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 "resolve-util.h"
83 #include "rlimit-util.h"
86 #include "seccomp-util.h"
88 #include "selinux-util.h"
89 #include "signal-util.h"
90 #include "socket-util.h"
91 #include "stat-util.h"
92 #include "stdio-util.h"
93 #include "string-table.h"
94 #include "string-util.h"
96 #include "sysctl-util.h"
97 #include "terminal-util.h"
98 #include "tmpfile-util.h"
99 #include "umask-util.h"
100 #include "unit-name.h"
101 #include "user-util.h"
104 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
105 * nspawn_notify_socket_path is relative to the container
106 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
107 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
109 #define EXIT_FORCE_RESTART 133
111 typedef enum ContainerStatus
{
112 CONTAINER_TERMINATED
,
116 static char *arg_directory
= NULL
;
117 static char *arg_template
= NULL
;
118 static char *arg_chdir
= NULL
;
119 static char *arg_pivot_root_new
= NULL
;
120 static char *arg_pivot_root_old
= NULL
;
121 static char *arg_user
= NULL
;
122 static uid_t arg_uid
= UID_INVALID
;
123 static gid_t arg_gid
= GID_INVALID
;
124 static gid_t
* arg_supplementary_gids
= NULL
;
125 static size_t arg_n_supplementary_gids
= 0;
126 static sd_id128_t arg_uuid
= {};
127 static char *arg_machine
= NULL
; /* The name used by the host to refer to this */
128 static char *arg_hostname
= NULL
; /* The name the payload sees by default */
129 static const char *arg_selinux_context
= NULL
;
130 static const char *arg_selinux_apifs_context
= NULL
;
131 static char *arg_slice
= NULL
;
132 static bool arg_private_network
= false;
133 static bool arg_read_only
= false;
134 static StartMode arg_start_mode
= START_PID1
;
135 static bool arg_ephemeral
= false;
136 static LinkJournal arg_link_journal
= LINK_AUTO
;
137 static bool arg_link_journal_try
= false;
138 static uint64_t arg_caps_retain
=
139 (1ULL << CAP_AUDIT_CONTROL
) |
140 (1ULL << CAP_AUDIT_WRITE
) |
141 (1ULL << CAP_CHOWN
) |
142 (1ULL << CAP_DAC_OVERRIDE
) |
143 (1ULL << CAP_DAC_READ_SEARCH
) |
144 (1ULL << CAP_FOWNER
) |
145 (1ULL << CAP_FSETID
) |
146 (1ULL << CAP_IPC_OWNER
) |
148 (1ULL << CAP_LEASE
) |
149 (1ULL << CAP_LINUX_IMMUTABLE
) |
150 (1ULL << CAP_MKNOD
) |
151 (1ULL << CAP_NET_BIND_SERVICE
) |
152 (1ULL << CAP_NET_BROADCAST
) |
153 (1ULL << CAP_NET_RAW
) |
154 (1ULL << CAP_SETFCAP
) |
155 (1ULL << CAP_SETGID
) |
156 (1ULL << CAP_SETPCAP
) |
157 (1ULL << CAP_SETUID
) |
158 (1ULL << CAP_SYS_ADMIN
) |
159 (1ULL << CAP_SYS_BOOT
) |
160 (1ULL << CAP_SYS_CHROOT
) |
161 (1ULL << CAP_SYS_NICE
) |
162 (1ULL << CAP_SYS_PTRACE
) |
163 (1ULL << CAP_SYS_RESOURCE
) |
164 (1ULL << CAP_SYS_TTY_CONFIG
);
165 static CapabilityQuintet arg_full_capabilities
= CAPABILITY_QUINTET_NULL
;
166 static CustomMount
*arg_custom_mounts
= NULL
;
167 static size_t arg_n_custom_mounts
= 0;
168 static char **arg_setenv
= NULL
;
169 static bool arg_quiet
= false;
170 static bool arg_register
= true;
171 static bool arg_keep_unit
= false;
172 static char **arg_network_interfaces
= NULL
;
173 static char **arg_network_macvlan
= NULL
;
174 static char **arg_network_ipvlan
= NULL
;
175 static bool arg_network_veth
= false;
176 static char **arg_network_veth_extra
= NULL
;
177 static char *arg_network_bridge
= NULL
;
178 static char *arg_network_zone
= NULL
;
179 static char *arg_network_namespace_path
= NULL
;
180 static PagerFlags arg_pager_flags
= 0;
181 static unsigned long arg_personality
= PERSONALITY_INVALID
;
182 static char *arg_image
= NULL
;
183 static char *arg_oci_bundle
= NULL
;
184 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
185 static ExposePort
*arg_expose_ports
= NULL
;
186 static char **arg_property
= NULL
;
187 static sd_bus_message
*arg_property_message
= NULL
;
188 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
189 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
190 static bool arg_userns_chown
= false;
191 static int arg_kill_signal
= 0;
192 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
193 static SettingsMask arg_settings_mask
= 0;
194 static int arg_settings_trusted
= -1;
195 static char **arg_parameters
= NULL
;
196 static const char *arg_container_service_name
= "systemd-nspawn";
197 static bool arg_notify_ready
= false;
198 static bool arg_use_cgns
= true;
199 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
200 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_TMPFS_TMP
;
201 static void *arg_root_hash
= NULL
;
202 static char *arg_verity_data
= NULL
;
203 static char *arg_root_hash_sig_path
= NULL
;
204 static void *arg_root_hash_sig
= NULL
;
205 static size_t arg_root_hash_sig_size
= 0;
206 static size_t arg_root_hash_size
= 0;
207 static char **arg_syscall_allow_list
= NULL
;
208 static char **arg_syscall_deny_list
= NULL
;
210 static scmp_filter_ctx arg_seccomp
= NULL
;
212 static struct rlimit
*arg_rlimit
[_RLIMIT_MAX
] = {};
213 static bool arg_no_new_privileges
= false;
214 static int arg_oom_score_adjust
= 0;
215 static bool arg_oom_score_adjust_set
= false;
216 static CPUSet arg_cpu_set
= {};
217 static ResolvConfMode arg_resolv_conf
= RESOLV_CONF_AUTO
;
218 static TimezoneMode arg_timezone
= TIMEZONE_AUTO
;
219 static unsigned arg_console_width
= (unsigned) -1, arg_console_height
= (unsigned) -1;
220 static DeviceNode
* arg_extra_nodes
= NULL
;
221 static size_t arg_n_extra_nodes
= 0;
222 static char **arg_sysctl
= NULL
;
223 static ConsoleMode arg_console_mode
= _CONSOLE_MODE_INVALID
;
225 STATIC_DESTRUCTOR_REGISTER(arg_directory
, freep
);
226 STATIC_DESTRUCTOR_REGISTER(arg_template
, freep
);
227 STATIC_DESTRUCTOR_REGISTER(arg_chdir
, freep
);
228 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new
, freep
);
229 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old
, freep
);
230 STATIC_DESTRUCTOR_REGISTER(arg_user
, freep
);
231 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids
, freep
);
232 STATIC_DESTRUCTOR_REGISTER(arg_machine
, freep
);
233 STATIC_DESTRUCTOR_REGISTER(arg_hostname
, freep
);
234 STATIC_DESTRUCTOR_REGISTER(arg_slice
, freep
);
235 STATIC_DESTRUCTOR_REGISTER(arg_setenv
, strv_freep
);
236 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces
, strv_freep
);
237 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan
, strv_freep
);
238 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan
, strv_freep
);
239 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra
, strv_freep
);
240 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge
, freep
);
241 STATIC_DESTRUCTOR_REGISTER(arg_network_zone
, freep
);
242 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path
, freep
);
243 STATIC_DESTRUCTOR_REGISTER(arg_image
, freep
);
244 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle
, freep
);
245 STATIC_DESTRUCTOR_REGISTER(arg_property
, strv_freep
);
246 STATIC_DESTRUCTOR_REGISTER(arg_property_message
, sd_bus_message_unrefp
);
247 STATIC_DESTRUCTOR_REGISTER(arg_parameters
, strv_freep
);
248 STATIC_DESTRUCTOR_REGISTER(arg_root_hash
, freep
);
249 STATIC_DESTRUCTOR_REGISTER(arg_verity_data
, freep
);
250 STATIC_DESTRUCTOR_REGISTER(arg_root_hash_sig_path
, freep
);
251 STATIC_DESTRUCTOR_REGISTER(arg_root_hash_sig
, freep
);
252 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list
, strv_freep
);
253 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list
, strv_freep
);
255 STATIC_DESTRUCTOR_REGISTER(arg_seccomp
, seccomp_releasep
);
257 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set
, cpu_set_reset
);
258 STATIC_DESTRUCTOR_REGISTER(arg_sysctl
, strv_freep
);
260 static int handle_arg_console(const char *arg
) {
261 if (streq(arg
, "help")) {
269 if (streq(arg
, "interactive"))
270 arg_console_mode
= CONSOLE_INTERACTIVE
;
271 else if (streq(arg
, "read-only"))
272 arg_console_mode
= CONSOLE_READ_ONLY
;
273 else if (streq(arg
, "passive"))
274 arg_console_mode
= CONSOLE_PASSIVE
;
275 else if (streq(arg
, "pipe"))
276 arg_console_mode
= CONSOLE_PIPE
;
278 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unknown console mode: %s", optarg
);
280 arg_settings_mask
|= SETTING_CONSOLE_MODE
;
284 static int help(void) {
285 _cleanup_free_
char *link
= NULL
;
288 (void) pager_open(arg_pager_flags
);
290 r
= terminal_urlify_man("systemd-nspawn", "1", &link
);
294 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
295 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
296 " -h --help Show this help\n"
297 " --version Print version string\n"
298 " -q --quiet Do not show status information\n"
299 " --no-pager Do not pipe output into a pager\n"
300 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
302 " -D --directory=PATH Root directory for the container\n"
303 " --template=PATH Initialize root directory from template directory,\n"
305 " -x --ephemeral Run container with snapshot of root directory, and\n"
306 " remove it after exit\n"
307 " -i --image=PATH Root file system disk image (or device node) for\n"
309 " --oci-bundle=PATH OCI bundle directory\n"
310 " --read-only Mount the root directory read-only\n"
311 " --volatile[=MODE] Run the system in volatile mode\n"
312 " --root-hash=HASH Specify verity root hash for root disk image\n"
313 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
314 " as a DER encoded PKCS7, either as a path to a file\n"
315 " or as an ASCII base64 encoded string prefixed by\n"
317 " --verity-data=PATH Specify hash device for verity\n"
318 " --pivot-root=PATH[:PATH]\n"
319 " Pivot root to given directory in the container\n\n"
320 "%3$sExecution:%4$s\n"
321 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
322 " -b --boot Boot up full system (i.e. invoke init)\n"
323 " --chdir=PATH Set working directory in the container\n"
324 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
325 " -u --user=USER Run the command under specified user or UID\n"
326 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
327 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
328 "%3$sSystem Identity:%4$s\n"
329 " -M --machine=NAME Set the machine name for the container\n"
330 " --hostname=NAME Override the hostname for the container\n"
331 " --uuid=UUID Set a specific machine UUID for the container\n\n"
332 "%3$sProperties:%4$s\n"
333 " -S --slice=SLICE Place the container in the specified slice\n"
334 " --property=NAME=VALUE Set scope unit property\n"
335 " --register=BOOLEAN Register container as machine\n"
336 " --keep-unit Do not register a scope for the machine, reuse\n"
337 " the service unit nspawn is running in\n\n"
338 "%3$sUser Namespacing:%4$s\n"
339 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
340 " --private-users[=UIDBASE[:NUIDS]]\n"
341 " Similar, but with user configured UID/GID range\n"
342 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
343 "%3$sNetworking:%4$s\n"
344 " --private-network Disable network in container\n"
345 " --network-interface=INTERFACE\n"
346 " Assign an existing network interface to the\n"
348 " --network-macvlan=INTERFACE\n"
349 " Create a macvlan network interface based on an\n"
350 " existing network interface to the container\n"
351 " --network-ipvlan=INTERFACE\n"
352 " Create a ipvlan network interface based on an\n"
353 " existing network interface to the container\n"
354 " -n --network-veth Add a virtual Ethernet connection between host\n"
356 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
357 " Add an additional virtual Ethernet link between\n"
358 " host and container\n"
359 " --network-bridge=INTERFACE\n"
360 " Add a virtual Ethernet connection to the container\n"
361 " and attach it to an existing bridge on the host\n"
362 " --network-zone=NAME Similar, but attach the new interface to an\n"
363 " an automatically managed bridge interface\n"
364 " --network-namespace-path=PATH\n"
365 " Set network namespace to the one represented by\n"
366 " the specified kernel namespace file node\n"
367 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
368 " Expose a container IP port on the host\n\n"
369 "%3$sSecurity:%4$s\n"
370 " --capability=CAP In addition to the default, retain specified\n"
372 " --drop-capability=CAP Drop the specified capability from the default set\n"
373 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
374 " --system-call-filter=LIST|~LIST\n"
375 " Permit/prohibit specific system calls\n"
376 " -Z --selinux-context=SECLABEL\n"
377 " Set the SELinux security context to be used by\n"
378 " processes in the container\n"
379 " -L --selinux-apifs-context=SECLABEL\n"
380 " Set the SELinux security context to be used by\n"
381 " API/tmpfs file systems in the container\n\n"
382 "%3$sResources:%4$s\n"
383 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
384 " --oom-score-adjust=VALUE\n"
385 " Adjust the OOM score value for the payload\n"
386 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
387 " --personality=ARCH Pick personality for this container\n\n"
388 "%3$sIntegration:%4$s\n"
389 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
390 " --timezone=MODE Select mode of /etc/localtime initialization\n"
391 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
392 " host, try-guest, try-host\n"
393 " -j Equivalent to --link-journal=try-guest\n\n"
395 " --bind=PATH[:PATH[:OPTIONS]]\n"
396 " Bind mount a file or directory from the host into\n"
398 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
399 " Similar, but creates a read-only bind mount\n"
400 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
402 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
403 " --overlay=PATH[:PATH...]:PATH\n"
404 " Create an overlay mount from the host to \n"
406 " --overlay-ro=PATH[:PATH...]:PATH\n"
407 " Similar, but creates a read-only overlay mount\n\n"
408 "%3$sInput/Output:%4$s\n"
409 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
410 " set up for the container.\n"
411 " -P --pipe Equivalent to --console=pipe\n"
412 "\nSee the %2$s for details.\n"
413 , program_invocation_short_name
415 , ansi_underline(), ansi_normal()
416 , ansi_highlight(), ansi_normal()
422 static int custom_mount_check_all(void) {
425 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
426 CustomMount
*m
= &arg_custom_mounts
[i
];
428 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
429 if (arg_userns_chown
)
430 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
431 "--private-users-chown may not be combined with custom root mounts.");
432 else if (arg_uid_shift
== UID_INVALID
)
433 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
434 "--private-users with automatic UID shift may not be combined with custom root mounts.");
441 static int detect_unified_cgroup_hierarchy_from_environment(void) {
442 const char *e
, *var
= "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
445 /* Allow the user to control whether the unified hierarchy is used */
449 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
450 var
= "UNIFIED_CGROUP_HIERARCHY";
455 r
= parse_boolean(e
);
457 return log_error_errno(r
, "Failed to parse $%s: %m", var
);
459 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
461 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
467 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
470 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
471 * in the image actually supports. */
472 r
= cg_all_unified();
474 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
476 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
477 * routine only detects 231, so we'll have a false negative here for 230. */
478 r
= systemd_installation_has_version(directory
, 230);
480 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
482 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
484 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
485 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
486 /* Mixed cgroup hierarchy support was added in 233 */
487 r
= systemd_installation_has_version(directory
, 233);
489 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
491 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
493 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
495 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
497 log_debug("Using %s hierarchy for container.",
498 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
499 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
504 static int parse_capability_spec(const char *spec
, uint64_t *ret_mask
) {
509 _cleanup_free_
char *t
= NULL
;
511 r
= extract_first_word(&spec
, &t
, ",", 0);
513 return log_error_errno(r
, "Failed to parse capability %s.", t
);
517 if (streq(t
, "help")) {
518 for (int i
= 0; i
< capability_list_length(); i
++) {
521 name
= capability_to_name(i
);
530 mask
= (uint64_t) -1;
532 r
= capability_from_name(t
);
534 return log_error_errno(r
, "Failed to parse capability %s.", t
);
541 return 1; /* continue */
544 static int parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
547 r
= getenv_bool(name
);
551 return log_error_errno(r
, "Failed to parse $%s: %m", name
);
553 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
554 arg_settings_mask
|= SETTING_CLONE_NS_FLAGS
;
558 static int parse_mount_settings_env(void) {
562 r
= getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
563 if (r
< 0 && r
!= -ENXIO
)
564 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
566 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_TMPFS_TMP
, r
> 0);
568 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
569 if (streq_ptr(e
, "network"))
570 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
573 r
= parse_boolean(e
);
575 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
577 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
578 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
584 static int parse_environment(void) {
588 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
591 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
594 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
597 r
= parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
601 r
= parse_mount_settings_env();
605 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
606 * even if it is supported. If not supported, it has no effect. */
607 if (!cg_ns_supported())
608 arg_use_cgns
= false;
610 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
613 return log_error_errno(r
, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
617 arg_use_cgns
= r
> 0;
618 arg_settings_mask
|= SETTING_USE_CGNS
;
622 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
624 arg_container_service_name
= e
;
626 return detect_unified_cgroup_hierarchy_from_environment();
629 static int parse_argv(int argc
, char *argv
[]) {
647 ARG_NETWORK_INTERFACE
,
652 ARG_NETWORK_VETH_EXTRA
,
653 ARG_NETWORK_NAMESPACE_PATH
,
663 ARG_PRIVATE_USERS_CHOWN
,
666 ARG_SYSTEM_CALL_FILTER
,
669 ARG_NO_NEW_PRIVILEGES
,
670 ARG_OOM_SCORE_ADJUST
,
682 static const struct option options
[] = {
683 { "help", no_argument
, NULL
, 'h' },
684 { "version", no_argument
, NULL
, ARG_VERSION
},
685 { "directory", required_argument
, NULL
, 'D' },
686 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
687 { "ephemeral", no_argument
, NULL
, 'x' },
688 { "user", required_argument
, NULL
, 'u' },
689 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
690 { "as-pid2", no_argument
, NULL
, 'a' },
691 { "boot", no_argument
, NULL
, 'b' },
692 { "uuid", required_argument
, NULL
, ARG_UUID
},
693 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
694 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
695 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
696 { "no-new-privileges", required_argument
, NULL
, ARG_NO_NEW_PRIVILEGES
},
697 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
698 { "bind", required_argument
, NULL
, ARG_BIND
},
699 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
700 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
701 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
702 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
703 { "inaccessible", required_argument
, NULL
, ARG_INACCESSIBLE
},
704 { "machine", required_argument
, NULL
, 'M' },
705 { "hostname", required_argument
, NULL
, ARG_HOSTNAME
},
706 { "slice", required_argument
, NULL
, 'S' },
707 { "setenv", required_argument
, NULL
, 'E' },
708 { "selinux-context", required_argument
, NULL
, 'Z' },
709 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
710 { "quiet", no_argument
, NULL
, 'q' },
711 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
712 { "register", required_argument
, NULL
, ARG_REGISTER
},
713 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
714 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
715 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
716 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
717 { "network-veth", no_argument
, NULL
, 'n' },
718 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
719 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
720 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
721 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
722 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
723 { "image", required_argument
, NULL
, 'i' },
724 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
725 { "port", required_argument
, NULL
, 'p' },
726 { "property", required_argument
, NULL
, ARG_PROPERTY
},
727 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
728 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
729 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
730 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
731 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
732 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
733 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
734 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
735 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
736 { "rlimit", required_argument
, NULL
, ARG_RLIMIT
},
737 { "oom-score-adjust", required_argument
, NULL
, ARG_OOM_SCORE_ADJUST
},
738 { "cpu-affinity", required_argument
, NULL
, ARG_CPU_AFFINITY
},
739 { "resolv-conf", required_argument
, NULL
, ARG_RESOLV_CONF
},
740 { "timezone", required_argument
, NULL
, ARG_TIMEZONE
},
741 { "console", required_argument
, NULL
, ARG_CONSOLE
},
742 { "pipe", no_argument
, NULL
, ARG_PIPE
},
743 { "oci-bundle", required_argument
, NULL
, ARG_OCI_BUNDLE
},
744 { "no-pager", no_argument
, NULL
, ARG_NO_PAGER
},
745 { "verity-data", required_argument
, NULL
, ARG_VERITY_DATA
},
746 { "root-hash-sig", required_argument
, NULL
, ARG_ROOT_HASH_SIG
},
751 uint64_t plus
= 0, minus
= 0;
752 bool mask_all_settings
= false, mask_no_settings
= false;
757 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options
, NULL
)) >= 0)
767 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
771 arg_settings_mask
|= SETTING_DIRECTORY
;
775 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
779 arg_settings_mask
|= SETTING_DIRECTORY
;
783 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
787 arg_settings_mask
|= SETTING_DIRECTORY
;
791 r
= parse_path_argument_and_warn(optarg
, false, &arg_oci_bundle
);
798 arg_ephemeral
= true;
799 arg_settings_mask
|= SETTING_EPHEMERAL
;
803 r
= free_and_strdup(&arg_user
, optarg
);
807 arg_settings_mask
|= SETTING_USER
;
810 case ARG_NETWORK_ZONE
: {
813 j
= strjoin("vz-", optarg
);
817 if (!ifname_valid(j
)) {
818 log_error("Network zone name not valid: %s", j
);
823 free_and_replace(arg_network_zone
, j
);
825 arg_network_veth
= true;
826 arg_private_network
= true;
827 arg_settings_mask
|= SETTING_NETWORK
;
831 case ARG_NETWORK_BRIDGE
:
833 if (!ifname_valid(optarg
))
834 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
835 "Bridge interface name not valid: %s", optarg
);
837 r
= free_and_strdup(&arg_network_bridge
, optarg
);
843 arg_network_veth
= true;
844 arg_private_network
= true;
845 arg_settings_mask
|= SETTING_NETWORK
;
848 case ARG_NETWORK_VETH_EXTRA
:
849 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
851 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
853 arg_private_network
= true;
854 arg_settings_mask
|= SETTING_NETWORK
;
857 case ARG_NETWORK_INTERFACE
:
858 if (!ifname_valid(optarg
))
859 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
860 "Network interface name not valid: %s", optarg
);
862 r
= test_network_interface_initialized(optarg
);
866 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
869 arg_private_network
= true;
870 arg_settings_mask
|= SETTING_NETWORK
;
873 case ARG_NETWORK_MACVLAN
:
875 if (!ifname_valid(optarg
))
876 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
877 "MACVLAN network interface name not valid: %s", optarg
);
879 r
= test_network_interface_initialized(optarg
);
883 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
886 arg_private_network
= true;
887 arg_settings_mask
|= SETTING_NETWORK
;
890 case ARG_NETWORK_IPVLAN
:
892 if (!ifname_valid(optarg
))
893 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
894 "IPVLAN network interface name not valid: %s", optarg
);
896 r
= test_network_interface_initialized(optarg
);
900 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
904 case ARG_PRIVATE_NETWORK
:
905 arg_private_network
= true;
906 arg_settings_mask
|= SETTING_NETWORK
;
909 case ARG_NETWORK_NAMESPACE_PATH
:
910 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
914 arg_settings_mask
|= SETTING_NETWORK
;
918 if (arg_start_mode
== START_PID2
)
919 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
920 "--boot and --as-pid2 may not be combined.");
922 arg_start_mode
= START_BOOT
;
923 arg_settings_mask
|= SETTING_START_MODE
;
927 if (arg_start_mode
== START_BOOT
)
928 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
929 "--boot and --as-pid2 may not be combined.");
931 arg_start_mode
= START_PID2
;
932 arg_settings_mask
|= SETTING_START_MODE
;
936 r
= sd_id128_from_string(optarg
, &arg_uuid
);
938 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
940 if (sd_id128_is_null(arg_uuid
))
941 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
942 "Machine UUID may not be all zeroes.");
944 arg_settings_mask
|= SETTING_MACHINE_ID
;
948 _cleanup_free_
char *mangled
= NULL
;
950 r
= unit_name_mangle_with_suffix(optarg
, NULL
, UNIT_NAME_MANGLE_WARN
, ".slice", &mangled
);
954 free_and_replace(arg_slice
, mangled
);
955 arg_settings_mask
|= SETTING_SLICE
;
961 arg_machine
= mfree(arg_machine
);
963 if (!machine_name_is_valid(optarg
))
964 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
965 "Invalid machine name: %s", optarg
);
967 r
= free_and_strdup(&arg_machine
, optarg
);
975 arg_hostname
= mfree(arg_hostname
);
977 if (!hostname_is_valid(optarg
, false))
978 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
979 "Invalid hostname: %s", optarg
);
981 r
= free_and_strdup(&arg_hostname
, optarg
);
986 arg_settings_mask
|= SETTING_HOSTNAME
;
990 arg_selinux_context
= optarg
;
994 arg_selinux_apifs_context
= optarg
;
998 arg_read_only
= true;
999 arg_settings_mask
|= SETTING_READ_ONLY
;
1002 case ARG_CAPABILITY
:
1003 case ARG_DROP_CAPABILITY
: {
1005 r
= parse_capability_spec(optarg
, &m
);
1009 if (c
== ARG_CAPABILITY
)
1013 arg_settings_mask
|= SETTING_CAPABILITY
;
1016 case ARG_NO_NEW_PRIVILEGES
:
1017 r
= parse_boolean(optarg
);
1019 return log_error_errno(r
, "Failed to parse --no-new-privileges= argument: %s", optarg
);
1021 arg_no_new_privileges
= r
;
1022 arg_settings_mask
|= SETTING_NO_NEW_PRIVILEGES
;
1026 arg_link_journal
= LINK_GUEST
;
1027 arg_link_journal_try
= true;
1028 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1031 case ARG_LINK_JOURNAL
:
1032 r
= parse_link_journal(optarg
, &arg_link_journal
, &arg_link_journal_try
);
1034 return log_error_errno(r
, "Failed to parse link journal mode %s", optarg
);
1036 arg_settings_mask
|= SETTING_LINK_JOURNAL
;
1041 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
1043 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
1045 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1049 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1051 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
1053 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1057 case ARG_OVERLAY_RO
:
1058 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
1059 if (r
== -EADDRNOTAVAIL
)
1060 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1062 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
1064 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1067 case ARG_INACCESSIBLE
:
1068 r
= inaccessible_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
1070 return log_error_errno(r
, "Failed to parse --inaccessible= argument %s: %m", optarg
);
1072 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
1078 if (!env_assignment_is_valid(optarg
))
1079 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1080 "Environment variable assignment '%s' is not valid.", optarg
);
1082 n
= strv_env_set(arg_setenv
, optarg
);
1086 strv_free_and_replace(arg_setenv
, n
);
1087 arg_settings_mask
|= SETTING_ENVIRONMENT
;
1095 case ARG_SHARE_SYSTEM
:
1096 /* We don't officially support this anymore, except for compat reasons. People should use the
1097 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1098 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1099 arg_clone_ns_flags
= 0;
1103 r
= parse_boolean(optarg
);
1105 log_error("Failed to parse --register= argument: %s", optarg
);
1113 arg_keep_unit
= true;
1116 case ARG_PERSONALITY
:
1118 arg_personality
= personality_from_string(optarg
);
1119 if (arg_personality
== PERSONALITY_INVALID
)
1120 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1121 "Unknown or unsupported personality '%s'.", optarg
);
1123 arg_settings_mask
|= SETTING_PERSONALITY
;
1129 arg_volatile_mode
= VOLATILE_YES
;
1130 else if (streq(optarg
, "help")) {
1131 DUMP_STRING_TABLE(volatile_mode
, VolatileMode
, _VOLATILE_MODE_MAX
);
1136 m
= volatile_mode_from_string(optarg
);
1138 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1139 "Failed to parse --volatile= argument: %s", optarg
);
1141 arg_volatile_mode
= m
;
1144 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
1148 r
= expose_port_parse(&arg_expose_ports
, optarg
);
1150 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
1152 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
1154 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
1158 if (strv_extend(&arg_property
, optarg
) < 0)
1163 case ARG_PRIVATE_USERS
: {
1168 else if (!in_charset(optarg
, DIGITS
))
1169 /* do *not* parse numbers as booleans */
1170 boolean
= parse_boolean(optarg
);
1172 if (boolean
== false) {
1173 /* no: User namespacing off */
1174 arg_userns_mode
= USER_NAMESPACE_NO
;
1175 arg_uid_shift
= UID_INVALID
;
1176 arg_uid_range
= UINT32_C(0x10000);
1177 } else if (boolean
== true) {
1178 /* yes: User namespacing on, UID range is read from root dir */
1179 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1180 arg_uid_shift
= UID_INVALID
;
1181 arg_uid_range
= UINT32_C(0x10000);
1182 } else if (streq(optarg
, "pick")) {
1183 /* pick: User namespacing on, UID range is picked randomly */
1184 arg_userns_mode
= USER_NAMESPACE_PICK
;
1185 arg_uid_shift
= UID_INVALID
;
1186 arg_uid_range
= UINT32_C(0x10000);
1188 _cleanup_free_
char *buffer
= NULL
;
1189 const char *range
, *shift
;
1191 /* anything else: User namespacing on, UID range is explicitly configured */
1193 range
= strchr(optarg
, ':');
1195 buffer
= strndup(optarg
, range
- optarg
);
1201 r
= safe_atou32(range
, &arg_uid_range
);
1203 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
1207 r
= parse_uid(shift
, &arg_uid_shift
);
1209 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
1211 arg_userns_mode
= USER_NAMESPACE_FIXED
;
1214 if (arg_uid_range
<= 0)
1215 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1216 "UID range cannot be 0.");
1218 arg_settings_mask
|= SETTING_USERNS
;
1223 if (userns_supported()) {
1224 arg_userns_mode
= USER_NAMESPACE_PICK
;
1225 arg_uid_shift
= UID_INVALID
;
1226 arg_uid_range
= UINT32_C(0x10000);
1228 arg_settings_mask
|= SETTING_USERNS
;
1233 case ARG_PRIVATE_USERS_CHOWN
:
1234 arg_userns_chown
= true;
1236 arg_settings_mask
|= SETTING_USERNS
;
1239 case ARG_KILL_SIGNAL
:
1240 if (streq(optarg
, "help")) {
1241 DUMP_STRING_TABLE(signal
, int, _NSIG
);
1245 arg_kill_signal
= signal_from_string(optarg
);
1246 if (arg_kill_signal
< 0)
1247 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1248 "Cannot parse signal: %s", optarg
);
1250 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
1255 /* no → do not read files
1256 * yes → read files, do not override cmdline, trust only subset
1257 * override → read files, override cmdline, trust only subset
1258 * trusted → read files, do not override cmdline, trust all
1261 r
= parse_boolean(optarg
);
1263 if (streq(optarg
, "trusted")) {
1264 mask_all_settings
= false;
1265 mask_no_settings
= false;
1266 arg_settings_trusted
= true;
1268 } else if (streq(optarg
, "override")) {
1269 mask_all_settings
= false;
1270 mask_no_settings
= true;
1271 arg_settings_trusted
= -1;
1273 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1276 mask_all_settings
= false;
1277 mask_no_settings
= false;
1278 arg_settings_trusted
= -1;
1281 mask_all_settings
= true;
1282 mask_no_settings
= false;
1283 arg_settings_trusted
= false;
1289 if (!path_is_absolute(optarg
))
1290 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1291 "Working directory %s is not an absolute path.", optarg
);
1293 r
= free_and_strdup(&arg_chdir
, optarg
);
1297 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1300 case ARG_PIVOT_ROOT
:
1301 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1303 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1305 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1308 case ARG_NOTIFY_READY
:
1309 r
= parse_boolean(optarg
);
1311 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1312 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1313 arg_notify_ready
= r
;
1314 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1317 case ARG_ROOT_HASH
: {
1321 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1323 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1324 if (l
< sizeof(sd_id128_t
)) {
1326 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Root hash must be at least 128bit long: %s", optarg
);
1329 free(arg_root_hash
);
1331 arg_root_hash_size
= l
;
1335 case ARG_VERITY_DATA
:
1336 r
= parse_path_argument_and_warn(optarg
, false, &arg_verity_data
);
1341 case ARG_ROOT_HASH_SIG
: {
1344 if ((value
= startswith(optarg
, "base64:"))) {
1348 r
= unbase64mem(value
, strlen(value
), &p
, &l
);
1350 return log_error_errno(r
, "Failed to parse root hash signature '%s': %m", optarg
);
1352 free_and_replace(arg_root_hash_sig
, p
);
1353 arg_root_hash_sig_size
= l
;
1354 arg_root_hash_sig_path
= mfree(arg_root_hash_sig_path
);
1356 r
= parse_path_argument_and_warn(optarg
, false, &arg_root_hash_sig_path
);
1359 arg_root_hash_sig
= mfree(arg_root_hash_sig
);
1360 arg_root_hash_sig_size
= 0;
1366 case ARG_SYSTEM_CALL_FILTER
: {
1370 negative
= optarg
[0] == '~';
1371 items
= negative
? optarg
+ 1 : optarg
;
1374 _cleanup_free_
char *word
= NULL
;
1376 r
= extract_first_word(&items
, &word
, NULL
, 0);
1382 return log_error_errno(r
, "Failed to parse system call filter: %m");
1385 r
= strv_extend(&arg_syscall_deny_list
, word
);
1387 r
= strv_extend(&arg_syscall_allow_list
, word
);
1392 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1398 _cleanup_free_
char *name
= NULL
;
1401 if (streq(optarg
, "help")) {
1402 DUMP_STRING_TABLE(rlimit
, int, _RLIMIT_MAX
);
1406 eq
= strchr(optarg
, '=');
1408 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1409 "--rlimit= expects an '=' assignment.");
1411 name
= strndup(optarg
, eq
- optarg
);
1415 rl
= rlimit_from_string_harder(name
);
1417 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1418 "Unknown resource limit: %s", name
);
1420 if (!arg_rlimit
[rl
]) {
1421 arg_rlimit
[rl
] = new0(struct rlimit
, 1);
1422 if (!arg_rlimit
[rl
])
1426 r
= rlimit_parse(rl
, eq
+ 1, arg_rlimit
[rl
]);
1428 return log_error_errno(r
, "Failed to parse resource limit: %s", eq
+ 1);
1430 arg_settings_mask
|= SETTING_RLIMIT_FIRST
<< rl
;
1434 case ARG_OOM_SCORE_ADJUST
:
1435 r
= parse_oom_score_adjust(optarg
, &arg_oom_score_adjust
);
1437 return log_error_errno(r
, "Failed to parse --oom-score-adjust= parameter: %s", optarg
);
1439 arg_oom_score_adjust_set
= true;
1440 arg_settings_mask
|= SETTING_OOM_SCORE_ADJUST
;
1443 case ARG_CPU_AFFINITY
: {
1446 r
= parse_cpu_set(optarg
, &cpuset
);
1448 return log_error_errno(r
, "Failed to parse CPU affinity mask %s: %m", optarg
);
1450 cpu_set_reset(&arg_cpu_set
);
1451 arg_cpu_set
= cpuset
;
1452 arg_settings_mask
|= SETTING_CPU_AFFINITY
;
1456 case ARG_RESOLV_CONF
:
1457 if (streq(optarg
, "help")) {
1458 DUMP_STRING_TABLE(resolv_conf_mode
, ResolvConfMode
, _RESOLV_CONF_MODE_MAX
);
1462 arg_resolv_conf
= resolv_conf_mode_from_string(optarg
);
1463 if (arg_resolv_conf
< 0)
1464 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1465 "Failed to parse /etc/resolv.conf mode: %s", optarg
);
1467 arg_settings_mask
|= SETTING_RESOLV_CONF
;
1471 if (streq(optarg
, "help")) {
1472 DUMP_STRING_TABLE(timezone_mode
, TimezoneMode
, _TIMEZONE_MODE_MAX
);
1476 arg_timezone
= timezone_mode_from_string(optarg
);
1477 if (arg_timezone
< 0)
1478 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1479 "Failed to parse /etc/localtime mode: %s", optarg
);
1481 arg_settings_mask
|= SETTING_TIMEZONE
;
1485 r
= handle_arg_console(optarg
);
1492 r
= handle_arg_console("pipe");
1498 arg_pager_flags
|= PAGER_DISABLE
;
1505 assert_not_reached("Unhandled option");
1508 if (argc
> optind
) {
1509 strv_free(arg_parameters
);
1510 arg_parameters
= strv_copy(argv
+ optind
);
1511 if (!arg_parameters
)
1514 arg_settings_mask
|= SETTING_START_MODE
;
1517 if (arg_ephemeral
&& arg_template
&& !arg_directory
)
1518 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1519 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1520 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1522 arg_directory
= TAKE_PTR(arg_template
);
1524 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? UINT64_C(1) << CAP_NET_ADMIN
: 0)) & ~minus
;
1526 /* Make sure to parse environment before we reset the settings mask below */
1527 r
= parse_environment();
1531 /* Load all settings from .nspawn files */
1532 if (mask_no_settings
)
1533 arg_settings_mask
= 0;
1535 /* Don't load any settings from .nspawn files */
1536 if (mask_all_settings
)
1537 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1542 static int verify_arguments(void) {
1545 if (arg_start_mode
== START_PID2
&& arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
1546 /* If we are running the stub init in the container, we don't need to look at what the init
1547 * in the container supports, because we are not using it. Let's immediately pick the right
1548 * setting based on the host system configuration.
1550 * We only do this, if the user didn't use an environment variable to override the detection.
1553 r
= cg_all_unified();
1555 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
1557 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
1558 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0)
1559 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
1561 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
1564 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1565 arg_mount_settings
|= MOUNT_USE_USERNS
;
1567 if (arg_private_network
)
1568 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1570 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1571 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1572 arg_register
= false;
1573 if (arg_start_mode
!= START_PID1
)
1574 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--boot cannot be used without namespacing.");
1577 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1578 arg_userns_chown
= true;
1580 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1581 arg_kill_signal
= SIGRTMIN
+3;
1583 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 */
1584 arg_read_only
= true;
1586 if (has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
))
1587 arg_read_only
= true;
1589 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0)
1590 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1591 * The latter is not technically a user session, but we don't need to labour the point. */
1592 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--keep-unit --register=yes may not be used when invoked from a user session.");
1594 if (arg_directory
&& arg_image
)
1595 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--directory= and --image= may not be combined.");
1597 if (arg_template
&& arg_image
)
1598 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= and --image= may not be combined.");
1600 if (arg_template
&& !(arg_directory
|| arg_machine
))
1601 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--template= needs --directory= or --machine=.");
1603 if (arg_ephemeral
&& arg_template
)
1604 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --template= may not be combined.");
1606 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
))
1607 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--ephemeral and --link-journal= may not be combined.");
1609 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported())
1610 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--private-users= is not supported, kernel compiled without user namespace support.");
1612 if (arg_userns_chown
&& arg_read_only
)
1613 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1614 "--read-only and --private-users-chown may not be combined.");
1616 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1617 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1618 * copy-up (in case of overlay) making the entire exercise pointless. */
1619 if (arg_userns_chown
&& arg_volatile_mode
!= VOLATILE_NO
)
1620 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--volatile= and --private-users-chown may not be combined.");
1622 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1623 * we need to error out, to avoid conflicts between different network options. */
1624 if (arg_network_namespace_path
&&
1625 (arg_network_interfaces
|| arg_network_macvlan
||
1626 arg_network_ipvlan
|| arg_network_veth_extra
||
1627 arg_network_bridge
|| arg_network_zone
||
1629 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "--network-namespace-path= cannot be combined with other network options.");
1631 if (arg_network_bridge
&& arg_network_zone
)
1632 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1633 "--network-bridge= and --network-zone= may not be combined.");
1635 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
)
1636 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1638 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
))
1639 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot combine --private-users with read-write mounts.");
1641 if (arg_expose_ports
&& !arg_private_network
)
1642 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot use --port= without private networking.");
1645 if (arg_expose_ports
)
1646 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "--port= is not supported, compiled without libiptc support.");
1649 r
= custom_mount_check_all();
1656 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1659 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1662 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1665 if (uid
!= UID_INVALID
) {
1666 uid
+= arg_uid_shift
;
1668 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1672 if (gid
!= GID_INVALID
) {
1673 gid
+= (gid_t
) arg_uid_shift
;
1675 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1679 if (lchown(p
, uid
, gid
) < 0)
1685 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1689 q
= prefix_roota(root
, path
);
1690 r
= mkdir_errno_wrapper(q
, mode
);
1696 return userns_lchown(q
, uid
, gid
);
1699 static const char *timezone_from_path(const char *path
) {
1700 return PATH_STARTSWITH_SET(
1702 "../usr/share/zoneinfo/",
1703 "/usr/share/zoneinfo/");
1706 static bool etc_writable(void) {
1707 return !arg_read_only
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_OVERLAY
);
1710 static int setup_timezone(const char *dest
) {
1711 _cleanup_free_
char *p
= NULL
, *etc
= NULL
;
1712 const char *where
, *check
;
1718 if (IN_SET(arg_timezone
, TIMEZONE_AUTO
, TIMEZONE_SYMLINK
)) {
1719 r
= readlink_malloc("/etc/localtime", &p
);
1720 if (r
== -ENOENT
&& arg_timezone
== TIMEZONE_AUTO
)
1721 m
= etc_writable() ? TIMEZONE_DELETE
: TIMEZONE_OFF
;
1722 else if (r
== -EINVAL
&& arg_timezone
== TIMEZONE_AUTO
) /* regular file? */
1723 m
= etc_writable() ? TIMEZONE_COPY
: TIMEZONE_BIND
;
1725 log_warning_errno(r
, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1726 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1730 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1733 } else if (arg_timezone
== TIMEZONE_AUTO
)
1734 m
= etc_writable() ? TIMEZONE_SYMLINK
: TIMEZONE_BIND
;
1740 if (m
== TIMEZONE_OFF
)
1743 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1745 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1749 where
= strjoina(etc
, "/localtime");
1753 case TIMEZONE_DELETE
:
1754 if (unlink(where
) < 0)
1755 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1759 case TIMEZONE_SYMLINK
: {
1760 _cleanup_free_
char *q
= NULL
;
1761 const char *z
, *what
;
1763 z
= timezone_from_path(p
);
1765 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1769 r
= readlink_malloc(where
, &q
);
1770 if (r
>= 0 && streq_ptr(timezone_from_path(q
), z
))
1771 return 0; /* Already pointing to the right place? Then do nothing .. */
1773 check
= strjoina(dest
, "/usr/share/zoneinfo/", z
);
1774 r
= chase_symlinks(check
, dest
, 0, NULL
, NULL
);
1776 log_debug_errno(r
, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z
);
1778 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1779 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1780 errno
, "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1784 what
= strjoina("../usr/share/zoneinfo/", z
);
1785 if (symlink(what
, where
) < 0) {
1786 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1787 errno
, "Failed to correct timezone of container, ignoring: %m");
1797 case TIMEZONE_BIND
: {
1798 _cleanup_free_
char *resolved
= NULL
;
1801 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1803 log_warning_errno(found
, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1807 if (found
== 0) /* missing? */
1808 (void) touch(resolved
);
1810 r
= mount_verbose(LOG_WARNING
, "/etc/localtime", resolved
, NULL
, MS_BIND
, NULL
);
1812 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1818 /* If mounting failed, try to copy */
1819 r
= copy_file_atomic("/etc/localtime", where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1821 log_full_errno(IN_SET(r
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1822 "Failed to copy /etc/localtime to %s, ignoring: %m", where
);
1829 assert_not_reached("unexpected mode");
1832 /* Fix permissions of the symlink or file copy we just created */
1833 r
= userns_lchown(where
, 0, 0);
1835 log_warning_errno(r
, "Failed to chown /etc/localtime, ignoring: %m");
1840 static int have_resolv_conf(const char *path
) {
1843 if (access(path
, F_OK
) < 0) {
1844 if (errno
== ENOENT
)
1847 return log_debug_errno(errno
, "Failed to determine whether '%s' is available: %m", path
);
1853 static int resolved_listening(void) {
1854 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1855 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1856 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1859 /* Check if resolved is listening */
1861 r
= sd_bus_open_system(&bus
);
1863 return log_debug_errno(r
, "Failed to open system bus: %m");
1865 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1867 return log_debug_errno(r
, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1871 r
= sd_bus_get_property_string(bus
,
1872 "org.freedesktop.resolve1",
1873 "/org/freedesktop/resolve1",
1874 "org.freedesktop.resolve1.Manager",
1877 &dns_stub_listener_mode
);
1879 return log_debug_errno(r
, "Failed to query DNSStubListener property: %s", bus_error_message(&error
, r
));
1881 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1884 static int setup_resolv_conf(const char *dest
) {
1885 _cleanup_free_
char *etc
= NULL
;
1886 const char *where
, *what
;
1892 if (arg_resolv_conf
== RESOLV_CONF_AUTO
) {
1893 if (arg_private_network
)
1894 m
= RESOLV_CONF_OFF
;
1895 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF
) > 0 && resolved_listening() > 0)
1896 m
= etc_writable() ? RESOLV_CONF_COPY_STUB
: RESOLV_CONF_BIND_STUB
;
1897 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1898 m
= etc_writable() ? RESOLV_CONF_COPY_HOST
: RESOLV_CONF_BIND_HOST
;
1900 m
= etc_writable() ? RESOLV_CONF_DELETE
: RESOLV_CONF_OFF
;
1903 m
= arg_resolv_conf
;
1905 if (m
== RESOLV_CONF_OFF
)
1908 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
, NULL
);
1910 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1914 where
= strjoina(etc
, "/resolv.conf");
1916 if (m
== RESOLV_CONF_DELETE
) {
1917 if (unlink(where
) < 0)
1918 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
, "Failed to remove '%s', ignoring: %m", where
);
1923 if (IN_SET(m
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_COPY_STATIC
))
1924 what
= PRIVATE_STATIC_RESOLV_CONF
;
1925 else if (IN_SET(m
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_COPY_UPLINK
))
1926 what
= PRIVATE_UPLINK_RESOLV_CONF
;
1927 else if (IN_SET(m
, RESOLV_CONF_BIND_STUB
, RESOLV_CONF_REPLACE_STUB
, RESOLV_CONF_COPY_STUB
))
1928 what
= PRIVATE_STUB_RESOLV_CONF
;
1930 what
= "/etc/resolv.conf";
1932 if (IN_SET(m
, RESOLV_CONF_BIND_HOST
, RESOLV_CONF_BIND_STATIC
, RESOLV_CONF_BIND_UPLINK
, RESOLV_CONF_BIND_STUB
)) {
1933 _cleanup_free_
char *resolved
= NULL
;
1936 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
, NULL
);
1938 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1942 if (found
== 0) /* missing? */
1943 (void) touch(resolved
);
1945 r
= mount_verbose(LOG_WARNING
, what
, resolved
, NULL
, MS_BIND
, NULL
);
1947 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1949 /* If that didn't work, let's copy the file */
1952 if (IN_SET(m
, RESOLV_CONF_REPLACE_HOST
, RESOLV_CONF_REPLACE_STATIC
, RESOLV_CONF_REPLACE_UPLINK
, RESOLV_CONF_REPLACE_STUB
))
1953 r
= copy_file_atomic(what
, where
, 0644, 0, 0, COPY_REFLINK
|COPY_REPLACE
);
1955 r
= copy_file(what
, where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, 0, COPY_REFLINK
);
1957 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1958 * resolved or something similar runs inside and the symlink points there.
1960 * If the disk image is read-only, there's also no point in complaining.
1962 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST
, RESOLV_CONF_COPY_STATIC
, RESOLV_CONF_COPY_UPLINK
, RESOLV_CONF_COPY_STUB
) &&
1963 IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1964 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1968 r
= userns_lchown(where
, 0, 0);
1970 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1975 static int setup_boot_id(void) {
1976 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1977 _cleanup_free_
char *path
= NULL
;
1978 sd_id128_t rnd
= SD_ID128_NULL
;
1982 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1984 r
= tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path
);
1986 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1988 r
= sd_id128_randomize(&rnd
);
1990 return log_error_errno(r
, "Failed to generate random boot id: %m");
1992 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1994 return log_error_errno(r
, "Failed to write boot id: %m");
1996 from
= TAKE_PTR(path
);
1997 to
= "/proc/sys/kernel/random/boot_id";
1999 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
2003 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
2006 static int copy_devnodes(const char *dest
) {
2007 static const char devnodes
[] =
2016 _cleanup_umask_ mode_t u
;
2024 /* Create /dev/net, so that we can create /dev/net/tun in it */
2025 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
2026 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
2028 NULSTR_FOREACH(d
, devnodes
) {
2029 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
2032 from
= path_join("/dev/", d
);
2036 to
= path_join(dest
, from
);
2040 if (stat(from
, &st
) < 0) {
2042 if (errno
!= ENOENT
)
2043 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
2045 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
2046 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2047 "%s is not a char or block device, cannot copy.", from
);
2049 _cleanup_free_
char *sl
= NULL
, *prefixed
= NULL
, *dn
= NULL
, *t
= NULL
;
2051 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
2052 /* Explicitly warn the user when /dev is already populated. */
2053 if (errno
== EEXIST
)
2054 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
);
2056 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
2058 /* Some systems abusively restrict mknod but allow bind mounts. */
2061 return log_error_errno(r
, "touch (%s) failed: %m", to
);
2062 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
2064 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
2067 r
= userns_lchown(to
, 0, 0);
2069 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
2071 dn
= path_join("/dev", S_ISCHR(st
.st_mode
) ? "char" : "block");
2075 r
= userns_mkdir(dest
, dn
, 0755, 0, 0);
2077 return log_error_errno(r
, "Failed to create '%s': %m", dn
);
2079 if (asprintf(&sl
, "%s/%u:%u", dn
, major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
2082 prefixed
= path_join(dest
, sl
);
2086 t
= path_join("..", d
);
2090 if (symlink(t
, prefixed
) < 0)
2091 log_debug_errno(errno
, "Failed to symlink '%s' to '%s': %m", t
, prefixed
);
2098 static int make_extra_nodes(const char *dest
) {
2099 _cleanup_umask_ mode_t u
;
2105 for (i
= 0; i
< arg_n_extra_nodes
; i
++) {
2106 _cleanup_free_
char *path
= NULL
;
2107 DeviceNode
*n
= arg_extra_nodes
+ i
;
2109 path
= path_join(dest
, n
->path
);
2113 if (mknod(path
, n
->mode
, S_ISCHR(n
->mode
) || S_ISBLK(n
->mode
) ? makedev(n
->major
, n
->minor
) : 0) < 0)
2114 return log_error_errno(errno
, "Failed to create device node '%s': %m", path
);
2116 r
= chmod_and_chown(path
, n
->mode
, n
->uid
, n
->gid
);
2118 return log_error_errno(r
, "Failed to adjust device node ownership of '%s': %m", path
);
2124 static int setup_pts(const char *dest
) {
2125 _cleanup_free_
char *options
= NULL
;
2130 if (arg_selinux_apifs_context
)
2131 (void) asprintf(&options
,
2132 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
2133 arg_uid_shift
+ TTY_GID
,
2134 arg_selinux_apifs_context
);
2137 (void) asprintf(&options
,
2138 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
2139 arg_uid_shift
+ TTY_GID
);
2144 /* Mount /dev/pts itself */
2145 p
= prefix_roota(dest
, "/dev/pts");
2146 r
= mkdir_errno_wrapper(p
, 0755);
2148 return log_error_errno(r
, "Failed to create /dev/pts: %m");
2150 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
2153 r
= userns_lchown(p
, 0, 0);
2155 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
2157 /* Create /dev/ptmx symlink */
2158 p
= prefix_roota(dest
, "/dev/ptmx");
2159 if (symlink("pts/ptmx", p
) < 0)
2160 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
2161 r
= userns_lchown(p
, 0, 0);
2163 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
2165 /* And fix /dev/pts/ptmx ownership */
2166 p
= prefix_roota(dest
, "/dev/pts/ptmx");
2167 r
= userns_lchown(p
, 0, 0);
2169 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
2174 static int setup_stdio_as_dev_console(void) {
2178 terminal
= open_terminal("/dev/console", O_RDWR
);
2180 return log_error_errno(terminal
, "Failed to open console: %m");
2182 /* Make sure we can continue logging to the original stderr, even if
2183 * stderr points elsewhere now */
2184 r
= log_dup_console();
2186 return log_error_errno(r
, "Failed to duplicate stderr: %m");
2188 /* invalidates 'terminal' on success and failure */
2189 r
= rearrange_stdio(terminal
, terminal
, terminal
);
2191 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2196 static int setup_dev_console(const char *console
) {
2197 _cleanup_free_
char *p
= NULL
;
2200 /* Create /dev/console symlink */
2201 r
= path_make_relative("/dev", console
, &p
);
2203 return log_error_errno(r
, "Failed to create relative path: %m");
2205 if (symlink(p
, "/dev/console") < 0)
2206 return log_error_errno(errno
, "Failed to create /dev/console symlink: %m");
2211 static int setup_keyring(void) {
2212 key_serial_t keyring
;
2214 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2215 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2216 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2217 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2218 * into the container. */
2220 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2221 if (keyring
== -1) {
2222 if (errno
== ENOSYS
)
2223 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2224 else if (IN_SET(errno
, EACCES
, EPERM
))
2225 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2227 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2233 static int setup_kmsg(int kmsg_socket
) {
2234 _cleanup_(unlink_and_freep
) char *from
= NULL
;
2235 _cleanup_free_
char *fifo
= NULL
;
2236 _cleanup_close_
int fd
= -1;
2237 _cleanup_umask_ mode_t u
;
2240 assert(kmsg_socket
>= 0);
2244 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2245 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2246 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2247 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2249 r
= tempfn_random_child("/run", "proc-kmsg", &fifo
);
2251 return log_error_errno(r
, "Failed to generate kmsg path: %m");
2253 if (mkfifo(fifo
, 0600) < 0)
2254 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
2256 from
= TAKE_PTR(fifo
);
2258 r
= mount_verbose(LOG_ERR
, from
, "/proc/kmsg", NULL
, MS_BIND
, NULL
);
2262 fd
= open(from
, O_RDWR
|O_NONBLOCK
|O_CLOEXEC
);
2264 return log_error_errno(errno
, "Failed to open fifo: %m");
2266 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2267 r
= send_one_fd(kmsg_socket
, fd
, 0);
2269 return log_error_errno(r
, "Failed to send FIFO fd: %m");
2274 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
2275 union in_addr_union
*exposed
= userdata
;
2281 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
2285 static int setup_hostname(void) {
2288 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
2291 r
= sethostname_idempotent(arg_hostname
?: arg_machine
);
2293 return log_error_errno(r
, "Failed to set hostname: %m");
2298 static int setup_journal(const char *directory
) {
2299 _cleanup_free_
char *d
= NULL
;
2300 char id
[SD_ID128_STRING_MAX
];
2301 const char *dirname
, *p
, *q
;
2306 /* Don't link journals in ephemeral mode */
2310 if (arg_link_journal
== LINK_NO
)
2313 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
2315 r
= sd_id128_get_machine(&this_id
);
2317 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
2319 if (sd_id128_equal(arg_uuid
, this_id
)) {
2320 log_full(try ? LOG_WARNING
: LOG_ERR
,
2321 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
2327 FOREACH_STRING(dirname
, "/var", "/var/log", "/var/log/journal") {
2328 r
= userns_mkdir(directory
, dirname
, 0755, 0, 0);
2330 bool ignore
= r
== -EROFS
&& try;
2331 log_full_errno(ignore
? LOG_DEBUG
: LOG_ERR
, r
,
2332 "Failed to create %s%s: %m", dirname
, ignore
? ", ignoring" : "");
2333 return ignore
? 0 : r
;
2337 (void) sd_id128_to_string(arg_uuid
, id
);
2339 p
= strjoina("/var/log/journal/", id
);
2340 q
= prefix_roota(directory
, p
);
2342 if (path_is_mount_point(p
, NULL
, 0) > 0) {
2346 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2347 "%s: already a mount point, refusing to use for journal", p
);
2350 if (path_is_mount_point(q
, NULL
, 0) > 0) {
2354 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
),
2355 "%s: already a mount point, refusing to use for journal", q
);
2358 r
= readlink_and_make_absolute(p
, &d
);
2360 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
2363 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2365 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2370 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
2371 } else if (r
== -EINVAL
) {
2373 if (arg_link_journal
== LINK_GUEST
&&
2376 if (errno
== ENOTDIR
) {
2377 log_error("%s already exists and is neither a symlink nor a directory", p
);
2380 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
2382 } else if (r
!= -ENOENT
)
2383 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
2385 if (arg_link_journal
== LINK_GUEST
) {
2387 if (symlink(q
, p
) < 0) {
2389 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
2392 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
2395 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2397 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
2401 if (arg_link_journal
== LINK_HOST
) {
2402 /* don't create parents here — if the host doesn't have
2403 * permanent journal set up, don't force it here */
2405 r
= mkdir_errno_wrapper(p
, 0755);
2406 if (r
< 0 && r
!= -EEXIST
) {
2408 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
2411 return log_error_errno(r
, "Failed to create %s: %m", p
);
2414 } else if (access(p
, F_OK
) < 0)
2417 if (dir_is_empty(q
) == 0)
2418 log_warning("%s is not empty, proceeding anyway.", q
);
2420 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
2422 return log_error_errno(r
, "Failed to create %s: %m", q
);
2424 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
2426 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
2431 static int drop_capabilities(uid_t uid
) {
2432 CapabilityQuintet q
;
2434 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2435 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2436 * arg_caps_retain. */
2438 if (capability_quintet_is_set(&arg_full_capabilities
)) {
2439 q
= arg_full_capabilities
;
2441 if (q
.bounding
== (uint64_t) -1)
2442 q
.bounding
= uid
== 0 ? arg_caps_retain
: 0;
2444 if (q
.effective
== (uint64_t) -1)
2445 q
.effective
= uid
== 0 ? q
.bounding
: 0;
2447 if (q
.inheritable
== (uint64_t) -1)
2448 q
.inheritable
= uid
== 0 ? q
.bounding
: 0;
2450 if (q
.permitted
== (uint64_t) -1)
2451 q
.permitted
= uid
== 0 ? q
.bounding
: 0;
2453 if (q
.ambient
== (uint64_t) -1 && ambient_capabilities_supported())
2456 if (capability_quintet_mangle(&q
))
2457 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Cannot set capabilities that are not in the current bounding set.");
2460 q
= (CapabilityQuintet
) {
2461 .bounding
= arg_caps_retain
,
2462 .effective
= uid
== 0 ? arg_caps_retain
: 0,
2463 .inheritable
= uid
== 0 ? arg_caps_retain
: 0,
2464 .permitted
= uid
== 0 ? arg_caps_retain
: 0,
2465 .ambient
= ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2468 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2469 * in order to maintain the same behavior as systemd < 242. */
2470 if (capability_quintet_mangle(&q
))
2471 log_full(arg_quiet
? LOG_DEBUG
: LOG_WARNING
,
2472 "Some capabilities will not be set because they are not in the current bounding set.");
2476 return capability_quintet_enforce(&q
);
2479 static int reset_audit_loginuid(void) {
2480 _cleanup_free_
char *p
= NULL
;
2483 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
2486 r
= read_one_line_file("/proc/self/loginuid", &p
);
2490 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
2492 /* Already reset? */
2493 if (streq(p
, "4294967295"))
2496 r
= write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER
);
2499 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2500 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2501 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2502 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2503 "using systemd-nspawn. Sleeping for 5s... (%m)");
2511 static int setup_propagate(const char *root
) {
2515 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2516 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2517 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
2518 (void) mkdir_p(p
, 0600);
2520 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
2522 return log_error_errno(r
, "Failed to create /run/systemd: %m");
2524 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
2526 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
2528 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2530 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
2532 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
2533 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
2537 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
2541 /* machined will MS_MOVE into that directory, and that's only
2542 * supported for non-shared mounts. */
2543 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
2546 static int setup_machine_id(const char *directory
) {
2547 const char *etc_machine_id
;
2551 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2552 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2553 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2554 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2555 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2556 * container behaves nicely). */
2558 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
2560 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
2562 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
2563 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
2565 if (sd_id128_is_null(arg_uuid
)) {
2566 r
= sd_id128_randomize(&arg_uuid
);
2568 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
2571 if (sd_id128_is_null(id
))
2572 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2573 "Machine ID in container image is zero, refusing.");
2581 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
2586 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
2589 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
2590 if (r
== -EOPNOTSUPP
)
2591 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.");
2593 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
2595 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
2597 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2599 log_debug("Patched directory tree to match UID/GID range.");
2606 * < 0 : wait_for_terminate() failed to get the state of the
2607 * container, the container was terminated by a signal, or
2608 * failed for an unknown reason. No change is made to the
2609 * container argument.
2610 * > 0 : The program executed in the container terminated with an
2611 * error. The exit code of the program executed in the
2612 * container is returned. The container argument has been set
2613 * to CONTAINER_TERMINATED.
2614 * 0 : The container is being rebooted, has been shut down or exited
2615 * successfully. The container argument has been set to either
2616 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2618 * That is, success is indicated by a return value of zero, and an
2619 * error is indicated by a non-zero value.
2621 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2625 r
= wait_for_terminate(pid
, &status
);
2627 return log_warning_errno(r
, "Failed to wait for container: %m");
2629 switch (status
.si_code
) {
2632 if (status
.si_status
== 0)
2633 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2635 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2637 *container
= CONTAINER_TERMINATED
;
2638 return status
.si_status
;
2641 if (status
.si_status
== SIGINT
) {
2642 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2643 *container
= CONTAINER_TERMINATED
;
2646 } else if (status
.si_status
== SIGHUP
) {
2647 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2648 *container
= CONTAINER_REBOOTED
;
2654 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2655 "Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2658 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
2659 "Container %s failed due to unknown reason.", arg_machine
);
2663 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2666 pid
= PTR_TO_PID(userdata
);
2668 if (kill(pid
, arg_kill_signal
) >= 0) {
2669 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2670 sd_event_source_set_userdata(s
, NULL
);
2675 sd_event_exit(sd_event_source_get_event(s
), 0);
2679 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2685 pid
= PTR_TO_PID(userdata
);
2690 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2691 return log_error_errno(errno
, "Failed to waitid(): %m");
2692 if (si
.si_pid
== 0) /* No pending children. */
2694 if (si
.si_pid
== pid
) {
2695 /* The main process we care for has exited. Return from
2696 * signal handler but leave the zombie. */
2697 sd_event_exit(sd_event_source_get_event(s
), 0);
2701 /* Reap all other children. */
2702 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2708 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2713 pid
= PTR_TO_PID(userdata
);
2715 if (arg_kill_signal
> 0) {
2716 log_info("Container termination requested. Attempting to halt container.");
2717 (void) kill(pid
, arg_kill_signal
);
2719 log_info("Container termination requested. Exiting.");
2720 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2726 static int determine_names(void) {
2729 if (arg_template
&& !arg_directory
&& arg_machine
) {
2731 /* If --template= was specified then we should not
2732 * search for a machine, but instead create a new one
2733 * in /var/lib/machine. */
2735 arg_directory
= path_join("/var/lib/machines", arg_machine
);
2740 if (!arg_image
&& !arg_directory
) {
2742 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2744 r
= image_find(IMAGE_MACHINE
, arg_machine
, &i
);
2746 return log_error_errno(r
, "No image for machine '%s'.", arg_machine
);
2748 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2750 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2751 r
= free_and_strdup(&arg_image
, i
->path
);
2753 r
= free_and_strdup(&arg_directory
, i
->path
);
2758 arg_read_only
= arg_read_only
|| i
->read_only
;
2760 r
= safe_getcwd(&arg_directory
);
2762 return log_error_errno(r
, "Failed to determine current directory: %m");
2765 if (!arg_directory
&& !arg_image
)
2766 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine path, please use -D or -i.");
2770 if (arg_directory
&& path_equal(arg_directory
, "/"))
2771 arg_machine
= gethostname_malloc();
2776 arg_machine
= strdup(basename(arg_image
));
2778 /* Truncate suffix if there is one */
2779 e
= endswith(arg_machine
, ".raw");
2783 arg_machine
= strdup(basename(arg_directory
));
2788 hostname_cleanup(arg_machine
);
2789 if (!machine_name_is_valid(arg_machine
))
2790 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine machine name automatically, please use -M.");
2792 if (arg_ephemeral
) {
2795 /* Add a random suffix when this is an
2796 * ephemeral machine, so that we can run many
2797 * instances at once without manually having
2798 * to specify -M each time. */
2800 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2811 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2820 r
= chase_symlinks(*p
, NULL
, flags
, &chased
, NULL
);
2822 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2824 return free_and_replace(*p
, chased
);
2827 static int determine_uid_shift(const char *directory
) {
2830 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2835 if (arg_uid_shift
== UID_INVALID
) {
2838 r
= stat(directory
, &st
);
2840 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2842 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2844 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000)))
2845 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2846 "UID and GID base of %s don't match.", directory
);
2848 arg_uid_range
= UINT32_C(0x10000);
2851 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
)
2852 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2853 "UID base too high for UID range.");
2858 static unsigned long effective_clone_ns_flags(void) {
2859 unsigned long flags
= arg_clone_ns_flags
;
2861 if (arg_private_network
)
2862 flags
|= CLONE_NEWNET
;
2864 flags
|= CLONE_NEWCGROUP
;
2865 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
2866 flags
|= CLONE_NEWUSER
;
2871 static int patch_sysctl(void) {
2873 /* This table is inspired by runc's sysctl() function */
2874 static const struct {
2877 unsigned long clone_flags
;
2879 { "kernel.hostname", false, CLONE_NEWUTS
},
2880 { "kernel.domainname", false, CLONE_NEWUTS
},
2881 { "kernel.msgmax", false, CLONE_NEWIPC
},
2882 { "kernel.msgmnb", false, CLONE_NEWIPC
},
2883 { "kernel.msgmni", false, CLONE_NEWIPC
},
2884 { "kernel.sem", false, CLONE_NEWIPC
},
2885 { "kernel.shmall", false, CLONE_NEWIPC
},
2886 { "kernel.shmmax", false, CLONE_NEWIPC
},
2887 { "kernel.shmmni", false, CLONE_NEWIPC
},
2888 { "fs.mqueue.", true, CLONE_NEWIPC
},
2889 { "net.", true, CLONE_NEWNET
},
2892 unsigned long flags
;
2896 flags
= effective_clone_ns_flags();
2898 STRV_FOREACH_PAIR(k
, v
, arg_sysctl
) {
2902 for (i
= 0; i
< ELEMENTSOF(safe_sysctl
); i
++) {
2904 if (!FLAGS_SET(flags
, safe_sysctl
[i
].clone_flags
))
2907 if (safe_sysctl
[i
].prefix
)
2908 good
= startswith(*k
, safe_sysctl
[i
].key
);
2910 good
= streq(*k
, safe_sysctl
[i
].key
);
2917 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k
);
2919 r
= sysctl_write(*k
, *v
);
2921 return log_error_errno(r
, "Failed to write sysctl '%s': %m", *k
);
2927 static int inner_child(
2929 const char *directory
,
2933 int master_pty_socket
,
2935 char **os_release_pairs
) {
2937 _cleanup_free_
char *home
= NULL
;
2938 char as_uuid
[ID128_UUID_STRING_MAX
];
2940 const char *envp
[] = {
2941 "PATH=" DEFAULT_PATH_COMPAT
,
2942 NULL
, /* container */
2947 NULL
, /* container_uuid */
2948 NULL
, /* LISTEN_FDS */
2949 NULL
, /* LISTEN_PID */
2950 NULL
, /* NOTIFY_SOCKET */
2953 const char *exec_target
;
2954 _cleanup_strv_free_
char **env_use
= NULL
;
2955 int r
, which_failed
;
2957 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2958 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2959 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2960 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2961 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2962 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2965 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2966 * unshare(). See below. */
2970 assert(kmsg_socket
>= 0);
2972 log_debug("Inner child is initializing.");
2974 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2975 /* Tell the parent, that it now can write the UID map. */
2976 (void) barrier_place(barrier
); /* #1 */
2978 /* Wait until the parent wrote the UID map */
2979 if (!barrier_place_and_sync(barrier
)) /* #2 */
2980 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
2981 "Parent died too early");
2984 r
= reset_uid_gid();
2986 return log_error_errno(r
, "Couldn't become new root: %m");
2989 arg_mount_settings
| MOUNT_IN_USERNS
,
2991 arg_selinux_apifs_context
);
2995 if (!arg_network_namespace_path
&& arg_private_network
) {
2996 r
= unshare(CLONE_NEWNET
);
2998 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
3000 /* Tell the parent that it can setup network interfaces. */
3001 (void) barrier_place(barrier
); /* #3 */
3004 r
= mount_sysfs(NULL
, arg_mount_settings
);
3008 /* Wait until we are cgroup-ified, so that we
3009 * can mount the right cgroup path writable */
3010 if (!barrier_place_and_sync(barrier
)) /* #4 */
3011 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3012 "Parent died too early");
3015 r
= unshare(CLONE_NEWCGROUP
);
3017 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
3020 arg_unified_cgroup_hierarchy
,
3021 arg_userns_mode
!= USER_NAMESPACE_NO
,
3024 arg_selinux_apifs_context
,
3027 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
3031 r
= setup_boot_id();
3035 r
= setup_kmsg(kmsg_socket
);
3038 kmsg_socket
= safe_close(kmsg_socket
);
3043 arg_n_custom_mounts
,
3045 arg_selinux_apifs_context
,
3046 MOUNT_NON_ROOT_ONLY
| MOUNT_IN_USERNS
);
3051 return log_error_errno(errno
, "setsid() failed: %m");
3053 if (arg_private_network
)
3054 (void) loopback_setup();
3056 if (arg_expose_ports
) {
3057 r
= expose_port_send_rtnl(rtnl_socket
);
3060 rtnl_socket
= safe_close(rtnl_socket
);
3063 if (arg_console_mode
!= CONSOLE_PIPE
) {
3064 _cleanup_close_
int master
= -1;
3065 _cleanup_free_
char *console
= NULL
;
3067 /* Allocate a pty and make it available as /dev/console. */
3068 master
= openpt_allocate(O_RDWR
|O_NONBLOCK
, &console
);
3070 return log_error_errno(master
, "Failed to allocate a pty: %m");
3072 r
= setup_dev_console(console
);
3074 return log_error_errno(r
, "Failed to set up /dev/console: %m");
3076 r
= send_one_fd(master_pty_socket
, master
, 0);
3078 return log_error_errno(r
, "Failed to send master fd: %m");
3079 master_pty_socket
= safe_close(master_pty_socket
);
3081 r
= setup_stdio_as_dev_console();
3090 if (arg_oom_score_adjust_set
) {
3091 r
= set_oom_score_adjust(arg_oom_score_adjust
);
3093 return log_error_errno(r
, "Failed to adjust OOM score: %m");
3096 if (arg_cpu_set
.set
)
3097 if (sched_setaffinity(0, arg_cpu_set
.allocated
, arg_cpu_set
.set
) < 0)
3098 return log_error_errno(errno
, "Failed to set CPU affinity: %m");
3100 (void) setup_hostname();
3102 if (arg_personality
!= PERSONALITY_INVALID
) {
3103 r
= safe_personality(arg_personality
);
3105 return log_error_errno(r
, "personality() failed: %m");
3106 } else if (secondary
) {
3107 r
= safe_personality(PER_LINUX32
);
3109 return log_error_errno(r
, "personality() failed: %m");
3112 r
= setrlimit_closest_all((const struct rlimit
*const*) arg_rlimit
, &which_failed
);
3114 return log_error_errno(r
, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3119 if (is_seccomp_available()) {
3121 r
= seccomp_load(arg_seccomp
);
3122 if (ERRNO_IS_SECCOMP_FATAL(r
))
3123 return log_error_errno(r
, "Failed to install seccomp filter: %m");
3125 log_debug_errno(r
, "Failed to install seccomp filter: %m");
3130 r
= setup_seccomp(arg_caps_retain
, arg_syscall_allow_list
, arg_syscall_deny_list
);
3136 if (arg_selinux_context
)
3137 if (setexeccon(arg_selinux_context
) < 0)
3138 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
3141 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3142 * if we need to later on. */
3143 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
3144 return log_error_errno(errno
, "Failed to set PR_SET_KEEPCAPS: %m");
3146 if (uid_is_valid(arg_uid
) || gid_is_valid(arg_gid
))
3147 r
= change_uid_gid_raw(arg_uid
, arg_gid
, arg_supplementary_gids
, arg_n_supplementary_gids
);
3149 r
= change_uid_gid(arg_user
, &home
);
3153 r
= drop_capabilities(getuid());
3155 return log_error_errno(r
, "Dropping capabilities failed: %m");
3157 if (arg_no_new_privileges
)
3158 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0)
3159 return log_error_errno(errno
, "Failed to disable new privileges: %m");
3161 /* LXC sets container=lxc, so follow the scheme here */
3162 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
3164 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
3168 if (home
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3169 if (asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
?: "/root") < 0)
3172 if (arg_user
|| !uid_is_valid(arg_uid
) || arg_uid
== 0)
3173 if (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
?: "root") < 0 ||
3174 asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0)
3177 assert(!sd_id128_is_null(arg_uuid
));
3179 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
3182 if (fdset_size(fds
) > 0) {
3183 r
= fdset_cloexec(fds
, false);
3185 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
3187 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
3188 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
3191 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
3194 env_use
= strv_env_merge(3, envp
, arg_setenv
, os_release_pairs
);
3198 /* Let the parent know that we are ready and
3199 * wait until the parent is ready with the
3201 if (!barrier_place_and_sync(barrier
)) /* #5 */
3202 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
),
3203 "Parent died too early");
3206 if (chdir(arg_chdir
) < 0)
3207 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
3209 if (arg_start_mode
== START_PID2
) {
3210 r
= stub_pid1(arg_uuid
);
3215 log_debug("Inner child completed, invoking payload.");
3217 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3218 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3219 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3221 log_set_open_when_needed(true);
3223 (void) fdset_close_others(fds
);
3225 if (arg_start_mode
== START_BOOT
) {
3229 /* Automatically search for the init system */
3231 m
= strv_length(arg_parameters
);
3232 a
= newa(char*, m
+ 2);
3233 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
3236 a
[0] = (char*) "/usr/lib/systemd/systemd";
3237 execve(a
[0], a
, env_use
);
3239 a
[0] = (char*) "/lib/systemd/systemd";
3240 execve(a
[0], a
, env_use
);
3242 a
[0] = (char*) "/sbin/init";
3243 execve(a
[0], a
, env_use
);
3245 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3246 } else if (!strv_isempty(arg_parameters
)) {
3247 const char *dollar_path
;
3249 exec_target
= arg_parameters
[0];
3251 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3253 dollar_path
= strv_env_get(env_use
, "PATH");
3255 if (putenv((char*) dollar_path
) != 0)
3256 return log_error_errno(errno
, "Failed to update $PATH: %m");
3259 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
3262 /* If we cannot change the directory, we'll end up in /, that is expected. */
3263 (void) chdir(home
?: "/root");
3265 execle("/bin/bash", "-bash", NULL
, env_use
);
3266 execle("/bin/sh", "-sh", NULL
, env_use
);
3268 exec_target
= "/bin/bash, /bin/sh";
3271 return log_error_errno(errno
, "execv(%s) failed: %m", exec_target
);
3274 static int setup_sd_notify_child(void) {
3275 _cleanup_close_
int fd
= -1;
3276 union sockaddr_union sa
= {
3277 .un
.sun_family
= AF_UNIX
,
3278 .un
.sun_path
= NSPAWN_NOTIFY_SOCKET_PATH
,
3282 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3284 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
3286 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
3287 (void) sockaddr_un_unlink(&sa
.un
);
3289 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
3291 return log_error_errno(errno
, "bind(" NSPAWN_NOTIFY_SOCKET_PATH
") failed: %m");
3293 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
3295 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
3297 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
3299 return log_error_errno(r
, "SO_PASSCRED failed: %m");
3304 static int outer_child(
3306 const char *directory
,
3307 DissectedImage
*dissected_image
,
3314 int uid_shift_socket
,
3315 int master_pty_socket
,
3316 int unified_cgroup_hierarchy_socket
,
3320 _cleanup_strv_free_
char **os_release_pairs
= NULL
;
3321 _cleanup_close_
int fd
= -1;
3327 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3328 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3329 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3330 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3334 assert(pid_socket
>= 0);
3335 assert(uuid_socket
>= 0);
3336 assert(notify_socket
>= 0);
3337 assert(master_pty_socket
>= 0);
3338 assert(kmsg_socket
>= 0);
3340 log_debug("Outer child is initializing.");
3342 r
= load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs
);
3344 log_debug_errno(r
, "Failed to read os-release from host for container, ignoring: %m");
3346 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
3347 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
3349 r
= reset_audit_loginuid();
3353 /* Mark everything as slave, so that we still
3354 * receive mounts from the real root, but don't
3355 * propagate mounts to the real root. */
3356 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
3360 if (dissected_image
) {
3361 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3362 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3363 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3364 * makes sure ESP partitions and userns are compatible. */
3366 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3367 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|
3368 (arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
)|
3369 (arg_start_mode
== START_BOOT
? DISSECT_IMAGE_VALIDATE_OS
: 0));
3371 return log_error_errno(r
, "File system check for image failed: %m");
3373 return log_error_errno(r
, "Failed to mount image root file system: %m");
3376 r
= determine_uid_shift(directory
);
3380 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3381 /* Let the parent know which UID shift we read from the image */
3382 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
3384 return log_error_errno(errno
, "Failed to send UID shift: %m");
3385 if (l
!= sizeof(arg_uid_shift
))
3386 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3387 "Short write while sending UID shift.");
3389 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3390 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3391 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3392 * not it will pick a different one, and send it back to us. */
3394 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
3396 return log_error_errno(errno
, "Failed to recv UID shift: %m");
3397 if (l
!= sizeof(arg_uid_shift
))
3398 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3399 "Short read while receiving UID shift.");
3402 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
3403 "Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
3406 if (path_equal(directory
, "/")) {
3407 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3408 * place, so that we can make changes to its mount structure (for example, to implement
3409 * --volatile=) without this interfering with our ability to access files such as
3410 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3411 * (instead of a temporary directory, since we are living in our own mount namspace here
3412 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3413 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3415 r
= mount_verbose(LOG_ERR
, "/", "/run/systemd/nspawn-root", NULL
, MS_BIND
|MS_REC
, NULL
);
3419 directory
= "/run/systemd/nspawn-root";
3422 r
= setup_pivot_root(
3425 arg_pivot_root_old
);
3429 r
= setup_volatile_mode(
3433 arg_selinux_apifs_context
);
3440 arg_n_custom_mounts
,
3442 arg_selinux_apifs_context
,
3447 /* Make sure we always have a mount that we can move to root later on. */
3448 if (!path_is_mount_point(directory
, NULL
, 0)) {
3449 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
3454 if (dissected_image
) {
3455 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3456 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
3457 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: DISSECT_IMAGE_FSCK
));
3459 return log_error_errno(r
, "File system check for image failed: %m");
3461 return log_error_errno(r
, "Failed to mount image file system: %m");
3464 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3465 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3467 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
3471 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
3473 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
3474 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
3475 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3476 "Short write while sending cgroup mode.");
3478 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
3481 /* Mark everything as shared so our mounts get propagated down. This is
3482 * required to make new bind mounts available in systemd services
3483 * inside the container that create a new mount namespace.
3484 * See https://github.com/systemd/systemd/issues/3860
3485 * Further submounts (such as /dev) done after this will inherit the
3486 * shared propagation mode.
3488 * IMPORTANT: Do not overmount the root directory anymore from now on to
3489 * enable moving the root directory mount to root later on.
3490 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3492 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
3496 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
3500 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
3504 if (arg_read_only
&& arg_volatile_mode
== VOLATILE_NO
&&
3505 !has_custom_root_mount(arg_custom_mounts
, arg_n_custom_mounts
)) {
3506 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
3508 return log_error_errno(r
, "Failed to make tree read-only: %m");
3511 r
= mount_all(directory
,
3514 arg_selinux_apifs_context
);
3518 r
= copy_devnodes(directory
);
3522 r
= make_extra_nodes(directory
);
3526 (void) dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
3528 p
= prefix_roota(directory
, "/run");
3529 (void) make_inaccessible_nodes(p
, arg_uid_shift
, arg_uid_shift
);
3531 r
= setup_pts(directory
);
3535 r
= setup_propagate(directory
);
3539 r
= setup_keyring();
3546 arg_n_custom_mounts
,
3548 arg_selinux_apifs_context
,
3549 MOUNT_NON_ROOT_ONLY
);
3553 r
= setup_timezone(directory
);
3557 r
= setup_resolv_conf(directory
);
3561 r
= setup_machine_id(directory
);
3565 r
= setup_journal(directory
);
3569 if (!arg_use_cgns
) {
3572 arg_unified_cgroup_hierarchy
,
3573 arg_userns_mode
!= USER_NAMESPACE_NO
,
3576 arg_selinux_apifs_context
,
3582 r
= mount_move_root(directory
);
3584 return log_error_errno(r
, "Failed to move root directory: %m");
3586 fd
= setup_sd_notify_child();
3590 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
3591 arg_clone_ns_flags
|
3592 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
3594 return log_error_errno(errno
, "Failed to fork inner child: %m");
3596 pid_socket
= safe_close(pid_socket
);
3597 uuid_socket
= safe_close(uuid_socket
);
3598 notify_socket
= safe_close(notify_socket
);
3599 uid_shift_socket
= safe_close(uid_shift_socket
);
3601 /* The inner child has all namespaces that are
3602 * requested, so that we all are owned by the user if
3603 * user namespaces are turned on. */
3605 if (arg_network_namespace_path
) {
3606 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
3608 return log_error_errno(r
, "Failed to join network namespace: %m");
3611 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, master_pty_socket
, fds
, os_release_pairs
);
3613 _exit(EXIT_FAILURE
);
3615 _exit(EXIT_SUCCESS
);
3618 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
3620 return log_error_errno(errno
, "Failed to send PID: %m");
3621 if (l
!= sizeof(pid
))
3622 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3623 "Short write while sending PID.");
3625 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
3627 return log_error_errno(errno
, "Failed to send machine ID: %m");
3628 if (l
!= sizeof(arg_uuid
))
3629 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
3630 "Short write while sending machine ID.");
3632 l
= send_one_fd(notify_socket
, fd
, 0);
3634 return log_error_errno(l
, "Failed to send notify fd: %m");
3636 pid_socket
= safe_close(pid_socket
);
3637 uuid_socket
= safe_close(uuid_socket
);
3638 notify_socket
= safe_close(notify_socket
);
3639 master_pty_socket
= safe_close(master_pty_socket
);
3640 kmsg_socket
= safe_close(kmsg_socket
);
3641 rtnl_socket
= safe_close(rtnl_socket
);
3642 netns_fd
= safe_close(netns_fd
);
3647 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
3648 bool tried_hashed
= false;
3649 unsigned n_tries
= 100;
3654 assert(ret_lock_file
);
3655 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
3656 assert(arg_uid_range
== 0x10000U
);
3660 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3663 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
3664 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
3669 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
3671 if ((candidate
& UINT32_C(0xFFFF)) != 0)
3674 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
3675 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
3676 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
3681 /* Make some superficial checks whether the range is currently known in the user database */
3682 if (getpwuid(candidate
))
3684 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
3686 if (getgrgid(candidate
))
3688 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
3691 *ret_lock_file
= lf
;
3692 lf
= (struct LockFile
) LOCK_FILE_INIT
;
3697 if (arg_machine
&& !tried_hashed
) {
3698 /* Try to hash the base from the container name */
3700 static const uint8_t hash_key
[] = {
3701 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3702 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3705 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
3707 tried_hashed
= true;
3709 random_bytes(&candidate
, sizeof(candidate
));
3711 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
3712 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
3716 static int setup_uid_map(pid_t pid
) {
3717 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
3722 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
3723 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
3724 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3726 return log_error_errno(r
, "Failed to write UID map: %m");
3728 /* We always assign the same UID and GID ranges */
3729 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
3730 r
= write_string_file(uid_map
, line
, WRITE_STRING_FILE_DISABLE_BUFFER
);
3732 return log_error_errno(r
, "Failed to write GID map: %m");
3737 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3738 char buf
[NOTIFY_BUFFER_MAX
+1];
3740 struct iovec iovec
= {
3742 .iov_len
= sizeof(buf
)-1,
3744 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
3745 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
3746 struct msghdr msghdr
= {
3749 .msg_control
= &control
,
3750 .msg_controllen
= sizeof(control
),
3752 struct ucred
*ucred
;
3754 pid_t inner_child_pid
;
3755 _cleanup_strv_free_
char **tags
= NULL
;
3759 inner_child_pid
= PTR_TO_PID(userdata
);
3761 if (revents
!= EPOLLIN
) {
3762 log_warning("Got unexpected poll event for notify fd.");
3766 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
3767 if (IN_SET(n
, -EAGAIN
, -EINTR
))
3770 return log_warning_errno(n
, "Couldn't read notification socket: %m");
3772 cmsg_close_all(&msghdr
);
3774 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
3775 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3776 log_debug("Received notify message without valid credentials. Ignoring.");
3780 if ((size_t) n
>= sizeof(buf
)) {
3781 log_warning("Received notify message exceeded maximum size. Ignoring.");
3786 tags
= strv_split(buf
, "\n\r");
3790 if (strv_find(tags
, "READY=1"))
3791 (void) sd_notifyf(false, "READY=1\n");
3793 p
= strv_find_startswith(tags
, "STATUS=");
3795 (void) sd_notifyf(false, "STATUS=Container running: %s", p
);
3800 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3803 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3805 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3807 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3812 static int merge_settings(Settings
*settings
, const char *path
) {
3818 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3819 * that this steals the fields of the Settings* structure, and hence modifies it. */
3821 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3822 settings
->start_mode
>= 0) {
3823 arg_start_mode
= settings
->start_mode
;
3824 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3827 if ((arg_settings_mask
& SETTING_EPHEMERAL
) == 0)
3828 arg_ephemeral
= settings
->ephemeral
;
3830 if ((arg_settings_mask
& SETTING_DIRECTORY
) == 0 &&
3833 if (!arg_settings_trusted
)
3834 log_warning("Ignoring root directory setting, file %s is not trusted.", path
);
3836 free_and_replace(arg_directory
, settings
->root
);
3839 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3840 settings
->pivot_root_new
) {
3841 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3842 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3845 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3846 settings
->working_directory
)
3847 free_and_replace(arg_chdir
, settings
->working_directory
);
3849 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3850 settings
->environment
)
3851 strv_free_and_replace(arg_setenv
, settings
->environment
);
3853 if ((arg_settings_mask
& SETTING_USER
) == 0) {
3856 free_and_replace(arg_user
, settings
->user
);
3858 if (uid_is_valid(settings
->uid
))
3859 arg_uid
= settings
->uid
;
3860 if (gid_is_valid(settings
->gid
))
3861 arg_gid
= settings
->gid
;
3862 if (settings
->n_supplementary_gids
> 0) {
3863 free_and_replace(arg_supplementary_gids
, settings
->supplementary_gids
);
3864 arg_n_supplementary_gids
= settings
->n_supplementary_gids
;
3868 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3869 uint64_t plus
, minus
;
3870 uint64_t network_minus
= 0;
3872 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3873 * Settings structure */
3875 plus
= settings
->capability
;
3876 minus
= settings
->drop_capability
;
3878 if ((arg_settings_mask
& SETTING_NETWORK
) == 0) {
3879 if (settings_private_network(settings
))
3880 plus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3882 network_minus
|= UINT64_C(1) << CAP_NET_ADMIN
;
3885 if (!arg_settings_trusted
&& plus
!= 0) {
3886 if (settings
->capability
!= 0)
3887 log_warning("Ignoring Capability= setting, file %s is not trusted.", path
);
3889 arg_caps_retain
&= ~network_minus
;
3890 arg_caps_retain
|= plus
;
3893 arg_caps_retain
&= ~minus
;
3895 /* Copy the full capabilities over too */
3896 if (capability_quintet_is_set(&settings
->full_capabilities
)) {
3897 if (!arg_settings_trusted
)
3898 log_warning("Ignoring capability settings, file %s is not trusted.", path
);
3900 arg_full_capabilities
= settings
->full_capabilities
;
3904 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3905 settings
->kill_signal
> 0)
3906 arg_kill_signal
= settings
->kill_signal
;
3908 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3909 settings
->personality
!= PERSONALITY_INVALID
)
3910 arg_personality
= settings
->personality
;
3912 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3913 !sd_id128_is_null(settings
->machine_id
)) {
3915 if (!arg_settings_trusted
)
3916 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path
);
3918 arg_uuid
= settings
->machine_id
;
3921 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3922 settings
->read_only
>= 0)
3923 arg_read_only
= settings
->read_only
;
3925 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3926 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3927 arg_volatile_mode
= settings
->volatile_mode
;
3929 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3930 settings
->n_custom_mounts
> 0) {
3932 if (!arg_settings_trusted
)
3933 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path
);
3935 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3936 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3937 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3938 settings
->n_custom_mounts
= 0;
3942 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3943 (settings
->private_network
>= 0 ||
3944 settings
->network_veth
>= 0 ||
3945 settings
->network_bridge
||
3946 settings
->network_zone
||
3947 settings
->network_interfaces
||
3948 settings
->network_macvlan
||
3949 settings
->network_ipvlan
||
3950 settings
->network_veth_extra
||
3951 settings
->network_namespace_path
)) {
3953 if (!arg_settings_trusted
)
3954 log_warning("Ignoring network settings, file %s is not trusted.", path
);
3956 arg_network_veth
= settings_network_veth(settings
);
3957 arg_private_network
= settings_private_network(settings
);
3959 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3960 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3961 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3962 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3964 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3965 free_and_replace(arg_network_zone
, settings
->network_zone
);
3967 free_and_replace(arg_network_namespace_path
, settings
->network_namespace_path
);
3971 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3972 settings
->expose_ports
) {
3974 if (!arg_settings_trusted
)
3975 log_warning("Ignoring Port= setting, file %s is not trusted.", path
);
3977 expose_port_free_all(arg_expose_ports
);
3978 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3982 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3983 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3985 if (!arg_settings_trusted
)
3986 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path
);
3988 arg_userns_mode
= settings
->userns_mode
;
3989 arg_uid_shift
= settings
->uid_shift
;
3990 arg_uid_range
= settings
->uid_range
;
3991 arg_userns_chown
= settings
->userns_chown
;
3995 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3996 arg_notify_ready
= settings
->notify_ready
;
3998 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
4000 if (!arg_settings_trusted
&& !strv_isempty(settings
->syscall_allow_list
))
4001 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path
);
4003 strv_free_and_replace(arg_syscall_allow_list
, settings
->syscall_allow_list
);
4004 strv_free_and_replace(arg_syscall_deny_list
, settings
->syscall_deny_list
);
4008 if (!arg_settings_trusted
&& settings
->seccomp
)
4009 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path
);
4011 seccomp_release(arg_seccomp
);
4012 arg_seccomp
= TAKE_PTR(settings
->seccomp
);
4017 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4018 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)))
4021 if (!settings
->rlimit
[rl
])
4024 if (!arg_settings_trusted
) {
4025 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl
), path
);
4029 free_and_replace(arg_rlimit
[rl
], settings
->rlimit
[rl
]);
4032 if ((arg_settings_mask
& SETTING_HOSTNAME
) == 0 &&
4034 free_and_replace(arg_hostname
, settings
->hostname
);
4036 if ((arg_settings_mask
& SETTING_NO_NEW_PRIVILEGES
) == 0 &&
4037 settings
->no_new_privileges
>= 0)
4038 arg_no_new_privileges
= settings
->no_new_privileges
;
4040 if ((arg_settings_mask
& SETTING_OOM_SCORE_ADJUST
) == 0 &&
4041 settings
->oom_score_adjust_set
) {
4043 if (!arg_settings_trusted
)
4044 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path
);
4046 arg_oom_score_adjust
= settings
->oom_score_adjust
;
4047 arg_oom_score_adjust_set
= true;
4051 if ((arg_settings_mask
& SETTING_CPU_AFFINITY
) == 0 &&
4052 settings
->cpu_set
.set
) {
4054 if (!arg_settings_trusted
)
4055 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path
);
4057 cpu_set_reset(&arg_cpu_set
);
4058 arg_cpu_set
= settings
->cpu_set
;
4059 settings
->cpu_set
= (CPUSet
) {};
4063 if ((arg_settings_mask
& SETTING_RESOLV_CONF
) == 0 &&
4064 settings
->resolv_conf
!= _RESOLV_CONF_MODE_INVALID
)
4065 arg_resolv_conf
= settings
->resolv_conf
;
4067 if ((arg_settings_mask
& SETTING_LINK_JOURNAL
) == 0 &&
4068 settings
->link_journal
!= _LINK_JOURNAL_INVALID
) {
4070 if (!arg_settings_trusted
)
4071 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path
);
4073 arg_link_journal
= settings
->link_journal
;
4074 arg_link_journal_try
= settings
->link_journal_try
;
4078 if ((arg_settings_mask
& SETTING_TIMEZONE
) == 0 &&
4079 settings
->timezone
!= _TIMEZONE_MODE_INVALID
)
4080 arg_timezone
= settings
->timezone
;
4082 if ((arg_settings_mask
& SETTING_SLICE
) == 0 &&
4085 if (!arg_settings_trusted
)
4086 log_warning("Ignoring slice setting, file '%s' is not trusted.", path
);
4088 free_and_replace(arg_slice
, settings
->slice
);
4091 if ((arg_settings_mask
& SETTING_USE_CGNS
) == 0 &&
4092 settings
->use_cgns
>= 0) {
4094 if (!arg_settings_trusted
)
4095 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path
);
4097 arg_use_cgns
= settings
->use_cgns
;
4100 if ((arg_settings_mask
& SETTING_CLONE_NS_FLAGS
) == 0 &&
4101 settings
->clone_ns_flags
!= (unsigned long) -1) {
4103 if (!arg_settings_trusted
)
4104 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path
);
4106 arg_clone_ns_flags
= settings
->clone_ns_flags
;
4109 if ((arg_settings_mask
& SETTING_CONSOLE_MODE
) == 0 &&
4110 settings
->console_mode
>= 0) {
4112 if (!arg_settings_trusted
)
4113 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path
);
4115 arg_console_mode
= settings
->console_mode
;
4118 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4119 * don't consult arg_settings_mask for them. */
4121 sd_bus_message_unref(arg_property_message
);
4122 arg_property_message
= TAKE_PTR(settings
->properties
);
4124 arg_console_width
= settings
->console_width
;
4125 arg_console_height
= settings
->console_height
;
4127 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
4128 arg_extra_nodes
= TAKE_PTR(settings
->extra_nodes
);
4129 arg_n_extra_nodes
= settings
->n_extra_nodes
;
4134 static int load_settings(void) {
4135 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4136 _cleanup_fclose_
FILE *f
= NULL
;
4137 _cleanup_free_
char *p
= NULL
;
4144 /* If all settings are masked, there's no point in looking for
4145 * the settings file */
4146 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
4149 fn
= strjoina(arg_machine
, ".nspawn");
4151 /* We first look in the admin's directories in /etc and /run */
4152 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4153 _cleanup_free_
char *j
= NULL
;
4155 j
= path_join(i
, fn
);
4163 /* By default, we trust configuration from /etc and /run */
4164 if (arg_settings_trusted
< 0)
4165 arg_settings_trusted
= true;
4170 if (errno
!= ENOENT
)
4171 return log_error_errno(errno
, "Failed to open %s: %m", j
);
4175 /* After that, let's look for a file next to the
4176 * actual image we shall boot. */
4179 p
= file_in_same_dir(arg_image
, fn
);
4182 } else if (arg_directory
&& !path_equal(arg_directory
, "/")) {
4183 p
= file_in_same_dir(arg_directory
, fn
);
4190 if (!f
&& errno
!= ENOENT
)
4191 return log_error_errno(errno
, "Failed to open %s: %m", p
);
4193 /* By default, we do not trust configuration from /var/lib/machines */
4194 if (arg_settings_trusted
< 0)
4195 arg_settings_trusted
= false;
4202 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
4204 r
= settings_load(f
, p
, &settings
);
4208 return merge_settings(settings
, p
);
4211 static int load_oci_bundle(void) {
4212 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
4215 if (!arg_oci_bundle
)
4218 /* By default let's trust OCI bundles */
4219 if (arg_settings_trusted
< 0)
4220 arg_settings_trusted
= true;
4222 r
= oci_load(NULL
, arg_oci_bundle
, &settings
);
4226 return merge_settings(settings
, arg_oci_bundle
);
4229 static int run_container(
4230 DissectedImage
*dissected_image
,
4233 char veth_name
[IFNAMSIZ
], bool *veth_created
,
4234 union in_addr_union
*exposed
,
4235 int *master
, pid_t
*pid
, int *ret
) {
4237 static const struct sigaction sa
= {
4238 .sa_handler
= nop_signal_handler
,
4239 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
4242 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
4243 _cleanup_close_
int etc_passwd_lock
= -1;
4244 _cleanup_close_pair_
int
4245 kmsg_socket_pair
[2] = { -1, -1 },
4246 rtnl_socket_pair
[2] = { -1, -1 },
4247 pid_socket_pair
[2] = { -1, -1 },
4248 uuid_socket_pair
[2] = { -1, -1 },
4249 notify_socket_pair
[2] = { -1, -1 },
4250 uid_shift_socket_pair
[2] = { -1, -1 },
4251 master_pty_socket_pair
[2] = { -1, -1 },
4252 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
4254 _cleanup_close_
int notify_socket
= -1;
4255 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
4256 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
4257 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
4258 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
4259 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
4260 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
4261 ContainerStatus container_status
= 0;
4265 _cleanup_close_
int child_netns_fd
= -1;
4267 assert_se(sigemptyset(&mask_chld
) == 0);
4268 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
4270 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4271 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4272 * check with getpwuid() if the specific user already exists. Note that /etc might be
4273 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4274 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4275 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4278 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
4279 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
4280 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
4283 r
= barrier_create(&barrier
);
4285 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
4287 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
4288 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
4290 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
4291 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
4293 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
4294 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
4296 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
4297 return log_error_errno(errno
, "Failed to create id socket pair: %m");
4299 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
4300 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
4302 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, master_pty_socket_pair
) < 0)
4303 return log_error_errno(errno
, "Failed to create console socket pair: %m");
4305 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
4306 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
4307 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
4309 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
4310 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
4311 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
4313 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4314 * parent's blocking calls and give it a chance to call wait() and terminate. */
4315 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
4317 return log_error_errno(errno
, "Failed to change the signal mask: %m");
4319 r
= sigaction(SIGCHLD
, &sa
, NULL
);
4321 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
4323 if (arg_network_namespace_path
) {
4324 child_netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
4325 if (child_netns_fd
< 0)
4326 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
4328 r
= fd_is_network_ns(child_netns_fd
);
4330 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
);
4332 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
4334 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
4335 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path
);
4338 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
4340 return log_error_errno(errno
, "clone() failed%s: %m",
4342 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4345 /* The outer child only has a file system namespace. */
4346 barrier_set_role(&barrier
, BARRIER_CHILD
);
4348 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
4349 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4350 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
4351 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
4352 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
4353 master_pty_socket_pair
[0] = safe_close(master_pty_socket_pair
[0]);
4354 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
4355 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
4357 (void) reset_all_signal_handlers();
4358 (void) reset_signal_mask();
4360 r
= outer_child(&barrier
,
4365 uuid_socket_pair
[1],
4366 notify_socket_pair
[1],
4367 kmsg_socket_pair
[1],
4368 rtnl_socket_pair
[1],
4369 uid_shift_socket_pair
[1],
4370 master_pty_socket_pair
[1],
4371 unified_cgroup_hierarchy_socket_pair
[1],
4375 _exit(EXIT_FAILURE
);
4377 _exit(EXIT_SUCCESS
);
4380 barrier_set_role(&barrier
, BARRIER_PARENT
);
4384 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
4385 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
4386 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
4387 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
4388 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
4389 master_pty_socket_pair
[1] = safe_close(master_pty_socket_pair
[1]);
4390 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
4391 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
4393 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4394 /* The child just let us know the UID shift it might have read from the image. */
4395 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
4397 return log_error_errno(errno
, "Failed to read UID shift: %m");
4398 if (l
!= sizeof arg_uid_shift
)
4399 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading UID shift.");
4401 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
4402 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4403 * image, but if that's already in use, pick a new one, and report back to the child,
4404 * which one we now picked. */
4406 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
4408 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
4410 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
4412 return log_error_errno(errno
, "Failed to send UID shift: %m");
4413 if (l
!= sizeof arg_uid_shift
)
4414 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while writing UID shift.");
4418 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
4419 /* The child let us know the support cgroup mode it might have read from the image. */
4420 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
4422 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
4423 if (l
!= sizeof(arg_unified_cgroup_hierarchy
))
4424 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading cgroup mode (%zu bytes).%s",
4425 l
, l
== 0 ? " The child is most likely dead." : "");
4428 /* Wait for the outer child. */
4429 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
4432 if (r
!= EXIT_SUCCESS
)
4435 /* And now retrieve the PID of the inner child. */
4436 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
4438 return log_error_errno(errno
, "Failed to read inner child PID: %m");
4439 if (l
!= sizeof *pid
)
4440 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading inner child PID.");
4442 /* We also retrieve container UUID in case it was generated by outer child */
4443 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
4445 return log_error_errno(errno
, "Failed to read container machine ID: %m");
4446 if (l
!= sizeof(arg_uuid
))
4447 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short read while reading container machined ID.");
4449 /* We also retrieve the socket used for notifications generated by outer child */
4450 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
4451 if (notify_socket
< 0)
4452 return log_error_errno(notify_socket
,
4453 "Failed to receive notification socket from the outer child: %m");
4455 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
4457 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
4458 if (!barrier_place_and_sync(&barrier
)) /* #1 */
4459 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4461 r
= setup_uid_map(*pid
);
4465 (void) barrier_place(&barrier
); /* #2 */
4468 if (arg_private_network
) {
4469 if (!arg_network_namespace_path
) {
4470 /* Wait until the child has unshared its network namespace. */
4471 if (!barrier_place_and_sync(&barrier
)) /* #3 */
4472 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early");
4475 if (child_netns_fd
< 0) {
4476 /* Make sure we have an open file descriptor to the child's network
4477 * namespace so it stays alive even if the child exits. */
4478 r
= namespace_open(*pid
, NULL
, NULL
, &child_netns_fd
, NULL
, NULL
);
4480 return log_error_errno(r
, "Failed to open child network namespace: %m");
4483 r
= move_network_interfaces(child_netns_fd
, arg_network_interfaces
);
4487 if (arg_network_veth
) {
4488 r
= setup_veth(arg_machine
, *pid
, veth_name
,
4489 arg_network_bridge
|| arg_network_zone
);
4495 if (arg_network_bridge
) {
4496 /* Add the interface to a bridge */
4497 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
4502 } else if (arg_network_zone
) {
4503 /* Add the interface to a bridge, possibly creating it */
4504 r
= setup_bridge(veth_name
, arg_network_zone
, true);
4512 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
4516 /* We created the primary and extra veth links now; let's remember this, so that we know to
4517 remove them later on. Note that we don't bother with removing veth links that were created
4518 here when their setup failed half-way, because in that case the kernel should be able to
4519 remove them on its own, since they cannot be referenced by anything yet. */
4520 *veth_created
= true;
4522 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
4526 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
4531 if (arg_register
|| !arg_keep_unit
) {
4532 r
= sd_bus_default_system(&bus
);
4534 return log_error_errno(r
, "Failed to open system bus: %m");
4536 r
= sd_bus_set_close_on_exit(bus
, false);
4538 return log_error_errno(r
, "Failed to disable close-on-exit behaviour: %m");
4541 if (!arg_keep_unit
) {
4542 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4543 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4544 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4546 r
= sd_bus_match_signal_async(
4549 "org.freedesktop.systemd1",
4551 "org.freedesktop.systemd1.Scope",
4553 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
4555 return log_error_errno(r
, "Failed to request RequestStop match: %m");
4559 r
= register_machine(
4567 arg_custom_mounts
, arg_n_custom_mounts
,
4570 arg_property_message
,
4572 arg_container_service_name
);
4576 } else if (!arg_keep_unit
) {
4582 arg_custom_mounts
, arg_n_custom_mounts
,
4585 arg_property_message
);
4589 } else if (arg_slice
|| arg_property
)
4590 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4592 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
4596 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4600 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
4604 /* Notify the child that the parent is ready with all
4605 * its setup (including cgroup-ification), and that
4606 * the child can now hand over control to the code to
4607 * run inside the container. */
4608 (void) barrier_place(&barrier
); /* #4 */
4610 /* Block SIGCHLD here, before notifying child.
4611 * process_pty() will handle it with the other signals. */
4612 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
4614 /* Reset signal to default */
4615 r
= default_signals(SIGCHLD
, -1);
4617 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
4619 r
= sd_event_new(&event
);
4621 return log_error_errno(r
, "Failed to get default event source: %m");
4623 (void) sd_event_set_watchdog(event
, true);
4626 r
= sd_bus_attach_event(bus
, event
, 0);
4628 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
4631 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
4635 /* Let the child know that we are ready and wait that the child is completely ready now. */
4636 if (!barrier_place_and_sync(&barrier
)) /* #5 */
4637 return log_error_errno(SYNTHETIC_ERRNO(ESRCH
), "Child died too early.");
4639 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
4640 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4641 etc_passwd_lock
= safe_close(etc_passwd_lock
);
4643 (void) sd_notifyf(false,
4644 "STATUS=Container running.\n"
4645 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
4646 if (!arg_notify_ready
)
4647 (void) sd_notify(false, "READY=1\n");
4649 if (arg_kill_signal
> 0) {
4650 /* Try to kill the init system on SIGINT or SIGTERM */
4651 (void) sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4652 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
4654 /* Immediately exit */
4655 (void) sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
4656 (void) sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
4659 /* Exit when the child exits */
4660 (void) sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
4662 if (arg_expose_ports
) {
4663 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
4667 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
4670 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
4672 if (arg_console_mode
!= CONSOLE_PIPE
) {
4673 _cleanup_close_
int fd
= -1;
4674 PTYForwardFlags flags
= 0;
4676 /* Retrieve the master pty allocated by inner child */
4677 fd
= receive_one_fd(master_pty_socket_pair
[0], 0);
4679 return log_error_errno(fd
, "Failed to receive master pty from the inner child: %m");
4681 switch (arg_console_mode
) {
4683 case CONSOLE_READ_ONLY
:
4684 flags
|= PTY_FORWARD_READ_ONLY
;
4688 case CONSOLE_INTERACTIVE
:
4689 flags
|= PTY_FORWARD_IGNORE_VHANGUP
;
4691 r
= pty_forward_new(event
, fd
, flags
, &forward
);
4693 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
4695 if (arg_console_width
!= (unsigned) -1 || arg_console_height
!= (unsigned) -1)
4696 (void) pty_forward_set_width_height(forward
,
4698 arg_console_height
);
4702 assert(arg_console_mode
== CONSOLE_PASSIVE
);
4705 *master
= TAKE_FD(fd
);
4708 r
= sd_event_loop(event
);
4710 return log_error_errno(r
, "Failed to run event loop: %m");
4715 (void) pty_forward_get_last_char(forward
, &last_char
);
4716 forward
= pty_forward_free(forward
);
4718 if (!arg_quiet
&& last_char
!= '\n')
4722 /* Kill if it is not dead yet anyway */
4723 if (!arg_register
&& !arg_keep_unit
&& bus
)
4724 terminate_scope(bus
, arg_machine
);
4726 /* Normally redundant, but better safe than sorry */
4727 (void) kill(*pid
, SIGKILL
);
4729 if (arg_private_network
) {
4730 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
4731 * to avoid having to move the parent to the child network namespace. */
4732 r
= safe_fork(NULL
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_WAIT
|FORK_LOG
, NULL
);
4737 _cleanup_close_
int parent_netns_fd
= -1;
4739 r
= namespace_open(getpid(), NULL
, NULL
, &parent_netns_fd
, NULL
, NULL
);
4741 log_error_errno(r
, "Failed to open parent network namespace: %m");
4742 _exit(EXIT_FAILURE
);
4745 r
= namespace_enter(-1, -1, child_netns_fd
, -1, -1);
4747 log_error_errno(r
, "Failed to enter child network namespace: %m");
4748 _exit(EXIT_FAILURE
);
4751 r
= move_network_interfaces(parent_netns_fd
, arg_network_interfaces
);
4753 log_error_errno(r
, "Failed to move network interfaces back to parent network namespace: %m");
4755 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
4759 r
= wait_for_container(*pid
, &container_status
);
4762 /* Tell machined that we are gone. */
4764 (void) unregister_machine(bus
, arg_machine
);
4767 /* We failed to wait for the container, or the container exited abnormally. */
4769 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
4770 /* r > 0 → The container exited with a non-zero status.
4771 * As a special case, we need to replace 133 with a different value,
4772 * because 133 is special-cased in the service file to reboot the container.
4773 * otherwise → The container exited with zero status and a reboot was not requested.
4775 if (r
== EXIT_FORCE_RESTART
)
4776 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
4778 return 0; /* finito */
4781 /* CONTAINER_REBOOTED, loop again */
4783 if (arg_keep_unit
) {
4784 /* Special handling if we are running as a service: instead of simply
4785 * restarting the machine we want to restart the entire service, so let's
4786 * inform systemd about this with the special exit code 133. The service
4787 * file uses RestartForceExitStatus=133 so that this results in a full
4788 * nspawn restart. This is necessary since we might have cgroup parameters
4789 * set we want to have flushed out. */
4790 *ret
= EXIT_FORCE_RESTART
;
4791 return 0; /* finito */
4794 expose_port_flush(arg_expose_ports
, exposed
);
4796 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4797 *veth_created
= false;
4798 return 1; /* loop again */
4801 static int initialize_rlimits(void) {
4802 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4803 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4804 * container execution environments. */
4806 static const struct rlimit kernel_defaults
[_RLIMIT_MAX
] = {
4807 [RLIMIT_AS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4808 [RLIMIT_CORE
] = { 0, RLIM_INFINITY
},
4809 [RLIMIT_CPU
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4810 [RLIMIT_DATA
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4811 [RLIMIT_FSIZE
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4812 [RLIMIT_LOCKS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4813 [RLIMIT_MEMLOCK
] = { 65536, 65536 },
4814 [RLIMIT_MSGQUEUE
] = { 819200, 819200 },
4815 [RLIMIT_NICE
] = { 0, 0 },
4816 [RLIMIT_NOFILE
] = { 1024, 4096 },
4817 [RLIMIT_RSS
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4818 [RLIMIT_RTPRIO
] = { 0, 0 },
4819 [RLIMIT_RTTIME
] = { RLIM_INFINITY
, RLIM_INFINITY
},
4820 [RLIMIT_STACK
] = { 8388608, RLIM_INFINITY
},
4822 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4823 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4824 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4825 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4826 * that PID 1 changes a number of other resource limits during early initialization which is why we
4827 * don't read the other limits from PID 1 but prefer the static table above. */
4832 for (rl
= 0; rl
< _RLIMIT_MAX
; rl
++) {
4833 /* Let's only fill in what the user hasn't explicitly configured anyway */
4834 if ((arg_settings_mask
& (SETTING_RLIMIT_FIRST
<< rl
)) == 0) {
4835 const struct rlimit
*v
;
4836 struct rlimit buffer
;
4838 if (IN_SET(rl
, RLIMIT_NPROC
, RLIMIT_SIGPENDING
)) {
4839 /* For these two let's read the limits off PID 1. See above for an explanation. */
4841 if (prlimit(1, rl
, NULL
, &buffer
) < 0)
4842 return log_error_errno(errno
, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl
));
4846 v
= kernel_defaults
+ rl
;
4848 arg_rlimit
[rl
] = newdup(struct rlimit
, v
, 1);
4849 if (!arg_rlimit
[rl
])
4853 if (DEBUG_LOGGING
) {
4854 _cleanup_free_
char *k
= NULL
;
4856 (void) rlimit_format(arg_rlimit
[rl
], &k
);
4857 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl
), k
);
4864 static int cant_be_in_netns(void) {
4865 union sockaddr_union sa
= {
4867 .sun_family
= AF_UNIX
,
4868 .sun_path
= "/run/udev/control",
4871 char udev_path
[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t
)];
4872 _cleanup_free_
char *udev_ns
= NULL
, *our_ns
= NULL
;
4873 _cleanup_close_
int fd
= -1;
4877 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
4878 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
4881 if (!arg_image
) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
4884 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_NONBLOCK
|SOCK_CLOEXEC
, 0);
4886 return log_error_errno(errno
, "Failed to allocate udev control socket: %m");
4888 if (connect(fd
, &sa
.un
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
4890 if (errno
== ENOENT
|| ERRNO_IS_DISCONNECT(errno
))
4891 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
4892 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
4894 return log_error_errno(errno
, "Failed to connect socket to udev control socket: %m");
4897 r
= getpeercred(fd
, &ucred
);
4899 return log_error_errno(r
, "Failed to determine peer of udev control socket: %m");
4901 xsprintf(udev_path
, "/proc/" PID_FMT
"/ns/net", ucred
.pid
);
4902 r
= readlink_malloc(udev_path
, &udev_ns
);
4904 return log_error_errno(r
, "Failed to read network namespace of udev: %m");
4906 r
= readlink_malloc("/proc/self/ns/net", &our_ns
);
4908 return log_error_errno(r
, "Failed to read our own network namespace: %m");
4910 if (!streq(our_ns
, udev_ns
))
4911 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
4912 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
4916 static int run(int argc
, char *argv
[]) {
4917 bool secondary
= false, remove_directory
= false, remove_image
= false,
4918 veth_created
= false, remove_tmprootdir
= false;
4919 _cleanup_close_
int master
= -1;
4920 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
4921 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
4922 char veth_name
[IFNAMSIZ
] = "";
4923 union in_addr_union exposed
= {};
4924 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
4925 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
4926 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
4927 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
4928 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4931 log_parse_environment();
4934 r
= parse_argv(argc
, argv
);
4942 r
= cant_be_in_netns();
4946 r
= initialize_rlimits();
4950 r
= load_oci_bundle();
4954 r
= determine_names();
4958 r
= load_settings();
4964 log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4968 r
= verify_arguments();
4972 /* Reapply environment settings. */
4973 (void) detect_unified_cgroup_hierarchy_from_environment();
4975 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4976 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4977 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4978 (void) ignore_signals(SIGPIPE
, -1);
4980 n_fd_passed
= sd_listen_fds(false);
4981 if (n_fd_passed
> 0) {
4982 r
= fdset_new_listen_fds(&fds
, false);
4984 log_error_errno(r
, "Failed to collect file descriptors: %m");
4989 /* The "default" umask. This is appropriate for most file and directory
4990 * operations performed by nspawn, and is the umask that will be used for
4991 * the child. Functions like copy_devnodes() change the umask temporarily. */
4994 if (arg_directory
) {
4997 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4998 * /var from the host will propagate into container dynamically (because bad things happen if
4999 * two systems write to the same /var). Let's allow it for the special cases where /var is
5000 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5001 if (path_equal(arg_directory
, "/") && !(arg_ephemeral
|| IN_SET(arg_volatile_mode
, VOLATILE_YES
, VOLATILE_STATE
))) {
5002 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5007 if (arg_ephemeral
) {
5008 _cleanup_free_
char *np
= NULL
;
5010 r
= chase_symlinks_and_update(&arg_directory
, 0);
5014 /* If the specified path is a mount point we generate the new snapshot immediately
5015 * inside it under a random name. However if the specified is not a mount point we
5016 * create the new snapshot in the parent directory, just next to it. */
5017 r
= path_is_mount_point(arg_directory
, NULL
, 0);
5019 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
5023 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
5025 r
= tempfn_random(arg_directory
, "machine.", &np
);
5027 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
5031 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5032 * only owned by us and no one else. */
5033 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5035 log_error_errno(r
, "Failed to lock %s: %m", np
);
5040 BLOCK_SIGNALS(SIGINT
);
5041 r
= btrfs_subvol_snapshot(arg_directory
, np
,
5042 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5043 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5044 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5045 BTRFS_SNAPSHOT_RECURSIVE
|
5046 BTRFS_SNAPSHOT_QUOTA
|
5047 BTRFS_SNAPSHOT_SIGINT
);
5050 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", np
);
5054 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
5058 free_and_replace(arg_directory
, np
);
5059 remove_directory
= true;
5061 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
5065 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5067 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
5071 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
5076 r
= chase_symlinks_and_update(&arg_template
, 0);
5081 BLOCK_SIGNALS(SIGINT
);
5082 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
5083 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
5084 BTRFS_SNAPSHOT_FALLBACK_COPY
|
5085 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
5086 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
5087 BTRFS_SNAPSHOT_RECURSIVE
|
5088 BTRFS_SNAPSHOT_QUOTA
|
5089 BTRFS_SNAPSHOT_SIGINT
);
5092 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5093 "Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
5094 else if (r
== -EINTR
) {
5095 log_error_errno(r
, "Interrupted while copying file system tree to %s, removed again.", arg_directory
);
5098 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
5101 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
,
5102 "Populated %s from template %s.", arg_directory
, arg_template
);
5106 if (arg_start_mode
== START_BOOT
) {
5109 if (arg_pivot_root_new
)
5110 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5114 if (path_is_os_tree(p
) <= 0) {
5115 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p
);
5122 if (arg_pivot_root_new
)
5123 p
= prefix_roota(arg_directory
, arg_pivot_root_new
);
5127 q
= strjoina(p
, "/usr/");
5129 if (laccess(q
, F_OK
) < 0) {
5130 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p
);
5137 DissectImageFlags dissect_image_flags
= DISSECT_IMAGE_REQUIRE_ROOT
| DISSECT_IMAGE_RELAX_VAR_CHECK
;
5139 assert(!arg_template
);
5141 r
= chase_symlinks_and_update(&arg_image
, 0);
5145 if (arg_ephemeral
) {
5146 _cleanup_free_
char *np
= NULL
;
5148 r
= tempfn_random(arg_image
, "machine.", &np
);
5150 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
5154 /* Always take an exclusive lock on our own ephemeral copy. */
5155 r
= image_path_lock(np
, LOCK_EX
|LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5157 r
= log_error_errno(r
, "Failed to create image lock: %m");
5162 BLOCK_SIGNALS(SIGINT
);
5163 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
);
5166 log_error_errno(r
, "Interrupted while copying image file to %s, removed again.", np
);
5170 r
= log_error_errno(r
, "Failed to copy image file: %m");
5174 free_and_replace(arg_image
, np
);
5175 remove_image
= true;
5177 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
5179 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
5183 r
= log_error_errno(r
, "Failed to create image lock: %m");
5187 r
= verity_metadata_load(arg_image
, NULL
, arg_root_hash
? NULL
: &arg_root_hash
, &arg_root_hash_size
,
5188 arg_verity_data
? NULL
: &arg_verity_data
,
5189 arg_root_hash_sig_path
|| arg_root_hash_sig
? NULL
: &arg_root_hash_sig_path
);
5191 log_error_errno(r
, "Failed to read verity artefacts for %s: %m", arg_image
);
5194 dissect_image_flags
|= arg_verity_data
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0;
5197 if (!mkdtemp(tmprootdir
)) {
5198 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
5202 remove_tmprootdir
= true;
5204 arg_directory
= strdup(tmprootdir
);
5205 if (!arg_directory
) {
5210 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, LO_FLAGS_PARTSCAN
, &loop
);
5212 log_error_errno(r
, "Failed to set up loopback block device: %m");
5216 r
= dissect_image_and_warn(
5219 arg_root_hash
, arg_root_hash_size
,
5221 dissect_image_flags
,
5224 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5225 log_notice("Note that the disk image needs to\n"
5226 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5227 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5228 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5229 " d) or contain a file system without a partition table\n"
5230 "in order to be bootable with systemd-nspawn.");
5236 if (!arg_root_hash
&& dissected_image
->can_verity
)
5237 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
5239 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, arg_verity_data
, arg_root_hash_sig_path
, arg_root_hash_sig
, arg_root_hash_sig_size
, 0, &decrypted_image
);
5243 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5244 if (remove_image
&& unlink(arg_image
) >= 0)
5245 remove_image
= false;
5248 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
5252 if (arg_console_mode
< 0)
5254 isatty(STDIN_FILENO
) > 0 &&
5255 isatty(STDOUT_FILENO
) > 0 ? CONSOLE_INTERACTIVE
: CONSOLE_READ_ONLY
;
5257 if (arg_console_mode
== CONSOLE_PIPE
) /* if we pass STDERR on to the container, don't add our own logs into it too */
5261 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5262 arg_machine
, arg_image
?: arg_directory
);
5264 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
5266 if (prctl(PR_SET_CHILD_SUBREAPER
, 1, 0, 0, 0) < 0) {
5267 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
5272 r
= run_container(dissected_image
,
5275 veth_name
, &veth_created
,
5283 (void) sd_notify(false,
5284 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
5285 "STOPPING=1\nSTATUS=Terminating...");
5288 (void) kill(pid
, SIGKILL
);
5290 /* Try to flush whatever is still queued in the pty */
5292 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
5293 master
= safe_close(master
);
5297 (void) wait_for_terminate(pid
, NULL
);
5301 if (remove_directory
&& arg_directory
) {
5304 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
5306 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
5309 if (remove_image
&& arg_image
) {
5310 if (unlink(arg_image
) < 0)
5311 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
5314 if (remove_tmprootdir
) {
5315 if (rmdir(tmprootdir
) < 0)
5316 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
5322 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
5323 (void) rm_rf(p
, REMOVE_ROOT
);
5326 expose_port_flush(arg_expose_ports
, &exposed
);
5329 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
5330 (void) remove_bridge(arg_network_zone
);
5332 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
5333 expose_port_free_all(arg_expose_ports
);
5334 rlimit_free_all(arg_rlimit
);
5335 device_node_array_free(arg_extra_nodes
, arg_n_extra_nodes
);
5343 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run
);