVALID
)
1097 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1098 "Unknown or unsupported personality '%s'.", optarg
);
1100 arg_settings_mask
|= SETTING_PERSONALITY
;
1106 arg_volatile_mode
= VOLATILE_YES
;
1107 else if (streq(optarg
, "help")) {
1108 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1113 m
= volatile_mode_from_string(optarg
);
1115 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1116 "Failed to parse --volatile= argument: %s", optarg
);
1118 arg_volatile_mode
= m
;
1121 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1125 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1127 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1129 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1131 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1135 if (strv_extend(&arg_property
, optarg
) < 0)
1140 case ARG_PRIVATE_USERS
: {
1145 else if (!in_charset(optarg
, DIGITS
))
1146 /* do *not* parse numbers as booleans */
1147 boolean
= parse_boolean(optarg
);
1149 if (boolean
== false) {
1150 /* no: User namespacing off */
1151 arg_userns_mode
= USER_NAMESPACE_NO
;
1152 arg_uid_shift
= UID_INVALID
;
1153 arg_uid_range
= UINT32_C(0x10000);
1154 } else if (boolean
== true) {
1155 /* yes: User namespacing on, UID range is read from root dir */
1156 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1157 arg_uid_shift
= UID_INVALID
;
1158 arg_uid_range
= UINT32_C(0x10000);
1159 } else if (streq(optarg
, "pick")) {
1160 /* pick: User namespacing on, UID range is picked randomly */
1161 arg_userns_mode
= USER_NAMESPACE_PICK
;
1162 arg_uid_shift
= UID_INVALID
;
1163 arg_uid_range
= UINT32_C(0x10000);
1165 _cleanup_free_
char *buffer
= NULL
;
1166 const char *range
, *shift
;
1168 /* anything else: User namespacing on, UID range is explicitly configured */
1170 range
= strchr(optarg
, ':');
1172 buffer
= strndup(optarg
, range
- optarg
);
1178 r
= safe_atou32(range
, &arg_uid_range
);
1180 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1184 r
= parse_uid(shift
, &arg_uid_shift
);
1186 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1188 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1191 if (arg_uid_range
<= 0)
1192 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1193 "UID range cannot be 0.");
1195 arg_settings_mask
|= SETTING_USERNS
;
1200 if (userns_supported()) {
1201 arg_userns_mode
= USER_NAMESPACE_PICK
;
1202 arg_uid_shift
= UID_INVALID
;
1203 arg_uid_range
= UINT32_C(0x10000);
1205 arg_settings_mask
|= SETTING_USERNS
;
1210 case ARG_PRIVATE_USERS_CHOWN
:
1211 arg_userns_chown
= true;
1213 arg_settings_mask
|= SETTING_USERNS
;
1216 case ARG_KILL_SIGNAL
:
1217 if (streq(optarg
, "help")) {
1218 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1222 arg_kill_signal
= signal_from_string(optarg
);
1223 if (arg_kill_signal
< 0)
1224 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1225 "Cannot parse signal: %s", optarg
);
1227 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1232 /* no → do not read files
1233 * yes → read files, do not override cmdline, trust only subset
1234 * override → read files, override cmdline, trust only subset
1235 * trusted → read files, do not override cmdline, trust all
1238 r
= parse_boolean(optarg
);
1240 if (streq(optarg
, "trusted")) {
1241 mask_all_settings
= false;
1242 mask_no_settings
= false;
1243 arg_settings_trusted
= true;
1245 } else if (streq(optarg
, "override")) {
1246 mask_all_settings
= false;
1247 mask_no_settings
= true;
1248 arg_settings_trusted
= -1;
1250 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1253 mask_all_settings
= false;
1254 mask_no_settings
= false;
1255 arg_settings_trusted
= -1;
1258 mask_all_settings
= true;
1259 mask_no_settings
= false;
1260 arg_settings_trusted
= false;
1266 if (!path_is_absolute(optarg
))
1267 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1268 "Working directory %s is not an absolute path.", optarg
);
1270 r
= free_and_strdup(&arg_chdir
, optarg
);
1274 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1277 case ARG_PIVOT_ROOT
:
1278 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1280 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1282 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1285 case ARG_NOTIFY_READY
:
1286 r
= parse_boolean(optarg
);
1288 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1289 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1290 arg_notify_ready
= r
;
1291 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1294 case ARG_ROOT_HASH
: {
1298 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1300 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1301 if (l
< sizeof(sd_id128_t
)) {
1303 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1306 free(arg_root_hash
);
1308 arg_root_hash_size
= l
;
1312 case ARG_SYSTEM_CALL_FILTER
: {
1316 negative
= optarg
[0] == '~';
1317 items
= negative
? optarg
+ 1 : optarg
;
1320 _cleanup_free_
char *word
= NULL
;
1322 r
= extract_first_word(&items
, &word
, NULL
, 0);
1328 return log_error_errno(r
, "Failed to parse system call filter: %m");
1331 r
= strv_extend(&arg_syscall_blacklist
, word
);
1333 r
= strv_extend(&arg_syscall_whitelist
, word
);
1338 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1344 _cleanup_free_
char *name
= NULL
;
1347 if (streq(optarg
, "help")) {
1348 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1352 eq
= strchr(optarg
, '=');
1354 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1355 "--rlimit= expects an '=' assignment.");
1357 name
= strndup(optarg
, eq
- optarg
);
1361 rl
= rlimit_from_string_harder(name
);
1363 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1364 "Unknown resource limit: %s", name
);
1366 if (!arg_rlimit
[rl
]) {
1367 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1368 if (!arg_rlimit
[rl
])
1372 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1374 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1376 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1380 case ARG_OOM_SCORE_ADJUST
:
1381 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1383 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1385 arg_oom_score_adjust_set
= true;
1386 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1389 case ARG_CPU_AFFINITY
: {
1392 r
= parse_cpu_set(optarg
, &cpuset
);
1394 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1396 cpu_set_reset(&arg_cpu_set
);
1397 arg_cpu_set
= cpuset
;
1398 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1402 case ARG_RESOLV_CONF
:
1403 if (streq(optarg
, "help")) {
1404 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1408 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1409 if (arg_resolv_conf
< 0)
1410 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1411 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1413 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1417 if (streq(optarg
, "help")) {
1418 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1422 arg_timezone
= timezone_mode_from_string(optarg
);
1423 if (arg_timezone
< 0)
1424 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1425 "Failed to parse /etc/localtime mode: %s", optarg
);
1427 arg_settings_mask
|= SETTING_TIMEZONE
;
1431 r
= handle_arg_console(optarg
);
1438 r
= handle_arg_console("pipe");
1444 arg_pager_flags
|= PAGER_DISABLE
;
1451 assert_not_reached("Unhandled option");
1454 if (argc
> optind
) {
1455 strv_free(arg_parameters
);
1456 arg_parameters
= strv_copy(argv
+ optind
);
1457 if (!arg_parameters
)
1460 arg_settings_mask
|= SETTING_START_MODE
;
1463 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1464 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1465 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1466 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1468 arg_directory
= TAKE_PTR(arg_template
);
1470 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1472 /* Make sure to parse environment before we reset the settings mask below */
1473 r
= parse_environment();
1477 /* Load all settings from .nspawn files */
1478 if (mask_no_settings
)
1479 arg_settings_mask
= 0;
1481 /* Don't load any settings from .nspawn files */
1482 if (mask_all_settings
)
1483 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1488 static int verify_arguments(void) {
1491 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1492 /* If we are running the stub init in the container, we don't need to look at what the init
1493 * in the container supports, because we are not using it. Let's immediately pick the right
1494 * setting based on the host system configuration.
1496 * We only do this, if the user didn't use an environment variable to override the detection.
1499 r
= cg_all_unified();
1501 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1503 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1504 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1505 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1507 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1510 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1511 arg_mount_settings
|= MOUNT_USE_USERNS
;
1513 if (arg_private_network
)
1514 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1516 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1517 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1518 arg_register
= false;
1519 if (arg_start_mode
!= START_PID1
)
1520 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1523 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1524 arg_userns_chown
= true;
1526 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1527 arg_kill_signal
= SIGRTMIN
+3;
1529 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 */
1530 arg_read_only
= true;
1532 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1533 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1534 * The latter is not technically a user session, but we don't need to labour the point. */
1535 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1537 if (arg_directory
&& arg_image
)
1538 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1540 if (arg_template
&& arg_image
)
1541 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1543 if (arg_template
&& !(arg_directory
|| arg_machine
))
1544 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1546 if (arg_ephemeral
&& arg_template
)
1547 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1549 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1550 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1552 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1553 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1555 if (arg_userns_chown
&& arg_read_only
)
1556 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1557 "--read-only and --private-users-chown may not be combined.");
1559 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1560 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1561 * copy-up (in case of overlay) making the entire exercise pointless. */
1562 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1563 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1565 /* If --network-namespace-path is given with any other network-related option, we need to error out,
1566 * to avoid conflicts between different network options. */
1567 if (arg_network_namespace_path
&&
1568 (arg_network_interfaces
|| arg_network_macvlan
||
1569 arg_network_ipvlan
|| arg_network_veth_extra
||
1570 arg_network_bridge
|| arg_network_zone
||
1571 arg_network_veth
|| arg_private_network
))
1572 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1574 if (arg_network_bridge
&& arg_network_zone
)
1575 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1576 "--network-bridge= and --network-zone= may not be combined.");
1578 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1579 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1581 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1582 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1584 if (arg_expose_ports
&& !arg_private_network
)
1585 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1588 if (arg_expose_ports
)
1589 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1592 r
= custom_mount_check_all();
1599 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1602 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1605 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1608 if (uid
!= UID_INVALID
) {
1609 uid
+= arg_uid_shift
;
1611 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1615 if (gid
!= GID_INVALID
) {
1616 gid
+= (gid_t
) arg_uid_shift
;
1618 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1622 if (lchown(p
, uid
, gid
) < 0)
1628 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1632 q
= prefix_roota(root
, path
);
1633 r
= mkdir_errno_wrapper(q
, mode
);
1639 return userns_lchown(q
, uid
, gid
);
1642 static const char *timezone_from_path(const char *path
) {
1643 return PATH_STARTSWITH_SET(
1645 "../usr/share/zoneinfo/",
1646 "/usr/share/zoneinfo/");
1649 static bool etc_writable(void) {
1650 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1653 static int setup_timezone(const char *dest
) {
1654 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1655 const char *where
, *check
;
1661 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1662 r
= readlink_malloc("/etc/localtime", &p
);
1663 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1664 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1665 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1666 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1668 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1669 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1673 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1676 } else if (arg_timezone
== TIMEZONE_AUTO
)
1677 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1683 if (m
== TIMEZONE_OFF
)
1686 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1688 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1692 where
= strjoina(etc
, "/localtime");
1696 case TIMEZONE_DELETE
:
1697 if (unlink(where
) < 0)
1698 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1702 case TIMEZONE_SYMLINK
: {
1703 _cleanup_free_
char *q
= NULL
;
1704 const char *z
, *what
;
1706 z
= timezone_from_path(p
);
1708 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1712 r
= readlink_malloc(where
, &q
);
1713 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1714 return 0; /* Already pointing to the right place? Then do nothing .. */
1716 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1717 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1719 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1721 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1722 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1723 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1727 what
= strjoina("../usr/share/zoneinfo/", z
);
1728 if (symlink(what
, where
) < 0) {
1729 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1730 errno
, "Failed to correct timezone of container, ignoring: %m");
1740 case TIMEZONE_BIND
: {
1741 _cleanup_free_
char *resolved
= NULL
;
1744 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1746 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1750 if (found
== 0) /* missing? */
1751 (void) touch(resolved
);
1753 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1755 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1761 /* If mounting failed, try to copy */
1762 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1764 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1765 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1772 assert_not_reached("unexpected mode");
1775 /* Fix permissions of the symlink or file copy we just created */
1776 r
= userns_lchown(where
, 0, 0);
1778 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1783 static int have_resolv_conf(const char *path
) {
1786 if (access(path
, F_OK
) < 0) {
1787 if (errno
== ENOENT
)
1790 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1796 static int resolved_listening(void) {
1797 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1798 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1799 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1802 /* Check if resolved is listening */
1804 r
= sd_bus_open_system(&bus
);
1806 return log_debug_errno(r
, "Failed to open system bus: %m");
1808 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1810 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1814 r
= sd_bus_get_property_string(bus
,
1815 "org.freedesktop.resolve1",
1816 "/org/freedesktop/resolve1",
1817 "org.freedesktop.resolve1.Manager",
1820 &dns_stub_listener_mode
);
1822 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1824 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1827 static int setup_resolv_conf(const char *dest
) {
1828 _cleanup_free_
char *etc
= NULL
;
1829 const char *where
, *what
;
1835 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1836 if (arg_private_network
)
1837 m
= RESOLV_CONF_OFF
;
1838 else if (have_resolv_conf(STATIC_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1839 m
= etc_writable() ? RESOLV_CONF_COPY_STATIC
: RESOLV_CONF_BIND_STATIC
;
1840 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1841 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1843 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1845 m
= arg_resolv_conf
;
1847 if (m
== RESOLV_CONF_OFF
)
1850 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1852 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1856 where
= strjoina(etc
, "/resolv.conf");
1858 if (m
== RESOLV_CONF_DELETE
) {
1859 if (unlink(where
) < 0)
1860 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1865 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_COPY_STATIC
))
1866 what
= STATIC_RESOLV_CONF
;
1868 what
= "/etc/resolv.conf";
1870 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
)) {
1871 _cleanup_free_
char *resolved
= NULL
;
1874 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1876 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1880 if (found
== 0) /* missing? */
1881 (void) touch(resolved
);
1883 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1885 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1888 /* If that didn't work, let's copy the file */
1889 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
1891 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1892 * resolved or something similar runs inside and the symlink points there.
1894 * If the disk image is read-only, there's also no point in complaining.
1896 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
) && IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1897 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1901 r
= userns_lchown(where
, 0, 0);
1903 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1908 static int setup_boot_id(void) {
1909 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1910 _cleanup_free_
char *path
= NULL
;
1911 sd_id128_t rnd
= SD_ID128_NULL
;
1915 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1917 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
1919 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1921 r
= sd_id128_randomize(&rnd
);
1923 return log_error_errno(r
, "Failed to generate random boot id: %m");
1925 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1927 return log_error_errno(r
, "Failed to write boot id: %m");
1929 from
= TAKE_PTR(path
);
1930 to
= "/proc/sys/kernel/random/boot_id";
1932 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1936 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1939 static int copy_devnodes(const char *dest
) {
1940 static const char devnodes
[] =
1949 _cleanup_umask_ mode_t u
;
1957 /* Create /dev/net, so that we can create /dev/net/tun in it */
1958 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1959 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1961 NULSTR_FOREACH(d
, devnodes
) {
1962 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1965 from
= path_join("/dev/", d
);
1969 to
= path_join(dest
, from
);
1973 if (stat(from
, &st
) < 0) {
1975 if (errno
!= ENOENT
)
1976 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1978 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1979 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
1980 "%s is not a char or block device, cannot copy.", from
);
1982 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
1984 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1985 /* Explicitly warn the user when /dev is already populated. */
1986 if (errno
== EEXIST
)
1987 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
);
1989 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
1991 /* Some systems abusively restrict mknod but allow bind mounts. */
1994 return log_error_errno(r
, "touch (%s) failed: %m", to
);
1995 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
1997 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2000 r
= userns_lchown(to
, 0, 0);
2002 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2004 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2008 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2010 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2012 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2015 prefixed
= path_join(dest
, sl
);
2019 t
= path_join("..", d
);
2023 if (symlink(t
, prefixed
) < 0)
2024 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2031 static int make_extra_nodes(const char *dest
) {
2032 _cleanup_umask_ mode_t u
;
2038 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2039 _cleanup_free_
char *path
= NULL
;
2040 DeviceNode
*n
= arg_extra_nodes
+ i
;
2042 path
= path_join(dest
, n
->path
);
2046 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2047 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2049 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2051 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2057 static int setup_pts(const char *dest
) {
2058 _cleanup_free_
char *options
= NULL
;
2063 if (arg_selinux_apifs_context
)
2064 (void) asprintf(&options
,
2065 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2066 arg_uid_shift
+ TTY_GID
,
2067 arg_selinux_apifs_context
);
2070 (void) asprintf(&options
,
2071 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2072 arg_uid_shift
+ TTY_GID
);
2077 /* Mount /dev/pts itself */
2078 p
= prefix_roota(dest
, "/dev/pts");
2079 r
= mkdir_errno_wrapper(p
, 0755);
2081 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2083 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2086 r
= userns_lchown(p
, 0, 0);
2088 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2090 /* Create /dev/ptmx symlink */
2091 p
= prefix_roota(dest
, "/dev/ptmx");
2092 if (symlink("pts/ptmx", p
) < 0)
2093 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2094 r
= userns_lchown(p
, 0, 0);
2096 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2098 /* And fix /dev/pts/ptmx ownership */
2099 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2100 r
= userns_lchown(p
, 0, 0);
2102 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2107 static int setup_stdio_as_dev_console(void) {
2111 terminal
= open_terminal("/dev/console", O_RDWR
);
2113 return log_error_errno(terminal
, "Failed to open console: %m");
2115 /* Make sure we can continue logging to the original stderr, even if
2116 * stderr points elsewhere now */
2117 r
= log_dup_console();
2119 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2121 /* invalidates 'terminal' on success and failure */
2122 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2124 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2129 static int setup_dev_console(const char *console
) {
2130 _cleanup_free_
char *p
= NULL
;
2133 /* Create /dev/console symlink */
2134 r
= path_make_relative("/dev", console
, &p
);
2136 return log_error_errno(r
, "Failed to create relative path: %m");
2138 if (symlink(p
, "/dev/console") < 0)
2139 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2144 static int setup_keyring(void) {
2145 key_serial_t keyring
;
2147 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2148 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2149 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2150 * these system calls let's make sure we don't leak anything into the container. */
2152 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2153 if (keyring
== -1) {
2154 if (errno
== ENOSYS
)
2155 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2156 else if (IN_SET(errno
, EACCES
, EPERM
))
2157 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2159 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2165 static int setup_kmsg(int kmsg_socket
) {
2166 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2167 _cleanup_free_
char *fifo
= NULL
;
2168 _cleanup_close_
int fd
= -1;
2169 _cleanup_umask_ mode_t u
;
2172 assert(kmsg_socket
>= 0);
2176 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2177 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2178 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2179 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2181 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2183 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2185 if (mkfifo(fifo
, 0600) < 0)
2186 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2188 from
= TAKE_PTR(fifo
);
2190 r
= mount_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2194 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2196 return log_error_errno(errno
, "Failed to open fifo: %m");
2198 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2199 r
= send_one_fd(kmsg_socket
, fd
, 0);
2201 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2206 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2207 union in_addr_union
*exposed
= userdata
;
2213 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2217 static int setup_hostname(void) {
2220 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2223 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2225 return log_error_errno(r
, "Failed to set hostname: %m");
2230 static int setup_journal(const char *directory
) {
2231 _cleanup_free_
char *d
= NULL
;
2232 const char *dirname
, *p
, *q
;
2238 /* Don't link journals in ephemeral mode */
2242 if (arg_link_journal
== LINK_NO
)
2245 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2247 r
= sd_id128_get_machine(&this_id
);
2249 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2251 if (sd_id128_equal(arg_uuid
, this_id
)) {
2252 log_full(try ? LOG_WARNING
: LOG_ERR
,
2253 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2259 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2260 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2262 bool ignore
= r
== -EROFS
&& try;
2263 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2264 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2265 return ignore
? 0 : r
;
2269 (void) sd_id128_to_string(arg_uuid
, id
);
2271 p
= strjoina("/var/log/journal/", id
);
2272 q
= prefix_roota(directory
, p
);
2274 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2278 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2279 "%s: already a mount point, refusing to use for journal", p
);
2282 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2286 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2287 "%s: already a mount point, refusing to use for journal", q
);
2290 r
= readlink_and_make_absolute(p
, &d
);
2292 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2295 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2297 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2302 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2303 } else if (r
== -EINVAL
) {
2305 if (arg_link_journal
== LINK_GUEST
&&
2308 if (errno
== ENOTDIR
) {
2309 log_error("%s already exists and is neither a symlink nor a directory", p
);
2312 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2314 } else if (r
!= -ENOENT
)
2315 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2317 if (arg_link_journal
== LINK_GUEST
) {
2319 if (symlink(q
, p
) < 0) {
2321 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2324 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2327 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2329 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2333 if (arg_link_journal
== LINK_HOST
) {
2334 /* don't create parents here — if the host doesn't have
2335 * permanent journal set up, don't force it here */
2337 r
= mkdir_errno_wrapper(p
, 0755);
2338 if (r
< 0 && r
!= -EEXIST
) {
2340 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2343 return log_error_errno(r
, "Failed to create %s: %m", p
);
2346 } else if (access(p
, F_OK
) < 0)
2349 if (dir_is_empty(q
) == 0)
2350 log_warning("%s is not empty, proceeding anyway.", q
);
2352 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2354 return log_error_errno(r
, "Failed to create %s: %m", q
);
2356 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2358 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2363 static int drop_capabilities(uid_t uid
) {
2364 CapabilityQuintet q
;
2366 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2367 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2368 * arg_caps_retain. */
2370 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2371 q
= arg_full_capabilities
;
2373 if (q
.bounding
== (uint64_t) -1)
2374 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2376 if (q
.effective
== (uint64_t) -1)
2377 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2379 if (q
.inheritable
== (uint64_t) -1)
2380 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2382 if (q
.permitted
== (uint64_t) -1)
2383 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2385 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2388 if (capability_quintet_mangle(&q
))
2389 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2392 q
= (CapabilityQuintet
) {
2393 .bounding
= arg_caps_retain
,
2394 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2395 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2396 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2397 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2400 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2401 * in order to maintain the same behavior as systemd < 242. */
2402 if (capability_quintet_mangle(&q
))
2403 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2404 "Some capabilities will not be set because they are not in the current bounding set.");
2408 return capability_quintet_enforce(&q
);
2411 static int reset_audit_loginuid(void) {
2412 _cleanup_free_
char *p
= NULL
;
2415 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2418 r
= read_one_line_file("/proc/self/loginuid", &p
);
2422 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2424 /* Already reset? */
2425 if (streq(p
, "4294967295"))
2428 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2431 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2432 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2433 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2434 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2435 "using systemd-nspawn. Sleeping for 5s... (%m)");
2443 static int setup_propagate(const char *root
) {
2447 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2448 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2449 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2450 (void) mkdir_p(p
, 0600);
2452 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2454 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2456 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2458 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2460 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2462 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2464 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2465 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2469 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2473 /* machined will MS_MOVE into that directory, and that's only
2474 * supported for non-shared mounts. */
2475 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2478 static int setup_machine_id(const char *directory
) {
2479 const char *etc_machine_id
;
2483 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2484 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2485 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2486 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2487 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2488 * container behaves nicely). */
2490 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2492 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2494 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2495 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2497 if (sd_id128_is_null(arg_uuid
)) {
2498 r
= sd_id128_randomize(&arg_uuid
);
2500 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2503 if (sd_id128_is_null(id
))
2504 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2505 "Machine ID in container image is zero, refusing.");
2513 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2518 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2521 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2522 if (r
== -EOPNOTSUPP
)
2523 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.");
2525 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2527 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2529 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2531 log_debug("Patched directory tree to match UID/GID range.");
2538 * < 0 : wait_for_terminate() failed to get the state of the
2539 * container, the container was terminated by a signal, or
2540 * failed for an unknown reason. No change is made to the
2541 * container argument.
2542 * > 0 : The program executed in the container terminated with an
2543 * error. The exit code of the program executed in the
2544 * container is returned. The container argument has been set
2545 * to CONTAINER_TERMINATED.
2546 * 0 : The container is being rebooted, has been shut down or exited
2547 * successfully. The container argument has been set to either
2548 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2550 * That is, success is indicated by a return value of zero, and an
2551 * error is indicated by a non-zero value.
2553 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2557 r
= wait_for_terminate(pid
, &status
);
2559 return log_warning_errno(r
, "Failed to wait for container: %m");
2561 switch (status
.si_code
) {
2564 if (status
.si_status
== 0)
2565 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2567 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2569 *container
= CONTAINER_TERMINATED
;
2570 return status
.si_status
;
2573 if (status
.si_status
== SIGINT
) {
2574 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2575 *container
= CONTAINER_TERMINATED
;
2578 } else if (status
.si_status
== SIGHUP
) {
2579 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2580 *container
= CONTAINER_REBOOTED
;
2586 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2587 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2590 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2591 "Container %s failed due to unknown reason.", arg_machine
);
2595 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2598 pid
= PTR_TO_PID(userdata
);
2600 if (kill(pid
, arg_kill_signal
) >= 0) {
2601 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2602 sd_event_source_set_userdata(s
, NULL
);
2607 sd_event_exit(sd_event_source_get_event(s
), 0);
2611 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2617 pid
= PTR_TO_PID(userdata
);
2622 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2623 return log_error_errno(errno
, "Failed to waitid(): %m");
2624 if (si
.si_pid
== 0) /* No pending children. */
2626 if (si
.si_pid
== pid
) {
2627 /* The main process we care for has exited. Return from
2628 * signal handler but leave the zombie. */
2629 sd_event_exit(sd_event_source_get_event(s
), 0);
2633 /* Reap all other children. */
2634 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2640 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2645 pid
= PTR_TO_PID(userdata
);
2647 if (arg_kill_signal
> 0) {
2648 log_info("Container termination requested. Attempting to halt container.");
2649 (void) kill(pid
, arg_kill_signal
);
2651 log_info("Container termination requested. Exiting.");
2652 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2658 static int determine_names(void) {
2661 if (arg_template
&& !arg_directory
&& arg_machine
) {
2663 /* If --template= was specified then we should not
2664 * search for a machine, but instead create a new one
2665 * in /var/lib/machine. */
2667 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2672 if (!arg_image
&& !arg_directory
) {
2674 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2676 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2678 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2680 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2682 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2683 r
= free_and_strdup(&arg_image
, i
->path
);
2685 r
= free_and_strdup(&arg_directory
, i
->path
);
2690 arg_read_only
= arg_read_only
|| i
->read_only
;
2692 r
= safe_getcwd(&arg_directory
);
2694 return log_error_errno(r
, "Failed to determine current directory: %m");
2697 if (!arg_directory
&& !arg_image
)
2698 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2702 if (arg_directory
&& path_equal(arg_directory
, "/"))
2703 arg_machine
= gethostname_malloc();
2708 arg_machine
= strdup(basename(arg_image
));
2710 /* Truncate suffix if there is one */
2711 e
= endswith(arg_machine
, ".raw");
2715 arg_machine
= strdup(basename(arg_directory
));
2720 hostname_cleanup(arg_machine
);
2721 if (!machine_name_is_valid(arg_machine
))
2722 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2724 if (arg_ephemeral
) {
2727 /* Add a random suffix when this is an
2728 * ephemeral machine, so that we can run many
2729 * instances at once without manually having
2730 * to specify -M each time. */
2732 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2743 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2752 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
2754 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2756 return free_and_replace(*p
, chased
);
2759 static int determine_uid_shift(const char *directory
) {
2762 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2767 if (arg_uid_shift
== UID_INVALID
) {
2770 r
= stat(directory
, &st
);
2772 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2774 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2776 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2777 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2778 "UID and GID base of %s don't match.", directory
);
2780 arg_uid_range
= UINT32_C(0x10000);
2783 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2784 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2785 "UID base too high for UID range.");
2790 static unsigned long effective_clone_ns_flags(void) {
2791 unsigned long flags
= arg_clone_ns_flags
;
2793 if (arg_private_network
)
2794 flags
|= CLONE_NEWNET
;
2796 flags
|= CLONE_NEWCGROUP
;
2797 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2798 flags
|= CLONE_NEWUSER
;
2803 static int patch_sysctl(void) {
2805 /* This table is inspired by runc's sysctl() function */
2806 static const struct {
2809 unsigned long clone_flags
;
2811 { "kernel.hostname", false, CLONE_NEWUTS
},
2812 { "kernel.domainname", false, CLONE_NEWUTS
},
2813 { "kernel.msgmax", false, CLONE_NEWIPC
},
2814 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2815 { "kernel.msgmni", false, CLONE_NEWIPC
},
2816 { "kernel.sem", false, CLONE_NEWIPC
},
2817 { "kernel.shmall", false, CLONE_NEWIPC
},
2818 { "kernel.shmmax", false, CLONE_NEWIPC
},
2819 { "kernel.shmmni", false, CLONE_NEWIPC
},
2820 { "fs.mqueue.", true, CLONE_NEWIPC
},
2821 { "net.", true, CLONE_NEWNET
},
2824 unsigned long flags
;
2828 flags
= effective_clone_ns_flags();
2830 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2834 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2836 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2839 if (safe_sysctl
[i
].prefix
)
2840 good
= startswith(*k
, safe_sysctl
[i
].key
);
2842 good
= streq(*k
, safe_sysctl
[i
].key
);
2849 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2851 r
= sysctl_write(*k
, *v
);
2853 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2859 static int inner_child(
2861 const char *directory
,
2865 int master_pty_socket
,
2868 _cleanup_free_
char *home
= NULL
;
2869 char as_uuid
[ID128_UUID_STRING_MAX
];
2871 const char *envp
[] = {
2872 "PATH=" DEFAULT_PATH_COMPAT
,
2873 NULL
, /* container */
2878 NULL
, /* container_uuid */
2879 NULL
, /* LISTEN_FDS */
2880 NULL
, /* LISTEN_PID */
2881 NULL
, /* NOTIFY_SOCKET */
2884 const char *exec_target
;
2885 _cleanup_strv_free_
char **env_use
= NULL
;
2886 int r
, which_failed
;
2888 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2889 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2890 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2891 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2892 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2893 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2896 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2897 * unshare(). See below. */
2901 assert(kmsg_socket
>= 0);
2903 log_debug("Inner child is initializing.");
2905 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2906 /* Tell the parent, that it now can write the UID map. */
2907 (void) barrier_place(barrier
); /* #1 */
2909 /* Wait until the parent wrote the UID map */
2910 if (!barrier_place_and_sync(barrier
)) /* #2 */
2911 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2912 "Parent died too early");
2915 r
= reset_uid_gid();
2917 return log_error_errno(r
, "Couldn't become new root: %m");
2920 arg_mount_settings
| MOUNT_IN_USERNS
,
2922 arg_selinux_apifs_context
);
2926 if (!arg_network_namespace_path
&& arg_private_network
) {
2927 r
= unshare(CLONE_NEWNET
);
2929 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2931 /* Tell the parent that it can setup network interfaces. */
2932 (void) barrier_place(barrier
); /* #3 */
2935 r
= mount_sysfs(NULL
, arg_mount_settings
);
2939 /* Wait until we are cgroup-ified, so that we
2940 * can mount the right cgroup path writable */
2941 if (!barrier_place_and_sync(barrier
)) /* #4 */
2942 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2943 "Parent died too early");
2946 r
= unshare(CLONE_NEWCGROUP
);
2948 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2951 arg_unified_cgroup_hierarchy
,
2952 arg_userns_mode
!= USER_NAMESPACE_NO
,
2955 arg_selinux_apifs_context
,
2960 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2965 r
= setup_boot_id();
2969 r
= setup_kmsg(kmsg_socket
);
2972 kmsg_socket
= safe_close(kmsg_socket
);
2977 arg_n_custom_mounts
,
2981 arg_selinux_apifs_context
,
2987 return log_error_errno(errno
, "setsid() failed: %m");
2989 if (arg_private_network
)
2992 if (arg_expose_ports
) {
2993 r
= expose_port_send_rtnl(rtnl_socket
);
2996 rtnl_socket
= safe_close(rtnl_socket
);
2999 if (arg_console_mode
!= CONSOLE_PIPE
) {
3000 _cleanup_close_
int master
= -1;
3001 _cleanup_free_
char *console
= NULL
;
3003 /* Allocate a pty and make it available as /dev/console. */
3004 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3006 return log_error_errno(master
, "Failed to allocate a pty: %m");
3008 r
= setup_dev_console(console
);
3010 return log_error_errno(r
, "Failed to setup /dev/console: %m");
3012 r
= send_one_fd(master_pty_socket
, master
, 0);
3014 return log_error_errno(r
, "Failed to send master fd: %m");
3015 master_pty_socket
= safe_close(master_pty_socket
);
3017 r
= setup_stdio_as_dev_console();
3026 if (arg_oom_score_adjust_set
) {
3027 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3029 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3032 if (arg_cpu_set
.set
)
3033 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3034 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3036 (void) setup_hostname();
3038 if (arg_personality
!= PERSONALITY_INVALID
) {
3039 r
= safe_personality(arg_personality
);
3041 return log_error_errno(r
, "personality() failed: %m");
3042 } else if (secondary
) {
3043 r
= safe_personality(PER_LINUX32
);
3045 return log_error_errno(r
, "personality() failed: %m");
3048 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3050 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3055 if (is_seccomp_available()) {
3057 r
= seccomp_load(arg_seccomp
);
3058 if (ERRNO_IS_SECCOMP_FATAL(r
))
3059 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3061 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3066 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
3072 if (arg_selinux_context
)
3073 if (setexeccon(arg_selinux_context
) < 0)
3074 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3077 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3078 * if we need to later on. */
3079 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3080 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3082 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3083 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
3085 r
= change_uid_gid(arg_user
, &home
);
3089 r
= drop_capabilities(getuid());
3091 return log_error_errno(r
, "Dropping capabilities failed: %m");
3093 if (arg_no_new_privileges
)
3094 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3095 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3097 /* LXC sets container=lxc, so follow the scheme here */
3098 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3100 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3104 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3105 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3108 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3109 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3110 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3113 assert(!sd_id128_is_null(arg_uuid
));
3115 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3118 if (fdset_size(fds
) > 0) {
3119 r
= fdset_cloexec(fds
, false);
3121 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3123 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3124 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3127 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3130 env_use
= strv_env_merge(2, envp
, arg_setenv
);
3134 /* Let the parent know that we are ready and
3135 * wait until the parent is ready with the
3137 if (!barrier_place_and_sync(barrier
)) /* #5 */
3138 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3139 "Parent died too early");
3142 if (chdir(arg_chdir
) < 0)
3143 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3145 if (arg_start_mode
== START_PID2
) {
3146 r
= stub_pid1(arg_uuid
);
3151 log_debug("Inner child completed, invoking payload.");
3153 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3154 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3155 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3157 log_set_open_when_needed(true);
3159 (void) fdset_close_others(fds
);
3161 if (arg_start_mode
== START_BOOT
) {
3165 /* Automatically search for the init system */
3167 m
= strv_length(arg_parameters
);
3168 a
= newa(char*, m
+ 2);
3169 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3172 a
[0] = (char*) "/usr/lib/systemd/systemd";
3173 execve(a
[0], a
, env_use
);
3175 a
[0] = (char*) "/lib/systemd/systemd";
3176 execve(a
[0], a
, env_use
);
3178 a
[0] = (char*) "/sbin/init";
3179 execve(a
[0], a
, env_use
);
3181 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3182 } else if (!strv_isempty(arg_parameters
)) {
3183 const char *dollar_path
;
3185 exec_target
= arg_parameters
[0];
3187 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3189 dollar_path
= strv_env_get(env_use
, "PATH");
3191 if (putenv((char*) dollar_path
) != 0)
3192 return log_error_errno(errno
, "Failed to update $PATH: %m");
3195 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3198 /* If we cannot change the directory, we'll end up in /, that is expected. */
3199 (void) chdir(home
?: "/root");
3201 execle("/bin/bash", "-bash", NULL
, env_use
);
3202 execle("/bin/sh", "-sh", NULL
, env_use
);
3204 exec_target
= "/bin/bash, /bin/sh";
3207 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3210 static int setup_sd_notify_child(void) {
3211 _cleanup_close_
int fd
= -1;
3212 union sockaddr_union sa
= {
3213 .un
.sun_family
= AF_UNIX
,
3214 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3218 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3220 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3222 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3223 (void) sockaddr_un_unlink(&sa
.un
);
3225 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3227 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3229 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3231 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3233 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3235 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3240 static int outer_child(
3242 const char *directory
,
3243 DissectedImage
*dissected_image
,
3250 int uid_shift_socket
,
3251 int master_pty_socket
,
3252 int unified_cgroup_hierarchy_socket
,
3256 _cleanup_close_
int fd
= -1;
3261 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3262 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3263 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3264 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3268 assert(pid_socket
>= 0);
3269 assert(uuid_socket
>= 0);
3270 assert(notify_socket
>= 0);
3271 assert(master_pty_socket
>= 0);
3272 assert(kmsg_socket
>= 0);
3274 log_debug("Outer child is initializing.");
3276 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3277 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3279 r
= reset_audit_loginuid();
3283 /* Mark everything as slave, so that we still
3284 * receive mounts from the real root, but don't
3285 * propagate mounts to the real root. */
3286 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3290 if (dissected_image
) {
3291 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3292 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3293 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3294 * makes sure ESP partitions and userns are compatible. */
3296 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3297 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3298 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0)|
3299 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3304 r
= determine_uid_shift(directory
);
3308 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3309 /* Let the parent know which UID shift we read from the image */
3310 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3312 return log_error_errno(errno
, "Failed to send UID shift: %m");
3313 if (l
!= sizeof(arg_uid_shift
))
3314 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3315 "Short write while sending UID shift.");
3317 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3318 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3319 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3320 * not it will pick a different one, and send it back to us. */
3322 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3324 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3325 if (l
!= sizeof(arg_uid_shift
))
3326 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3327 "Short read while receiving UID shift.");
3330 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3331 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3334 if (path_equal(directory
, "/")) {
3335 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3336 * place, so that we can make changes to its mount structure (for example, to implement
3337 * --volatile=) without this interfering with our ability to access files such as
3338 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3339 * (instead of a temporary directory, since we are living in our own mount namspace here
3340 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3341 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3343 r
= mount_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3347 directory
= "/run/systemd/nspawn-root";
3349 } else if (!dissected_image
) {
3350 /* Turn directory into bind mount (we need that so that we can move the bind mount to root
3352 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3357 r
= setup_pivot_root(
3360 arg_pivot_root_old
);
3364 r
= setup_volatile_mode(
3367 arg_userns_mode
!= USER_NAMESPACE_NO
,
3370 arg_selinux_apifs_context
);
3374 if (dissected_image
) {
3375 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3376 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3377 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
3382 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3383 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3385 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3389 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3391 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3392 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3393 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3394 "Short write while sending cgroup mode.");
3396 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3399 /* Mark everything as shared so our mounts get propagated down. This is
3400 * required to make new bind mounts available in systemd services
3401 * inside the container that create a new mount namespace.
3402 * See https://github.com/systemd/systemd/issues/3860
3403 * Further submounts (such as /dev) done after this will inherit the
3404 * shared propagation mode. */
3405 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3409 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3413 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3417 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
) {
3418 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3420 return log_error_errno(r
, "Failed to make tree read-only: %m");
3423 r
= mount_all(directory
,
3426 arg_selinux_apifs_context
);
3430 r
= copy_devnodes(directory
);
3434 r
= make_extra_nodes(directory
);
3438 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3439 (void) make_inaccessible_nodes(directory
, arg_uid_shift
, arg_uid_shift
);
3441 r
= setup_pts(directory
);
3445 r
= setup_propagate(directory
);
3449 r
= setup_keyring();
3453 r
= setup_timezone(directory
);
3457 r
= setup_resolv_conf(directory
);
3461 r
= setup_machine_id(directory
);
3465 r
= setup_journal(directory
);
3472 arg_n_custom_mounts
,
3473 arg_userns_mode
!= USER_NAMESPACE_NO
,
3476 arg_selinux_apifs_context
,
3481 if (!arg_use_cgns
) {
3484 arg_unified_cgroup_hierarchy
,
3485 arg_userns_mode
!= USER_NAMESPACE_NO
,
3488 arg_selinux_apifs_context
,
3494 r
= mount_move_root(directory
);
3496 return log_error_errno(r
, "Failed to move root directory: %m");
3498 fd
= setup_sd_notify_child();
3502 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3503 arg_clone_ns_flags
|
3504 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3506 return log_error_errno(errno
, "Failed to fork inner child: %m");
3508 pid_socket
= safe_close(pid_socket
);
3509 uuid_socket
= safe_close(uuid_socket
);
3510 notify_socket
= safe_close(notify_socket
);
3511 uid_shift_socket
= safe_close(uid_shift_socket
);
3513 /* The inner child has all namespaces that are
3514 * requested, so that we all are owned by the user if
3515 * user namespaces are turned on. */
3517 if (arg_network_namespace_path
) {
3518 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3520 return log_error_errno(r
, "Failed to join network namespace: %m");
3523 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
);
3525 _exit(EXIT_FAILURE
);
3527 _exit(EXIT_SUCCESS
);
3530 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3532 return log_error_errno(errno
, "Failed to send PID: %m");
3533 if (l
!= sizeof(pid
))
3534 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3535 "Short write while sending PID.");
3537 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3539 return log_error_errno(errno
, "Failed to send machine ID: %m");
3540 if (l
!= sizeof(arg_uuid
))
3541 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3542 "Short write while sending machine ID.");
3544 l
= send_one_fd(notify_socket
, fd
, 0);
3546 return log_error_errno(l
, "Failed to send notify fd: %m");
3548 pid_socket
= safe_close(pid_socket
);
3549 uuid_socket
= safe_close(uuid_socket
);
3550 notify_socket
= safe_close(notify_socket
);
3551 master_pty_socket
= safe_close(master_pty_socket
);
3552 kmsg_socket
= safe_close(kmsg_socket
);
3553 rtnl_socket
= safe_close(rtnl_socket
);
3554 netns_fd
= safe_close(netns_fd
);
3559 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3560 bool tried_hashed
= false;
3561 unsigned n_tries
= 100;
3566 assert(ret_lock_file
);
3567 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3568 assert(arg_uid_range
== 0x10000U
);
3572 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3575 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3576 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3581 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3583 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3586 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3587 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3588 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3593 /* Make some superficial checks whether the range is currently known in the user database */
3594 if (getpwuid(candidate
))
3596 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3598 if (getgrgid(candidate
))
3600 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3603 *ret_lock_file
= lf
;
3604 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3609 if (arg_machine
&& !tried_hashed
) {
3610 /* Try to hash the base from the container name */
3612 static const uint8_t hash_key
[] = {
3613 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3614 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3617 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3619 tried_hashed
= true;
3621 random_bytes(&candidate
, sizeof(candidate
));
3623 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3624 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3628 static int setup_uid_map(pid_t pid
) {
3629 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3634 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3635 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3636 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3638 return log_error_errno(r
, "Failed to write UID map: %m");
3640 /* We always assign the same UID and GID ranges */
3641 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3642 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3644 return log_error_errno(r
, "Failed to write GID map: %m");
3649 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3650 char buf
[NOTIFY_BUFFER_MAX
+1];
3652 struct iovec iovec
= {
3654 .iov_len
= sizeof(buf
)-1,
3657 struct cmsghdr cmsghdr
;
3658 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
3659 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
3661 struct msghdr msghdr
= {
3664 .msg_control
= &control
,
3665 .msg_controllen
= sizeof(control
),
3667 struct cmsghdr
*cmsg
;
3668 struct ucred
*ucred
= NULL
;
3670 pid_t inner_child_pid
;
3671 _cleanup_strv_free_
char **tags
= NULL
;
3675 inner_child_pid
= PTR_TO_PID(userdata
);
3677 if (revents
!= EPOLLIN
) {
3678 log_warning("Got unexpected poll event for notify fd.");
3682 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3684 if (IN_SET(errno
, EAGAIN
, EINTR
))
3687 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
3689 cmsg_close_all(&msghdr
);
3691 CMSG_FOREACH(cmsg
, &msghdr
) {
3692 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
3693 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
3694 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
3696 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
3700 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3701 log_debug("Received notify message without valid credentials. Ignoring.");
3705 if ((size_t) n
>= sizeof(buf
)) {
3706 log_warning("Received notify message exceeded maximum size. Ignoring.");
3711 tags
= strv_split(buf
, "\n\r");
3715 if (strv_find(tags
, "READY=1"))
3716 (void) sd_notifyf(false, "READY=1\n");
3718 p
= strv_find_startswith(tags
, "STATUS=");
3720 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3725 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3728 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3730 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3732 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3737 static int merge_settings(Settings
*settings
, const char *path
) {
3743 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3744 * that this steals the fields of the Settings* structure, and hence modifies it. */
3746 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3747 settings
->start_mode
>= 0) {
3748 arg_start_mode
= settings
->start_mode
;
3749 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3752 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3753 arg_ephemeral
= settings
->ephemeral
;
3755 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3758 if (!arg_settings_trusted
)
3759 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3761 free_and_replace(arg_directory
, settings
->root
);
3764 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3765 settings
->pivot_root_new
) {
3766 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3767 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3770 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3771 settings
->working_directory
)
3772 free_and_replace(arg_chdir
, settings
->working_directory
);
3774 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3775 settings
->environment
)
3776 strv_free_and_replace(arg_setenv
, settings
->environment
);
3778 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3781 free_and_replace(arg_user
, settings
->user
);
3783 if (uid_is_valid(settings
->uid
))
3784 arg_uid
= settings
->uid
;
3785 if (gid_is_valid(settings
->gid
))
3786 arg_gid
= settings
->gid
;
3787 if (settings
->n_supplementary_gids
> 0) {
3788 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3789 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3793 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3794 uint64_t plus
, minus
;
3795 uint64_t network_minus
= 0;
3797 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3798 * Settings structure */
3800 plus
= settings
->capability
;
3801 minus
= settings
->drop_capability
;
3803 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3804 if (settings_private_network(settings
))
3805 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3807 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3810 if (!arg_settings_trusted
&& plus
!= 0) {
3811 if (settings
->capability
!= 0)
3812 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3814 arg_caps_retain
&= ~network_minus
;
3815 arg_caps_retain
|= plus
;
3818 arg_caps_retain
&= ~minus
;
3820 /* Copy the full capabilities over too */
3821 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3822 if (!arg_settings_trusted
)
3823 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
3825 arg_full_capabilities
= settings
->full_capabilities
;
3829 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3830 settings
->kill_signal
> 0)
3831 arg_kill_signal
= settings
->kill_signal
;
3833 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3834 settings
->personality
!= PERSONALITY_INVALID
)
3835 arg_personality
= settings
->personality
;
3837 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3838 !sd_id128_is_null(settings
->machine_id
)) {
3840 if (!arg_settings_trusted
)
3841 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3843 arg_uuid
= settings
->machine_id
;
3846 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3847 settings
->read_only
>= 0)
3848 arg_read_only
= settings
->read_only
;
3850 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3851 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3852 arg_volatile_mode
= settings
->volatile_mode
;
3854 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3855 settings
->n_custom_mounts
> 0) {
3857 if (!arg_settings_trusted
)
3858 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3860 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3861 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3862 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3863 settings
->n_custom_mounts
= 0;
3867 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3868 (settings
->private_network
>= 0 ||
3869 settings
->network_veth
>= 0 ||
3870 settings
->network_bridge
||
3871 settings
->network_zone
||
3872 settings
->network_interfaces
||
3873 settings
->network_macvlan
||
3874 settings
->network_ipvlan
||
3875 settings
->network_veth_extra
||
3876 settings
->network_namespace_path
)) {
3878 if (!arg_settings_trusted
)
3879 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3881 arg_network_veth
= settings_network_veth(settings
);
3882 arg_private_network
= settings_private_network(settings
);
3884 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3885 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3886 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3887 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3889 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3890 free_and_replace(arg_network_zone
, settings
->network_zone
);
3892 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3896 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3897 settings
->expose_ports
) {
3899 if (!arg_settings_trusted
)
3900 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3902 expose_port_free_all(arg_expose_ports
);
3903 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3907 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3908 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3910 if (!arg_settings_trusted
)
3911 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3913 arg_userns_mode
= settings
->userns_mode
;
3914 arg_uid_shift
= settings
->uid_shift
;
3915 arg_uid_range
= settings
->uid_range
;
3916 arg_userns_chown
= settings
->userns_chown
;
3920 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3921 arg_notify_ready
= settings
->notify_ready
;
3923 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3925 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_whitelist
))
3926 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3928 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3929 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3933 if (!arg_settings_trusted
&& settings
->seccomp
)
3934 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
3936 seccomp_release(arg_seccomp
);
3937 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
3942 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3943 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3946 if (!settings
->rlimit
[rl
])
3949 if (!arg_settings_trusted
) {
3950 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3954 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3957 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3959 free_and_replace(arg_hostname
, settings
->hostname
);
3961 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3962 settings
->no_new_privileges
>= 0)
3963 arg_no_new_privileges
= settings
->no_new_privileges
;
3965 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3966 settings
->oom_score_adjust_set
) {
3968 if (!arg_settings_trusted
)
3969 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
3971 arg_oom_score_adjust
= settings
->oom_score_adjust
;
3972 arg_oom_score_adjust_set
= true;
3976 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
3977 settings
->cpu_set
.set
) {
3979 if (!arg_settings_trusted
)
3980 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
3982 cpu_set_reset(&arg_cpu_set
);
3983 arg_cpu_set
= settings
->cpu_set
;
3984 settings
->cpu_set
= (CPUSet
) {};
3988 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
3989 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
3990 arg_resolv_conf
= settings
->resolv_conf
;
3992 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
3993 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
3995 if (!arg_settings_trusted
)
3996 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
3998 arg_link_journal
= settings
->link_journal
;
3999 arg_link_journal_try
= settings
->link_journal_try
;
4003 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4004 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4005 arg_timezone
= settings
->timezone
;
4007 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4010 if (!arg_settings_trusted
)
4011 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4013 free_and_replace(arg_slice
, settings
->slice
);
4016 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4017 settings
->use_cgns
>= 0) {
4019 if (!arg_settings_trusted
)
4020 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4022 arg_use_cgns
= settings
->use_cgns
;
4025 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4026 settings
->clone_ns_flags
!= (unsigned long) -1) {
4028 if (!arg_settings_trusted
)
4029 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4031 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4034 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4035 settings
->console_mode
>= 0) {
4037 if (!arg_settings_trusted
)
4038 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4040 arg_console_mode
= settings
->console_mode
;
4043 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4044 * don't consult arg_settings_mask for them. */
4046 sd_bus_message_unref(arg_property_message
);
4047 arg_property_message
= TAKE_PTR(settings
->properties
);
4049 arg_console_width
= settings
->console_width
;
4050 arg_console_height
= settings
->console_height
;
4052 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4053 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4054 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4059 static int load_settings(void) {
4060 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4061 _cleanup_fclose_
FILE *f
= NULL
;
4062 _cleanup_free_
char *p
= NULL
;
4069 /* If all settings are masked, there's no point in looking for
4070 * the settings file */
4071 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
4074 fn
= strjoina(arg_machine
, ".nspawn");
4076 /* We first look in the admin's directories in /etc and /run */
4077 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4078 _cleanup_free_
char *j
= NULL
;
4080 j
= path_join(i
, fn
);
4088 /* By default, we trust configuration from /etc and /run */
4089 if (arg_settings_trusted
< 0)
4090 arg_settings_trusted
= true;
4095 if (errno
!= ENOENT
)
4096 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4100 /* After that, let's look for a file next to the
4101 * actual image we shall boot. */
4104 p
= file_in_same_dir(arg_image
, fn
);
4107 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4108 p
= file_in_same_dir(arg_directory
, fn
);
4115 if (!f
&& errno
!= ENOENT
)
4116 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4118 /* By default, we do not trust configuration from /var/lib/machines */
4119 if (arg_settings_trusted
< 0)
4120 arg_settings_trusted
= false;
4127 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4129 r
= settings_load(f
, p
, &settings
);
4133 return merge_settings(settings
, p
);
4136 static int load_oci_bundle(void) {
4137 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4140 if (!arg_oci_bundle
)
4143 /* By default let's trust OCI bundles */
4144 if (arg_settings_trusted
< 0)
4145 arg_settings_trusted
= true;
4147 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4151 return merge_settings(settings
, arg_oci_bundle
);
4154 static int run_container(
4155 DissectedImage
*dissected_image
,
4158 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4159 union in_addr_union
*exposed
,
4160 int *master
, pid_t
*pid
, int *ret
) {
4162 static const struct sigaction sa
= {
4163 .sa_handler
= nop_signal_handler
,
4164 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4167 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4168 _cleanup_close_
int etc_passwd_lock
= -1;
4169 _cleanup_close_pair_
int
4170 kmsg_socket_pair
[2] = { -1, -1 },
4171 rtnl_socket_pair
[2] = { -1, -1 },
4172 pid_socket_pair
[2] = { -1, -1 },
4173 uuid_socket_pair
[2] = { -1, -1 },
4174 notify_socket_pair
[2] = { -1, -1 },
4175 uid_shift_socket_pair
[2] = { -1, -1 },
4176 master_pty_socket_pair
[2] = { -1, -1 },
4177 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4179 _cleanup_close_
int notify_socket
= -1;
4180 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4181 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4182 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4183 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4184 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4185 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4186 ContainerStatus container_status
= 0;
4190 _cleanup_close_
int netns_fd
= -1;
4192 assert_se(sigemptyset(&mask_chld
) == 0);
4193 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4195 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4196 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4197 * check with getpwuid() if the specific user already exists. Note that /etc might be
4198 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4199 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4200 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4203 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4204 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4205 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4208 r
= barrier_create(&barrier
);
4210 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4212 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4213 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4215 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4216 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4218 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4219 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4221 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4222 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4224 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4225 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4227 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4228 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4230 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4231 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4232 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4234 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4235 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4236 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4238 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4239 * parent's blocking calls and give it a chance to call wait() and terminate. */
4240 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4242 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4244 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4246 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4248 if (arg_network_namespace_path
) {
4249 netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4251 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4253 r
= fd_is_network_ns(netns_fd
);
4255 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
);
4257 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4259 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4260 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4263 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4265 return log_error_errno(errno
, "clone() failed%s: %m",
4267 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4270 /* The outer child only has a file system namespace. */
4271 barrier_set_role(&barrier
, BARRIER_CHILD
);
4273 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4274 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4275 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4276 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4277 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4278 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4279 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4280 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4282 (void) reset_all_signal_handlers();
4283 (void) reset_signal_mask();
4285 r
= outer_child(&barrier
,
4290 uuid_socket_pair
[1],
4291 notify_socket_pair
[1],
4292 kmsg_socket_pair
[1],
4293 rtnl_socket_pair
[1],
4294 uid_shift_socket_pair
[1],
4295 master_pty_socket_pair
[1],
4296 unified_cgroup_hierarchy_socket_pair
[1],
4300 _exit(EXIT_FAILURE
);
4302 _exit(EXIT_SUCCESS
);
4305 barrier_set_role(&barrier
, BARRIER_PARENT
);
4309 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4310 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4311 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4312 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4313 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4314 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4315 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4316 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4318 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4319 /* The child just let us know the UID shift it might have read from the image. */
4320 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4322 return log_error_errno(errno
, "Failed to read UID shift: %m");
4323 if (l
!= sizeof arg_uid_shift
)
4324 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4326 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4327 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4328 * image, but if that's already in use, pick a new one, and report back to the child,
4329 * which one we now picked. */
4331 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4333 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4335 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4337 return log_error_errno(errno
, "Failed to send UID shift: %m");
4338 if (l
!= sizeof arg_uid_shift
)
4339 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4343 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4344 /* The child let us know the support cgroup mode it might have read from the image. */
4345 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4347 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4348 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4349 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4350 l
, l
== 0 ? " The child is most likely dead." : "");
4353 /* Wait for the outer child. */
4354 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4357 if (r
!= EXIT_SUCCESS
)
4360 /* And now retrieve the PID of the inner child. */
4361 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4363 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4364 if (l
!= sizeof *pid
)
4365 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4367 /* We also retrieve container UUID in case it was generated by outer child */
4368 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4370 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4371 if (l
!= sizeof(arg_uuid
))
4372 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4374 /* We also retrieve the socket used for notifications generated by outer child */
4375 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4376 if (notify_socket
< 0)
4377 return log_error_errno(notify_socket
,
4378 "Failed to receive notification socket from the outer child: %m");
4380 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4382 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4383 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4384 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4386 r
= setup_uid_map(*pid
);
4390 (void) barrier_place(&barrier
); /* #2 */
4393 if (arg_private_network
) {
4394 if (!arg_network_namespace_path
) {
4395 /* Wait until the child has unshared its network namespace. */
4396 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4397 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4400 r
= move_network_interfaces(*pid
, arg_network_interfaces
);
4404 if (arg_network_veth
) {
4405 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4406 arg_network_bridge
|| arg_network_zone
);
4412 if (arg_network_bridge
) {
4413 /* Add the interface to a bridge */
4414 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4419 } else if (arg_network_zone
) {
4420 /* Add the interface to a bridge, possibly creating it */
4421 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4429 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4433 /* We created the primary and extra veth links now; let's remember this, so that we know to
4434 remove them later on. Note that we don't bother with removing veth links that were created
4435 here when their setup failed half-way, because in that case the kernel should be able to
4436 remove them on its own, since they cannot be referenced by anything yet. */
4437 *veth_created
= true;
4439 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4443 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4448 if (arg_register
|| !arg_keep_unit
) {
4449 r
= sd_bus_default_system(&bus
);
4451 return log_error_errno(r
, "Failed to open system bus: %m");
4453 r
= sd_bus_set_close_on_exit(bus
, false);
4455 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4458 if (!arg_keep_unit
) {
4459 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4460 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4461 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4463 r
= sd_bus_match_signal_async(
4466 "org.freedesktop.systemd1",
4468 "org.freedesktop.systemd1.Scope",
4470 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4472 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4476 r
= register_machine(
4484 arg_custom_mounts
, arg_n_custom_mounts
,
4487 arg_property_message
,
4489 arg_container_service_name
);
4493 } else if (!arg_keep_unit
) {
4499 arg_custom_mounts
, arg_n_custom_mounts
,
4502 arg_property_message
);
4506 } else if (arg_slice
|| arg_property
)
4507 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4509 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4513 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4517 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4521 /* Notify the child that the parent is ready with all
4522 * its setup (including cgroup-ification), and that
4523 * the child can now hand over control to the code to
4524 * run inside the container. */
4525 (void) barrier_place(&barrier
); /* #4 */
4527 /* Block SIGCHLD here, before notifying child.
4528 * process_pty() will handle it with the other signals. */
4529 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4531 /* Reset signal to default */
4532 r
= default_signals(SIGCHLD
, -1);
4534 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4536 r
= sd_event_new(&event
);
4538 return log_error_errno(r
, "Failed to get default event source: %m");
4540 (void) sd_event_set_watchdog(event
, true);
4543 r
= sd_bus_attach_event(bus
, event
, 0);
4545 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4548 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4552 /* Let the child know that we are ready and wait that the child is completely ready now. */
4553 if (!barrier_place_and_sync(&barrier
)) /* #5 */
4554 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4556 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4557 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4558 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4560 (void) sd_notifyf(false,
4561 "STATUS=Container running.\n"
4562 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4563 if (!arg_notify_ready
)
4564 (void) sd_notify(false, "READY=1\n");
4566 if (arg_kill_signal
> 0) {
4567 /* Try to kill the init system on SIGINT or SIGTERM */
4568 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4569 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4571 /* Immediately exit */
4572 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4573 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4576 /* Exit when the child exits */
4577 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4579 if (arg_expose_ports
) {
4580 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4584 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4587 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4589 if (arg_console_mode
!= CONSOLE_PIPE
) {
4590 _cleanup_close_
int fd
= -1;
4591 PTYForwardFlags flags
= 0;
4593 /* Retrieve the master pty allocated by inner child */
4594 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
4596 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
4598 switch (arg_console_mode
) {
4600 case CONSOLE_READ_ONLY
:
4601 flags
|= PTY_FORWARD_READ_ONLY
;
4605 case CONSOLE_INTERACTIVE
:
4606 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
4608 r
= pty_forward_new(event
, fd
, flags
, &forward
);
4610 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4612 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4613 (void) pty_forward_set_width_height(forward
,
4615 arg_console_height
);
4619 assert(arg_console_mode
== CONSOLE_PASSIVE
);
4622 *master
= TAKE_FD(fd
);
4625 r
= sd_event_loop(event
);
4627 return log_error_errno(r
, "Failed to run event loop: %m");
4632 (void) pty_forward_get_last_char(forward
, &last_char
);
4633 forward
= pty_forward_free(forward
);
4635 if (!arg_quiet
&& last_char
!= '\n')
4639 /* Kill if it is not dead yet anyway */
4640 if (!arg_register
&& !arg_keep_unit
&& bus
)
4641 terminate_scope(bus
, arg_machine
);
4643 /* Normally redundant, but better safe than sorry */
4644 (void) kill(*pid
, SIGKILL
);
4646 r
= wait_for_container(*pid
, &container_status
);
4649 /* Tell machined that we are gone. */
4651 (void) unregister_machine(bus
, arg_machine
);
4654 /* We failed to wait for the container, or the container exited abnormally. */
4656 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4657 /* r > 0 → The container exited with a non-zero status.
4658 * As a special case, we need to replace 133 with a different value,
4659 * because 133 is special-cased in the service file to reboot the container.
4660 * otherwise → The container exited with zero status and a reboot was not requested.
4662 if (r
== EXIT_FORCE_RESTART
)
4663 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4665 return 0; /* finito */
4668 /* CONTAINER_REBOOTED, loop again */
4670 if (arg_keep_unit
) {
4671 /* Special handling if we are running as a service: instead of simply
4672 * restarting the machine we want to restart the entire service, so let's
4673 * inform systemd about this with the special exit code 133. The service
4674 * file uses RestartForceExitStatus=133 so that this results in a full
4675 * nspawn restart. This is necessary since we might have cgroup parameters
4676 * set we want to have flushed out. */
4677 *ret
= EXIT_FORCE_RESTART
;
4678 return 0; /* finito */
4681 expose_port_flush(arg_expose_ports
, exposed
);
4683 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4684 *veth_created
= false;
4685 return 1; /* loop again */
4688 static int initialize_rlimits(void) {
4689 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4690 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4691 * container execution environments. */
4693 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4694 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4695 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4696 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4697 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4698 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4699 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4700 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4701 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4702 [RLIMIT_NICE
] = { 0, 0 },
4703 [RLIMIT_NOFILE
] = { 1024, 4096 },
4704 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4705 [RLIMIT_RTPRIO
] = { 0, 0 },
4706 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4707 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4709 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4710 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4711 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4712 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4713 * that PID 1 changes a number of other resource limits during early initialization which is why we
4714 * don't read the other limits from PID 1 but prefer the static table above. */
4719 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4720 /* Let's only fill in what the user hasn't explicitly configured anyway */
4721 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4722 const struct rlimit
*v
;
4723 struct rlimit buffer
;
4725 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4726 /* For these two let's read the limits off PID 1. See above for an explanation. */
4728 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4729 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4733 v
= kernel_defaults
+ rl
;
4735 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4736 if (!arg_rlimit
[rl
])
4740 if (DEBUG_LOGGING
) {
4741 _cleanup_free_
char *k
= NULL
;
4743 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4744 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4751 static int run(int argc
, char *argv
[]) {
4752 bool secondary
= false, remove_directory
= false, remove_image
= false,
4753 veth_created
= false, remove_tmprootdir
= false;
4754 _cleanup_close_
int master
= -1;
4755 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4756 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4757 char veth_name
[IFNAMSIZ
] = "";
4758 union in_addr_union exposed
= {};
4759 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4760 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4761 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4762 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4763 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4766 log_parse_environment();
4769 r
= parse_argv(argc
, argv
);
4777 r
= initialize_rlimits();
4781 r
= load_oci_bundle();
4785 r
= determine_names();
4789 r
= load_settings();
4795 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4799 r
= verify_arguments();
4803 /* Reapply environment settings. */
4804 (void) detect_unified_cgroup_hierarchy_from_environment();
4806 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4807 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4808 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4809 (void) ignore_signals(SIGPIPE
, -1);
4811 n_fd_passed
= sd_listen_fds(false);
4812 if (n_fd_passed
> 0) {
4813 r
= fdset_new_listen_fds(&fds
, false);
4815 log_error_errno(r
, "Failed to collect file descriptors: %m");
4820 /* The "default" umask. This is appropriate for most file and directory
4821 * operations performed by nspawn, and is the umask that will be used for
4822 * the child. Functions like copy_devnodes() change the umask temporarily. */
4825 if (arg_directory
) {
4828 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4829 * /var from the host will propagate into container dynamically (because bad things happen if
4830 * two systems write to the same /var). Let's allow it for the special cases where /var is
4831 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4832 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
4833 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4838 if (arg_ephemeral
) {
4839 _cleanup_free_
char *np
= NULL
;
4841 r
= chase_symlinks_and_update(&arg_directory
, 0);
4845 /* If the specified path is a mount point we generate the new snapshot immediately
4846 * inside it under a random name. However if the specified is not a mount point we
4847 * create the new snapshot in the parent directory, just next to it. */
4848 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4850 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4854 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4856 r
= tempfn_random(arg_directory
, "machine.", &np
);
4858 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4862 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4863 * only owned by us and noone else. */
4864 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4866 log_error_errno(r
, "Failed to lock %s: %m", np
);
4871 BLOCK_SIGNALS(SIGINT
);
4872 r
= btrfs_subvol_snapshot(arg_directory
, np
,
4873 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4874 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4875 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4876 BTRFS_SNAPSHOT_RECURSIVE
|
4877 BTRFS_SNAPSHOT_QUOTA
|
4878 BTRFS_SNAPSHOT_SIGINT
);
4881 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
4885 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
4889 free_and_replace(arg_directory
, np
);
4890 remove_directory
= true;
4892 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
4896 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4898 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
4902 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
4907 r
= chase_symlinks_and_update(&arg_template
, 0);
4912 BLOCK_SIGNALS(SIGINT
);
4913 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
4914 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4915 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4916 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4917 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
4918 BTRFS_SNAPSHOT_RECURSIVE
|
4919 BTRFS_SNAPSHOT_QUOTA
|
4920 BTRFS_SNAPSHOT_SIGINT
);
4923 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4924 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
4925 else if (r
== -EINTR
) {
4926 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
4929 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
4932 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4933 "Populated %s from template %s.", arg_directory
, arg_template
);
4937 if (arg_start_mode
== START_BOOT
) {
4940 if (arg_pivot_root_new
)
4941 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4945 if (path_is_os_tree(p
) <= 0) {
4946 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
4953 if (arg_pivot_root_new
)
4954 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
4958 q
= strjoina(p
, "/usr/");
4960 if (laccess(q
, F_OK
) < 0) {
4961 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
4969 assert(!arg_template
);
4971 r
= chase_symlinks_and_update(&arg_image
, 0);
4975 if (arg_ephemeral
) {
4976 _cleanup_free_
char *np
= NULL
;
4978 r
= tempfn_random(arg_image
, "machine.", &np
);
4980 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
4984 /* Always take an exclusive lock on our own ephemeral copy. */
4985 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4987 r
= log_error_errno(r
, "Failed to create image lock: %m");
4992 BLOCK_SIGNALS(SIGINT
);
4993 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
);
4996 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5000 r
= log_error_errno(r
, "Failed to copy image file: %m");
5004 free_and_replace(arg_image
, np
);
5005 remove_image
= true;
5007 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5009 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5013 r
= log_error_errno(r
, "Failed to create image lock: %m");
5017 if (!arg_root_hash
) {
5018 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
5020 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
5026 if (!mkdtemp(tmprootdir
)) {
5027 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5031 remove_tmprootdir
= true;
5033 arg_directory
= strdup(tmprootdir
);
5034 if (!arg_directory
) {
5039 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, LO_FLAGS_PARTSCAN
, &loop
);
5041 log_error_errno(r
, "Failed to set up loopback block device: %m");
5045 r
= dissect_image_and_warn(
5048 arg_root_hash
, arg_root_hash_size
,
5049 DISSECT_IMAGE_REQUIRE_ROOT
,
5052 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5053 log_notice("Note that the disk image needs to\n"
5054 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5055 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5056 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
5057 " d) or contain a file system without a partition table\n"
5058 "in order to be bootable with systemd-nspawn.");
5064 if (!arg_root_hash
&& dissected_image
->can_verity
)
5065 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5067 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
5071 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5072 if (remove_image
&& unlink(arg_image
) >= 0)
5073 remove_image
= false;
5076 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5080 if (arg_console_mode
< 0)
5082 isatty(STDIN_FILENO
) > 0 &&
5083 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5085 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5089 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5090 arg_machine
, arg_image
?: arg_directory
);
5092 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5094 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5095 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5100 r
= run_container(dissected_image
,
5103 veth_name
, &veth_created
,
5111 (void) sd_notify(false,
5112 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5113 "STOPPING=1\nSTATUS=Terminating...");
5116 (void) kill(pid
, SIGKILL
);
5118 /* Try to flush whatever is still queued in the pty */
5120 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5121 master
= safe_close(master
);
5125 (void) wait_for_terminate(pid
, NULL
);
5129 if (remove_directory
&& arg_directory
) {
5132 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5134 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5137 if (remove_image
&& arg_image
) {
5138 if (unlink(arg_image
) < 0)
5139 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5142 if (remove_tmprootdir
) {
5143 if (rmdir(tmprootdir
) < 0)
5144 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5150 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5151 (void) rm_rf(p
, REMOVE_ROOT
);
5154 expose_port_flush(arg_expose_ports
, &exposed
);
5157 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5158 (void) remove_bridge(arg_network_zone
);
5160 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5161 expose_port_free_all(arg_expose_ports
);
5162 rlimit_free_all(arg_rlimit
);
5163 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5171 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);