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 r
= test_network_interface_initialized(optarg
);
855 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
858 arg_private_network
= true;
859 arg_settings_mask
|= SETTING_NETWORK
;
862 case ARG_NETWORK_MACVLAN
:
864 if (!ifname_valid(optarg
))
865 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
866 "MACVLAN network interface name not valid: %s", optarg
);
868 r
= test_network_interface_initialized(optarg
);
872 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
875 arg_private_network
= true;
876 arg_settings_mask
|= SETTING_NETWORK
;
879 case ARG_NETWORK_IPVLAN
:
881 if (!ifname_valid(optarg
))
882 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
883 "IPVLAN network interface name not valid: %s", optarg
);
885 r
= test_network_interface_initialized(optarg
);
889 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
893 case ARG_PRIVATE_NETWORK
:
894 arg_private_network
= true;
895 arg_settings_mask
|= SETTING_NETWORK
;
898 case ARG_NETWORK_NAMESPACE_PATH
:
899 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
903 arg_settings_mask
|= SETTING_NETWORK
;
907 if (arg_start_mode
== START_PID2
)
908 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
909 "--boot and --as-pid2 may not be combined.");
911 arg_start_mode
= START_BOOT
;
912 arg_settings_mask
|= SETTING_START_MODE
;
916 if (arg_start_mode
== START_BOOT
)
917 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
918 "--boot and --as-pid2 may not be combined.");
920 arg_start_mode
= START_PID2
;
921 arg_settings_mask
|= SETTING_START_MODE
;
925 r
= sd_id128_from_string(optarg
, &arg_uuid
);
927 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
929 if (sd_id128_is_null(arg_uuid
))
930 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
931 "Machine UUID may not be all zeroes.");
933 arg_settings_mask
|= SETTING_MACHINE_ID
;
937 _cleanup_free_
char *mangled
= NULL
;
939 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
943 free_and_replace(arg_slice
, mangled
);
944 arg_settings_mask
|= SETTING_SLICE
;
950 arg_machine
= mfree(arg_machine
);
952 if (!machine_name_is_valid(optarg
))
953 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
954 "Invalid machine name: %s", optarg
);
956 r
= free_and_strdup(&arg_machine
, optarg
);
964 arg_hostname
= mfree(arg_hostname
);
966 if (!hostname_is_valid(optarg
, false))
967 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
968 "Invalid hostname: %s", optarg
);
970 r
= free_and_strdup(&arg_hostname
, optarg
);
975 arg_settings_mask
|= SETTING_HOSTNAME
;
979 arg_selinux_context
= optarg
;
983 arg_selinux_apifs_context
= optarg
;
987 arg_read_only
= true;
988 arg_settings_mask
|= SETTING_READ_ONLY
;
992 case ARG_DROP_CAPABILITY
: {
994 r
= parse_capability_spec(optarg
, &m
);
998 if (c
== ARG_CAPABILITY
)
1002 arg_settings_mask
|= SETTING_CAPABILITY
;
1005 case ARG_NO_NEW_PRIVILEGES
:
1006 r
= parse_boolean(optarg
);
1008 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1010 arg_no_new_privileges
= r
;
1011 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1015 arg_link_journal
= LINK_GUEST
;
1016 arg_link_journal_try
= true;
1017 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1020 case ARG_LINK_JOURNAL
:
1021 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1023 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1025 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1030 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1032 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1034 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1038 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1040 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1042 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1046 case ARG_OVERLAY_RO
:
1047 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1048 if (r
== -EADDRNOTAVAIL
)
1049 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1051 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1053 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1056 case ARG_INACCESSIBLE
:
1057 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1059 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1061 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1067 if (!env_assignment_is_valid(optarg
))
1068 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1069 "Environment variable assignment '%s' is not valid.", optarg
);
1071 n
= strv_env_set(arg_setenv
, optarg
);
1075 strv_free_and_replace(arg_setenv
, n
);
1076 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1084 case ARG_SHARE_SYSTEM
:
1085 /* We don't officially support this anymore, except for compat reasons. People should use the
1086 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1087 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1088 arg_clone_ns_flags
= 0;
1092 r
= parse_boolean(optarg
);
1094 log_error("Failed to parse --register= argument: %s", optarg
);
1102 arg_keep_unit
= true;
1105 case ARG_PERSONALITY
:
1107 arg_personality
= personality_from_string(optarg
);
1108 if (arg_personality
== PERSONALITY_INVALID
)
1109 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1110 "Unknown or unsupported personality '%s'.", optarg
);
1112 arg_settings_mask
|= SETTING_PERSONALITY
;
1118 arg_volatile_mode
= VOLATILE_YES
;
1119 else if (streq(optarg
, "help")) {
1120 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1125 m
= volatile_mode_from_string(optarg
);
1127 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1128 "Failed to parse --volatile= argument: %s", optarg
);
1130 arg_volatile_mode
= m
;
1133 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1137 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1139 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1141 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1143 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1147 if (strv_extend(&arg_property
, optarg
) < 0)
1152 case ARG_PRIVATE_USERS
: {
1157 else if (!in_charset(optarg
, DIGITS
))
1158 /* do *not* parse numbers as booleans */
1159 boolean
= parse_boolean(optarg
);
1161 if (boolean
== false) {
1162 /* no: User namespacing off */
1163 arg_userns_mode
= USER_NAMESPACE_NO
;
1164 arg_uid_shift
= UID_INVALID
;
1165 arg_uid_range
= UINT32_C(0x10000);
1166 } else if (boolean
== true) {
1167 /* yes: User namespacing on, UID range is read from root dir */
1168 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1169 arg_uid_shift
= UID_INVALID
;
1170 arg_uid_range
= UINT32_C(0x10000);
1171 } else if (streq(optarg
, "pick")) {
1172 /* pick: User namespacing on, UID range is picked randomly */
1173 arg_userns_mode
= USER_NAMESPACE_PICK
;
1174 arg_uid_shift
= UID_INVALID
;
1175 arg_uid_range
= UINT32_C(0x10000);
1177 _cleanup_free_
char *buffer
= NULL
;
1178 const char *range
, *shift
;
1180 /* anything else: User namespacing on, UID range is explicitly configured */
1182 range
= strchr(optarg
, ':');
1184 buffer
= strndup(optarg
, range
- optarg
);
1190 r
= safe_atou32(range
, &arg_uid_range
);
1192 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1196 r
= parse_uid(shift
, &arg_uid_shift
);
1198 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1200 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1203 if (arg_uid_range
<= 0)
1204 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1205 "UID range cannot be 0.");
1207 arg_settings_mask
|= SETTING_USERNS
;
1212 if (userns_supported()) {
1213 arg_userns_mode
= USER_NAMESPACE_PICK
;
1214 arg_uid_shift
= UID_INVALID
;
1215 arg_uid_range
= UINT32_C(0x10000);
1217 arg_settings_mask
|= SETTING_USERNS
;
1222 case ARG_PRIVATE_USERS_CHOWN
:
1223 arg_userns_chown
= true;
1225 arg_settings_mask
|= SETTING_USERNS
;
1228 case ARG_KILL_SIGNAL
:
1229 if (streq(optarg
, "help")) {
1230 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1234 arg_kill_signal
= signal_from_string(optarg
);
1235 if (arg_kill_signal
< 0)
1236 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1237 "Cannot parse signal: %s", optarg
);
1239 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1244 /* no → do not read files
1245 * yes → read files, do not override cmdline, trust only subset
1246 * override → read files, override cmdline, trust only subset
1247 * trusted → read files, do not override cmdline, trust all
1250 r
= parse_boolean(optarg
);
1252 if (streq(optarg
, "trusted")) {
1253 mask_all_settings
= false;
1254 mask_no_settings
= false;
1255 arg_settings_trusted
= true;
1257 } else if (streq(optarg
, "override")) {
1258 mask_all_settings
= false;
1259 mask_no_settings
= true;
1260 arg_settings_trusted
= -1;
1262 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1265 mask_all_settings
= false;
1266 mask_no_settings
= false;
1267 arg_settings_trusted
= -1;
1270 mask_all_settings
= true;
1271 mask_no_settings
= false;
1272 arg_settings_trusted
= false;
1278 if (!path_is_absolute(optarg
))
1279 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1280 "Working directory %s is not an absolute path.", optarg
);
1282 r
= free_and_strdup(&arg_chdir
, optarg
);
1286 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1289 case ARG_PIVOT_ROOT
:
1290 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1292 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1294 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1297 case ARG_NOTIFY_READY
:
1298 r
= parse_boolean(optarg
);
1300 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1301 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1302 arg_notify_ready
= r
;
1303 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1306 case ARG_ROOT_HASH
: {
1310 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1312 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1313 if (l
< sizeof(sd_id128_t
)) {
1315 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1318 free(arg_root_hash
);
1320 arg_root_hash_size
= l
;
1324 case ARG_SYSTEM_CALL_FILTER
: {
1328 negative
= optarg
[0] == '~';
1329 items
= negative
? optarg
+ 1 : optarg
;
1332 _cleanup_free_
char *word
= NULL
;
1334 r
= extract_first_word(&items
, &word
, NULL
, 0);
1340 return log_error_errno(r
, "Failed to parse system call filter: %m");
1343 r
= strv_extend(&arg_syscall_blacklist
, word
);
1345 r
= strv_extend(&arg_syscall_whitelist
, word
);
1350 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1356 _cleanup_free_
char *name
= NULL
;
1359 if (streq(optarg
, "help")) {
1360 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1364 eq
= strchr(optarg
, '=');
1366 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1367 "--rlimit= expects an '=' assignment.");
1369 name
= strndup(optarg
, eq
- optarg
);
1373 rl
= rlimit_from_string_harder(name
);
1375 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1376 "Unknown resource limit: %s", name
);
1378 if (!arg_rlimit
[rl
]) {
1379 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1380 if (!arg_rlimit
[rl
])
1384 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1386 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1388 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1392 case ARG_OOM_SCORE_ADJUST
:
1393 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1395 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1397 arg_oom_score_adjust_set
= true;
1398 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1401 case ARG_CPU_AFFINITY
: {
1404 r
= parse_cpu_set(optarg
, &cpuset
);
1406 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1408 cpu_set_reset(&arg_cpu_set
);
1409 arg_cpu_set
= cpuset
;
1410 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1414 case ARG_RESOLV_CONF
:
1415 if (streq(optarg
, "help")) {
1416 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1420 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1421 if (arg_resolv_conf
< 0)
1422 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1423 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1425 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1429 if (streq(optarg
, "help")) {
1430 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1434 arg_timezone
= timezone_mode_from_string(optarg
);
1435 if (arg_timezone
< 0)
1436 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1437 "Failed to parse /etc/localtime mode: %s", optarg
);
1439 arg_settings_mask
|= SETTING_TIMEZONE
;
1443 r
= handle_arg_console(optarg
);
1450 r
= handle_arg_console("pipe");
1456 arg_pager_flags
|= PAGER_DISABLE
;
1463 assert_not_reached("Unhandled option");
1466 if (argc
> optind
) {
1467 strv_free(arg_parameters
);
1468 arg_parameters
= strv_copy(argv
+ optind
);
1469 if (!arg_parameters
)
1472 arg_settings_mask
|= SETTING_START_MODE
;
1475 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1476 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1477 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1478 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1480 arg_directory
= TAKE_PTR(arg_template
);
1482 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1484 /* Make sure to parse environment before we reset the settings mask below */
1485 r
= parse_environment();
1489 /* Load all settings from .nspawn files */
1490 if (mask_no_settings
)
1491 arg_settings_mask
= 0;
1493 /* Don't load any settings from .nspawn files */
1494 if (mask_all_settings
)
1495 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1500 static int verify_arguments(void) {
1503 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1504 /* If we are running the stub init in the container, we don't need to look at what the init
1505 * in the container supports, because we are not using it. Let's immediately pick the right
1506 * setting based on the host system configuration.
1508 * We only do this, if the user didn't use an environment variable to override the detection.
1511 r
= cg_all_unified();
1513 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1515 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1516 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1517 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1519 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1522 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1523 arg_mount_settings
|= MOUNT_USE_USERNS
;
1525 if (arg_private_network
)
1526 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1528 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1529 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1530 arg_register
= false;
1531 if (arg_start_mode
!= START_PID1
)
1532 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1535 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1536 arg_userns_chown
= true;
1538 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1539 arg_kill_signal
= SIGRTMIN
+3;
1541 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 */
1542 arg_read_only
= true;
1544 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1545 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1546 * The latter is not technically a user session, but we don't need to labour the point. */
1547 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1549 if (arg_directory
&& arg_image
)
1550 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1552 if (arg_template
&& arg_image
)
1553 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1555 if (arg_template
&& !(arg_directory
|| arg_machine
))
1556 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1558 if (arg_ephemeral
&& arg_template
)
1559 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1561 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1562 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1564 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1565 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1567 if (arg_userns_chown
&& arg_read_only
)
1568 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1569 "--read-only and --private-users-chown may not be combined.");
1571 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1572 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1573 * copy-up (in case of overlay) making the entire exercise pointless. */
1574 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1575 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1577 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1578 * we need to error out, to avoid conflicts between different network options. */
1579 if (arg_network_namespace_path
&&
1580 (arg_network_interfaces
|| arg_network_macvlan
||
1581 arg_network_ipvlan
|| arg_network_veth_extra
||
1582 arg_network_bridge
|| arg_network_zone
||
1584 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1586 if (arg_network_bridge
&& arg_network_zone
)
1587 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1588 "--network-bridge= and --network-zone= may not be combined.");
1590 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1591 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1593 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1594 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1596 if (arg_expose_ports
&& !arg_private_network
)
1597 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1600 if (arg_expose_ports
)
1601 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1604 r
= custom_mount_check_all();
1611 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1614 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1617 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1620 if (uid
!= UID_INVALID
) {
1621 uid
+= arg_uid_shift
;
1623 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1627 if (gid
!= GID_INVALID
) {
1628 gid
+= (gid_t
) arg_uid_shift
;
1630 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1634 if (lchown(p
, uid
, gid
) < 0)
1640 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1644 q
= prefix_roota(root
, path
);
1645 r
= mkdir_errno_wrapper(q
, mode
);
1651 return userns_lchown(q
, uid
, gid
);
1654 static const char *timezone_from_path(const char *path
) {
1655 return PATH_STARTSWITH_SET(
1657 "../usr/share/zoneinfo/",
1658 "/usr/share/zoneinfo/");
1661 static bool etc_writable(void) {
1662 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1665 static int setup_timezone(const char *dest
) {
1666 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1667 const char *where
, *check
;
1673 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1674 r
= readlink_malloc("/etc/localtime", &p
);
1675 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1676 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1677 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1678 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1680 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1681 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1685 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1688 } else if (arg_timezone
== TIMEZONE_AUTO
)
1689 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1695 if (m
== TIMEZONE_OFF
)
1698 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1700 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1704 where
= strjoina(etc
, "/localtime");
1708 case TIMEZONE_DELETE
:
1709 if (unlink(where
) < 0)
1710 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1714 case TIMEZONE_SYMLINK
: {
1715 _cleanup_free_
char *q
= NULL
;
1716 const char *z
, *what
;
1718 z
= timezone_from_path(p
);
1720 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1724 r
= readlink_malloc(where
, &q
);
1725 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1726 return 0; /* Already pointing to the right place? Then do nothing .. */
1728 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1729 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1731 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1733 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1734 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1735 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1739 what
= strjoina("../usr/share/zoneinfo/", z
);
1740 if (symlink(what
, where
) < 0) {
1741 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1742 errno
, "Failed to correct timezone of container, ignoring: %m");
1752 case TIMEZONE_BIND
: {
1753 _cleanup_free_
char *resolved
= NULL
;
1756 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1758 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1762 if (found
== 0) /* missing? */
1763 (void) touch(resolved
);
1765 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1767 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1773 /* If mounting failed, try to copy */
1774 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1776 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1777 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1784 assert_not_reached("unexpected mode");
1787 /* Fix permissions of the symlink or file copy we just created */
1788 r
= userns_lchown(where
, 0, 0);
1790 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1795 static int have_resolv_conf(const char *path
) {
1798 if (access(path
, F_OK
) < 0) {
1799 if (errno
== ENOENT
)
1802 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1808 static int resolved_listening(void) {
1809 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1810 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1811 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1814 /* Check if resolved is listening */
1816 r
= sd_bus_open_system(&bus
);
1818 return log_debug_errno(r
, "Failed to open system bus: %m");
1820 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1822 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1826 r
= sd_bus_get_property_string(bus
,
1827 "org.freedesktop.resolve1",
1828 "/org/freedesktop/resolve1",
1829 "org.freedesktop.resolve1.Manager",
1832 &dns_stub_listener_mode
);
1834 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1836 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1839 static int setup_resolv_conf(const char *dest
) {
1840 _cleanup_free_
char *etc
= NULL
;
1841 const char *where
, *what
;
1847 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1848 if (arg_private_network
)
1849 m
= RESOLV_CONF_OFF
;
1850 else if (have_resolv_conf(STATIC_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1851 m
= etc_writable() ? RESOLV_CONF_COPY_STATIC
: RESOLV_CONF_BIND_STATIC
;
1852 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1853 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1855 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1857 m
= arg_resolv_conf
;
1859 if (m
== RESOLV_CONF_OFF
)
1862 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1864 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1868 where
= strjoina(etc
, "/resolv.conf");
1870 if (m
== RESOLV_CONF_DELETE
) {
1871 if (unlink(where
) < 0)
1872 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1877 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_COPY_STATIC
))
1878 what
= STATIC_RESOLV_CONF
;
1880 what
= "/etc/resolv.conf";
1882 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
)) {
1883 _cleanup_free_
char *resolved
= NULL
;
1886 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1888 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1892 if (found
== 0) /* missing? */
1893 (void) touch(resolved
);
1895 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1897 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1900 /* If that didn't work, let's copy the file */
1901 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
1903 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1904 * resolved or something similar runs inside and the symlink points there.
1906 * If the disk image is read-only, there's also no point in complaining.
1908 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
,
1909 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1913 r
= userns_lchown(where
, 0, 0);
1915 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1920 static int setup_boot_id(void) {
1921 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1922 _cleanup_free_
char *path
= NULL
;
1923 sd_id128_t rnd
= SD_ID128_NULL
;
1927 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1929 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
1931 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1933 r
= sd_id128_randomize(&rnd
);
1935 return log_error_errno(r
, "Failed to generate random boot id: %m");
1937 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1939 return log_error_errno(r
, "Failed to write boot id: %m");
1941 from
= TAKE_PTR(path
);
1942 to
= "/proc/sys/kernel/random/boot_id";
1944 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1948 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1951 static int copy_devnodes(const char *dest
) {
1952 static const char devnodes
[] =
1961 _cleanup_umask_ mode_t u
;
1969 /* Create /dev/net, so that we can create /dev/net/tun in it */
1970 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1971 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1973 NULSTR_FOREACH(d
, devnodes
) {
1974 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1977 from
= path_join("/dev/", d
);
1981 to
= path_join(dest
, from
);
1985 if (stat(from
, &st
) < 0) {
1987 if (errno
!= ENOENT
)
1988 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1990 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1991 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
1992 "%s is not a char or block device, cannot copy.", from
);
1994 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
1996 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1997 /* Explicitly warn the user when /dev is already populated. */
1998 if (errno
== EEXIST
)
1999 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
);
2001 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2003 /* Some systems abusively restrict mknod but allow bind mounts. */
2006 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2007 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2009 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2012 r
= userns_lchown(to
, 0, 0);
2014 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2016 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2020 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2022 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2024 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2027 prefixed
= path_join(dest
, sl
);
2031 t
= path_join("..", d
);
2035 if (symlink(t
, prefixed
) < 0)
2036 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2043 static int make_extra_nodes(const char *dest
) {
2044 _cleanup_umask_ mode_t u
;
2050 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2051 _cleanup_free_
char *path
= NULL
;
2052 DeviceNode
*n
= arg_extra_nodes
+ i
;
2054 path
= path_join(dest
, n
->path
);
2058 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2059 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2061 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2063 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2069 static int setup_pts(const char *dest
) {
2070 _cleanup_free_
char *options
= NULL
;
2075 if (arg_selinux_apifs_context
)
2076 (void) asprintf(&options
,
2077 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2078 arg_uid_shift
+ TTY_GID
,
2079 arg_selinux_apifs_context
);
2082 (void) asprintf(&options
,
2083 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2084 arg_uid_shift
+ TTY_GID
);
2089 /* Mount /dev/pts itself */
2090 p
= prefix_roota(dest
, "/dev/pts");
2091 r
= mkdir_errno_wrapper(p
, 0755);
2093 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2095 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2098 r
= userns_lchown(p
, 0, 0);
2100 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2102 /* Create /dev/ptmx symlink */
2103 p
= prefix_roota(dest
, "/dev/ptmx");
2104 if (symlink("pts/ptmx", p
) < 0)
2105 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2106 r
= userns_lchown(p
, 0, 0);
2108 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2110 /* And fix /dev/pts/ptmx ownership */
2111 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2112 r
= userns_lchown(p
, 0, 0);
2114 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2119 static int setup_stdio_as_dev_console(void) {
2123 terminal
= open_terminal("/dev/console", O_RDWR
);
2125 return log_error_errno(terminal
, "Failed to open console: %m");
2127 /* Make sure we can continue logging to the original stderr, even if
2128 * stderr points elsewhere now */
2129 r
= log_dup_console();
2131 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2133 /* invalidates 'terminal' on success and failure */
2134 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2136 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2141 static int setup_dev_console(const char *console
) {
2142 _cleanup_free_
char *p
= NULL
;
2145 /* Create /dev/console symlink */
2146 r
= path_make_relative("/dev", console
, &p
);
2148 return log_error_errno(r
, "Failed to create relative path: %m");
2150 if (symlink(p
, "/dev/console") < 0)
2151 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2156 static int setup_keyring(void) {
2157 key_serial_t keyring
;
2159 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2160 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2161 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2162 * these system calls let's make sure we don't leak anything into the container. */
2164 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2165 if (keyring
== -1) {
2166 if (errno
== ENOSYS
)
2167 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2168 else if (IN_SET(errno
, EACCES
, EPERM
))
2169 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2171 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2177 static int setup_kmsg(int kmsg_socket
) {
2178 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2179 _cleanup_free_
char *fifo
= NULL
;
2180 _cleanup_close_
int fd
= -1;
2181 _cleanup_umask_ mode_t u
;
2184 assert(kmsg_socket
>= 0);
2188 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2189 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2190 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2191 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2193 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2195 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2197 if (mkfifo(fifo
, 0600) < 0)
2198 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2200 from
= TAKE_PTR(fifo
);
2202 r
= mount_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2206 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2208 return log_error_errno(errno
, "Failed to open fifo: %m");
2210 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2211 r
= send_one_fd(kmsg_socket
, fd
, 0);
2213 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2218 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2219 union in_addr_union
*exposed
= userdata
;
2225 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2229 static int setup_hostname(void) {
2232 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2235 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2237 return log_error_errno(r
, "Failed to set hostname: %m");
2242 static int setup_journal(const char *directory
) {
2243 _cleanup_free_
char *d
= NULL
;
2244 char id
[SD_ID128_STRING_MAX
];
2245 const char *dirname
, *p
, *q
;
2250 /* Don't link journals in ephemeral mode */
2254 if (arg_link_journal
== LINK_NO
)
2257 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2259 r
= sd_id128_get_machine(&this_id
);
2261 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2263 if (sd_id128_equal(arg_uuid
, this_id
)) {
2264 log_full(try ? LOG_WARNING
: LOG_ERR
,
2265 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2271 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2272 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2274 bool ignore
= r
== -EROFS
&& try;
2275 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2276 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2277 return ignore
? 0 : r
;
2281 (void) sd_id128_to_string(arg_uuid
, id
);
2283 p
= strjoina("/var/log/journal/", id
);
2284 q
= prefix_roota(directory
, p
);
2286 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2290 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2291 "%s: already a mount point, refusing to use for journal", p
);
2294 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2298 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2299 "%s: already a mount point, refusing to use for journal", q
);
2302 r
= readlink_and_make_absolute(p
, &d
);
2304 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2307 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2309 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2314 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2315 } else if (r
== -EINVAL
) {
2317 if (arg_link_journal
== LINK_GUEST
&&
2320 if (errno
== ENOTDIR
) {
2321 log_error("%s already exists and is neither a symlink nor a directory", p
);
2324 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2326 } else if (r
!= -ENOENT
)
2327 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2329 if (arg_link_journal
== LINK_GUEST
) {
2331 if (symlink(q
, p
) < 0) {
2333 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2336 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2339 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2341 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2345 if (arg_link_journal
== LINK_HOST
) {
2346 /* don't create parents here — if the host doesn't have
2347 * permanent journal set up, don't force it here */
2349 r
= mkdir_errno_wrapper(p
, 0755);
2350 if (r
< 0 && r
!= -EEXIST
) {
2352 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2355 return log_error_errno(r
, "Failed to create %s: %m", p
);
2358 } else if (access(p
, F_OK
) < 0)
2361 if (dir_is_empty(q
) == 0)
2362 log_warning("%s is not empty, proceeding anyway.", q
);
2364 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2366 return log_error_errno(r
, "Failed to create %s: %m", q
);
2368 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2370 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2375 static int drop_capabilities(uid_t uid
) {
2376 CapabilityQuintet q
;
2378 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2379 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2380 * arg_caps_retain. */
2382 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2383 q
= arg_full_capabilities
;
2385 if (q
.bounding
== (uint64_t) -1)
2386 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2388 if (q
.effective
== (uint64_t) -1)
2389 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2391 if (q
.inheritable
== (uint64_t) -1)
2392 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2394 if (q
.permitted
== (uint64_t) -1)
2395 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2397 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2400 if (capability_quintet_mangle(&q
))
2401 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2404 q
= (CapabilityQuintet
) {
2405 .bounding
= arg_caps_retain
,
2406 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2407 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2408 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2409 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2412 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2413 * in order to maintain the same behavior as systemd < 242. */
2414 if (capability_quintet_mangle(&q
))
2415 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2416 "Some capabilities will not be set because they are not in the current bounding set.");
2420 return capability_quintet_enforce(&q
);
2423 static int reset_audit_loginuid(void) {
2424 _cleanup_free_
char *p
= NULL
;
2427 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2430 r
= read_one_line_file("/proc/self/loginuid", &p
);
2434 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2436 /* Already reset? */
2437 if (streq(p
, "4294967295"))
2440 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2443 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2444 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2445 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2446 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2447 "using systemd-nspawn. Sleeping for 5s... (%m)");
2455 static int setup_propagate(const char *root
) {
2459 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2460 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2461 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2462 (void) mkdir_p(p
, 0600);
2464 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2466 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2468 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2470 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2472 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2474 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2476 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2477 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2481 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2485 /* machined will MS_MOVE into that directory, and that's only
2486 * supported for non-shared mounts. */
2487 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2490 static int setup_machine_id(const char *directory
) {
2491 const char *etc_machine_id
;
2495 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2496 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2497 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2498 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2499 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2500 * container behaves nicely). */
2502 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2504 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2506 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2507 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2509 if (sd_id128_is_null(arg_uuid
)) {
2510 r
= sd_id128_randomize(&arg_uuid
);
2512 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2515 if (sd_id128_is_null(id
))
2516 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2517 "Machine ID in container image is zero, refusing.");
2525 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2530 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2533 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2534 if (r
== -EOPNOTSUPP
)
2535 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.");
2537 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2539 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2541 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2543 log_debug("Patched directory tree to match UID/GID range.");
2550 * < 0 : wait_for_terminate() failed to get the state of the
2551 * container, the container was terminated by a signal, or
2552 * failed for an unknown reason. No change is made to the
2553 * container argument.
2554 * > 0 : The program executed in the container terminated with an
2555 * error. The exit code of the program executed in the
2556 * container is returned. The container argument has been set
2557 * to CONTAINER_TERMINATED.
2558 * 0 : The container is being rebooted, has been shut down or exited
2559 * successfully. The container argument has been set to either
2560 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2562 * That is, success is indicated by a return value of zero, and an
2563 * error is indicated by a non-zero value.
2565 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2569 r
= wait_for_terminate(pid
, &status
);
2571 return log_warning_errno(r
, "Failed to wait for container: %m");
2573 switch (status
.si_code
) {
2576 if (status
.si_status
== 0)
2577 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2579 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2581 *container
= CONTAINER_TERMINATED
;
2582 return status
.si_status
;
2585 if (status
.si_status
== SIGINT
) {
2586 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2587 *container
= CONTAINER_TERMINATED
;
2590 } else if (status
.si_status
== SIGHUP
) {
2591 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2592 *container
= CONTAINER_REBOOTED
;
2598 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2599 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2602 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2603 "Container %s failed due to unknown reason.", arg_machine
);
2607 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2610 pid
= PTR_TO_PID(userdata
);
2612 if (kill(pid
, arg_kill_signal
) >= 0) {
2613 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2614 sd_event_source_set_userdata(s
, NULL
);
2619 sd_event_exit(sd_event_source_get_event(s
), 0);
2623 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2629 pid
= PTR_TO_PID(userdata
);
2634 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2635 return log_error_errno(errno
, "Failed to waitid(): %m");
2636 if (si
.si_pid
== 0) /* No pending children. */
2638 if (si
.si_pid
== pid
) {
2639 /* The main process we care for has exited. Return from
2640 * signal handler but leave the zombie. */
2641 sd_event_exit(sd_event_source_get_event(s
), 0);
2645 /* Reap all other children. */
2646 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2652 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2657 pid
= PTR_TO_PID(userdata
);
2659 if (arg_kill_signal
> 0) {
2660 log_info("Container termination requested. Attempting to halt container.");
2661 (void) kill(pid
, arg_kill_signal
);
2663 log_info("Container termination requested. Exiting.");
2664 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2670 static int determine_names(void) {
2673 if (arg_template
&& !arg_directory
&& arg_machine
) {
2675 /* If --template= was specified then we should not
2676 * search for a machine, but instead create a new one
2677 * in /var/lib/machine. */
2679 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2684 if (!arg_image
&& !arg_directory
) {
2686 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2688 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2690 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2692 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2694 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2695 r
= free_and_strdup(&arg_image
, i
->path
);
2697 r
= free_and_strdup(&arg_directory
, i
->path
);
2702 arg_read_only
= arg_read_only
|| i
->read_only
;
2704 r
= safe_getcwd(&arg_directory
);
2706 return log_error_errno(r
, "Failed to determine current directory: %m");
2709 if (!arg_directory
&& !arg_image
)
2710 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2714 if (arg_directory
&& path_equal(arg_directory
, "/"))
2715 arg_machine
= gethostname_malloc();
2720 arg_machine
= strdup(basename(arg_image
));
2722 /* Truncate suffix if there is one */
2723 e
= endswith(arg_machine
, ".raw");
2727 arg_machine
= strdup(basename(arg_directory
));
2732 hostname_cleanup(arg_machine
);
2733 if (!machine_name_is_valid(arg_machine
))
2734 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2736 if (arg_ephemeral
) {
2739 /* Add a random suffix when this is an
2740 * ephemeral machine, so that we can run many
2741 * instances at once without manually having
2742 * to specify -M each time. */
2744 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2755 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2764 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
2766 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2768 return free_and_replace(*p
, chased
);
2771 static int determine_uid_shift(const char *directory
) {
2774 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2779 if (arg_uid_shift
== UID_INVALID
) {
2782 r
= stat(directory
, &st
);
2784 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2786 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2788 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2789 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2790 "UID and GID base of %s don't match.", directory
);
2792 arg_uid_range
= UINT32_C(0x10000);
2795 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2796 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2797 "UID base too high for UID range.");
2802 static unsigned long effective_clone_ns_flags(void) {
2803 unsigned long flags
= arg_clone_ns_flags
;
2805 if (arg_private_network
)
2806 flags
|= CLONE_NEWNET
;
2808 flags
|= CLONE_NEWCGROUP
;
2809 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2810 flags
|= CLONE_NEWUSER
;
2815 static int patch_sysctl(void) {
2817 /* This table is inspired by runc's sysctl() function */
2818 static const struct {
2821 unsigned long clone_flags
;
2823 { "kernel.hostname", false, CLONE_NEWUTS
},
2824 { "kernel.domainname", false, CLONE_NEWUTS
},
2825 { "kernel.msgmax", false, CLONE_NEWIPC
},
2826 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2827 { "kernel.msgmni", false, CLONE_NEWIPC
},
2828 { "kernel.sem", false, CLONE_NEWIPC
},
2829 { "kernel.shmall", false, CLONE_NEWIPC
},
2830 { "kernel.shmmax", false, CLONE_NEWIPC
},
2831 { "kernel.shmmni", false, CLONE_NEWIPC
},
2832 { "fs.mqueue.", true, CLONE_NEWIPC
},
2833 { "net.", true, CLONE_NEWNET
},
2836 unsigned long flags
;
2840 flags
= effective_clone_ns_flags();
2842 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2846 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2848 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2851 if (safe_sysctl
[i
].prefix
)
2852 good
= startswith(*k
, safe_sysctl
[i
].key
);
2854 good
= streq(*k
, safe_sysctl
[i
].key
);
2861 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2863 r
= sysctl_write(*k
, *v
);
2865 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2871 static int inner_child(
2873 const char *directory
,
2877 int master_pty_socket
,
2880 _cleanup_free_
char *home
= NULL
;
2881 char as_uuid
[ID128_UUID_STRING_MAX
];
2883 const char *envp
[] = {
2884 "PATH=" DEFAULT_PATH_COMPAT
,
2885 NULL
, /* container */
2890 NULL
, /* container_uuid */
2891 NULL
, /* LISTEN_FDS */
2892 NULL
, /* LISTEN_PID */
2893 NULL
, /* NOTIFY_SOCKET */
2896 const char *exec_target
;
2897 _cleanup_strv_free_
char **env_use
= NULL
;
2898 int r
, which_failed
;
2900 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2901 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2902 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2903 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2904 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2905 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2908 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2909 * unshare(). See below. */
2913 assert(kmsg_socket
>= 0);
2915 log_debug("Inner child is initializing.");
2917 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2918 /* Tell the parent, that it now can write the UID map. */
2919 (void) barrier_place(barrier
); /* #1 */
2921 /* Wait until the parent wrote the UID map */
2922 if (!barrier_place_and_sync(barrier
)) /* #2 */
2923 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2924 "Parent died too early");
2927 r
= reset_uid_gid();
2929 return log_error_errno(r
, "Couldn't become new root: %m");
2932 arg_mount_settings
| MOUNT_IN_USERNS
,
2934 arg_selinux_apifs_context
);
2938 if (!arg_network_namespace_path
&& arg_private_network
) {
2939 r
= unshare(CLONE_NEWNET
);
2941 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2943 /* Tell the parent that it can setup network interfaces. */
2944 (void) barrier_place(barrier
); /* #3 */
2947 r
= mount_sysfs(NULL
, arg_mount_settings
);
2951 /* Wait until we are cgroup-ified, so that we
2952 * can mount the right cgroup path writable */
2953 if (!barrier_place_and_sync(barrier
)) /* #4 */
2954 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2955 "Parent died too early");
2958 r
= unshare(CLONE_NEWCGROUP
);
2960 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2963 arg_unified_cgroup_hierarchy
,
2964 arg_userns_mode
!= USER_NAMESPACE_NO
,
2967 arg_selinux_apifs_context
,
2972 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2977 r
= setup_boot_id();
2981 r
= setup_kmsg(kmsg_socket
);
2984 kmsg_socket
= safe_close(kmsg_socket
);
2989 arg_n_custom_mounts
,
2991 arg_selinux_apifs_context
,
2992 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
2997 return log_error_errno(errno
, "setsid() failed: %m");
2999 if (arg_private_network
)
3002 if (arg_expose_ports
) {
3003 r
= expose_port_send_rtnl(rtnl_socket
);
3006 rtnl_socket
= safe_close(rtnl_socket
);
3009 if (arg_console_mode
!= CONSOLE_PIPE
) {
3010 _cleanup_close_
int master
= -1;
3011 _cleanup_free_
char *console
= NULL
;
3013 /* Allocate a pty and make it available as /dev/console. */
3014 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3016 return log_error_errno(master
, "Failed to allocate a pty: %m");
3018 r
= setup_dev_console(console
);
3020 return log_error_errno(r
, "Failed to setup /dev/console: %m");
3022 r
= send_one_fd(master_pty_socket
, master
, 0);
3024 return log_error_errno(r
, "Failed to send master fd: %m");
3025 master_pty_socket
= safe_close(master_pty_socket
);
3027 r
= setup_stdio_as_dev_console();
3036 if (arg_oom_score_adjust_set
) {
3037 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3039 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3042 if (arg_cpu_set
.set
)
3043 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3044 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3046 (void) setup_hostname();
3048 if (arg_personality
!= PERSONALITY_INVALID
) {
3049 r
= safe_personality(arg_personality
);
3051 return log_error_errno(r
, "personality() failed: %m");
3052 } else if (secondary
) {
3053 r
= safe_personality(PER_LINUX32
);
3055 return log_error_errno(r
, "personality() failed: %m");
3058 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3060 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3065 if (is_seccomp_available()) {
3067 r
= seccomp_load(arg_seccomp
);
3068 if (ERRNO_IS_SECCOMP_FATAL(r
))
3069 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3071 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3076 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
3082 if (arg_selinux_context
)
3083 if (setexeccon(arg_selinux_context
) < 0)
3084 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3087 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3088 * if we need to later on. */
3089 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3090 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3092 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3093 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
3095 r
= change_uid_gid(arg_user
, &home
);
3099 r
= drop_capabilities(getuid());
3101 return log_error_errno(r
, "Dropping capabilities failed: %m");
3103 if (arg_no_new_privileges
)
3104 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3105 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3107 /* LXC sets container=lxc, so follow the scheme here */
3108 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3110 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3114 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3115 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3118 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3119 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3120 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3123 assert(!sd_id128_is_null(arg_uuid
));
3125 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3128 if (fdset_size(fds
) > 0) {
3129 r
= fdset_cloexec(fds
, false);
3131 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3133 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3134 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3137 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3140 env_use
= strv_env_merge(2, envp
, arg_setenv
);
3144 /* Let the parent know that we are ready and
3145 * wait until the parent is ready with the
3147 if (!barrier_place_and_sync(barrier
)) /* #5 */
3148 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3149 "Parent died too early");
3152 if (chdir(arg_chdir
) < 0)
3153 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3155 if (arg_start_mode
== START_PID2
) {
3156 r
= stub_pid1(arg_uuid
);
3161 log_debug("Inner child completed, invoking payload.");
3163 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3164 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3165 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3167 log_set_open_when_needed(true);
3169 (void) fdset_close_others(fds
);
3171 if (arg_start_mode
== START_BOOT
) {
3175 /* Automatically search for the init system */
3177 m
= strv_length(arg_parameters
);
3178 a
= newa(char*, m
+ 2);
3179 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3182 a
[0] = (char*) "/usr/lib/systemd/systemd";
3183 execve(a
[0], a
, env_use
);
3185 a
[0] = (char*) "/lib/systemd/systemd";
3186 execve(a
[0], a
, env_use
);
3188 a
[0] = (char*) "/sbin/init";
3189 execve(a
[0], a
, env_use
);
3191 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3192 } else if (!strv_isempty(arg_parameters
)) {
3193 const char *dollar_path
;
3195 exec_target
= arg_parameters
[0];
3197 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3199 dollar_path
= strv_env_get(env_use
, "PATH");
3201 if (putenv((char*) dollar_path
) != 0)
3202 return log_error_errno(errno
, "Failed to update $PATH: %m");
3205 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3208 /* If we cannot change the directory, we'll end up in /, that is expected. */
3209 (void) chdir(home
?: "/root");
3211 execle("/bin/bash", "-bash", NULL
, env_use
);
3212 execle("/bin/sh", "-sh", NULL
, env_use
);
3214 exec_target
= "/bin/bash, /bin/sh";
3217 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3220 static int setup_sd_notify_child(void) {
3221 _cleanup_close_
int fd
= -1;
3222 union sockaddr_union sa
= {
3223 .un
.sun_family
= AF_UNIX
,
3224 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3228 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3230 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3232 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3233 (void) sockaddr_un_unlink(&sa
.un
);
3235 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3237 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3239 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3241 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3243 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3245 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3250 static int outer_child(
3252 const char *directory
,
3253 DissectedImage
*dissected_image
,
3260 int uid_shift_socket
,
3261 int master_pty_socket
,
3262 int unified_cgroup_hierarchy_socket
,
3266 _cleanup_close_
int fd
= -1;
3272 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3273 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3274 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3275 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3279 assert(pid_socket
>= 0);
3280 assert(uuid_socket
>= 0);
3281 assert(notify_socket
>= 0);
3282 assert(master_pty_socket
>= 0);
3283 assert(kmsg_socket
>= 0);
3285 log_debug("Outer child is initializing.");
3287 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3288 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3290 r
= reset_audit_loginuid();
3294 /* Mark everything as slave, so that we still
3295 * receive mounts from the real root, but don't
3296 * propagate mounts to the real root. */
3297 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3301 if (dissected_image
) {
3302 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3303 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3304 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3305 * makes sure ESP partitions and userns are compatible. */
3307 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3308 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3309 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0)|
3310 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3315 r
= determine_uid_shift(directory
);
3319 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3320 /* Let the parent know which UID shift we read from the image */
3321 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3323 return log_error_errno(errno
, "Failed to send UID shift: %m");
3324 if (l
!= sizeof(arg_uid_shift
))
3325 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3326 "Short write while sending UID shift.");
3328 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3329 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3330 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3331 * not it will pick a different one, and send it back to us. */
3333 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3335 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3336 if (l
!= sizeof(arg_uid_shift
))
3337 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3338 "Short read while receiving UID shift.");
3341 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3342 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3345 if (path_equal(directory
, "/")) {
3346 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3347 * place, so that we can make changes to its mount structure (for example, to implement
3348 * --volatile=) without this interfering with our ability to access files such as
3349 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3350 * (instead of a temporary directory, since we are living in our own mount namspace here
3351 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3352 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3354 r
= mount_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3358 directory
= "/run/systemd/nspawn-root";
3361 r
= setup_pivot_root(
3364 arg_pivot_root_old
);
3368 r
= setup_volatile_mode(
3372 arg_selinux_apifs_context
);
3379 arg_n_custom_mounts
,
3381 arg_selinux_apifs_context
,
3386 /* Make sure we always have a mount that we can move to root later on. */
3387 if (!path_is_mount_point(directory
, NULL
, 0)) {
3388 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3393 if (dissected_image
) {
3394 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3395 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3396 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
3401 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3402 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3404 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3408 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3410 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3411 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3412 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3413 "Short write while sending cgroup mode.");
3415 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3418 /* Mark everything as shared so our mounts get propagated down. This is
3419 * required to make new bind mounts available in systemd services
3420 * inside the container that create a new mount namespace.
3421 * See https://github.com/systemd/systemd/issues/3860
3422 * Further submounts (such as /dev) done after this will inherit the
3423 * shared propagation mode.
3425 * IMPORTANT: Do not overmount the root directory anymore from now on to
3426 * enable moving the root directory mount to root later on.
3427 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3429 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3433 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3437 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3441 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
) {
3442 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3444 return log_error_errno(r
, "Failed to make tree read-only: %m");
3447 r
= mount_all(directory
,
3450 arg_selinux_apifs_context
);
3454 r
= copy_devnodes(directory
);
3458 r
= make_extra_nodes(directory
);
3462 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3464 p
= prefix_roota(directory
, "/run/systemd");
3465 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3467 r
= setup_pts(directory
);
3471 r
= setup_propagate(directory
);
3475 r
= setup_keyring();
3479 r
= setup_timezone(directory
);
3483 r
= setup_resolv_conf(directory
);
3487 r
= setup_machine_id(directory
);
3491 r
= setup_journal(directory
);
3498 arg_n_custom_mounts
,
3500 arg_selinux_apifs_context
,
3501 MOUNT_NON_ROOT_ONLY
);
3505 if (!arg_use_cgns
) {
3508 arg_unified_cgroup_hierarchy
,
3509 arg_userns_mode
!= USER_NAMESPACE_NO
,
3512 arg_selinux_apifs_context
,
3518 r
= mount_move_root(directory
);
3520 return log_error_errno(r
, "Failed to move root directory: %m");
3522 fd
= setup_sd_notify_child();
3526 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3527 arg_clone_ns_flags
|
3528 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3530 return log_error_errno(errno
, "Failed to fork inner child: %m");
3532 pid_socket
= safe_close(pid_socket
);
3533 uuid_socket
= safe_close(uuid_socket
);
3534 notify_socket
= safe_close(notify_socket
);
3535 uid_shift_socket
= safe_close(uid_shift_socket
);
3537 /* The inner child has all namespaces that are
3538 * requested, so that we all are owned by the user if
3539 * user namespaces are turned on. */
3541 if (arg_network_namespace_path
) {
3542 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3544 return log_error_errno(r
, "Failed to join network namespace: %m");
3547 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
);
3549 _exit(EXIT_FAILURE
);
3551 _exit(EXIT_SUCCESS
);
3554 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3556 return log_error_errno(errno
, "Failed to send PID: %m");
3557 if (l
!= sizeof(pid
))
3558 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3559 "Short write while sending PID.");
3561 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3563 return log_error_errno(errno
, "Failed to send machine ID: %m");
3564 if (l
!= sizeof(arg_uuid
))
3565 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3566 "Short write while sending machine ID.");
3568 l
= send_one_fd(notify_socket
, fd
, 0);
3570 return log_error_errno(l
, "Failed to send notify fd: %m");
3572 pid_socket
= safe_close(pid_socket
);
3573 uuid_socket
= safe_close(uuid_socket
);
3574 notify_socket
= safe_close(notify_socket
);
3575 master_pty_socket
= safe_close(master_pty_socket
);
3576 kmsg_socket
= safe_close(kmsg_socket
);
3577 rtnl_socket
= safe_close(rtnl_socket
);
3578 netns_fd
= safe_close(netns_fd
);
3583 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3584 bool tried_hashed
= false;
3585 unsigned n_tries
= 100;
3590 assert(ret_lock_file
);
3591 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3592 assert(arg_uid_range
== 0x10000U
);
3596 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3599 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3600 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3605 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3607 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3610 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3611 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3612 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3617 /* Make some superficial checks whether the range is currently known in the user database */
3618 if (getpwuid(candidate
))
3620 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3622 if (getgrgid(candidate
))
3624 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3627 *ret_lock_file
= lf
;
3628 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3633 if (arg_machine
&& !tried_hashed
) {
3634 /* Try to hash the base from the container name */
3636 static const uint8_t hash_key
[] = {
3637 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3638 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3641 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3643 tried_hashed
= true;
3645 random_bytes(&candidate
, sizeof(candidate
));
3647 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3648 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3652 static int setup_uid_map(pid_t pid
) {
3653 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3658 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3659 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3660 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3662 return log_error_errno(r
, "Failed to write UID map: %m");
3664 /* We always assign the same UID and GID ranges */
3665 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3666 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3668 return log_error_errno(r
, "Failed to write GID map: %m");
3673 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3674 char buf
[NOTIFY_BUFFER_MAX
+1];
3676 struct iovec iovec
= {
3678 .iov_len
= sizeof(buf
)-1,
3681 struct cmsghdr cmsghdr
;
3682 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
3683 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
3685 struct msghdr msghdr
= {
3688 .msg_control
= &control
,
3689 .msg_controllen
= sizeof(control
),
3691 struct cmsghdr
*cmsg
;
3692 struct ucred
*ucred
= NULL
;
3694 pid_t inner_child_pid
;
3695 _cleanup_strv_free_
char **tags
= NULL
;
3699 inner_child_pid
= PTR_TO_PID(userdata
);
3701 if (revents
!= EPOLLIN
) {
3702 log_warning("Got unexpected poll event for notify fd.");
3706 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3708 if (IN_SET(errno
, EAGAIN
, EINTR
))
3711 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
3713 cmsg_close_all(&msghdr
);
3715 CMSG_FOREACH(cmsg
, &msghdr
) {
3716 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
3717 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
3718 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
3720 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
3724 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3725 log_debug("Received notify message without valid credentials. Ignoring.");
3729 if ((size_t) n
>= sizeof(buf
)) {
3730 log_warning("Received notify message exceeded maximum size. Ignoring.");
3735 tags
= strv_split(buf
, "\n\r");
3739 if (strv_find(tags
, "READY=1"))
3740 (void) sd_notifyf(false, "READY=1\n");
3742 p
= strv_find_startswith(tags
, "STATUS=");
3744 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3749 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3752 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3754 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3756 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3761 static int merge_settings(Settings
*settings
, const char *path
) {
3767 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3768 * that this steals the fields of the Settings* structure, and hence modifies it. */
3770 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3771 settings
->start_mode
>= 0) {
3772 arg_start_mode
= settings
->start_mode
;
3773 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3776 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3777 arg_ephemeral
= settings
->ephemeral
;
3779 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3782 if (!arg_settings_trusted
)
3783 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3785 free_and_replace(arg_directory
, settings
->root
);
3788 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3789 settings
->pivot_root_new
) {
3790 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3791 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3794 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3795 settings
->working_directory
)
3796 free_and_replace(arg_chdir
, settings
->working_directory
);
3798 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3799 settings
->environment
)
3800 strv_free_and_replace(arg_setenv
, settings
->environment
);
3802 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3805 free_and_replace(arg_user
, settings
->user
);
3807 if (uid_is_valid(settings
->uid
))
3808 arg_uid
= settings
->uid
;
3809 if (gid_is_valid(settings
->gid
))
3810 arg_gid
= settings
->gid
;
3811 if (settings
->n_supplementary_gids
> 0) {
3812 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3813 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3817 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3818 uint64_t plus
, minus
;
3819 uint64_t network_minus
= 0;
3821 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3822 * Settings structure */
3824 plus
= settings
->capability
;
3825 minus
= settings
->drop_capability
;
3827 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3828 if (settings_private_network(settings
))
3829 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3831 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3834 if (!arg_settings_trusted
&& plus
!= 0) {
3835 if (settings
->capability
!= 0)
3836 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3838 arg_caps_retain
&= ~network_minus
;
3839 arg_caps_retain
|= plus
;
3842 arg_caps_retain
&= ~minus
;
3844 /* Copy the full capabilities over too */
3845 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3846 if (!arg_settings_trusted
)
3847 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
3849 arg_full_capabilities
= settings
->full_capabilities
;
3853 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3854 settings
->kill_signal
> 0)
3855 arg_kill_signal
= settings
->kill_signal
;
3857 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3858 settings
->personality
!= PERSONALITY_INVALID
)
3859 arg_personality
= settings
->personality
;
3861 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3862 !sd_id128_is_null(settings
->machine_id
)) {
3864 if (!arg_settings_trusted
)
3865 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3867 arg_uuid
= settings
->machine_id
;
3870 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3871 settings
->read_only
>= 0)
3872 arg_read_only
= settings
->read_only
;
3874 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3875 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3876 arg_volatile_mode
= settings
->volatile_mode
;
3878 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3879 settings
->n_custom_mounts
> 0) {
3881 if (!arg_settings_trusted
)
3882 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3884 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3885 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3886 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3887 settings
->n_custom_mounts
= 0;
3891 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3892 (settings
->private_network
>= 0 ||
3893 settings
->network_veth
>= 0 ||
3894 settings
->network_bridge
||
3895 settings
->network_zone
||
3896 settings
->network_interfaces
||
3897 settings
->network_macvlan
||
3898 settings
->network_ipvlan
||
3899 settings
->network_veth_extra
||
3900 settings
->network_namespace_path
)) {
3902 if (!arg_settings_trusted
)
3903 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3905 arg_network_veth
= settings_network_veth(settings
);
3906 arg_private_network
= settings_private_network(settings
);
3908 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3909 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3910 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3911 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3913 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3914 free_and_replace(arg_network_zone
, settings
->network_zone
);
3916 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3920 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3921 settings
->expose_ports
) {
3923 if (!arg_settings_trusted
)
3924 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3926 expose_port_free_all(arg_expose_ports
);
3927 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3931 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3932 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3934 if (!arg_settings_trusted
)
3935 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3937 arg_userns_mode
= settings
->userns_mode
;
3938 arg_uid_shift
= settings
->uid_shift
;
3939 arg_uid_range
= settings
->uid_range
;
3940 arg_userns_chown
= settings
->userns_chown
;
3944 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3945 arg_notify_ready
= settings
->notify_ready
;
3947 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3949 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_whitelist
))
3950 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
3952 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3953 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3957 if (!arg_settings_trusted
&& settings
->seccomp
)
3958 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
3960 seccomp_release(arg_seccomp
);
3961 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
3966 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
3967 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
3970 if (!settings
->rlimit
[rl
])
3973 if (!arg_settings_trusted
) {
3974 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
3978 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
3981 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
3983 free_and_replace(arg_hostname
, settings
->hostname
);
3985 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
3986 settings
->no_new_privileges
>= 0)
3987 arg_no_new_privileges
= settings
->no_new_privileges
;
3989 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
3990 settings
->oom_score_adjust_set
) {
3992 if (!arg_settings_trusted
)
3993 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
3995 arg_oom_score_adjust
= settings
->oom_score_adjust
;
3996 arg_oom_score_adjust_set
= true;
4000 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4001 settings
->cpu_set
.set
) {
4003 if (!arg_settings_trusted
)
4004 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4006 cpu_set_reset(&arg_cpu_set
);
4007 arg_cpu_set
= settings
->cpu_set
;
4008 settings
->cpu_set
= (CPUSet
) {};
4012 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4013 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4014 arg_resolv_conf
= settings
->resolv_conf
;
4016 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4017 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4019 if (!arg_settings_trusted
)
4020 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4022 arg_link_journal
= settings
->link_journal
;
4023 arg_link_journal_try
= settings
->link_journal_try
;
4027 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4028 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4029 arg_timezone
= settings
->timezone
;
4031 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4034 if (!arg_settings_trusted
)
4035 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4037 free_and_replace(arg_slice
, settings
->slice
);
4040 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4041 settings
->use_cgns
>= 0) {
4043 if (!arg_settings_trusted
)
4044 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4046 arg_use_cgns
= settings
->use_cgns
;
4049 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4050 settings
->clone_ns_flags
!= (unsigned long) -1) {
4052 if (!arg_settings_trusted
)
4053 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4055 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4058 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4059 settings
->console_mode
>= 0) {
4061 if (!arg_settings_trusted
)
4062 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4064 arg_console_mode
= settings
->console_mode
;
4067 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4068 * don't consult arg_settings_mask for them. */
4070 sd_bus_message_unref(arg_property_message
);
4071 arg_property_message
= TAKE_PTR(settings
->properties
);
4073 arg_console_width
= settings
->console_width
;
4074 arg_console_height
= settings
->console_height
;
4076 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4077 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4078 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4083 static int load_settings(void) {
4084 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4085 _cleanup_fclose_
FILE *f
= NULL
;
4086 _cleanup_free_
char *p
= NULL
;
4093 /* If all settings are masked, there's no point in looking for
4094 * the settings file */
4095 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
4098 fn
= strjoina(arg_machine
, ".nspawn");
4100 /* We first look in the admin's directories in /etc and /run */
4101 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4102 _cleanup_free_
char *j
= NULL
;
4104 j
= path_join(i
, fn
);
4112 /* By default, we trust configuration from /etc and /run */
4113 if (arg_settings_trusted
< 0)
4114 arg_settings_trusted
= true;
4119 if (errno
!= ENOENT
)
4120 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4124 /* After that, let's look for a file next to the
4125 * actual image we shall boot. */
4128 p
= file_in_same_dir(arg_image
, fn
);
4131 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4132 p
= file_in_same_dir(arg_directory
, fn
);
4139 if (!f
&& errno
!= ENOENT
)
4140 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4142 /* By default, we do not trust configuration from /var/lib/machines */
4143 if (arg_settings_trusted
< 0)
4144 arg_settings_trusted
= false;
4151 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4153 r
= settings_load(f
, p
, &settings
);
4157 return merge_settings(settings
, p
);
4160 static int load_oci_bundle(void) {
4161 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4164 if (!arg_oci_bundle
)
4167 /* By default let's trust OCI bundles */
4168 if (arg_settings_trusted
< 0)
4169 arg_settings_trusted
= true;
4171 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4175 return merge_settings(settings
, arg_oci_bundle
);
4178 static int run_container(
4179 DissectedImage
*dissected_image
,
4182 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4183 union in_addr_union
*exposed
,
4184 int *master
, pid_t
*pid
, int *ret
) {
4186 static const struct sigaction sa
= {
4187 .sa_handler
= nop_signal_handler
,
4188 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4191 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4192 _cleanup_close_
int etc_passwd_lock
= -1;
4193 _cleanup_close_pair_
int
4194 kmsg_socket_pair
[2] = { -1, -1 },
4195 rtnl_socket_pair
[2] = { -1, -1 },
4196 pid_socket_pair
[2] = { -1, -1 },
4197 uuid_socket_pair
[2] = { -1, -1 },
4198 notify_socket_pair
[2] = { -1, -1 },
4199 uid_shift_socket_pair
[2] = { -1, -1 },
4200 master_pty_socket_pair
[2] = { -1, -1 },
4201 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4203 _cleanup_close_
int notify_socket
= -1;
4204 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4205 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4206 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4207 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4208 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4209 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4210 ContainerStatus container_status
= 0;
4214 _cleanup_close_
int child_netns_fd
= -1;
4216 assert_se(sigemptyset(&mask_chld
) == 0);
4217 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4219 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4220 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4221 * check with getpwuid() if the specific user already exists. Note that /etc might be
4222 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4223 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4224 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4227 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4228 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4229 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4232 r
= barrier_create(&barrier
);
4234 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4236 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4237 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4239 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4240 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4242 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4243 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4245 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4246 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4248 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4249 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4251 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4252 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4254 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4255 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4256 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4258 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4259 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4260 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4262 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4263 * parent's blocking calls and give it a chance to call wait() and terminate. */
4264 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4266 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4268 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4270 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4272 if (arg_network_namespace_path
) {
4273 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4274 if (child_netns_fd
< 0)
4275 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4277 r
= fd_is_network_ns(child_netns_fd
);
4279 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
);
4281 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4283 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4284 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4287 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4289 return log_error_errno(errno
, "clone() failed%s: %m",
4291 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4294 /* The outer child only has a file system namespace. */
4295 barrier_set_role(&barrier
, BARRIER_CHILD
);
4297 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4298 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4299 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4300 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4301 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4302 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4303 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4304 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4306 (void) reset_all_signal_handlers();
4307 (void) reset_signal_mask();
4309 r
= outer_child(&barrier
,
4314 uuid_socket_pair
[1],
4315 notify_socket_pair
[1],
4316 kmsg_socket_pair
[1],
4317 rtnl_socket_pair
[1],
4318 uid_shift_socket_pair
[1],
4319 master_pty_socket_pair
[1],
4320 unified_cgroup_hierarchy_socket_pair
[1],
4324 _exit(EXIT_FAILURE
);
4326 _exit(EXIT_SUCCESS
);
4329 barrier_set_role(&barrier
, BARRIER_PARENT
);
4333 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4334 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4335 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4336 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4337 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4338 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4339 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4340 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4342 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4343 /* The child just let us know the UID shift it might have read from the image. */
4344 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4346 return log_error_errno(errno
, "Failed to read UID shift: %m");
4347 if (l
!= sizeof arg_uid_shift
)
4348 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4350 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4351 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4352 * image, but if that's already in use, pick a new one, and report back to the child,
4353 * which one we now picked. */
4355 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4357 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4359 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4361 return log_error_errno(errno
, "Failed to send UID shift: %m");
4362 if (l
!= sizeof arg_uid_shift
)
4363 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4367 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4368 /* The child let us know the support cgroup mode it might have read from the image. */
4369 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4371 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4372 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4373 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4374 l
, l
== 0 ? " The child is most likely dead." : "");
4377 /* Wait for the outer child. */
4378 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4381 if (r
!= EXIT_SUCCESS
)
4384 /* And now retrieve the PID of the inner child. */
4385 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4387 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4388 if (l
!= sizeof *pid
)
4389 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4391 /* We also retrieve container UUID in case it was generated by outer child */
4392 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4394 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4395 if (l
!= sizeof(arg_uuid
))
4396 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4398 /* We also retrieve the socket used for notifications generated by outer child */
4399 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4400 if (notify_socket
< 0)
4401 return log_error_errno(notify_socket
,
4402 "Failed to receive notification socket from the outer child: %m");
4404 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4406 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4407 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4408 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4410 r
= setup_uid_map(*pid
);
4414 (void) barrier_place(&barrier
); /* #2 */
4417 if (arg_private_network
) {
4418 if (!arg_network_namespace_path
) {
4419 /* Wait until the child has unshared its network namespace. */
4420 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4421 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4424 if (child_netns_fd
< 0) {
4425 /* Make sure we have an open file descriptor to the child's network
4426 * namespace so it stays alive even if the child exits. */
4427 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4429 return log_error_errno(r
, "Failed to open child network namespace: %m");
4432 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4436 if (arg_network_veth
) {
4437 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4438 arg_network_bridge
|| arg_network_zone
);
4444 if (arg_network_bridge
) {
4445 /* Add the interface to a bridge */
4446 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4451 } else if (arg_network_zone
) {
4452 /* Add the interface to a bridge, possibly creating it */
4453 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4461 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4465 /* We created the primary and extra veth links now; let's remember this, so that we know to
4466 remove them later on. Note that we don't bother with removing veth links that were created
4467 here when their setup failed half-way, because in that case the kernel should be able to
4468 remove them on its own, since they cannot be referenced by anything yet. */
4469 *veth_created
= true;
4471 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4475 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4480 if (arg_register
|| !arg_keep_unit
) {
4481 r
= sd_bus_default_system(&bus
);
4483 return log_error_errno(r
, "Failed to open system bus: %m");
4485 r
= sd_bus_set_close_on_exit(bus
, false);
4487 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4490 if (!arg_keep_unit
) {
4491 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4492 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4493 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4495 r
= sd_bus_match_signal_async(
4498 "org.freedesktop.systemd1",
4500 "org.freedesktop.systemd1.Scope",
4502 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4504 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4508 r
= register_machine(
4516 arg_custom_mounts
, arg_n_custom_mounts
,
4519 arg_property_message
,
4521 arg_container_service_name
);
4525 } else if (!arg_keep_unit
) {
4531 arg_custom_mounts
, arg_n_custom_mounts
,
4534 arg_property_message
);
4538 } else if (arg_slice
|| arg_property
)
4539 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4541 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4545 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4549 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4553 /* Notify the child that the parent is ready with all
4554 * its setup (including cgroup-ification), and that
4555 * the child can now hand over control to the code to
4556 * run inside the container. */
4557 (void) barrier_place(&barrier
); /* #4 */
4559 /* Block SIGCHLD here, before notifying child.
4560 * process_pty() will handle it with the other signals. */
4561 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4563 /* Reset signal to default */
4564 r
= default_signals(SIGCHLD
, -1);
4566 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4568 r
= sd_event_new(&event
);
4570 return log_error_errno(r
, "Failed to get default event source: %m");
4572 (void) sd_event_set_watchdog(event
, true);
4575 r
= sd_bus_attach_event(bus
, event
, 0);
4577 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4580 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4584 /* Let the child know that we are ready and wait that the child is completely ready now. */
4585 if (!barrier_place_and_sync(&barrier
)) /* #5 */
4586 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4588 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4589 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4590 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4592 (void) sd_notifyf(false,
4593 "STATUS=Container running.\n"
4594 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4595 if (!arg_notify_ready
)
4596 (void) sd_notify(false, "READY=1\n");
4598 if (arg_kill_signal
> 0) {
4599 /* Try to kill the init system on SIGINT or SIGTERM */
4600 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4601 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4603 /* Immediately exit */
4604 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4605 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4608 /* Exit when the child exits */
4609 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4611 if (arg_expose_ports
) {
4612 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4616 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4619 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4621 if (arg_console_mode
!= CONSOLE_PIPE
) {
4622 _cleanup_close_
int fd
= -1;
4623 PTYForwardFlags flags
= 0;
4625 /* Retrieve the master pty allocated by inner child */
4626 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
4628 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
4630 switch (arg_console_mode
) {
4632 case CONSOLE_READ_ONLY
:
4633 flags
|= PTY_FORWARD_READ_ONLY
;
4637 case CONSOLE_INTERACTIVE
:
4638 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
4640 r
= pty_forward_new(event
, fd
, flags
, &forward
);
4642 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4644 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4645 (void) pty_forward_set_width_height(forward
,
4647 arg_console_height
);
4651 assert(arg_console_mode
== CONSOLE_PASSIVE
);
4654 *master
= TAKE_FD(fd
);
4657 r
= sd_event_loop(event
);
4659 return log_error_errno(r
, "Failed to run event loop: %m");
4664 (void) pty_forward_get_last_char(forward
, &last_char
);
4665 forward
= pty_forward_free(forward
);
4667 if (!arg_quiet
&& last_char
!= '\n')
4671 /* Kill if it is not dead yet anyway */
4672 if (!arg_register
&& !arg_keep_unit
&& bus
)
4673 terminate_scope(bus
, arg_machine
);
4675 /* Normally redundant, but better safe than sorry */
4676 (void) kill(*pid
, SIGKILL
);
4678 if (arg_private_network
) {
4679 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
4680 * to avoid having to move the parent to the child network namespace. */
4681 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
4686 _cleanup_close_
int parent_netns_fd
= -1;
4688 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
4690 log_error_errno(r
, "Failed to open parent network namespace: %m");
4691 _exit(EXIT_FAILURE
);
4694 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
4696 log_error_errno(r
, "Failed to enter child network namespace: %m");
4697 _exit(EXIT_FAILURE
);
4700 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
4702 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
4704 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
4708 r
= wait_for_container(*pid
, &container_status
);
4711 /* Tell machined that we are gone. */
4713 (void) unregister_machine(bus
, arg_machine
);
4716 /* We failed to wait for the container, or the container exited abnormally. */
4718 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4719 /* r > 0 → The container exited with a non-zero status.
4720 * As a special case, we need to replace 133 with a different value,
4721 * because 133 is special-cased in the service file to reboot the container.
4722 * otherwise → The container exited with zero status and a reboot was not requested.
4724 if (r
== EXIT_FORCE_RESTART
)
4725 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4727 return 0; /* finito */
4730 /* CONTAINER_REBOOTED, loop again */
4732 if (arg_keep_unit
) {
4733 /* Special handling if we are running as a service: instead of simply
4734 * restarting the machine we want to restart the entire service, so let's
4735 * inform systemd about this with the special exit code 133. The service
4736 * file uses RestartForceExitStatus=133 so that this results in a full
4737 * nspawn restart. This is necessary since we might have cgroup parameters
4738 * set we want to have flushed out. */
4739 *ret
= EXIT_FORCE_RESTART
;
4740 return 0; /* finito */
4743 expose_port_flush(arg_expose_ports
, exposed
);
4745 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4746 *veth_created
= false;
4747 return 1; /* loop again */
4750 static int initialize_rlimits(void) {
4751 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4752 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4753 * container execution environments. */
4755 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4756 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4757 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4758 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4759 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4760 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4761 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4762 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4763 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4764 [RLIMIT_NICE
] = { 0, 0 },
4765 [RLIMIT_NOFILE
] = { 1024, 4096 },
4766 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4767 [RLIMIT_RTPRIO
] = { 0, 0 },
4768 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4769 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4771 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4772 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4773 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4774 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4775 * that PID 1 changes a number of other resource limits during early initialization which is why we
4776 * don't read the other limits from PID 1 but prefer the static table above. */
4781 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4782 /* Let's only fill in what the user hasn't explicitly configured anyway */
4783 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4784 const struct rlimit
*v
;
4785 struct rlimit buffer
;
4787 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4788 /* For these two let's read the limits off PID 1. See above for an explanation. */
4790 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4791 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4795 v
= kernel_defaults
+ rl
;
4797 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4798 if (!arg_rlimit
[rl
])
4802 if (DEBUG_LOGGING
) {
4803 _cleanup_free_
char *k
= NULL
;
4805 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4806 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4813 static int run(int argc
, char *argv
[]) {
4814 bool secondary
= false, remove_directory
= false, remove_image
= false,
4815 veth_created
= false, remove_tmprootdir
= false;
4816 _cleanup_close_
int master
= -1;
4817 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4818 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4819 char veth_name
[IFNAMSIZ
] = "";
4820 union in_addr_union exposed
= {};
4821 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4822 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4823 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4824 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4825 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4828 log_parse_environment();
4831 r
= parse_argv(argc
, argv
);
4839 r
= initialize_rlimits();
4843 r
= load_oci_bundle();
4847 r
= determine_names();
4851 r
= load_settings();
4857 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4861 r
= verify_arguments();
4865 /* Reapply environment settings. */
4866 (void) detect_unified_cgroup_hierarchy_from_environment();
4868 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4869 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4870 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4871 (void) ignore_signals(SIGPIPE
, -1);
4873 n_fd_passed
= sd_listen_fds(false);
4874 if (n_fd_passed
> 0) {
4875 r
= fdset_new_listen_fds(&fds
, false);
4877 log_error_errno(r
, "Failed to collect file descriptors: %m");
4882 /* The "default" umask. This is appropriate for most file and directory
4883 * operations performed by nspawn, and is the umask that will be used for
4884 * the child. Functions like copy_devnodes() change the umask temporarily. */
4887 if (arg_directory
) {
4890 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4891 * /var from the host will propagate into container dynamically (because bad things happen if
4892 * two systems write to the same /var). Let's allow it for the special cases where /var is
4893 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4894 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
4895 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4900 if (arg_ephemeral
) {
4901 _cleanup_free_
char *np
= NULL
;
4903 r
= chase_symlinks_and_update(&arg_directory
, 0);
4907 /* If the specified path is a mount point we generate the new snapshot immediately
4908 * inside it under a random name. However if the specified is not a mount point we
4909 * create the new snapshot in the parent directory, just next to it. */
4910 r
= path_is_mount_point(arg_directory
, NULL
, 0);
4912 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
4916 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
4918 r
= tempfn_random(arg_directory
, "machine.", &np
);
4920 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
4924 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4925 * only owned by us and noone else. */
4926 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4928 log_error_errno(r
, "Failed to lock %s: %m", np
);
4933 BLOCK_SIGNALS(SIGINT
);
4934 r
= btrfs_subvol_snapshot(arg_directory
, np
,
4935 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4936 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4937 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4938 BTRFS_SNAPSHOT_RECURSIVE
|
4939 BTRFS_SNAPSHOT_QUOTA
|
4940 BTRFS_SNAPSHOT_SIGINT
);
4943 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
4947 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
4951 free_and_replace(arg_directory
, np
);
4952 remove_directory
= true;
4954 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
4958 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
4960 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
4964 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
4969 r
= chase_symlinks_and_update(&arg_template
, 0);
4974 BLOCK_SIGNALS(SIGINT
);
4975 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
4976 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
4977 BTRFS_SNAPSHOT_FALLBACK_COPY
|
4978 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
4979 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
4980 BTRFS_SNAPSHOT_RECURSIVE
|
4981 BTRFS_SNAPSHOT_QUOTA
|
4982 BTRFS_SNAPSHOT_SIGINT
);
4985 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4986 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
4987 else if (r
== -EINTR
) {
4988 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
4991 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
4994 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
4995 "Populated %s from template %s.", arg_directory
, arg_template
);
4999 if (arg_start_mode
== START_BOOT
) {
5002 if (arg_pivot_root_new
)
5003 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5007 if (path_is_os_tree(p
) <= 0) {
5008 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5015 if (arg_pivot_root_new
)
5016 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5020 q
= strjoina(p
, "/usr/");
5022 if (laccess(q
, F_OK
) < 0) {
5023 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
5031 assert(!arg_template
);
5033 r
= chase_symlinks_and_update(&arg_image
, 0);
5037 if (arg_ephemeral
) {
5038 _cleanup_free_
char *np
= NULL
;
5040 r
= tempfn_random(arg_image
, "machine.", &np
);
5042 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5046 /* Always take an exclusive lock on our own ephemeral copy. */
5047 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5049 r
= log_error_errno(r
, "Failed to create image lock: %m");
5054 BLOCK_SIGNALS(SIGINT
);
5055 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
);
5058 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5062 r
= log_error_errno(r
, "Failed to copy image file: %m");
5066 free_and_replace(arg_image
, np
);
5067 remove_image
= true;
5069 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5071 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5075 r
= log_error_errno(r
, "Failed to create image lock: %m");
5079 if (!arg_root_hash
) {
5080 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
5082 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
5088 if (!mkdtemp(tmprootdir
)) {
5089 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5093 remove_tmprootdir
= true;
5095 arg_directory
= strdup(tmprootdir
);
5096 if (!arg_directory
) {
5101 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, LO_FLAGS_PARTSCAN
, &loop
);
5103 log_error_errno(r
, "Failed to set up loopback block device: %m");
5107 r
= dissect_image_and_warn(
5110 arg_root_hash
, arg_root_hash_size
,
5111 DISSECT_IMAGE_REQUIRE_ROOT
,
5114 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5115 log_notice("Note that the disk image needs to\n"
5116 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5117 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5118 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
5119 " d) or contain a file system without a partition table\n"
5120 "in order to be bootable with systemd-nspawn.");
5126 if (!arg_root_hash
&& dissected_image
->can_verity
)
5127 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5129 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
5133 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5134 if (remove_image
&& unlink(arg_image
) >= 0)
5135 remove_image
= false;
5138 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5142 if (arg_console_mode
< 0)
5144 isatty(STDIN_FILENO
) > 0 &&
5145 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5147 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5151 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5152 arg_machine
, arg_image
?: arg_directory
);
5154 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5156 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5157 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5162 r
= run_container(dissected_image
,
5165 veth_name
, &veth_created
,
5173 (void) sd_notify(false,
5174 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5175 "STOPPING=1\nSTATUS=Terminating...");
5178 (void) kill(pid
, SIGKILL
);
5180 /* Try to flush whatever is still queued in the pty */
5182 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5183 master
= safe_close(master
);
5187 (void) wait_for_terminate(pid
, NULL
);
5191 if (remove_directory
&& arg_directory
) {
5194 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5196 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5199 if (remove_image
&& arg_image
) {
5200 if (unlink(arg_image
) < 0)
5201 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5204 if (remove_tmprootdir
) {
5205 if (rmdir(tmprootdir
) < 0)
5206 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5212 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5213 (void) rm_rf(p
, REMOVE_ROOT
);
5216 expose_port_flush(arg_expose_ports
, &exposed
);
5219 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5220 (void) remove_bridge(arg_network_zone
);
5222 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5223 expose_port_free_all(arg_expose_ports
);
5224 rlimit_free_all(arg_rlimit
);
5225 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5233 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);