1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
14 #include <linux/loop.h>
18 #include <selinux/selinux.h>
25 #include <sys/mount.h>
26 #include <sys/personality.h>
27 #include <sys/prctl.h>
28 #include <sys/types.h>
33 #include "sd-daemon.h"
36 #include "alloc-util.h"
38 #include "base-filesystem.h"
39 #include "blkid-util.h"
40 #include "btrfs-util.h"
43 #include "capability-util.h"
44 #include "cgroup-util.h"
46 #include "dev-setup.h"
47 #include "dissect-image.h"
52 #include "format-util.h"
55 #include "hexdecoct.h"
56 #include "hostname-util.h"
57 #include "id128-util.h"
59 #include "loop-util.h"
60 #include "loopback-setup.h"
61 #include "machine-image.h"
65 #include "mount-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-def.h"
69 #include "nspawn-expose-ports.h"
70 #include "nspawn-mount.h"
71 #include "nspawn-network.h"
72 #include "nspawn-patch-uid.h"
73 #include "nspawn-register.h"
74 #include "nspawn-seccomp.h"
75 #include "nspawn-settings.h"
76 #include "nspawn-setuid.h"
77 #include "nspawn-stub-pid1.h"
78 #include "parse-util.h"
79 #include "path-util.h"
80 #include "process-util.h"
82 #include "random-util.h"
83 #include "raw-clone.h"
85 #include "selinux-util.h"
86 #include "signal-util.h"
87 #include "socket-util.h"
88 #include "stat-util.h"
89 #include "stdio-util.h"
90 #include "string-util.h"
92 #include "terminal-util.h"
93 #include "udev-util.h"
94 #include "umask-util.h"
95 #include "user-util.h"
99 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
101 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
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 typedef enum LinkJournal
{
123 static char *arg_directory
= NULL
;
124 static char *arg_template
= NULL
;
125 static char *arg_chdir
= NULL
;
126 static char *arg_pivot_root_new
= NULL
;
127 static char *arg_pivot_root_old
= NULL
;
128 static char *arg_user
= NULL
;
129 static sd_id128_t arg_uuid
= {};
130 static char *arg_machine
= NULL
;
131 static const char *arg_selinux_context
= NULL
;
132 static const char *arg_selinux_apifs_context
= NULL
;
133 static const char *arg_slice
= NULL
;
134 static bool arg_private_network
= false;
135 static bool arg_read_only
= false;
136 static StartMode arg_start_mode
= START_PID1
;
137 static bool arg_ephemeral
= false;
138 static LinkJournal arg_link_journal
= LINK_AUTO
;
139 static bool arg_link_journal_try
= false;
140 static uint64_t arg_caps_retain
=
141 (1ULL << CAP_AUDIT_CONTROL
) |
142 (1ULL << CAP_AUDIT_WRITE
) |
143 (1ULL << CAP_CHOWN
) |
144 (1ULL << CAP_DAC_OVERRIDE
) |
145 (1ULL << CAP_DAC_READ_SEARCH
) |
146 (1ULL << CAP_FOWNER
) |
147 (1ULL << CAP_FSETID
) |
148 (1ULL << CAP_IPC_OWNER
) |
150 (1ULL << CAP_LEASE
) |
151 (1ULL << CAP_LINUX_IMMUTABLE
) |
152 (1ULL << CAP_MKNOD
) |
153 (1ULL << CAP_NET_BIND_SERVICE
) |
154 (1ULL << CAP_NET_BROADCAST
) |
155 (1ULL << CAP_NET_RAW
) |
156 (1ULL << CAP_SETFCAP
) |
157 (1ULL << CAP_SETGID
) |
158 (1ULL << CAP_SETPCAP
) |
159 (1ULL << CAP_SETUID
) |
160 (1ULL << CAP_SYS_ADMIN
) |
161 (1ULL << CAP_SYS_BOOT
) |
162 (1ULL << CAP_SYS_CHROOT
) |
163 (1ULL << CAP_SYS_NICE
) |
164 (1ULL << CAP_SYS_PTRACE
) |
165 (1ULL << CAP_SYS_RESOURCE
) |
166 (1ULL << CAP_SYS_TTY_CONFIG
);
167 static CustomMount
*arg_custom_mounts
= NULL
;
168 static size_t arg_n_custom_mounts
= 0;
169 static char **arg_setenv
= NULL
;
170 static bool arg_quiet
= false;
171 static bool arg_register
= true;
172 static bool arg_keep_unit
= false;
173 static char **arg_network_interfaces
= NULL
;
174 static char **arg_network_macvlan
= NULL
;
175 static char **arg_network_ipvlan
= NULL
;
176 static bool arg_network_veth
= false;
177 static char **arg_network_veth_extra
= NULL
;
178 static char *arg_network_bridge
= NULL
;
179 static char *arg_network_zone
= NULL
;
180 static char *arg_network_namespace_path
= NULL
;
181 static unsigned long arg_personality
= PERSONALITY_INVALID
;
182 static char *arg_image
= NULL
;
183 static VolatileMode arg_volatile_mode
= VOLATILE_NO
;
184 static ExposePort
*arg_expose_ports
= NULL
;
185 static char **arg_property
= NULL
;
186 static UserNamespaceMode arg_userns_mode
= USER_NAMESPACE_NO
;
187 static uid_t arg_uid_shift
= UID_INVALID
, arg_uid_range
= 0x10000U
;
188 static bool arg_userns_chown
= false;
189 static int arg_kill_signal
= 0;
190 static CGroupUnified arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_UNKNOWN
;
191 static SettingsMask arg_settings_mask
= 0;
192 static int arg_settings_trusted
= -1;
193 static char **arg_parameters
= NULL
;
194 static const char *arg_container_service_name
= "systemd-nspawn";
195 static bool arg_notify_ready
= false;
196 static bool arg_use_cgns
= true;
197 static unsigned long arg_clone_ns_flags
= CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
;
198 static MountSettingsMask arg_mount_settings
= MOUNT_APPLY_APIVFS_RO
;
199 static void *arg_root_hash
= NULL
;
200 static size_t arg_root_hash_size
= 0;
201 static char **arg_syscall_whitelist
= NULL
;
202 static char **arg_syscall_blacklist
= NULL
;
204 static void help(void) {
205 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
206 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
207 " -h --help Show this help\n"
208 " --version Print version string\n"
209 " -q --quiet Do not show status information\n"
210 " -D --directory=PATH Root directory for the container\n"
211 " --template=PATH Initialize root directory from template directory,\n"
213 " -x --ephemeral Run container with snapshot of root directory, and\n"
214 " remove it after exit\n"
215 " -i --image=PATH File system device or disk image for the container\n"
216 " --root-hash=HASH Specify verity root hash\n"
217 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
218 " -b --boot Boot up full system (i.e. invoke init)\n"
219 " --chdir=PATH Set working directory in the container\n"
220 " --pivot-root=PATH[:PATH]\n"
221 " Pivot root to given directory in the container\n"
222 " -u --user=USER Run the command under specified user or uid\n"
223 " -M --machine=NAME Set the machine name for the container\n"
224 " --uuid=UUID Set a specific machine UUID for the container\n"
225 " -S --slice=SLICE Place the container in the specified slice\n"
226 " --property=NAME=VALUE Set scope unit property\n"
227 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
228 " --private-users[=UIDBASE[:NUIDS]]\n"
229 " Similar, but with user configured UID/GID range\n"
230 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
231 " --private-network Disable network in container\n"
232 " --network-interface=INTERFACE\n"
233 " Assign an existing network interface to the\n"
235 " --network-macvlan=INTERFACE\n"
236 " Create a macvlan network interface based on an\n"
237 " existing network interface to the container\n"
238 " --network-ipvlan=INTERFACE\n"
239 " Create a ipvlan network interface based on an\n"
240 " existing network interface to the container\n"
241 " -n --network-veth Add a virtual Ethernet connection between host\n"
243 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
244 " Add an additional virtual Ethernet link between\n"
245 " host and container\n"
246 " --network-bridge=INTERFACE\n"
247 " Add a virtual Ethernet connection to the container\n"
248 " and attach it to an existing bridge on the host\n"
249 " --network-zone=NAME Similar, but attach the new interface to an\n"
250 " an automatically managed bridge interface\n"
251 " --network-namespace-path=PATH\n"
252 " Set network namespace to the one represented by\n"
253 " the specified kernel namespace file node\n"
254 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
255 " Expose a container IP port on the host\n"
256 " -Z --selinux-context=SECLABEL\n"
257 " Set the SELinux security context to be used by\n"
258 " processes in the container\n"
259 " -L --selinux-apifs-context=SECLABEL\n"
260 " Set the SELinux security context to be used by\n"
261 " API/tmpfs file systems in the container\n"
262 " --capability=CAP In addition to the default, retain specified\n"
264 " --drop-capability=CAP Drop the specified capability from the default set\n"
265 " --system-call-filter=LIST|~LIST\n"
266 " Permit/prohibit specific system calls\n"
267 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
268 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
269 " host, try-guest, try-host\n"
270 " -j Equivalent to --link-journal=try-guest\n"
271 " --read-only Mount the root directory read-only\n"
272 " --bind=PATH[:PATH[:OPTIONS]]\n"
273 " Bind mount a file or directory from the host into\n"
275 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
276 " Similar, but creates a read-only bind mount\n"
277 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
278 " --overlay=PATH[:PATH...]:PATH\n"
279 " Create an overlay mount from the host to \n"
281 " --overlay-ro=PATH[:PATH...]:PATH\n"
282 " Similar, but creates a read-only overlay mount\n"
283 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
284 " --register=BOOLEAN Register container as machine\n"
285 " --keep-unit Do not register a scope for the machine, reuse\n"
286 " the service unit nspawn is running in\n"
287 " --volatile[=MODE] Run the system in volatile mode\n"
288 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
289 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
290 , program_invocation_short_name
);
293 static int custom_mount_check_all(void) {
296 for (i
= 0; i
< arg_n_custom_mounts
; i
++) {
297 CustomMount
*m
= &arg_custom_mounts
[i
];
299 if (path_equal(m
->destination
, "/") && arg_userns_mode
!= USER_NAMESPACE_NO
) {
301 if (arg_userns_chown
) {
302 log_error("--private-users-chown may not be combined with custom root mounts.");
304 } else if (arg_uid_shift
== UID_INVALID
) {
305 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
314 static int detect_unified_cgroup_hierarchy_from_environment(void) {
318 /* Allow the user to control whether the unified hierarchy is used */
319 e
= getenv("UNIFIED_CGROUP_HIERARCHY");
321 r
= parse_boolean(e
);
323 return log_error_errno(r
, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
325 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
327 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
333 static int detect_unified_cgroup_hierarchy_from_image(const char *directory
) {
336 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
337 * image actually supports. */
338 r
= cg_all_unified();
340 return log_error_errno(r
, "Failed to determine whether we are in all unified mode.");
342 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
343 * routine only detects 231, so we'll have a false negative here for 230. */
344 r
= systemd_installation_has_version(directory
, 230);
346 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
348 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_ALL
;
350 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
351 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
) > 0) {
352 /* Mixed cgroup hierarchy support was added in 233 */
353 r
= systemd_installation_has_version(directory
, 233);
355 return log_error_errno(r
, "Failed to determine systemd version in container: %m");
357 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_SYSTEMD
;
359 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
361 arg_unified_cgroup_hierarchy
= CGROUP_UNIFIED_NONE
;
363 log_debug("Using %s hierarchy for container.",
364 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_NONE
? "legacy" :
365 arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_SYSTEMD
? "hybrid" : "unified");
370 static void parse_share_ns_env(const char *name
, unsigned long ns_flag
) {
373 r
= getenv_bool(name
);
377 log_warning_errno(r
, "Failed to parse %s from environment, defaulting to false.", name
);
378 arg_clone_ns_flags
= (arg_clone_ns_flags
& ~ns_flag
) | (r
> 0 ? 0 : ns_flag
);
381 static void parse_mount_settings_env(void) {
385 e
= getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
389 if (streq(e
, "network")) {
390 arg_mount_settings
|= MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
;
394 r
= parse_boolean(e
);
396 log_warning_errno(r
, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
400 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_RO
, r
== 0);
401 SET_FLAG(arg_mount_settings
, MOUNT_APPLY_APIVFS_NETNS
, false);
404 static int parse_argv(int argc
, char *argv
[]) {
422 ARG_NETWORK_INTERFACE
,
427 ARG_NETWORK_VETH_EXTRA
,
428 ARG_NETWORK_NAMESPACE_PATH
,
438 ARG_PRIVATE_USERS_CHOWN
,
441 ARG_SYSTEM_CALL_FILTER
,
444 static const struct option options
[] = {
445 { "help", no_argument
, NULL
, 'h' },
446 { "version", no_argument
, NULL
, ARG_VERSION
},
447 { "directory", required_argument
, NULL
, 'D' },
448 { "template", required_argument
, NULL
, ARG_TEMPLATE
},
449 { "ephemeral", no_argument
, NULL
, 'x' },
450 { "user", required_argument
, NULL
, 'u' },
451 { "private-network", no_argument
, NULL
, ARG_PRIVATE_NETWORK
},
452 { "as-pid2", no_argument
, NULL
, 'a' },
453 { "boot", no_argument
, NULL
, 'b' },
454 { "uuid", required_argument
, NULL
, ARG_UUID
},
455 { "read-only", no_argument
, NULL
, ARG_READ_ONLY
},
456 { "capability", required_argument
, NULL
, ARG_CAPABILITY
},
457 { "drop-capability", required_argument
, NULL
, ARG_DROP_CAPABILITY
},
458 { "link-journal", required_argument
, NULL
, ARG_LINK_JOURNAL
},
459 { "bind", required_argument
, NULL
, ARG_BIND
},
460 { "bind-ro", required_argument
, NULL
, ARG_BIND_RO
},
461 { "tmpfs", required_argument
, NULL
, ARG_TMPFS
},
462 { "overlay", required_argument
, NULL
, ARG_OVERLAY
},
463 { "overlay-ro", required_argument
, NULL
, ARG_OVERLAY_RO
},
464 { "machine", required_argument
, NULL
, 'M' },
465 { "slice", required_argument
, NULL
, 'S' },
466 { "setenv", required_argument
, NULL
, 'E' },
467 { "selinux-context", required_argument
, NULL
, 'Z' },
468 { "selinux-apifs-context", required_argument
, NULL
, 'L' },
469 { "quiet", no_argument
, NULL
, 'q' },
470 { "share-system", no_argument
, NULL
, ARG_SHARE_SYSTEM
}, /* not documented */
471 { "register", required_argument
, NULL
, ARG_REGISTER
},
472 { "keep-unit", no_argument
, NULL
, ARG_KEEP_UNIT
},
473 { "network-interface", required_argument
, NULL
, ARG_NETWORK_INTERFACE
},
474 { "network-macvlan", required_argument
, NULL
, ARG_NETWORK_MACVLAN
},
475 { "network-ipvlan", required_argument
, NULL
, ARG_NETWORK_IPVLAN
},
476 { "network-veth", no_argument
, NULL
, 'n' },
477 { "network-veth-extra", required_argument
, NULL
, ARG_NETWORK_VETH_EXTRA
},
478 { "network-bridge", required_argument
, NULL
, ARG_NETWORK_BRIDGE
},
479 { "network-zone", required_argument
, NULL
, ARG_NETWORK_ZONE
},
480 { "network-namespace-path", required_argument
, NULL
, ARG_NETWORK_NAMESPACE_PATH
},
481 { "personality", required_argument
, NULL
, ARG_PERSONALITY
},
482 { "image", required_argument
, NULL
, 'i' },
483 { "volatile", optional_argument
, NULL
, ARG_VOLATILE
},
484 { "port", required_argument
, NULL
, 'p' },
485 { "property", required_argument
, NULL
, ARG_PROPERTY
},
486 { "private-users", optional_argument
, NULL
, ARG_PRIVATE_USERS
},
487 { "private-users-chown", optional_argument
, NULL
, ARG_PRIVATE_USERS_CHOWN
},
488 { "kill-signal", required_argument
, NULL
, ARG_KILL_SIGNAL
},
489 { "settings", required_argument
, NULL
, ARG_SETTINGS
},
490 { "chdir", required_argument
, NULL
, ARG_CHDIR
},
491 { "pivot-root", required_argument
, NULL
, ARG_PIVOT_ROOT
},
492 { "notify-ready", required_argument
, NULL
, ARG_NOTIFY_READY
},
493 { "root-hash", required_argument
, NULL
, ARG_ROOT_HASH
},
494 { "system-call-filter", required_argument
, NULL
, ARG_SYSTEM_CALL_FILTER
},
500 uint64_t plus
= 0, minus
= 0;
501 bool mask_all_settings
= false, mask_no_settings
= false;
506 while ((c
= getopt_long(argc
, argv
, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options
, NULL
)) >= 0)
518 r
= parse_path_argument_and_warn(optarg
, false, &arg_directory
);
524 r
= parse_path_argument_and_warn(optarg
, false, &arg_template
);
530 r
= parse_path_argument_and_warn(optarg
, false, &arg_image
);
536 arg_ephemeral
= true;
540 r
= free_and_strdup(&arg_user
, optarg
);
544 arg_settings_mask
|= SETTING_USER
;
547 case ARG_NETWORK_ZONE
: {
550 j
= strappend("vz-", optarg
);
554 if (!ifname_valid(j
)) {
555 log_error("Network zone name not valid: %s", j
);
560 free(arg_network_zone
);
561 arg_network_zone
= j
;
563 arg_network_veth
= true;
564 arg_private_network
= true;
565 arg_settings_mask
|= SETTING_NETWORK
;
569 case ARG_NETWORK_BRIDGE
:
571 if (!ifname_valid(optarg
)) {
572 log_error("Bridge interface name not valid: %s", optarg
);
576 r
= free_and_strdup(&arg_network_bridge
, optarg
);
582 arg_network_veth
= true;
583 arg_private_network
= true;
584 arg_settings_mask
|= SETTING_NETWORK
;
587 case ARG_NETWORK_VETH_EXTRA
:
588 r
= veth_extra_parse(&arg_network_veth_extra
, optarg
);
590 return log_error_errno(r
, "Failed to parse --network-veth-extra= parameter: %s", optarg
);
592 arg_private_network
= true;
593 arg_settings_mask
|= SETTING_NETWORK
;
596 case ARG_NETWORK_INTERFACE
:
598 if (!ifname_valid(optarg
)) {
599 log_error("Network interface name not valid: %s", optarg
);
603 if (strv_extend(&arg_network_interfaces
, optarg
) < 0)
606 arg_private_network
= true;
607 arg_settings_mask
|= SETTING_NETWORK
;
610 case ARG_NETWORK_MACVLAN
:
612 if (!ifname_valid(optarg
)) {
613 log_error("MACVLAN network interface name not valid: %s", optarg
);
617 if (strv_extend(&arg_network_macvlan
, optarg
) < 0)
620 arg_private_network
= true;
621 arg_settings_mask
|= SETTING_NETWORK
;
624 case ARG_NETWORK_IPVLAN
:
626 if (!ifname_valid(optarg
)) {
627 log_error("IPVLAN network interface name not valid: %s", optarg
);
631 if (strv_extend(&arg_network_ipvlan
, optarg
) < 0)
635 case ARG_PRIVATE_NETWORK
:
636 arg_private_network
= true;
637 arg_settings_mask
|= SETTING_NETWORK
;
640 case ARG_NETWORK_NAMESPACE_PATH
:
641 r
= parse_path_argument_and_warn(optarg
, false, &arg_network_namespace_path
);
648 if (arg_start_mode
== START_PID2
) {
649 log_error("--boot and --as-pid2 may not be combined.");
653 arg_start_mode
= START_BOOT
;
654 arg_settings_mask
|= SETTING_START_MODE
;
658 if (arg_start_mode
== START_BOOT
) {
659 log_error("--boot and --as-pid2 may not be combined.");
663 arg_start_mode
= START_PID2
;
664 arg_settings_mask
|= SETTING_START_MODE
;
668 r
= sd_id128_from_string(optarg
, &arg_uuid
);
670 return log_error_errno(r
, "Invalid UUID: %s", optarg
);
672 if (sd_id128_is_null(arg_uuid
)) {
673 log_error("Machine UUID may not be all zeroes.");
677 arg_settings_mask
|= SETTING_MACHINE_ID
;
686 arg_machine
= mfree(arg_machine
);
688 if (!machine_name_is_valid(optarg
)) {
689 log_error("Invalid machine name: %s", optarg
);
693 r
= free_and_strdup(&arg_machine
, optarg
);
700 arg_selinux_context
= optarg
;
704 arg_selinux_apifs_context
= optarg
;
708 arg_read_only
= true;
709 arg_settings_mask
|= SETTING_READ_ONLY
;
713 case ARG_DROP_CAPABILITY
: {
716 _cleanup_free_
char *t
= NULL
;
718 r
= extract_first_word(&p
, &t
, ",", 0);
720 return log_error_errno(r
, "Failed to parse capability %s.", t
);
725 if (streq(t
, "all")) {
726 if (c
== ARG_CAPABILITY
)
727 plus
= (uint64_t) -1;
729 minus
= (uint64_t) -1;
733 cap
= capability_from_name(t
);
735 log_error("Failed to parse capability %s.", t
);
739 if (c
== ARG_CAPABILITY
)
740 plus
|= 1ULL << (uint64_t) cap
;
742 minus
|= 1ULL << (uint64_t) cap
;
746 arg_settings_mask
|= SETTING_CAPABILITY
;
751 arg_link_journal
= LINK_GUEST
;
752 arg_link_journal_try
= true;
755 case ARG_LINK_JOURNAL
:
756 if (streq(optarg
, "auto")) {
757 arg_link_journal
= LINK_AUTO
;
758 arg_link_journal_try
= false;
759 } else if (streq(optarg
, "no")) {
760 arg_link_journal
= LINK_NO
;
761 arg_link_journal_try
= false;
762 } else if (streq(optarg
, "guest")) {
763 arg_link_journal
= LINK_GUEST
;
764 arg_link_journal_try
= false;
765 } else if (streq(optarg
, "host")) {
766 arg_link_journal
= LINK_HOST
;
767 arg_link_journal_try
= false;
768 } else if (streq(optarg
, "try-guest")) {
769 arg_link_journal
= LINK_GUEST
;
770 arg_link_journal_try
= true;
771 } else if (streq(optarg
, "try-host")) {
772 arg_link_journal
= LINK_HOST
;
773 arg_link_journal_try
= true;
775 log_error("Failed to parse link journal mode %s", optarg
);
783 r
= bind_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_BIND_RO
);
785 return log_error_errno(r
, "Failed to parse --bind(-ro)= argument %s: %m", optarg
);
787 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
791 r
= tmpfs_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
);
793 return log_error_errno(r
, "Failed to parse --tmpfs= argument %s: %m", optarg
);
795 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
800 r
= overlay_mount_parse(&arg_custom_mounts
, &arg_n_custom_mounts
, optarg
, c
== ARG_OVERLAY_RO
);
801 if (r
== -EADDRNOTAVAIL
)
802 return log_error_errno(r
, "--overlay(-ro)= needs at least two colon-separated directories specified.");
804 return log_error_errno(r
, "Failed to parse --overlay(-ro)= argument %s: %m", optarg
);
806 arg_settings_mask
|= SETTING_CUSTOM_MOUNTS
;
812 if (!env_assignment_is_valid(optarg
)) {
813 log_error("Environment variable assignment '%s' is not valid.", optarg
);
817 n
= strv_env_set(arg_setenv
, optarg
);
821 strv_free_and_replace(arg_setenv
, n
);
822 arg_settings_mask
|= SETTING_ENVIRONMENT
;
830 case ARG_SHARE_SYSTEM
:
831 /* We don't officially support this anymore, except for compat reasons. People should use the
832 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
833 arg_clone_ns_flags
= 0;
837 r
= parse_boolean(optarg
);
839 log_error("Failed to parse --register= argument: %s", optarg
);
847 arg_keep_unit
= true;
850 case ARG_PERSONALITY
:
852 arg_personality
= personality_from_string(optarg
);
853 if (arg_personality
== PERSONALITY_INVALID
) {
854 log_error("Unknown or unsupported personality '%s'.", optarg
);
858 arg_settings_mask
|= SETTING_PERSONALITY
;
864 arg_volatile_mode
= VOLATILE_YES
;
868 m
= volatile_mode_from_string(optarg
);
870 log_error("Failed to parse --volatile= argument: %s", optarg
);
873 arg_volatile_mode
= m
;
876 arg_settings_mask
|= SETTING_VOLATILE_MODE
;
880 r
= expose_port_parse(&arg_expose_ports
, optarg
);
882 return log_error_errno(r
, "Duplicate port specification: %s", optarg
);
884 return log_error_errno(r
, "Failed to parse host port %s: %m", optarg
);
886 arg_settings_mask
|= SETTING_EXPOSE_PORTS
;
890 if (strv_extend(&arg_property
, optarg
) < 0)
895 case ARG_PRIVATE_USERS
: {
900 else if (!in_charset(optarg
, DIGITS
))
901 /* do *not* parse numbers as booleans */
902 boolean
= parse_boolean(optarg
);
904 if (boolean
== false) {
905 /* no: User namespacing off */
906 arg_userns_mode
= USER_NAMESPACE_NO
;
907 arg_uid_shift
= UID_INVALID
;
908 arg_uid_range
= UINT32_C(0x10000);
909 } else if (boolean
== true) {
910 /* yes: User namespacing on, UID range is read from root dir */
911 arg_userns_mode
= USER_NAMESPACE_FIXED
;
912 arg_uid_shift
= UID_INVALID
;
913 arg_uid_range
= UINT32_C(0x10000);
914 } else if (streq(optarg
, "pick")) {
915 /* pick: User namespacing on, UID range is picked randomly */
916 arg_userns_mode
= USER_NAMESPACE_PICK
;
917 arg_uid_shift
= UID_INVALID
;
918 arg_uid_range
= UINT32_C(0x10000);
920 _cleanup_free_
char *buffer
= NULL
;
921 const char *range
, *shift
;
923 /* anything else: User namespacing on, UID range is explicitly configured */
925 range
= strchr(optarg
, ':');
927 buffer
= strndup(optarg
, range
- optarg
);
933 r
= safe_atou32(range
, &arg_uid_range
);
935 return log_error_errno(r
, "Failed to parse UID range \"%s\": %m", range
);
939 r
= parse_uid(shift
, &arg_uid_shift
);
941 return log_error_errno(r
, "Failed to parse UID \"%s\": %m", optarg
);
943 arg_userns_mode
= USER_NAMESPACE_FIXED
;
946 if (arg_uid_range
<= 0) {
947 log_error("UID range cannot be 0.");
951 arg_settings_mask
|= SETTING_USERNS
;
956 if (userns_supported()) {
957 arg_userns_mode
= USER_NAMESPACE_PICK
;
958 arg_uid_shift
= UID_INVALID
;
959 arg_uid_range
= UINT32_C(0x10000);
961 arg_settings_mask
|= SETTING_USERNS
;
966 case ARG_PRIVATE_USERS_CHOWN
:
967 arg_userns_chown
= true;
969 arg_settings_mask
|= SETTING_USERNS
;
972 case ARG_KILL_SIGNAL
:
973 arg_kill_signal
= signal_from_string(optarg
);
974 if (arg_kill_signal
< 0) {
975 log_error("Cannot parse signal: %s", optarg
);
979 arg_settings_mask
|= SETTING_KILL_SIGNAL
;
984 /* no → do not read files
985 * yes → read files, do not override cmdline, trust only subset
986 * override → read files, override cmdline, trust only subset
987 * trusted → read files, do not override cmdline, trust all
990 r
= parse_boolean(optarg
);
992 if (streq(optarg
, "trusted")) {
993 mask_all_settings
= false;
994 mask_no_settings
= false;
995 arg_settings_trusted
= true;
997 } else if (streq(optarg
, "override")) {
998 mask_all_settings
= false;
999 mask_no_settings
= true;
1000 arg_settings_trusted
= -1;
1002 return log_error_errno(r
, "Failed to parse --settings= argument: %s", optarg
);
1005 mask_all_settings
= false;
1006 mask_no_settings
= false;
1007 arg_settings_trusted
= -1;
1010 mask_all_settings
= true;
1011 mask_no_settings
= false;
1012 arg_settings_trusted
= false;
1018 if (!path_is_absolute(optarg
)) {
1019 log_error("Working directory %s is not an absolute path.", optarg
);
1023 r
= free_and_strdup(&arg_chdir
, optarg
);
1027 arg_settings_mask
|= SETTING_WORKING_DIRECTORY
;
1030 case ARG_PIVOT_ROOT
:
1031 r
= pivot_root_parse(&arg_pivot_root_new
, &arg_pivot_root_old
, optarg
);
1033 return log_error_errno(r
, "Failed to parse --pivot-root= argument %s: %m", optarg
);
1035 arg_settings_mask
|= SETTING_PIVOT_ROOT
;
1038 case ARG_NOTIFY_READY
:
1039 r
= parse_boolean(optarg
);
1041 log_error("%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg
);
1044 arg_notify_ready
= r
;
1045 arg_settings_mask
|= SETTING_NOTIFY_READY
;
1048 case ARG_ROOT_HASH
: {
1052 r
= unhexmem(optarg
, strlen(optarg
), &k
, &l
);
1054 return log_error_errno(r
, "Failed to parse root hash: %s", optarg
);
1055 if (l
< sizeof(sd_id128_t
)) {
1056 log_error("Root hash must be at least 128bit long: %s", optarg
);
1061 free(arg_root_hash
);
1063 arg_root_hash_size
= l
;
1067 case ARG_SYSTEM_CALL_FILTER
: {
1071 negative
= optarg
[0] == '~';
1072 items
= negative
? optarg
+ 1 : optarg
;
1075 _cleanup_free_
char *word
= NULL
;
1077 r
= extract_first_word(&items
, &word
, NULL
, 0);
1083 return log_error_errno(r
, "Failed to parse system call filter: %m");
1086 r
= strv_extend(&arg_syscall_blacklist
, word
);
1088 r
= strv_extend(&arg_syscall_whitelist
, word
);
1093 arg_settings_mask
|= SETTING_SYSCALL_FILTER
;
1101 assert_not_reached("Unhandled option");
1104 /* If --network-namespace-path is given with any other network-related option,
1105 * we need to error out, to avoid conflicts between different network options. */
1106 if (arg_network_namespace_path
&&
1107 (arg_network_interfaces
|| arg_network_macvlan
||
1108 arg_network_ipvlan
|| arg_network_veth_extra
||
1109 arg_network_bridge
|| arg_network_zone
||
1110 arg_network_veth
|| arg_private_network
)) {
1111 log_error("--network-namespace-path cannot be combined with other network options.");
1115 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC
);
1116 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID
);
1117 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS
);
1118 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC
|CLONE_NEWPID
|CLONE_NEWUTS
);
1120 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
1121 arg_mount_settings
|= MOUNT_USE_USERNS
;
1123 if (arg_private_network
)
1124 arg_mount_settings
|= MOUNT_APPLY_APIVFS_NETNS
;
1126 parse_mount_settings_env();
1128 if (!(arg_clone_ns_flags
& CLONE_NEWPID
) ||
1129 !(arg_clone_ns_flags
& CLONE_NEWUTS
)) {
1130 arg_register
= false;
1131 if (arg_start_mode
!= START_PID1
) {
1132 log_error("--boot cannot be used without namespacing.");
1137 if (arg_userns_mode
== USER_NAMESPACE_PICK
)
1138 arg_userns_chown
= true;
1140 if (arg_keep_unit
&& arg_register
&& cg_pid_get_owner_uid(0, NULL
) >= 0) {
1141 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1142 * The latter is not technically a user session, but we don't need to labour the point. */
1143 log_error("--keep-unit --register=yes may not be used when invoked from a user session.");
1147 if (arg_directory
&& arg_image
) {
1148 log_error("--directory= and --image= may not be combined.");
1152 if (arg_template
&& arg_image
) {
1153 log_error("--template= and --image= may not be combined.");
1157 if (arg_ephemeral
&& arg_template
&& !arg_directory
) {
1158 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1159 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1160 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1163 arg_directory
= TAKE_PTR(arg_template
);
1166 if (arg_template
&& !(arg_directory
|| arg_machine
)) {
1167 log_error("--template= needs --directory= or --machine=.");
1171 if (arg_ephemeral
&& arg_template
) {
1172 log_error("--ephemeral and --template= may not be combined.");
1176 if (arg_ephemeral
&& !IN_SET(arg_link_journal
, LINK_NO
, LINK_AUTO
)) {
1177 log_error("--ephemeral and --link-journal= may not be combined.");
1181 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !userns_supported()) {
1182 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1186 if (arg_userns_chown
&& arg_read_only
) {
1187 log_error("--read-only and --private-users-chown may not be combined.");
1191 if (arg_network_bridge
&& arg_network_zone
) {
1192 log_error("--network-bridge= and --network-zone= may not be combined.");
1196 if (argc
> optind
) {
1197 arg_parameters
= strv_copy(argv
+ optind
);
1198 if (!arg_parameters
)
1201 arg_settings_mask
|= SETTING_START_MODE
;
1204 /* Load all settings from .nspawn files */
1205 if (mask_no_settings
)
1206 arg_settings_mask
= 0;
1208 /* Don't load any settings from .nspawn files */
1209 if (mask_all_settings
)
1210 arg_settings_mask
= _SETTINGS_MASK_ALL
;
1212 arg_caps_retain
= (arg_caps_retain
| plus
| (arg_private_network
? 1ULL << CAP_NET_ADMIN
: 0)) & ~minus
;
1214 r
= cg_unified_flush();
1216 return log_error_errno(r
, "Failed to determine whether the unified cgroups hierarchy is used: %m");
1218 e
= getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1220 arg_container_service_name
= e
;
1222 r
= getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1224 arg_use_cgns
= cg_ns_supported();
1228 r
= custom_mount_check_all();
1235 static int verify_arguments(void) {
1236 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& (arg_mount_settings
& MOUNT_APPLY_APIVFS_NETNS
) && !arg_private_network
) {
1237 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1241 if (arg_userns_mode
!= USER_NAMESPACE_NO
&& !(arg_mount_settings
& MOUNT_APPLY_APIVFS_RO
)) {
1242 log_error("Cannot combine --private-users with read-write mounts.");
1246 if (arg_volatile_mode
!= VOLATILE_NO
&& arg_read_only
) {
1247 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1251 if (arg_expose_ports
&& !arg_private_network
) {
1252 log_error("Cannot use --port= without private networking.");
1257 if (arg_expose_ports
) {
1258 log_error("--port= is not supported, compiled without libiptc support.");
1263 if (arg_start_mode
== START_BOOT
&& arg_kill_signal
<= 0)
1264 arg_kill_signal
= SIGRTMIN
+3;
1269 static int userns_lchown(const char *p
, uid_t uid
, gid_t gid
) {
1272 if (arg_userns_mode
== USER_NAMESPACE_NO
)
1275 if (uid
== UID_INVALID
&& gid
== GID_INVALID
)
1278 if (uid
!= UID_INVALID
) {
1279 uid
+= arg_uid_shift
;
1281 if (uid
< arg_uid_shift
|| uid
>= arg_uid_shift
+ arg_uid_range
)
1285 if (gid
!= GID_INVALID
) {
1286 gid
+= (gid_t
) arg_uid_shift
;
1288 if (gid
< (gid_t
) arg_uid_shift
|| gid
>= (gid_t
) (arg_uid_shift
+ arg_uid_range
))
1292 if (lchown(p
, uid
, gid
) < 0)
1298 static int userns_mkdir(const char *root
, const char *path
, mode_t mode
, uid_t uid
, gid_t gid
) {
1302 q
= prefix_roota(root
, path
);
1303 r
= mkdir_errno_wrapper(q
, mode
);
1309 return userns_lchown(q
, uid
, gid
);
1312 static int setup_timezone(const char *dest
) {
1313 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
1314 const char *where
, *check
, *what
;
1320 /* Fix the timezone, if possible */
1321 r
= readlink_malloc("/etc/localtime", &p
);
1323 log_warning("host's /etc/localtime is not a symlink, not updating container timezone.");
1324 /* to handle warning, delete /etc/localtime and replace it
1325 * with a symbolic link to a time zone data file.
1328 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1333 z
= path_startswith(p
, "../usr/share/zoneinfo/");
1335 z
= path_startswith(p
, "/usr/share/zoneinfo/");
1337 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1341 where
= prefix_roota(dest
, "/etc/localtime");
1342 r
= readlink_malloc(where
, &q
);
1344 y
= path_startswith(q
, "../usr/share/zoneinfo/");
1346 y
= path_startswith(q
, "/usr/share/zoneinfo/");
1348 /* Already pointing to the right place? Then do nothing .. */
1349 if (y
&& streq(y
, z
))
1353 check
= strjoina("/usr/share/zoneinfo/", z
);
1354 check
= prefix_roota(dest
, check
);
1355 if (laccess(check
, F_OK
) < 0) {
1356 log_warning("Timezone %s does not exist in container, not updating container timezone.", z
);
1360 if (unlink(where
) < 0 && errno
!= ENOENT
) {
1361 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
, /* Don't complain on read-only images */
1363 "Failed to remove existing timezone info %s in container, ignoring: %m", where
);
1367 what
= strjoina("../usr/share/zoneinfo/", z
);
1368 if (symlink(what
, where
) < 0) {
1369 log_full_errno(IN_SET(errno
, EROFS
, EACCES
, EPERM
) ? LOG_DEBUG
: LOG_WARNING
,
1371 "Failed to correct timezone of container, ignoring: %m");
1375 r
= userns_lchown(where
, 0, 0);
1377 return log_warning_errno(r
, "Failed to chown /etc/localtime: %m");
1382 static int resolved_listening(void) {
1383 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
1384 _cleanup_free_
char *dns_stub_listener_mode
= NULL
;
1387 /* Check if resolved is listening */
1389 r
= sd_bus_open_system(&bus
);
1393 r
= bus_name_has_owner(bus
, "org.freedesktop.resolve1", NULL
);
1397 r
= sd_bus_get_property_string(bus
,
1398 "org.freedesktop.resolve1",
1399 "/org/freedesktop/resolve1",
1400 "org.freedesktop.resolve1.Manager",
1403 &dns_stub_listener_mode
);
1407 return STR_IN_SET(dns_stub_listener_mode
, "udp", "yes");
1410 static int setup_resolv_conf(const char *dest
) {
1411 _cleanup_free_
char *resolved
= NULL
, *etc
= NULL
;
1417 if (arg_private_network
)
1420 r
= chase_symlinks("/etc", dest
, CHASE_PREFIX_ROOT
, &etc
);
1422 log_warning_errno(r
, "Failed to resolve /etc path in container, ignoring: %m");
1426 where
= strjoina(etc
, "/resolv.conf");
1427 found
= chase_symlinks(where
, dest
, CHASE_NONEXISTENT
, &resolved
);
1429 log_warning_errno(found
, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1433 if (access(STATIC_RESOLV_CONF
, F_OK
) >= 0 &&
1434 resolved_listening() > 0) {
1436 /* resolved is enabled on the host. In this, case bind mount its static resolv.conf file into the
1437 * container, so that the container can use the host's resolver. Given that network namespacing is
1438 * disabled it's only natural of the container also uses the host's resolver. It also has the big
1439 * advantage that the container will be able to follow the host's DNS server configuration changes
1442 if (found
== 0) /* missing? */
1443 (void) touch(resolved
);
1445 r
= mount_verbose(LOG_DEBUG
, STATIC_RESOLV_CONF
, resolved
, NULL
, MS_BIND
, NULL
);
1447 return mount_verbose(LOG_ERR
, NULL
, resolved
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NODEV
, NULL
);
1450 /* If that didn't work, let's copy the file */
1451 r
= copy_file("/etc/resolv.conf", where
, O_TRUNC
|O_NOFOLLOW
, 0644, 0, COPY_REFLINK
);
1453 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1454 * resolved or something similar runs inside and the symlink points there.
1456 * If the disk image is read-only, there's also no point in complaining.
1458 log_full_errno(IN_SET(r
, -ELOOP
, -EROFS
, -EACCES
, -EPERM
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1459 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where
);
1463 r
= userns_lchown(where
, 0, 0);
1465 log_warning_errno(r
, "Failed to chown /etc/resolv.conf, ignoring: %m");
1470 static int setup_boot_id(void) {
1471 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1472 _cleanup_free_
char *path
= NULL
;
1473 sd_id128_t rnd
= SD_ID128_NULL
;
1477 /* Generate a new randomized boot ID, so that each boot-up of
1478 * the container gets a new one */
1480 r
= tempfn_random_child(NULL
, "proc-sys-kernel-random-boot-id", &path
);
1482 return log_error_errno(r
, "Failed to generate random boot ID path: %m");
1484 r
= sd_id128_randomize(&rnd
);
1486 return log_error_errno(r
, "Failed to generate random boot id: %m");
1488 r
= id128_write(path
, ID128_UUID
, rnd
, false);
1490 return log_error_errno(r
, "Failed to write boot id: %m");
1492 from
= TAKE_PTR(path
);
1493 to
= "/proc/sys/kernel/random/boot_id";
1495 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1499 return mount_verbose(LOG_ERR
, NULL
, to
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1502 static int copy_devnodes(const char *dest
) {
1504 static const char devnodes
[] =
1515 _cleanup_umask_ mode_t u
;
1521 /* Create /dev/net, so that we can create /dev/net/tun in it */
1522 if (userns_mkdir(dest
, "/dev/net", 0755, 0, 0) < 0)
1523 return log_error_errno(r
, "Failed to create /dev/net directory: %m");
1525 NULSTR_FOREACH(d
, devnodes
) {
1526 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
1529 from
= strappend("/dev/", d
);
1530 to
= prefix_root(dest
, from
);
1532 if (stat(from
, &st
) < 0) {
1534 if (errno
!= ENOENT
)
1535 return log_error_errno(errno
, "Failed to stat %s: %m", from
);
1537 } else if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
1539 log_error("%s is not a char or block device, cannot copy.", from
);
1543 if (mknod(to
, st
.st_mode
, st
.st_rdev
) < 0) {
1544 /* Explicitly warn the user when /dev is already populated. */
1545 if (errno
== EEXIST
)
1546 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
);
1548 return log_error_errno(errno
, "mknod(%s) failed: %m", to
);
1550 /* Some systems abusively restrict mknod but
1551 * allow bind mounts. */
1554 return log_error_errno(r
, "touch (%s) failed: %m", to
);
1555 r
= mount_verbose(LOG_DEBUG
, from
, to
, NULL
, MS_BIND
, NULL
);
1557 return log_error_errno(r
, "Both mknod and bind mount (%s) failed: %m", to
);
1560 r
= userns_lchown(to
, 0, 0);
1562 return log_error_errno(r
, "chown() of device node %s failed: %m", to
);
1569 static int setup_pts(const char *dest
) {
1570 _cleanup_free_
char *options
= NULL
;
1575 if (arg_selinux_apifs_context
)
1576 (void) asprintf(&options
,
1577 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
",context=\"%s\"",
1578 arg_uid_shift
+ TTY_GID
,
1579 arg_selinux_apifs_context
);
1582 (void) asprintf(&options
,
1583 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT
,
1584 arg_uid_shift
+ TTY_GID
);
1589 /* Mount /dev/pts itself */
1590 p
= prefix_roota(dest
, "/dev/pts");
1591 r
= mkdir_errno_wrapper(p
, 0755);
1593 return log_error_errno(r
, "Failed to create /dev/pts: %m");
1595 r
= mount_verbose(LOG_ERR
, "devpts", p
, "devpts", MS_NOSUID
|MS_NOEXEC
, options
);
1598 r
= userns_lchown(p
, 0, 0);
1600 return log_error_errno(r
, "Failed to chown /dev/pts: %m");
1602 /* Create /dev/ptmx symlink */
1603 p
= prefix_roota(dest
, "/dev/ptmx");
1604 if (symlink("pts/ptmx", p
) < 0)
1605 return log_error_errno(errno
, "Failed to create /dev/ptmx symlink: %m");
1606 r
= userns_lchown(p
, 0, 0);
1608 return log_error_errno(r
, "Failed to chown /dev/ptmx: %m");
1610 /* And fix /dev/pts/ptmx ownership */
1611 p
= prefix_roota(dest
, "/dev/pts/ptmx");
1612 r
= userns_lchown(p
, 0, 0);
1614 return log_error_errno(r
, "Failed to chown /dev/pts/ptmx: %m");
1619 static int setup_dev_console(const char *dest
, const char *console
) {
1620 _cleanup_umask_ mode_t u
;
1629 r
= chmod_and_chown(console
, 0600, arg_uid_shift
, arg_uid_shift
);
1631 return log_error_errno(r
, "Failed to correct access mode for TTY: %m");
1633 /* We need to bind mount the right tty to /dev/console since
1634 * ptys can only exist on pts file systems. To have something
1635 * to bind mount things on we create a empty regular file. */
1637 to
= prefix_roota(dest
, "/dev/console");
1640 return log_error_errno(r
, "touch() for /dev/console failed: %m");
1642 return mount_verbose(LOG_ERR
, console
, to
, NULL
, MS_BIND
, NULL
);
1645 static int setup_keyring(void) {
1646 key_serial_t keyring
;
1648 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
1649 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
1650 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
1651 * these system calls let's make sure we don't leak anything into the container. */
1653 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
1654 if (keyring
== -1) {
1655 if (errno
== ENOSYS
)
1656 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
1657 else if (IN_SET(errno
, EACCES
, EPERM
))
1658 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
1660 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
1666 static int setup_kmsg(int kmsg_socket
) {
1667 _cleanup_(unlink_and_freep
) char *from
= NULL
;
1668 _cleanup_free_
char *fifo
= NULL
;
1669 _cleanup_close_
int fd
= -1;
1670 _cleanup_umask_ mode_t u
;
1674 assert(kmsg_socket
>= 0);
1678 /* We create the kmsg FIFO as as temporary file in /tmp, but immediately delete it after bind mounting it to
1679 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
1680 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
1681 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
1683 r
= tempfn_random_child(NULL
, "proc-kmsg", &fifo
);
1685 return log_error_errno(r
, "Failed to generate kmsg path: %m");
1687 if (mkfifo(fifo
, 0600) < 0)
1688 return log_error_errno(errno
, "mkfifo() for /run/kmsg failed: %m");
1690 from
= TAKE_PTR(fifo
);
1693 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
1697 fd
= open(from
, O_RDWR
|O_NDELAY
|O_CLOEXEC
);
1699 return log_error_errno(errno
, "Failed to open fifo: %m");
1701 /* Store away the fd in the socket, so that it stays open as long as we run the child */
1702 r
= send_one_fd(kmsg_socket
, fd
, 0);
1704 return log_error_errno(r
, "Failed to send FIFO fd: %m");
1709 static int on_address_change(sd_netlink
*rtnl
, sd_netlink_message
*m
, void *userdata
) {
1710 union in_addr_union
*exposed
= userdata
;
1716 expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
1720 static int setup_hostname(void) {
1722 if ((arg_clone_ns_flags
& CLONE_NEWUTS
) == 0)
1725 if (sethostname_idempotent(arg_machine
) < 0)
1731 static int setup_journal(const char *directory
) {
1733 _cleanup_free_
char *d
= NULL
;
1739 /* Don't link journals in ephemeral mode */
1743 if (arg_link_journal
== LINK_NO
)
1746 try = arg_link_journal_try
|| arg_link_journal
== LINK_AUTO
;
1748 r
= sd_id128_get_machine(&this_id
);
1750 return log_error_errno(r
, "Failed to retrieve machine ID: %m");
1752 if (sd_id128_equal(arg_uuid
, this_id
)) {
1753 log_full(try ? LOG_WARNING
: LOG_ERR
,
1754 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid
, id
));
1760 r
= userns_mkdir(directory
, "/var", 0755, 0, 0);
1762 return log_error_errno(r
, "Failed to create /var: %m");
1764 r
= userns_mkdir(directory
, "/var/log", 0755, 0, 0);
1766 return log_error_errno(r
, "Failed to create /var/log: %m");
1768 r
= userns_mkdir(directory
, "/var/log/journal", 0755, 0, 0);
1770 return log_error_errno(r
, "Failed to create /var/log/journal: %m");
1772 (void) sd_id128_to_string(arg_uuid
, id
);
1774 p
= strjoina("/var/log/journal/", id
);
1775 q
= prefix_roota(directory
, p
);
1777 if (path_is_mount_point(p
, NULL
, 0) > 0) {
1781 log_error("%s: already a mount point, refusing to use for journal", p
);
1785 if (path_is_mount_point(q
, NULL
, 0) > 0) {
1789 log_error("%s: already a mount point, refusing to use for journal", q
);
1793 r
= readlink_and_make_absolute(p
, &d
);
1795 if (IN_SET(arg_link_journal
, LINK_GUEST
, LINK_AUTO
) &&
1798 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
1800 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
1805 return log_error_errno(errno
, "Failed to remove symlink %s: %m", p
);
1806 } else if (r
== -EINVAL
) {
1808 if (arg_link_journal
== LINK_GUEST
&&
1811 if (errno
== ENOTDIR
) {
1812 log_error("%s already exists and is neither a symlink nor a directory", p
);
1815 return log_error_errno(errno
, "Failed to remove %s: %m", p
);
1817 } else if (r
!= -ENOENT
)
1818 return log_error_errno(r
, "readlink(%s) failed: %m", p
);
1820 if (arg_link_journal
== LINK_GUEST
) {
1822 if (symlink(q
, p
) < 0) {
1824 log_debug_errno(errno
, "Failed to symlink %s to %s, skipping journal setup: %m", q
, p
);
1827 return log_error_errno(errno
, "Failed to symlink %s to %s: %m", q
, p
);
1830 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
1832 log_warning_errno(r
, "Failed to create directory %s: %m", q
);
1836 if (arg_link_journal
== LINK_HOST
) {
1837 /* don't create parents here — if the host doesn't have
1838 * permanent journal set up, don't force it here */
1840 r
= mkdir_errno_wrapper(p
, 0755);
1841 if (r
< 0 && r
!= -EEXIST
) {
1843 log_debug_errno(r
, "Failed to create %s, skipping journal setup: %m", p
);
1846 return log_error_errno(r
, "Failed to create %s: %m", p
);
1849 } else if (access(p
, F_OK
) < 0)
1852 if (dir_is_empty(q
) == 0)
1853 log_warning("%s is not empty, proceeding anyway.", q
);
1855 r
= userns_mkdir(directory
, p
, 0755, 0, 0);
1857 return log_error_errno(r
, "Failed to create %s: %m", q
);
1859 r
= mount_verbose(LOG_DEBUG
, p
, q
, NULL
, MS_BIND
, NULL
);
1861 return log_error_errno(errno
, "Failed to bind mount journal from host into guest: %m");
1866 static int drop_capabilities(void) {
1867 return capability_bounding_set_drop(arg_caps_retain
, false);
1870 static int reset_audit_loginuid(void) {
1871 _cleanup_free_
char *p
= NULL
;
1874 if ((arg_clone_ns_flags
& CLONE_NEWPID
) == 0)
1877 r
= read_one_line_file("/proc/self/loginuid", &p
);
1881 return log_error_errno(r
, "Failed to read /proc/self/loginuid: %m");
1883 /* Already reset? */
1884 if (streq(p
, "4294967295"))
1887 r
= write_string_file("/proc/self/loginuid", "4294967295", 0);
1890 "Failed to reset audit login UID. This probably means that your kernel is too\n"
1891 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1892 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1893 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1894 "using systemd-nspawn. Sleeping for 5s... (%m)");
1902 static int setup_propagate(const char *root
) {
1906 (void) mkdir_p("/run/systemd/nspawn/", 0755);
1907 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
1908 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
1909 (void) mkdir_p(p
, 0600);
1911 r
= userns_mkdir(root
, "/run/systemd", 0755, 0, 0);
1913 return log_error_errno(r
, "Failed to create /run/systemd: %m");
1915 r
= userns_mkdir(root
, "/run/systemd/nspawn", 0755, 0, 0);
1917 return log_error_errno(r
, "Failed to create /run/systemd/nspawn: %m");
1919 r
= userns_mkdir(root
, "/run/systemd/nspawn/incoming", 0600, 0, 0);
1921 return log_error_errno(r
, "Failed to create /run/systemd/nspawn/incoming: %m");
1923 q
= prefix_roota(root
, "/run/systemd/nspawn/incoming");
1924 r
= mount_verbose(LOG_ERR
, p
, q
, NULL
, MS_BIND
, NULL
);
1928 r
= mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
, NULL
);
1932 /* machined will MS_MOVE into that directory, and that's only
1933 * supported for non-shared mounts. */
1934 return mount_verbose(LOG_ERR
, NULL
, q
, NULL
, MS_SLAVE
, NULL
);
1937 static int setup_machine_id(const char *directory
) {
1938 const char *etc_machine_id
;
1942 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
1943 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
1944 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
1945 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
1946 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
1947 * container behaves nicely). */
1949 etc_machine_id
= prefix_roota(directory
, "/etc/machine-id");
1951 r
= id128_read(etc_machine_id
, ID128_PLAIN
, &id
);
1953 if (!IN_SET(r
, -ENOENT
, -ENOMEDIUM
)) /* If the file is missing or empty, we don't mind */
1954 return log_error_errno(r
, "Failed to read machine ID from container image: %m");
1956 if (sd_id128_is_null(arg_uuid
)) {
1957 r
= sd_id128_randomize(&arg_uuid
);
1959 return log_error_errno(r
, "Failed to acquire randomized machine UUID: %m");
1962 if (sd_id128_is_null(id
)) {
1963 log_error("Machine ID in container image is zero, refusing.");
1973 static int recursive_chown(const char *directory
, uid_t shift
, uid_t range
) {
1978 if (arg_userns_mode
== USER_NAMESPACE_NO
|| !arg_userns_chown
)
1981 r
= path_patch_uid(directory
, arg_uid_shift
, arg_uid_range
);
1982 if (r
== -EOPNOTSUPP
)
1983 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.");
1985 return log_error_errno(r
, "Upper 16 bits of root directory UID and GID do not match.");
1987 return log_error_errno(r
, "Failed to adjust UID/GID shift of OS tree: %m");
1989 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
1991 log_debug("Patched directory tree to match UID/GID range.");
1998 * < 0 : wait_for_terminate() failed to get the state of the
1999 * container, the container was terminated by a signal, or
2000 * failed for an unknown reason. No change is made to the
2001 * container argument.
2002 * > 0 : The program executed in the container terminated with an
2003 * error. The exit code of the program executed in the
2004 * container is returned. The container argument has been set
2005 * to CONTAINER_TERMINATED.
2006 * 0 : The container is being rebooted, has been shut down or exited
2007 * successfully. The container argument has been set to either
2008 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2010 * That is, success is indicated by a return value of zero, and an
2011 * error is indicated by a non-zero value.
2013 static int wait_for_container(pid_t pid
, ContainerStatus
*container
) {
2017 r
= wait_for_terminate(pid
, &status
);
2019 return log_warning_errno(r
, "Failed to wait for container: %m");
2021 switch (status
.si_code
) {
2024 if (status
.si_status
== 0)
2025 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s exited successfully.", arg_machine
);
2027 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s failed with error code %i.", arg_machine
, status
.si_status
);
2029 *container
= CONTAINER_TERMINATED
;
2030 return status
.si_status
;
2033 if (status
.si_status
== SIGINT
) {
2034 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s has been shut down.", arg_machine
);
2035 *container
= CONTAINER_TERMINATED
;
2038 } else if (status
.si_status
== SIGHUP
) {
2039 log_full(arg_quiet
? LOG_DEBUG
: LOG_INFO
, "Container %s is being rebooted.", arg_machine
);
2040 *container
= CONTAINER_REBOOTED
;
2046 log_error("Container %s terminated by signal %s.", arg_machine
, signal_to_string(status
.si_status
));
2050 log_error("Container %s failed due to unknown reason.", arg_machine
);
2055 static int on_orderly_shutdown(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
2058 pid
= PTR_TO_PID(userdata
);
2060 if (kill(pid
, arg_kill_signal
) >= 0) {
2061 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2062 sd_event_source_set_userdata(s
, NULL
);
2067 sd_event_exit(sd_event_source_get_event(s
), 0);
2071 static int on_sigchld(sd_event_source
*s
, const struct signalfd_siginfo
*ssi
, void *userdata
) {
2077 pid
= PTR_TO_PID(userdata
);
2082 if (waitid(P_ALL
, 0, &si
, WNOHANG
|WNOWAIT
|WEXITED
) < 0)
2083 return log_error_errno(errno
, "Failed to waitid(): %m");
2084 if (si
.si_pid
== 0) /* No pending children. */
2086 if (si
.si_pid
== pid
) {
2087 /* The main process we care for has exited. Return from
2088 * signal handler but leave the zombie. */
2089 sd_event_exit(sd_event_source_get_event(s
), 0);
2093 /* Reap all other children. */
2094 (void) waitid(P_PID
, si
.si_pid
, &si
, WNOHANG
|WEXITED
);
2100 static int on_request_stop(sd_bus_message
*m
, void *userdata
, sd_bus_error
*error
) {
2105 pid
= PTR_TO_PID(userdata
);
2107 if (arg_kill_signal
> 0) {
2108 log_info("Container termination requested. Attempting to halt container.");
2109 (void) kill(pid
, arg_kill_signal
);
2111 log_info("Container termination requested. Exiting.");
2112 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m
)), 0);
2118 static int determine_names(void) {
2121 if (arg_template
&& !arg_directory
&& arg_machine
) {
2123 /* If --template= was specified then we should not
2124 * search for a machine, but instead create a new one
2125 * in /var/lib/machine. */
2127 arg_directory
= strjoin("/var/lib/machines/", arg_machine
);
2132 if (!arg_image
&& !arg_directory
) {
2134 _cleanup_(image_unrefp
) Image
*i
= NULL
;
2136 r
= image_find(arg_machine
, &i
);
2138 return log_error_errno(r
, "Failed to find image for machine '%s': %m", arg_machine
);
2140 log_error("No image for machine '%s'.", arg_machine
);
2144 if (IN_SET(i
->type
, IMAGE_RAW
, IMAGE_BLOCK
))
2145 r
= free_and_strdup(&arg_image
, i
->path
);
2147 r
= free_and_strdup(&arg_directory
, i
->path
);
2152 arg_read_only
= arg_read_only
|| i
->read_only
;
2154 r
= safe_getcwd(&arg_directory
);
2156 return log_error_errno(r
, "Failed to determine current directory: %m");
2159 if (!arg_directory
&& !arg_image
) {
2160 log_error("Failed to determine path, please use -D or -i.");
2167 if (arg_directory
&& path_equal(arg_directory
, "/"))
2168 arg_machine
= gethostname_malloc();
2173 arg_machine
= strdup(basename(arg_image
));
2175 /* Truncate suffix if there is one */
2176 e
= endswith(arg_machine
, ".raw");
2180 arg_machine
= strdup(basename(arg_directory
));
2185 hostname_cleanup(arg_machine
);
2186 if (!machine_name_is_valid(arg_machine
)) {
2187 log_error("Failed to determine machine name automatically, please use -M.");
2191 if (arg_ephemeral
) {
2194 /* Add a random suffix when this is an
2195 * ephemeral machine, so that we can run many
2196 * instances at once without manually having
2197 * to specify -M each time. */
2199 if (asprintf(&b
, "%s-%016" PRIx64
, arg_machine
, random_u64()) < 0)
2210 static int chase_symlinks_and_update(char **p
, unsigned flags
) {
2219 r
= chase_symlinks(*p
, NULL
, flags
, &chased
);
2221 return log_error_errno(r
, "Failed to resolve path %s: %m", *p
);
2223 free_and_replace(*p
, chased
);
2224 return r
; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2227 static int determine_uid_shift(const char *directory
) {
2230 if (arg_userns_mode
== USER_NAMESPACE_NO
) {
2235 if (arg_uid_shift
== UID_INVALID
) {
2238 r
= stat(directory
, &st
);
2240 return log_error_errno(errno
, "Failed to determine UID base of %s: %m", directory
);
2242 arg_uid_shift
= st
.st_uid
& UINT32_C(0xffff0000);
2244 if (arg_uid_shift
!= (st
.st_gid
& UINT32_C(0xffff0000))) {
2245 log_error("UID and GID base of %s don't match.", directory
);
2249 arg_uid_range
= UINT32_C(0x10000);
2252 if (arg_uid_shift
> (uid_t
) -1 - arg_uid_range
) {
2253 log_error("UID base too high for UID range.");
2260 static int inner_child(
2262 const char *directory
,
2268 _cleanup_free_
char *home
= NULL
;
2271 const char *envp
[] = {
2272 "PATH=" DEFAULT_PATH_COMPAT
,
2273 NULL
, /* container */
2278 NULL
, /* container_uuid */
2279 NULL
, /* LISTEN_FDS */
2280 NULL
, /* LISTEN_PID */
2281 NULL
, /* NOTIFY_SOCKET */
2284 const char *exec_target
;
2286 _cleanup_strv_free_
char **env_use
= NULL
;
2291 assert(kmsg_socket
>= 0);
2293 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2294 /* Tell the parent, that it now can write the UID map. */
2295 (void) barrier_place(barrier
); /* #1 */
2297 /* Wait until the parent wrote the UID map */
2298 if (!barrier_place_and_sync(barrier
)) { /* #2 */
2299 log_error("Parent died too early");
2304 r
= reset_uid_gid();
2306 return log_error_errno(r
, "Couldn't become new root: %m");
2309 arg_mount_settings
| MOUNT_IN_USERNS
,
2312 arg_selinux_apifs_context
);
2316 if (!arg_network_namespace_path
&& arg_private_network
) {
2317 r
= unshare(CLONE_NEWNET
);
2319 return log_error_errno(errno
, "Failed to unshare network namespace: %m");
2321 /* Tell the parent that it can setup network interfaces. */
2322 (void) barrier_place(barrier
); /* #3 */
2325 r
= mount_sysfs(NULL
, arg_mount_settings
);
2329 /* Wait until we are cgroup-ified, so that we
2330 * can mount the right cgroup path writable */
2331 if (!barrier_place_and_sync(barrier
)) { /* #4 */
2332 log_error("Parent died too early");
2336 if (arg_use_cgns
&& cg_ns_supported()) {
2337 r
= unshare(CLONE_NEWCGROUP
);
2339 return log_error_errno(errno
, "Failed to unshare cgroup namespace: %m");
2342 arg_unified_cgroup_hierarchy
,
2343 arg_userns_mode
!= USER_NAMESPACE_NO
,
2346 arg_selinux_apifs_context
,
2351 r
= mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy
);
2356 r
= setup_boot_id();
2360 r
= setup_kmsg(kmsg_socket
);
2363 kmsg_socket
= safe_close(kmsg_socket
);
2368 return log_error_errno(errno
, "setsid() failed: %m");
2370 if (arg_private_network
)
2373 if (arg_expose_ports
) {
2374 r
= expose_port_send_rtnl(rtnl_socket
);
2377 rtnl_socket
= safe_close(rtnl_socket
);
2380 r
= drop_capabilities();
2382 return log_error_errno(r
, "drop_capabilities() failed: %m");
2386 if (arg_personality
!= PERSONALITY_INVALID
) {
2387 r
= safe_personality(arg_personality
);
2389 return log_error_errno(r
, "personality() failed: %m");
2390 } else if (secondary
) {
2391 r
= safe_personality(PER_LINUX32
);
2393 return log_error_errno(r
, "personality() failed: %m");
2397 if (arg_selinux_context
)
2398 if (setexeccon(arg_selinux_context
) < 0)
2399 return log_error_errno(errno
, "setexeccon(\"%s\") failed: %m", arg_selinux_context
);
2402 r
= change_uid_gid(arg_user
, &home
);
2406 /* LXC sets container=lxc, so follow the scheme here */
2407 envp
[n_env
++] = strjoina("container=", arg_container_service_name
);
2409 envp
[n_env
] = strv_find_prefix(environ
, "TERM=");
2413 if ((asprintf((char**)(envp
+ n_env
++), "HOME=%s", home
? home
: "/root") < 0) ||
2414 (asprintf((char**)(envp
+ n_env
++), "USER=%s", arg_user
? arg_user
: "root") < 0) ||
2415 (asprintf((char**)(envp
+ n_env
++), "LOGNAME=%s", arg_user
? arg_user
: "root") < 0))
2418 assert(!sd_id128_is_null(arg_uuid
));
2420 if (asprintf((char**)(envp
+ n_env
++), "container_uuid=%s", id128_to_uuid_string(arg_uuid
, as_uuid
)) < 0)
2423 if (fdset_size(fds
) > 0) {
2424 r
= fdset_cloexec(fds
, false);
2426 return log_error_errno(r
, "Failed to unset O_CLOEXEC for file descriptors.");
2428 if ((asprintf((char **)(envp
+ n_env
++), "LISTEN_FDS=%u", fdset_size(fds
)) < 0) ||
2429 (asprintf((char **)(envp
+ n_env
++), "LISTEN_PID=1") < 0))
2432 if (asprintf((char **)(envp
+ n_env
++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH
) < 0)
2435 env_use
= strv_env_merge(2, envp
, arg_setenv
);
2439 /* Let the parent know that we are ready and
2440 * wait until the parent is ready with the
2442 if (!barrier_place_and_sync(barrier
)) { /* #5 */
2443 log_error("Parent died too early");
2448 if (chdir(arg_chdir
) < 0)
2449 return log_error_errno(errno
, "Failed to change to specified working directory %s: %m", arg_chdir
);
2451 if (arg_start_mode
== START_PID2
) {
2452 r
= stub_pid1(arg_uuid
);
2457 /* Now, explicitly close the log, so that we
2458 * then can close all remaining fds. Closing
2459 * the log explicitly first has the benefit
2460 * that the logging subsystem knows about it,
2461 * and is thus ready to be reopened should we
2462 * need it again. Note that the other fds
2463 * closed here are at least the locking and
2466 (void) fdset_close_others(fds
);
2468 if (arg_start_mode
== START_BOOT
) {
2472 /* Automatically search for the init system */
2474 m
= strv_length(arg_parameters
);
2475 a
= newa(char*, m
+ 2);
2476 memcpy_safe(a
+ 1, arg_parameters
, m
* sizeof(char*));
2479 a
[0] = (char*) "/usr/lib/systemd/systemd";
2480 execve(a
[0], a
, env_use
);
2482 a
[0] = (char*) "/lib/systemd/systemd";
2483 execve(a
[0], a
, env_use
);
2485 a
[0] = (char*) "/sbin/init";
2486 execve(a
[0], a
, env_use
);
2488 exec_target
= "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2489 } else if (!strv_isempty(arg_parameters
)) {
2490 exec_target
= arg_parameters
[0];
2491 execvpe(arg_parameters
[0], arg_parameters
, env_use
);
2494 /* If we cannot change the directory, we'll end up in /, that is expected. */
2495 (void) chdir(home
?: "/root");
2497 execle("/bin/bash", "-bash", NULL
, env_use
);
2498 execle("/bin/sh", "-sh", NULL
, env_use
);
2500 exec_target
= "/bin/bash, /bin/sh";
2505 return log_error_errno(r
, "execv(%s) failed: %m", exec_target
);
2508 static int setup_sd_notify_child(void) {
2509 static const int one
= 1;
2511 union sockaddr_union sa
= {
2512 .sa
.sa_family
= AF_UNIX
,
2516 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2518 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
2520 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH
, 0755);
2521 (void) unlink(NSPAWN_NOTIFY_SOCKET_PATH
);
2523 strncpy(sa
.un
.sun_path
, NSPAWN_NOTIFY_SOCKET_PATH
, sizeof(sa
.un
.sun_path
)-1);
2524 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
2527 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
2530 r
= userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH
, 0, 0);
2533 return log_error_errno(r
, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH
": %m");
2536 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
2539 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
2545 static int outer_child(
2547 const char *directory
,
2548 const char *console
,
2549 DissectedImage
*dissected_image
,
2557 int uid_shift_socket
,
2558 int unified_cgroup_hierarchy_socket
,
2565 _cleanup_close_
int fd
= -1;
2570 assert(pid_socket
>= 0);
2571 assert(uuid_socket
>= 0);
2572 assert(notify_socket
>= 0);
2573 assert(kmsg_socket
>= 0);
2575 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
) < 0)
2576 return log_error_errno(errno
, "PR_SET_PDEATHSIG failed: %m");
2581 terminal
= open_terminal(console
, O_RDWR
);
2583 return log_error_errno(terminal
, "Failed to open console: %m");
2585 r
= rearrange_stdio(terminal
, terminal
, terminal
); /* invalidates 'terminal' on success and failure */
2587 return log_error_errno(r
, "Failed to move console to stdin/stdout/stderr: %m");
2590 r
= reset_audit_loginuid();
2594 /* Mark everything as slave, so that we still
2595 * receive mounts from the real root, but don't
2596 * propagate mounts to the real root. */
2597 r
= mount_verbose(LOG_ERR
, NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
);
2601 if (dissected_image
) {
2602 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
2603 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
2604 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
2605 * makes sure ESP partitions and userns are compatible. */
2607 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
2608 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
2613 r
= determine_uid_shift(directory
);
2617 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
2618 /* Let the parent know which UID shift we read from the image */
2619 l
= send(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), MSG_NOSIGNAL
);
2621 return log_error_errno(errno
, "Failed to send UID shift: %m");
2622 if (l
!= sizeof(arg_uid_shift
)) {
2623 log_error("Short write while sending UID shift.");
2627 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
2628 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2629 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2630 * not it will pick a different one, and send it back to us. */
2632 l
= recv(uid_shift_socket
, &arg_uid_shift
, sizeof(arg_uid_shift
), 0);
2634 return log_error_errno(errno
, "Failed to recv UID shift: %m");
2635 if (l
!= sizeof(arg_uid_shift
)) {
2636 log_error("Short read while receiving UID shift.");
2641 log_info("Selected user namespace base " UID_FMT
" and range " UID_FMT
".", arg_uid_shift
, arg_uid_range
);
2644 if (dissected_image
) {
2645 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
2646 r
= dissected_image_mount(dissected_image
, directory
, arg_uid_shift
,
2647 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
|DISSECT_IMAGE_DISCARD_ON_LOOP
|(arg_read_only
? DISSECT_IMAGE_READ_ONLY
: 0));
2652 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
2653 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
2655 r
= detect_unified_cgroup_hierarchy_from_image(directory
);
2659 l
= send(unified_cgroup_hierarchy_socket
, &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), MSG_NOSIGNAL
);
2661 return log_error_errno(errno
, "Failed to send cgroup mode: %m");
2662 if (l
!= sizeof(arg_unified_cgroup_hierarchy
)) {
2663 log_error("Short write while sending cgroup mode: %m");
2667 unified_cgroup_hierarchy_socket
= safe_close(unified_cgroup_hierarchy_socket
);
2670 /* Turn directory into bind mount */
2671 r
= mount_verbose(LOG_ERR
, directory
, directory
, NULL
, MS_BIND
|MS_REC
, NULL
);
2675 r
= setup_pivot_root(
2678 arg_pivot_root_old
);
2685 arg_userns_mode
!= USER_NAMESPACE_NO
,
2688 arg_selinux_context
);
2692 r
= setup_volatile_state(
2695 arg_userns_mode
!= USER_NAMESPACE_NO
,
2698 arg_selinux_context
);
2702 /* Mark everything as shared so our mounts get propagated down. This is
2703 * required to make new bind mounts available in systemd services
2704 * inside the containter that create a new mount namespace.
2705 * See https://github.com/systemd/systemd/issues/3860
2706 * Further submounts (such as /dev) done after this will inherit the
2707 * shared propagation mode. */
2708 r
= mount_verbose(LOG_ERR
, NULL
, directory
, NULL
, MS_SHARED
|MS_REC
, NULL
);
2712 r
= recursive_chown(directory
, arg_uid_shift
, arg_uid_range
);
2716 r
= base_filesystem_create(directory
, arg_uid_shift
, (gid_t
) arg_uid_shift
);
2720 if (arg_read_only
) {
2721 r
= bind_remount_recursive(directory
, true, NULL
);
2723 return log_error_errno(r
, "Failed to make tree read-only: %m");
2726 r
= mount_all(directory
,
2730 arg_selinux_apifs_context
);
2734 r
= copy_devnodes(directory
);
2738 dev_setup(directory
, arg_uid_shift
, arg_uid_shift
);
2740 r
= setup_pts(directory
);
2744 r
= setup_propagate(directory
);
2748 r
= setup_dev_console(directory
, console
);
2752 r
= setup_keyring();
2756 r
= setup_seccomp(arg_caps_retain
, arg_syscall_whitelist
, arg_syscall_blacklist
);
2760 r
= setup_timezone(directory
);
2764 r
= setup_resolv_conf(directory
);
2768 r
= setup_machine_id(directory
);
2772 r
= setup_journal(directory
);
2779 arg_n_custom_mounts
,
2780 arg_userns_mode
!= USER_NAMESPACE_NO
,
2783 arg_selinux_apifs_context
);
2787 if (!arg_use_cgns
|| !cg_ns_supported()) {
2790 arg_unified_cgroup_hierarchy
,
2791 arg_userns_mode
!= USER_NAMESPACE_NO
,
2794 arg_selinux_apifs_context
,
2800 r
= mount_move_root(directory
);
2802 return log_error_errno(r
, "Failed to move root directory: %m");
2804 fd
= setup_sd_notify_child();
2808 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
|
2809 arg_clone_ns_flags
|
2810 (arg_userns_mode
!= USER_NAMESPACE_NO
? CLONE_NEWUSER
: 0));
2812 return log_error_errno(errno
, "Failed to fork inner child: %m");
2814 pid_socket
= safe_close(pid_socket
);
2815 uuid_socket
= safe_close(uuid_socket
);
2816 notify_socket
= safe_close(notify_socket
);
2817 uid_shift_socket
= safe_close(uid_shift_socket
);
2819 /* The inner child has all namespaces that are
2820 * requested, so that we all are owned by the user if
2821 * user namespaces are turned on. */
2823 if (arg_network_namespace_path
) {
2824 r
= namespace_enter(-1, -1, netns_fd
, -1, -1);
2829 r
= inner_child(barrier
, directory
, secondary
, kmsg_socket
, rtnl_socket
, fds
);
2831 _exit(EXIT_FAILURE
);
2833 _exit(EXIT_SUCCESS
);
2836 l
= send(pid_socket
, &pid
, sizeof(pid
), MSG_NOSIGNAL
);
2838 return log_error_errno(errno
, "Failed to send PID: %m");
2839 if (l
!= sizeof(pid
)) {
2840 log_error("Short write while sending PID.");
2844 l
= send(uuid_socket
, &arg_uuid
, sizeof(arg_uuid
), MSG_NOSIGNAL
);
2846 return log_error_errno(errno
, "Failed to send machine ID: %m");
2847 if (l
!= sizeof(arg_uuid
)) {
2848 log_error("Short write while sending machine ID.");
2852 l
= send_one_fd(notify_socket
, fd
, 0);
2854 return log_error_errno(errno
, "Failed to send notify fd: %m");
2856 pid_socket
= safe_close(pid_socket
);
2857 uuid_socket
= safe_close(uuid_socket
);
2858 notify_socket
= safe_close(notify_socket
);
2859 kmsg_socket
= safe_close(kmsg_socket
);
2860 rtnl_socket
= safe_close(rtnl_socket
);
2861 netns_fd
= safe_close(netns_fd
);
2866 static int uid_shift_pick(uid_t
*shift
, LockFile
*ret_lock_file
) {
2867 bool tried_hashed
= false;
2868 unsigned n_tries
= 100;
2873 assert(ret_lock_file
);
2874 assert(arg_userns_mode
== USER_NAMESPACE_PICK
);
2875 assert(arg_uid_range
== 0x10000U
);
2879 (void) mkdir("/run/systemd/nspawn-uid", 0755);
2882 char lock_path
[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t
) + 1];
2883 _cleanup_(release_lock_file
) LockFile lf
= LOCK_FILE_INIT
;
2888 if (candidate
< CONTAINER_UID_BASE_MIN
|| candidate
> CONTAINER_UID_BASE_MAX
)
2890 if ((candidate
& UINT32_C(0xFFFF)) != 0)
2893 xsprintf(lock_path
, "/run/systemd/nspawn-uid/" UID_FMT
, candidate
);
2894 r
= make_lock_file(lock_path
, LOCK_EX
|LOCK_NB
, &lf
);
2895 if (r
== -EBUSY
) /* Range already taken by another nspawn instance */
2900 /* Make some superficial checks whether the range is currently known in the user database */
2901 if (getpwuid(candidate
))
2903 if (getpwuid(candidate
+ UINT32_C(0xFFFE)))
2905 if (getgrgid(candidate
))
2907 if (getgrgid(candidate
+ UINT32_C(0xFFFE)))
2910 *ret_lock_file
= lf
;
2911 lf
= (struct LockFile
) LOCK_FILE_INIT
;
2916 if (arg_machine
&& !tried_hashed
) {
2917 /* Try to hash the base from the container name */
2919 static const uint8_t hash_key
[] = {
2920 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
2921 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
2924 candidate
= (uid_t
) siphash24(arg_machine
, strlen(arg_machine
), hash_key
);
2926 tried_hashed
= true;
2928 random_bytes(&candidate
, sizeof(candidate
));
2930 candidate
= (candidate
% (CONTAINER_UID_BASE_MAX
- CONTAINER_UID_BASE_MIN
)) + CONTAINER_UID_BASE_MIN
;
2931 candidate
&= (uid_t
) UINT32_C(0xFFFF0000);
2935 static int setup_uid_map(pid_t pid
) {
2936 char uid_map
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t
) + 1], line
[DECIMAL_STR_MAX(uid_t
)*3+3+1];
2941 xsprintf(uid_map
, "/proc/" PID_FMT
"/uid_map", pid
);
2942 xsprintf(line
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", 0, arg_uid_shift
, arg_uid_range
);
2943 r
= write_string_file(uid_map
, line
, 0);
2945 return log_error_errno(r
, "Failed to write UID map: %m");
2947 /* We always assign the same UID and GID ranges */
2948 xsprintf(uid_map
, "/proc/" PID_FMT
"/gid_map", pid
);
2949 r
= write_string_file(uid_map
, line
, 0);
2951 return log_error_errno(r
, "Failed to write GID map: %m");
2956 static int nspawn_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2957 char buf
[NOTIFY_BUFFER_MAX
+1];
2959 struct iovec iovec
= {
2961 .iov_len
= sizeof(buf
)-1,
2964 struct cmsghdr cmsghdr
;
2965 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2966 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2968 struct msghdr msghdr
= {
2971 .msg_control
= &control
,
2972 .msg_controllen
= sizeof(control
),
2974 struct cmsghdr
*cmsg
;
2975 struct ucred
*ucred
= NULL
;
2977 pid_t inner_child_pid
;
2978 _cleanup_strv_free_
char **tags
= NULL
;
2982 inner_child_pid
= PTR_TO_PID(userdata
);
2984 if (revents
!= EPOLLIN
) {
2985 log_warning("Got unexpected poll event for notify fd.");
2989 n
= recvmsg(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
2991 if (IN_SET(errno
, EAGAIN
, EINTR
))
2994 return log_warning_errno(errno
, "Couldn't read notification socket: %m");
2996 cmsg_close_all(&msghdr
);
2998 CMSG_FOREACH(cmsg
, &msghdr
) {
2999 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
3000 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
3001 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
3003 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
3007 if (!ucred
|| ucred
->pid
!= inner_child_pid
) {
3008 log_debug("Received notify message without valid credentials. Ignoring.");
3012 if ((size_t) n
>= sizeof(buf
)) {
3013 log_warning("Received notify message exceeded maximum size. Ignoring.");
3018 tags
= strv_split(buf
, "\n\r");
3022 if (strv_find(tags
, "READY=1"))
3023 sd_notifyf(false, "READY=1\n");
3025 p
= strv_find_startswith(tags
, "STATUS=");
3027 sd_notifyf(false, "STATUS=Container running: %s", p
);
3032 static int setup_sd_notify_parent(sd_event
*event
, int fd
, pid_t
*inner_child_pid
, sd_event_source
**notify_event_source
) {
3035 r
= sd_event_add_io(event
, notify_event_source
, fd
, EPOLLIN
, nspawn_dispatch_notify_fd
, inner_child_pid
);
3037 return log_error_errno(r
, "Failed to allocate notify event source: %m");
3039 (void) sd_event_source_set_description(*notify_event_source
, "nspawn-notify");
3044 static int load_settings(void) {
3045 _cleanup_(settings_freep
) Settings
*settings
= NULL
;
3046 _cleanup_fclose_
FILE *f
= NULL
;
3047 _cleanup_free_
char *p
= NULL
;
3051 /* If all settings are masked, there's no point in looking for
3052 * the settings file */
3053 if ((arg_settings_mask
& _SETTINGS_MASK_ALL
) == _SETTINGS_MASK_ALL
)
3056 fn
= strjoina(arg_machine
, ".nspawn");
3058 /* We first look in the admin's directories in /etc and /run */
3059 FOREACH_STRING(i
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3060 _cleanup_free_
char *j
= NULL
;
3062 j
= strjoin(i
, "/", fn
);
3070 /* By default, we trust configuration from /etc and /run */
3071 if (arg_settings_trusted
< 0)
3072 arg_settings_trusted
= true;
3077 if (errno
!= ENOENT
)
3078 return log_error_errno(errno
, "Failed to open %s: %m", j
);
3082 /* After that, let's look for a file next to the
3083 * actual image we shall boot. */
3086 p
= file_in_same_dir(arg_image
, fn
);
3089 } else if (arg_directory
) {
3090 p
= file_in_same_dir(arg_directory
, fn
);
3097 if (!f
&& errno
!= ENOENT
)
3098 return log_error_errno(errno
, "Failed to open %s: %m", p
);
3100 /* By default, we do not trust configuration from /var/lib/machines */
3101 if (arg_settings_trusted
< 0)
3102 arg_settings_trusted
= false;
3109 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted
));
3111 r
= settings_load(f
, p
, &settings
);
3115 /* Copy over bits from the settings, unless they have been
3116 * explicitly masked by command line switches. */
3118 if ((arg_settings_mask
& SETTING_START_MODE
) == 0 &&
3119 settings
->start_mode
>= 0) {
3120 arg_start_mode
= settings
->start_mode
;
3121 strv_free_and_replace(arg_parameters
, settings
->parameters
);
3124 if ((arg_settings_mask
& SETTING_PIVOT_ROOT
) == 0 &&
3125 settings
->pivot_root_new
) {
3126 free_and_replace(arg_pivot_root_new
, settings
->pivot_root_new
);
3127 free_and_replace(arg_pivot_root_old
, settings
->pivot_root_old
);
3130 if ((arg_settings_mask
& SETTING_WORKING_DIRECTORY
) == 0 &&
3131 settings
->working_directory
)
3132 free_and_replace(arg_chdir
, settings
->working_directory
);
3134 if ((arg_settings_mask
& SETTING_ENVIRONMENT
) == 0 &&
3135 settings
->environment
)
3136 strv_free_and_replace(arg_setenv
, settings
->environment
);
3138 if ((arg_settings_mask
& SETTING_USER
) == 0 &&
3140 free_and_replace(arg_user
, settings
->user
);
3142 if ((arg_settings_mask
& SETTING_CAPABILITY
) == 0) {
3145 plus
= settings
->capability
;
3146 if (settings_private_network(settings
))
3147 plus
|= (1ULL << CAP_NET_ADMIN
);
3149 if (!arg_settings_trusted
&& plus
!= 0) {
3150 if (settings
->capability
!= 0)
3151 log_warning("Ignoring Capability= setting, file %s is not trusted.", p
);
3153 arg_caps_retain
|= plus
;
3155 arg_caps_retain
&= ~settings
->drop_capability
;
3158 if ((arg_settings_mask
& SETTING_KILL_SIGNAL
) == 0 &&
3159 settings
->kill_signal
> 0)
3160 arg_kill_signal
= settings
->kill_signal
;
3162 if ((arg_settings_mask
& SETTING_PERSONALITY
) == 0 &&
3163 settings
->personality
!= PERSONALITY_INVALID
)
3164 arg_personality
= settings
->personality
;
3166 if ((arg_settings_mask
& SETTING_MACHINE_ID
) == 0 &&
3167 !sd_id128_is_null(settings
->machine_id
)) {
3169 if (!arg_settings_trusted
)
3170 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p
);
3172 arg_uuid
= settings
->machine_id
;
3175 if ((arg_settings_mask
& SETTING_READ_ONLY
) == 0 &&
3176 settings
->read_only
>= 0)
3177 arg_read_only
= settings
->read_only
;
3179 if ((arg_settings_mask
& SETTING_VOLATILE_MODE
) == 0 &&
3180 settings
->volatile_mode
!= _VOLATILE_MODE_INVALID
)
3181 arg_volatile_mode
= settings
->volatile_mode
;
3183 if ((arg_settings_mask
& SETTING_CUSTOM_MOUNTS
) == 0 &&
3184 settings
->n_custom_mounts
> 0) {
3186 if (!arg_settings_trusted
)
3187 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p
);
3189 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
3190 arg_custom_mounts
= TAKE_PTR(settings
->custom_mounts
);
3191 arg_n_custom_mounts
= settings
->n_custom_mounts
;
3192 settings
->n_custom_mounts
= 0;
3196 if ((arg_settings_mask
& SETTING_NETWORK
) == 0 &&
3197 (settings
->private_network
>= 0 ||
3198 settings
->network_veth
>= 0 ||
3199 settings
->network_bridge
||
3200 settings
->network_zone
||
3201 settings
->network_interfaces
||
3202 settings
->network_macvlan
||
3203 settings
->network_ipvlan
||
3204 settings
->network_veth_extra
)) {
3206 if (!arg_settings_trusted
)
3207 log_warning("Ignoring network settings, file %s is not trusted.", p
);
3209 arg_network_veth
= settings_network_veth(settings
);
3210 arg_private_network
= settings_private_network(settings
);
3212 strv_free_and_replace(arg_network_interfaces
, settings
->network_interfaces
);
3213 strv_free_and_replace(arg_network_macvlan
, settings
->network_macvlan
);
3214 strv_free_and_replace(arg_network_ipvlan
, settings
->network_ipvlan
);
3215 strv_free_and_replace(arg_network_veth_extra
, settings
->network_veth_extra
);
3217 free_and_replace(arg_network_bridge
, settings
->network_bridge
);
3218 free_and_replace(arg_network_zone
, settings
->network_zone
);
3222 if ((arg_settings_mask
& SETTING_EXPOSE_PORTS
) == 0 &&
3223 settings
->expose_ports
) {
3225 if (!arg_settings_trusted
)
3226 log_warning("Ignoring Port= setting, file %s is not trusted.", p
);
3228 expose_port_free_all(arg_expose_ports
);
3229 arg_expose_ports
= TAKE_PTR(settings
->expose_ports
);
3233 if ((arg_settings_mask
& SETTING_USERNS
) == 0 &&
3234 settings
->userns_mode
!= _USER_NAMESPACE_MODE_INVALID
) {
3236 if (!arg_settings_trusted
)
3237 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", p
);
3239 arg_userns_mode
= settings
->userns_mode
;
3240 arg_uid_shift
= settings
->uid_shift
;
3241 arg_uid_range
= settings
->uid_range
;
3242 arg_userns_chown
= settings
->userns_chown
;
3246 if ((arg_settings_mask
& SETTING_NOTIFY_READY
) == 0)
3247 arg_notify_ready
= settings
->notify_ready
;
3249 if ((arg_settings_mask
& SETTING_SYSCALL_FILTER
) == 0) {
3251 if (!arg_settings_trusted
&& !strv_isempty(arg_syscall_whitelist
))
3252 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", p
);
3254 strv_free_and_replace(arg_syscall_whitelist
, settings
->syscall_whitelist
);
3255 strv_free_and_replace(arg_syscall_blacklist
, settings
->syscall_blacklist
);
3262 static int run(int master
,
3263 const char* console
,
3264 DissectedImage
*dissected_image
,
3268 char veth_name
[IFNAMSIZ
], bool *veth_created
,
3269 union in_addr_union
*exposed
,
3270 pid_t
*pid
, int *ret
) {
3272 static const struct sigaction sa
= {
3273 .sa_handler
= nop_signal_handler
,
3274 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
3277 _cleanup_(release_lock_file
) LockFile uid_shift_lock
= LOCK_FILE_INIT
;
3278 _cleanup_close_
int etc_passwd_lock
= -1;
3279 _cleanup_close_pair_
int
3280 kmsg_socket_pair
[2] = { -1, -1 },
3281 rtnl_socket_pair
[2] = { -1, -1 },
3282 pid_socket_pair
[2] = { -1, -1 },
3283 uuid_socket_pair
[2] = { -1, -1 },
3284 notify_socket_pair
[2] = { -1, -1 },
3285 uid_shift_socket_pair
[2] = { -1, -1 },
3286 unified_cgroup_hierarchy_socket_pair
[2] = { -1, -1};
3288 _cleanup_close_
int notify_socket
= -1;
3289 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
3290 _cleanup_(sd_event_source_unrefp
) sd_event_source
*notify_event_source
= NULL
;
3291 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
3292 _cleanup_(pty_forward_freep
) PTYForward
*forward
= NULL
;
3293 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
3294 _cleanup_(sd_bus_flush_close_unrefp
) sd_bus
*bus
= NULL
;
3295 ContainerStatus container_status
= 0;
3300 _cleanup_close_
int netns_fd
= -1;
3302 assert_se(sigemptyset(&mask_chld
) == 0);
3303 assert_se(sigaddset(&mask_chld
, SIGCHLD
) == 0);
3305 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3306 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3307 * check with getpwuid() if the specific user already exists. Note that /etc might be
3308 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3309 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3310 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3313 etc_passwd_lock
= take_etc_passwd_lock(NULL
);
3314 if (etc_passwd_lock
< 0 && etc_passwd_lock
!= -EROFS
)
3315 return log_error_errno(etc_passwd_lock
, "Failed to take /etc/passwd lock: %m");
3318 r
= barrier_create(&barrier
);
3320 return log_error_errno(r
, "Cannot initialize IPC barrier: %m");
3322 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, kmsg_socket_pair
) < 0)
3323 return log_error_errno(errno
, "Failed to create kmsg socket pair: %m");
3325 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, rtnl_socket_pair
) < 0)
3326 return log_error_errno(errno
, "Failed to create rtnl socket pair: %m");
3328 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, pid_socket_pair
) < 0)
3329 return log_error_errno(errno
, "Failed to create pid socket pair: %m");
3331 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uuid_socket_pair
) < 0)
3332 return log_error_errno(errno
, "Failed to create id socket pair: %m");
3334 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, notify_socket_pair
) < 0)
3335 return log_error_errno(errno
, "Failed to create notify socket pair: %m");
3337 if (arg_userns_mode
!= USER_NAMESPACE_NO
)
3338 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, uid_shift_socket_pair
) < 0)
3339 return log_error_errno(errno
, "Failed to create uid shift socket pair: %m");
3341 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
)
3342 if (socketpair(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0, unified_cgroup_hierarchy_socket_pair
) < 0)
3343 return log_error_errno(errno
, "Failed to create unified cgroup socket pair: %m");
3345 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3346 * parent's blocking calls and give it a chance to call wait() and terminate. */
3347 r
= sigprocmask(SIG_UNBLOCK
, &mask_chld
, NULL
);
3349 return log_error_errno(errno
, "Failed to change the signal mask: %m");
3351 r
= sigaction(SIGCHLD
, &sa
, NULL
);
3353 return log_error_errno(errno
, "Failed to install SIGCHLD handler: %m");
3355 if (arg_network_namespace_path
) {
3356 netns_fd
= open(arg_network_namespace_path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
3358 return log_error_errno(errno
, "Cannot open file %s: %m", arg_network_namespace_path
);
3360 r
= fd_is_network_ns(netns_fd
);
3361 if (r
< 0 && r
!= -ENOTTY
)
3362 return log_error_errno(r
, "Failed to check %s fs type: %m", arg_network_namespace_path
);
3364 log_error("Path %s doesn't refer to a network namespace", arg_network_namespace_path
);
3369 *pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
3371 return log_error_errno(errno
, "clone() failed%s: %m",
3373 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3376 /* The outer child only has a file system namespace. */
3377 barrier_set_role(&barrier
, BARRIER_CHILD
);
3379 master
= safe_close(master
);
3381 kmsg_socket_pair
[0] = safe_close(kmsg_socket_pair
[0]);
3382 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
3383 pid_socket_pair
[0] = safe_close(pid_socket_pair
[0]);
3384 uuid_socket_pair
[0] = safe_close(uuid_socket_pair
[0]);
3385 notify_socket_pair
[0] = safe_close(notify_socket_pair
[0]);
3386 uid_shift_socket_pair
[0] = safe_close(uid_shift_socket_pair
[0]);
3387 unified_cgroup_hierarchy_socket_pair
[0] = safe_close(unified_cgroup_hierarchy_socket_pair
[0]);
3389 (void) reset_all_signal_handlers();
3390 (void) reset_signal_mask();
3392 r
= outer_child(&barrier
,
3399 uuid_socket_pair
[1],
3400 notify_socket_pair
[1],
3401 kmsg_socket_pair
[1],
3402 rtnl_socket_pair
[1],
3403 uid_shift_socket_pair
[1],
3404 unified_cgroup_hierarchy_socket_pair
[1],
3408 _exit(EXIT_FAILURE
);
3410 _exit(EXIT_SUCCESS
);
3413 barrier_set_role(&barrier
, BARRIER_PARENT
);
3415 fds
= fdset_free(fds
);
3417 kmsg_socket_pair
[1] = safe_close(kmsg_socket_pair
[1]);
3418 rtnl_socket_pair
[1] = safe_close(rtnl_socket_pair
[1]);
3419 pid_socket_pair
[1] = safe_close(pid_socket_pair
[1]);
3420 uuid_socket_pair
[1] = safe_close(uuid_socket_pair
[1]);
3421 notify_socket_pair
[1] = safe_close(notify_socket_pair
[1]);
3422 uid_shift_socket_pair
[1] = safe_close(uid_shift_socket_pair
[1]);
3423 unified_cgroup_hierarchy_socket_pair
[1] = safe_close(unified_cgroup_hierarchy_socket_pair
[1]);
3425 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3426 /* The child just let us know the UID shift it might have read from the image. */
3427 l
= recv(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, 0);
3429 return log_error_errno(errno
, "Failed to read UID shift: %m");
3430 if (l
!= sizeof arg_uid_shift
) {
3431 log_error("Short read while reading UID shift.");
3435 if (arg_userns_mode
== USER_NAMESPACE_PICK
) {
3436 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3437 * image, but if that's already in use, pick a new one, and report back to the child,
3438 * which one we now picked. */
3440 r
= uid_shift_pick(&arg_uid_shift
, &uid_shift_lock
);
3442 return log_error_errno(r
, "Failed to pick suitable UID/GID range: %m");
3444 l
= send(uid_shift_socket_pair
[0], &arg_uid_shift
, sizeof arg_uid_shift
, MSG_NOSIGNAL
);
3446 return log_error_errno(errno
, "Failed to send UID shift: %m");
3447 if (l
!= sizeof arg_uid_shift
) {
3448 log_error("Short write while writing UID shift.");
3454 if (arg_unified_cgroup_hierarchy
== CGROUP_UNIFIED_UNKNOWN
) {
3455 /* The child let us know the support cgroup mode it might have read from the image. */
3456 l
= recv(unified_cgroup_hierarchy_socket_pair
[0], &arg_unified_cgroup_hierarchy
, sizeof(arg_unified_cgroup_hierarchy
), 0);
3458 return log_error_errno(errno
, "Failed to read cgroup mode: %m");
3459 if (l
!= sizeof(arg_unified_cgroup_hierarchy
)) {
3460 log_error("Short read while reading cgroup mode.");
3465 /* Wait for the outer child. */
3466 r
= wait_for_terminate_and_check("(sd-namespace)", *pid
, WAIT_LOG_ABNORMAL
);
3469 if (r
!= EXIT_SUCCESS
)
3472 /* And now retrieve the PID of the inner child. */
3473 l
= recv(pid_socket_pair
[0], pid
, sizeof *pid
, 0);
3475 return log_error_errno(errno
, "Failed to read inner child PID: %m");
3476 if (l
!= sizeof *pid
) {
3477 log_error("Short read while reading inner child PID.");
3481 /* We also retrieve container UUID in case it was generated by outer child */
3482 l
= recv(uuid_socket_pair
[0], &arg_uuid
, sizeof arg_uuid
, 0);
3484 return log_error_errno(errno
, "Failed to read container machine ID: %m");
3485 if (l
!= sizeof(arg_uuid
)) {
3486 log_error("Short read while reading container machined ID.");
3490 /* We also retrieve the socket used for notifications generated by outer child */
3491 notify_socket
= receive_one_fd(notify_socket_pair
[0], 0);
3492 if (notify_socket
< 0)
3493 return log_error_errno(notify_socket
,
3494 "Failed to receive notification socket from the outer child: %m");
3496 log_debug("Init process invoked as PID "PID_FMT
, *pid
);
3498 if (arg_userns_mode
!= USER_NAMESPACE_NO
) {
3499 if (!barrier_place_and_sync(&barrier
)) { /* #1 */
3500 log_error("Child died too early.");
3504 r
= setup_uid_map(*pid
);
3508 (void) barrier_place(&barrier
); /* #2 */
3511 if (arg_private_network
) {
3513 if (!arg_network_namespace_path
) {
3514 /* Wait until the child has unshared its network namespace. */
3515 if (!barrier_place_and_sync(&barrier
)) { /* #3 */
3516 log_error("Child died too early");
3521 r
= move_network_interfaces(*pid
, arg_network_interfaces
);
3525 if (arg_network_veth
) {
3526 r
= setup_veth(arg_machine
, *pid
, veth_name
,
3527 arg_network_bridge
|| arg_network_zone
);
3533 if (arg_network_bridge
) {
3534 /* Add the interface to a bridge */
3535 r
= setup_bridge(veth_name
, arg_network_bridge
, false);
3540 } else if (arg_network_zone
) {
3541 /* Add the interface to a bridge, possibly creating it */
3542 r
= setup_bridge(veth_name
, arg_network_zone
, true);
3550 r
= setup_veth_extra(arg_machine
, *pid
, arg_network_veth_extra
);
3554 /* We created the primary and extra veth links now; let's remember this, so that we know to
3555 remove them later on. Note that we don't bother with removing veth links that were created
3556 here when their setup failed half-way, because in that case the kernel should be able to
3557 remove them on its own, since they cannot be referenced by anything yet. */
3558 *veth_created
= true;
3560 r
= setup_macvlan(arg_machine
, *pid
, arg_network_macvlan
);
3564 r
= setup_ipvlan(arg_machine
, *pid
, arg_network_ipvlan
);
3569 if (arg_register
|| !arg_keep_unit
) {
3570 r
= sd_bus_default_system(&bus
);
3572 return log_error_errno(r
, "Failed to open system bus: %m");
3575 if (!arg_keep_unit
) {
3576 /* When a new scope is created for this container, then we'll be registered as its controller, in which
3577 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
3578 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
3580 r
= sd_bus_match_signal_async(
3583 "org.freedesktop.systemd1",
3585 "org.freedesktop.systemd1.Scope",
3587 on_request_stop
, NULL
, PID_TO_PTR(*pid
));
3589 return log_error_errno(r
, "Failed to request RequestStop match: %m");
3594 r
= register_machine(
3602 arg_custom_mounts
, arg_n_custom_mounts
,
3606 arg_container_service_name
);
3610 } else if (!arg_keep_unit
) {
3617 arg_custom_mounts
, arg_n_custom_mounts
,
3623 } else if (arg_slice
|| arg_property
)
3624 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
3626 r
= sync_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
3630 r
= create_subcgroup(*pid
, arg_keep_unit
, arg_unified_cgroup_hierarchy
);
3634 r
= chown_cgroup(*pid
, arg_unified_cgroup_hierarchy
, arg_uid_shift
);
3638 /* Notify the child that the parent is ready with all
3639 * its setup (including cgroup-ification), and that
3640 * the child can now hand over control to the code to
3641 * run inside the container. */
3642 (void) barrier_place(&barrier
); /* #4 */
3644 /* Block SIGCHLD here, before notifying child.
3645 * process_pty() will handle it with the other signals. */
3646 assert_se(sigprocmask(SIG_BLOCK
, &mask_chld
, NULL
) >= 0);
3648 /* Reset signal to default */
3649 r
= default_signals(SIGCHLD
, -1);
3651 return log_error_errno(r
, "Failed to reset SIGCHLD: %m");
3653 r
= sd_event_new(&event
);
3655 return log_error_errno(r
, "Failed to get default event source: %m");
3657 (void) sd_event_set_watchdog(event
, true);
3660 r
= sd_bus_attach_event(bus
, event
, 0);
3662 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
3665 r
= setup_sd_notify_parent(event
, notify_socket
, PID_TO_PTR(*pid
), ¬ify_event_source
);
3669 /* Let the child know that we are ready and wait that the child is completely ready now. */
3670 if (!barrier_place_and_sync(&barrier
)) { /* #5 */
3671 log_error("Child died too early.");
3675 /* At this point we have made use of the UID we picked, and thus nss-mymachines
3676 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
3677 etc_passwd_lock
= safe_close(etc_passwd_lock
);
3680 "STATUS=Container running.\n"
3681 "X_NSPAWN_LEADER_PID=" PID_FMT
, *pid
);
3682 if (!arg_notify_ready
)
3683 sd_notify(false, "READY=1\n");
3685 if (arg_kill_signal
> 0) {
3686 /* Try to kill the init system on SIGINT or SIGTERM */
3687 sd_event_add_signal(event
, NULL
, SIGINT
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
3688 sd_event_add_signal(event
, NULL
, SIGTERM
, on_orderly_shutdown
, PID_TO_PTR(*pid
));
3690 /* Immediately exit */
3691 sd_event_add_signal(event
, NULL
, SIGINT
, NULL
, NULL
);
3692 sd_event_add_signal(event
, NULL
, SIGTERM
, NULL
, NULL
);
3695 /* Exit when the child exits */
3696 sd_event_add_signal(event
, NULL
, SIGCHLD
, on_sigchld
, PID_TO_PTR(*pid
));
3698 if (arg_expose_ports
) {
3699 r
= expose_port_watch_rtnl(event
, rtnl_socket_pair
[0], on_address_change
, exposed
, &rtnl
);
3703 (void) expose_port_execute(rtnl
, arg_expose_ports
, exposed
);
3706 rtnl_socket_pair
[0] = safe_close(rtnl_socket_pair
[0]);
3708 r
= pty_forward_new(event
, master
,
3709 PTY_FORWARD_IGNORE_VHANGUP
| (interactive
? 0 : PTY_FORWARD_READ_ONLY
),
3712 return log_error_errno(r
, "Failed to create PTY forwarder: %m");
3714 r
= sd_event_loop(event
);
3716 return log_error_errno(r
, "Failed to run event loop: %m");
3718 pty_forward_get_last_char(forward
, &last_char
);
3720 forward
= pty_forward_free(forward
);
3722 if (!arg_quiet
&& last_char
!= '\n')
3725 /* Kill if it is not dead yet anyway */
3726 if (arg_register
&& !arg_keep_unit
&& bus
)
3727 terminate_machine(bus
, *pid
);
3729 /* Normally redundant, but better safe than sorry */
3730 (void) kill(*pid
, SIGKILL
);
3732 r
= wait_for_container(*pid
, &container_status
);
3736 /* We failed to wait for the container, or the container exited abnormally. */
3738 if (r
> 0 || container_status
== CONTAINER_TERMINATED
) {
3739 /* r > 0 → The container exited with a non-zero status.
3740 * As a special case, we need to replace 133 with a different value,
3741 * because 133 is special-cased in the service file to reboot the container.
3742 * otherwise → The container exited with zero status and a reboot was not requested.
3744 if (r
== EXIT_FORCE_RESTART
)
3745 r
= EXIT_FAILURE
; /* replace 133 with the general failure code */
3747 return 0; /* finito */
3750 /* CONTAINER_REBOOTED, loop again */
3752 if (arg_keep_unit
) {
3753 /* Special handling if we are running as a service: instead of simply
3754 * restarting the machine we want to restart the entire service, so let's
3755 * inform systemd about this with the special exit code 133. The service
3756 * file uses RestartForceExitStatus=133 so that this results in a full
3757 * nspawn restart. This is necessary since we might have cgroup parameters
3758 * set we want to have flushed out. */
3759 *ret
= EXIT_FORCE_RESTART
;
3760 return 0; /* finito */
3763 expose_port_flush(arg_expose_ports
, exposed
);
3765 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
3766 *veth_created
= false;
3767 return 1; /* loop again */
3770 int main(int argc
, char *argv
[]) {
3772 _cleanup_free_
char *console
= NULL
;
3773 _cleanup_close_
int master
= -1;
3774 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3775 int r
, n_fd_passed
, ret
= EXIT_SUCCESS
;
3776 char veth_name
[IFNAMSIZ
] = "";
3777 bool secondary
= false, remove_directory
= false, remove_image
= false;
3779 union in_addr_union exposed
= {};
3780 _cleanup_(release_lock_file
) LockFile tree_global_lock
= LOCK_FILE_INIT
, tree_local_lock
= LOCK_FILE_INIT
;
3781 bool interactive
, veth_created
= false, remove_tmprootdir
= false;
3782 char tmprootdir
[] = "/tmp/nspawn-root-XXXXXX";
3783 _cleanup_(loop_device_unrefp
) LoopDevice
*loop
= NULL
;
3784 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
3785 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3787 log_parse_environment();
3790 /* Make sure rename_process() in the stub init process can work */
3794 r
= parse_argv(argc
, argv
);
3802 r
= determine_names();
3806 r
= load_settings();
3810 r
= verify_arguments();
3814 r
= detect_unified_cgroup_hierarchy_from_environment();
3818 n_fd_passed
= sd_listen_fds(false);
3819 if (n_fd_passed
> 0) {
3820 r
= fdset_new_listen_fds(&fds
, false);
3822 log_error_errno(r
, "Failed to collect file descriptors: %m");
3827 if (arg_directory
) {
3830 if (path_equal(arg_directory
, "/") && !arg_ephemeral
) {
3831 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3836 if (arg_ephemeral
) {
3837 _cleanup_free_
char *np
= NULL
;
3839 r
= chase_symlinks_and_update(&arg_directory
, 0);
3843 /* If the specified path is a mount point we
3844 * generate the new snapshot immediately
3845 * inside it under a random name. However if
3846 * the specified is not a mount point we
3847 * create the new snapshot in the parent
3848 * directory, just next to it. */
3849 r
= path_is_mount_point(arg_directory
, NULL
, 0);
3851 log_error_errno(r
, "Failed to determine whether directory %s is mount point: %m", arg_directory
);
3855 r
= tempfn_random_child(arg_directory
, "machine.", &np
);
3857 r
= tempfn_random(arg_directory
, "machine.", &np
);
3859 log_error_errno(r
, "Failed to generate name for directory snapshot: %m");
3863 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3865 log_error_errno(r
, "Failed to lock %s: %m", np
);
3869 r
= btrfs_subvol_snapshot(arg_directory
, np
,
3870 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
3871 BTRFS_SNAPSHOT_FALLBACK_COPY
|
3872 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
3873 BTRFS_SNAPSHOT_RECURSIVE
|
3874 BTRFS_SNAPSHOT_QUOTA
);
3876 log_error_errno(r
, "Failed to create snapshot %s from %s: %m", np
, arg_directory
);
3880 free_and_replace(arg_directory
, np
);
3882 remove_directory
= true;
3885 r
= chase_symlinks_and_update(&arg_directory
, arg_template
? CHASE_NONEXISTENT
: 0);
3889 r
= image_path_lock(arg_directory
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3891 log_error_errno(r
, "Directory tree %s is currently busy.", arg_directory
);
3895 log_error_errno(r
, "Failed to lock %s: %m", arg_directory
);
3900 r
= chase_symlinks_and_update(&arg_template
, 0);
3904 r
= btrfs_subvol_snapshot(arg_template
, arg_directory
,
3905 (arg_read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
3906 BTRFS_SNAPSHOT_FALLBACK_COPY
|
3907 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
3908 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
3909 BTRFS_SNAPSHOT_RECURSIVE
|
3910 BTRFS_SNAPSHOT_QUOTA
);
3913 log_info("Directory %s already exists, not populating from template %s.", arg_directory
, arg_template
);
3915 log_error_errno(r
, "Couldn't create snapshot %s from %s: %m", arg_directory
, arg_template
);
3919 log_info("Populated %s from template %s.", arg_directory
, arg_template
);
3924 if (arg_start_mode
== START_BOOT
) {
3925 if (path_is_os_tree(arg_directory
) <= 0) {
3926 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory
);
3933 p
= strjoina(arg_directory
, "/usr/");
3934 if (laccess(p
, F_OK
) < 0) {
3935 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", arg_directory
);
3943 assert(!arg_template
);
3945 r
= chase_symlinks_and_update(&arg_image
, 0);
3949 if (arg_ephemeral
) {
3950 _cleanup_free_
char *np
= NULL
;
3952 r
= tempfn_random(arg_image
, "machine.", &np
);
3954 log_error_errno(r
, "Failed to generate name for image snapshot: %m");
3958 r
= image_path_lock(np
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3960 r
= log_error_errno(r
, "Failed to create image lock: %m");
3964 r
= copy_file(arg_image
, np
, O_EXCL
, arg_read_only
? 0400 : 0600, FS_NOCOW_FL
, COPY_REFLINK
);
3966 r
= log_error_errno(r
, "Failed to copy image file: %m");
3970 free_and_replace(arg_image
, np
);
3972 remove_image
= true;
3974 r
= image_path_lock(arg_image
, (arg_read_only
? LOCK_SH
: LOCK_EX
) | LOCK_NB
, &tree_global_lock
, &tree_local_lock
);
3976 r
= log_error_errno(r
, "Disk image %s is currently busy.", arg_image
);
3980 r
= log_error_errno(r
, "Failed to create image lock: %m");
3984 if (!arg_root_hash
) {
3985 r
= root_hash_load(arg_image
, &arg_root_hash
, &arg_root_hash_size
);
3987 log_error_errno(r
, "Failed to load root hash file for %s: %m", arg_image
);
3993 if (!mkdtemp(tmprootdir
)) {
3994 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
3998 remove_tmprootdir
= true;
4000 arg_directory
= strdup(tmprootdir
);
4001 if (!arg_directory
) {
4006 r
= loop_device_make_by_path(arg_image
, arg_read_only
? O_RDONLY
: O_RDWR
, &loop
);
4008 log_error_errno(r
, "Failed to set up loopback block device: %m");
4012 r
= dissect_image_and_warn(
4015 arg_root_hash
, arg_root_hash_size
,
4016 DISSECT_IMAGE_REQUIRE_ROOT
,
4019 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4020 log_notice("Note that the disk image needs to\n"
4021 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4022 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4023 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4024 " d) or contain a file system without a partition table\n"
4025 "in order to be bootable with systemd-nspawn.");
4031 if (!arg_root_hash
&& dissected_image
->can_verity
)
4032 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image
);
4034 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, arg_root_hash
, arg_root_hash_size
, 0, &decrypted_image
);
4038 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4039 if (remove_image
&& unlink(arg_image
) >= 0)
4040 remove_image
= false;
4043 r
= custom_mount_prepare_all(arg_directory
, arg_custom_mounts
, arg_n_custom_mounts
);
4048 isatty(STDIN_FILENO
) > 0 &&
4049 isatty(STDOUT_FILENO
) > 0;
4051 master
= posix_openpt(O_RDWR
|O_NOCTTY
|O_CLOEXEC
|O_NDELAY
);
4053 r
= log_error_errno(errno
, "Failed to acquire pseudo tty: %m");
4057 r
= ptsname_malloc(master
, &console
);
4059 r
= log_error_errno(r
, "Failed to determine tty name: %m");
4063 if (arg_selinux_apifs_context
) {
4064 r
= mac_selinux_apply(console
, arg_selinux_apifs_context
);
4069 if (unlockpt(master
) < 0) {
4070 r
= log_error_errno(errno
, "Failed to unlock tty: %m");
4075 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4076 arg_machine
, arg_image
?: arg_directory
);
4078 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGCHLD
, SIGWINCH
, SIGTERM
, SIGINT
, -1) >= 0);
4080 if (prctl(PR_SET_CHILD_SUBREAPER
, 1) < 0) {
4081 r
= log_error_errno(errno
, "Failed to become subreaper: %m");
4089 interactive
, secondary
,
4091 veth_name
, &veth_created
,
4100 r
== 0 && ret
== EXIT_FORCE_RESTART
? "STOPPING=1\nSTATUS=Restarting..." :
4101 "STOPPING=1\nSTATUS=Terminating...");
4104 (void) kill(pid
, SIGKILL
);
4106 /* Try to flush whatever is still queued in the pty */
4108 (void) copy_bytes(master
, STDOUT_FILENO
, (uint64_t) -1, 0);
4109 master
= safe_close(master
);
4113 (void) wait_for_terminate(pid
, NULL
);
4115 if (remove_directory
&& arg_directory
) {
4118 k
= rm_rf(arg_directory
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
4120 log_warning_errno(k
, "Cannot remove '%s', ignoring: %m", arg_directory
);
4123 if (remove_image
&& arg_image
) {
4124 if (unlink(arg_image
) < 0)
4125 log_warning_errno(errno
, "Can't remove image file '%s', ignoring: %m", arg_image
);
4128 if (remove_tmprootdir
) {
4129 if (rmdir(tmprootdir
) < 0)
4130 log_debug_errno(errno
, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir
);
4136 p
= strjoina("/run/systemd/nspawn/propagate/", arg_machine
);
4137 (void) rm_rf(p
, REMOVE_ROOT
);
4140 expose_port_flush(arg_expose_ports
, &exposed
);
4143 (void) remove_veth_links(veth_name
, arg_network_veth_extra
);
4144 (void) remove_bridge(arg_network_zone
);
4146 free(arg_directory
);
4151 free(arg_pivot_root_new
);
4152 free(arg_pivot_root_old
);
4154 strv_free(arg_setenv
);
4155 free(arg_network_bridge
);
4156 strv_free(arg_network_interfaces
);
4157 strv_free(arg_network_macvlan
);
4158 strv_free(arg_network_ipvlan
);
4159 strv_free(arg_network_veth_extra
);
4160 strv_free(arg_parameters
);
4161 custom_mount_free_all(arg_custom_mounts
, arg_n_custom_mounts
);
4162 expose_port_free_all(arg_expose_ports
);
4163 free(arg_root_hash
);
4165 return r
< 0 ? EXIT_FAILURE
: ret
;