]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:


ff4ccde1f88a52fc5f056511ecd9ca8d40571cb4
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #if HAVE_BLKID
4 #endif
5 #include <errno.h>
6 #include <getopt.h>
7 #include <linux/fs.h>
8 #include <linux/loop.h>
9 #if HAVE_SELINUX
10 #include <selinux/selinux.h>
11 #endif
12 #include <stdlib.h>
13 #include <sys/file.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.h>
18 #include <sys/wait.h>
19 #include <termios.h>
20 #include <unistd.h>
21
22 #include "sd-bus.h"
23 #include "sd-daemon.h"
24 #include "sd-id128.h"
25
26 #include "alloc-util.h"
27 #include "barrier.h"
28 #include "base-filesystem.h"
29 #include "blkid-util.h"
30 #include "btrfs-util.h"
31 #include "bus-error.h"
32 #include "bus-util.h"
33 #include "cap-list.h"
34 #include "capability-util.h"
35 #include "cgroup-util.h"
36 #include "chase-symlinks.h"
37 #include "copy.h"
38 #include "cpu-set-util.h"
39 #include "creds-util.h"
40 #include "dev-setup.h"
41 #include "discover-image.h"
42 #include "dissect-image.h"
43 #include "env-util.h"
44 #include "escape.h"
45 #include "fd-util.h"
46 #include "fdset.h"
47 #include "fileio.h"
48 #include "format-util.h"
49 #include "fs-util.h"
50 #include "gpt.h"
51 #include "hexdecoct.h"
52 #include "hostname-setup.h"
53 #include "hostname-util.h"
54 #include "id128-util.h"
55 #include "io-util.h"
56 #include "log.h"
57 #include "loop-util.h"
58 #include "loopback-setup.h"
59 #include "macro.h"
60 #include "main-func.h"
61 #include "missing_sched.h"
62 #include "mkdir.h"
63 #include "mount-util.h"
64 #include "mountpoint-util.h"
65 #include "namespace-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-bind-user.h"
68 #include "nspawn-cgroup.h"
69 #include "nspawn-creds.h"
70 #include "nspawn-def.h"
71 #include "nspawn-expose-ports.h"
72 #include "nspawn-mount.h"
73 #include "nspawn-network.h"
74 #include "nspawn-oci.h"
75 #include "nspawn-patch-uid.h"
76 #include "nspawn-register.h"
77 #include "nspawn-seccomp.h"
78 #include "nspawn-settings.h"
79 #include "nspawn-setuid.h"
80 #include "nspawn-stub-pid1.h"
81 #include "nspawn.h"
82 #include "nulstr-util.h"
83 #include "os-util.h"
84 #include "pager.h"
85 #include "parse-argument.h"
86 #include "parse-util.h"
87 #include "path-util.h"
88 #include "pretty-print.h"
89 #include "process-util.h"
90 #include "ptyfwd.h"
91 #include "random-util.h"
92 #include "raw-clone.h"
93 #include "resolve-util.h"
94 #include "rlimit-util.h"
95 #include "rm-rf.h"
96 #if HAVE_SECCOMP
97 #include "seccomp-util.h"
98 #endif
99 #include "selinux-util.h"
100 #include "signal-util.h"
101 #include "socket-util.h"
102 #include "stat-util.h"
103 #include "stdio-util.h"
104 #include "string-table.h"
105 #include "string-util.h"
106 #include "strv.h"
107 #include "sysctl-util.h"
108 #include "terminal-util.h"
109 #include "tmpfile-util.h"
110 #include "umask-util.h"
111 #include "unit-name.h"
112 #include "user-util.h"
113 #include "util.h"
114
115 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
116 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
117
118 #define EXIT_FORCE_RESTART 133
119
120 typedef enum ContainerStatus {
121 CONTAINER_TERMINATED,
122 CONTAINER_REBOOTED,
123 } ContainerStatus;
124
125 static char *arg_directory = NULL;
126 static char *arg_template = NULL;
127 static char *arg_chdir = NULL;
128 static char *arg_pivot_root_new = NULL;
129 static char *arg_pivot_root_old = NULL;
130 static char *arg_user = NULL;
131 static uid_t arg_uid = UID_INVALID;
132 static gid_t arg_gid = GID_INVALID;
133 static gid_t* arg_supplementary_gids = NULL;
134 static size_t arg_n_supplementary_gids = 0;
135 static sd_id128_t arg_uuid = {};
136 static char *arg_machine = NULL; /* The name used by the host to refer to this */
137 static char *arg_hostname = NULL; /* The name the payload sees by default */
138 static const char *arg_selinux_context = NULL;
139 static const char *arg_selinux_apifs_context = NULL;
140 static char *arg_slice = NULL;
141 static bool arg_private_network = false;
142 static bool arg_read_only = false;
143 static StartMode arg_start_mode = START_PID1;
144 static bool arg_ephemeral = false;
145 static LinkJournal arg_link_journal = LINK_AUTO;
146 static bool arg_link_journal_try = false;
147 static uint64_t arg_caps_retain =
148 (1ULL << CAP_AUDIT_CONTROL) |
149 (1ULL << CAP_AUDIT_WRITE) |
150 (1ULL << CAP_CHOWN) |
151 (1ULL << CAP_DAC_OVERRIDE) |
152 (1ULL << CAP_DAC_READ_SEARCH) |
153 (1ULL << CAP_FOWNER) |
154 (1ULL << CAP_FSETID) |
155 (1ULL << CAP_IPC_OWNER) |
156 (1ULL << CAP_KILL) |
157 (1ULL << CAP_LEASE) |
158 (1ULL << CAP_LINUX_IMMUTABLE) |
159 (1ULL << CAP_MKNOD) |
160 (1ULL << CAP_NET_BIND_SERVICE) |
161 (1ULL << CAP_NET_BROADCAST) |
162 (1ULL << CAP_NET_RAW) |
163 (1ULL << CAP_SETFCAP) |
164 (1ULL << CAP_SETGID) |
165 (1ULL << CAP_SETPCAP) |
166 (1ULL << CAP_SETUID) |
167 (1ULL << CAP_SYS_ADMIN) |
168 (1ULL << CAP_SYS_BOOT) |
169 (1ULL << CAP_SYS_CHROOT) |
170 (1ULL << CAP_SYS_NICE) |
171 (1ULL << CAP_SYS_PTRACE) |
172 (1ULL << CAP_SYS_RESOURCE) |
173 (1ULL << CAP_SYS_TTY_CONFIG);
174 static uint64_t arg_caps_ambient = 0;
175 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
176 static CustomMount *arg_custom_mounts = NULL;
177 static size_t arg_n_custom_mounts = 0;
178 static char **arg_setenv = NULL;
179 static bool arg_quiet = false;
180 static bool arg_register = true;
181 static bool arg_keep_unit = false;
182 static char **arg_network_interfaces = NULL;
183 static char **arg_network_macvlan = NULL;
184 static char **arg_network_ipvlan = NULL;
185 static bool arg_network_veth = false;
186 static char **arg_network_veth_extra = NULL;
187 static char *arg_network_bridge = NULL;
188 static char *arg_network_zone = NULL;
189 static char *arg_network_namespace_path = NULL;
190 static PagerFlags arg_pager_flags = 0;
191 static unsigned long arg_personality = PERSONALITY_INVALID;
192 static char *arg_image = NULL;
193 static char *arg_oci_bundle = NULL;
194 static VolatileMode arg_volatile_mode = VOLATILE_NO;
195 static ExposePort *arg_expose_ports = NULL;
196 static char **arg_property = NULL;
197 static sd_bus_message *arg_property_message = NULL;
198 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
199 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
200 static UserNamespaceOwnership arg_userns_ownership = _USER_NAMESPACE_OWNERSHIP_INVALID;
201 static int arg_kill_signal = 0;
202 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
203 static SettingsMask arg_settings_mask = 0;
204 static int arg_settings_trusted = -1;
205 static char **arg_parameters = NULL;
206 static const char *arg_container_service_name = "systemd-nspawn";
207 static bool arg_notify_ready = false;
208 static bool arg_use_cgns = true;
209 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
210 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
211 static VeritySettings arg_verity_settings = VERITY_SETTINGS_DEFAULT;
212 static char **arg_syscall_allow_list = NULL;
213 static char **arg_syscall_deny_list = NULL;
214 #if HAVE_SECCOMP
215 static scmp_filter_ctx arg_seccomp = NULL;
216 #endif
217 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
218 static bool arg_no_new_privileges = false;
219 static int arg_oom_score_adjust = 0;
220 static bool arg_oom_score_adjust_set = false;
221 static CPUSet arg_cpu_set = {};
222 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
223 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
224 static unsigned arg_console_width = UINT_MAX, arg_console_height = UINT_MAX;
225 static DeviceNode* arg_extra_nodes = NULL;
226 static size_t arg_n_extra_nodes = 0;
227 static char **arg_sysctl = NULL;
228 static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
229 static Credential *arg_credentials = NULL;
230 static size_t arg_n_credentials = 0;
231 static char **arg_bind_user = NULL;
232 static bool arg_suppress_sync = false;
233
234 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
235 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
236 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
237 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
238 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
239 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
240 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
241 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
252 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
253 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
254 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
255 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
256 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
257 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings, verity_settings_done);
258 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list, strv_freep);
259 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list, strv_freep);
260 #if HAVE_SECCOMP
261 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
262 #endif
263 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
264 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
265 STATIC_DESTRUCTOR_REGISTER(arg_bind_user, strv_freep);
266
267 static int handle_arg_console(const char *arg) {
268 if (streq(arg, "help")) {
269 puts("autopipe\n"
270 "interactive\n"
271 "passive\n"
272 "pipe\n"
273 "read-only");
274 return 0;
275 }
276
277 if (streq(arg, "interactive"))
278 arg_console_mode = CONSOLE_INTERACTIVE;
279 else if (streq(arg, "read-only"))
280 arg_console_mode = CONSOLE_READ_ONLY;
281 else if (streq(arg, "passive"))
282 arg_console_mode = CONSOLE_PASSIVE;
283 else if (streq(arg, "pipe")) {
284 if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
285 log_full(arg_quiet ? LOG_DEBUG : LOG_NOTICE,
286 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
287 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
288 "Proceeding anyway.");
289
290 arg_console_mode = CONSOLE_PIPE;
291 } else if (streq(arg, "autopipe")) {
292 if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
293 arg_console_mode = CONSOLE_INTERACTIVE;
294 else
295 arg_console_mode = CONSOLE_PIPE;
296 } else
297 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
298
299 arg_settings_mask |= SETTING_CONSOLE_MODE;
300 return 1;
301 }
302
303 static int help(void) {
304 _cleanup_free_ char *link = NULL;
305 int r;
306
307 pager_open(arg_pager_flags);
308
309 r = terminal_urlify_man("systemd-nspawn", "1", &link);
310 if (r < 0)
311 return log_oom();
312
313 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
314 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
315 " -h --help Show this help\n"
316 " --version Print version string\n"
317 " -q --quiet Do not show status information\n"
318 " --no-pager Do not pipe output into a pager\n"
319 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
320 "%3$sImage:%4$s\n"
321 " -D --directory=PATH Root directory for the container\n"
322 " --template=PATH Initialize root directory from template directory,\n"
323 " if missing\n"
324 " -x --ephemeral Run container with snapshot of root directory, and\n"
325 " remove it after exit\n"
326 " -i --image=PATH Root file system disk image (or device node) for\n"
327 " the container\n"
328 " --oci-bundle=PATH OCI bundle directory\n"
329 " --read-only Mount the root directory read-only\n"
330 " --volatile[=MODE] Run the system in volatile mode\n"
331 " --root-hash=HASH Specify verity root hash for root disk image\n"
332 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
333 " as a DER encoded PKCS7, either as a path to a file\n"
334 " or as an ASCII base64 encoded string prefixed by\n"
335 " 'base64:'\n"
336 " --verity-data=PATH Specify hash device for verity\n"
337 " --pivot-root=PATH[:PATH]\n"
338 " Pivot root to given directory in the container\n\n"
339 "%3$sExecution:%4$s\n"
340 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
341 " -b --boot Boot up full system (i.e. invoke init)\n"
342 " --chdir=PATH Set working directory in the container\n"
343 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
344 " -u --user=USER Run the command under specified user or UID\n"
345 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
346 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
347 " --suppress-sync=BOOLEAN\n"
348 " Suppress any form of disk data synchronization\n\n"
349 "%3$sSystem Identity:%4$s\n"
350 " -M --machine=NAME Set the machine name for the container\n"
351 " --hostname=NAME Override the hostname for the container\n"
352 " --uuid=UUID Set a specific machine UUID for the container\n\n"
353 "%3$sProperties:%4$s\n"
354 " -S --slice=SLICE Place the container in the specified slice\n"
355 " --property=NAME=VALUE Set scope unit property\n"
356 " --register=BOOLEAN Register container as machine\n"
357 " --keep-unit Do not register a scope for the machine, reuse\n"
358 " the service unit nspawn is running in\n\n"
359 "%3$sUser Namespacing:%4$s\n"
360 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
361 " --private-users[=UIDBASE[:NUIDS]]\n"
362 " Similar, but with user configured UID/GID range\n"
363 " --private-users-ownership=MODE\n"
364 " Adjust ('chown') or map ('map') OS tree ownership\n"
365 " to private UID/GID range\n\n"
366 "%3$sNetworking:%4$s\n"
367 " --private-network Disable network in container\n"
368 " --network-interface=INTERFACE\n"
369 " Assign an existing network interface to the\n"
370 " container\n"
371 " --network-macvlan=INTERFACE\n"
372 " Create a macvlan network interface based on an\n"
373 " existing network interface to the container\n"
374 " --network-ipvlan=INTERFACE\n"
375 " Create an ipvlan network interface based on an\n"
376 " existing network interface to the container\n"
377 " -n --network-veth Add a virtual Ethernet connection between host\n"
378 " and container\n"
379 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
380 " Add an additional virtual Ethernet link between\n"
381 " host and container\n"
382 " --network-bridge=INTERFACE\n"
383 " Add a virtual Ethernet connection to the container\n"
384 " and attach it to an existing bridge on the host\n"
385 " --network-zone=NAME Similar, but attach the new interface to an\n"
386 " an automatically managed bridge interface\n"
387 " --network-namespace-path=PATH\n"
388 " Set network namespace to the one represented by\n"
389 " the specified kernel namespace file node\n"
390 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
391 " Expose a container IP port on the host\n\n"
392 "%3$sSecurity:%4$s\n"
393 " --capability=CAP In addition to the default, retain specified\n"
394 " capability\n"
395 " --drop-capability=CAP Drop the specified capability from the default set\n"
396 " --ambient-capability=CAP\n"
397 " Sets the specified capability for the started\n"
398 " process. Not useful if booting a machine.\n"
399 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
400 " --system-call-filter=LIST|~LIST\n"
401 " Permit/prohibit specific system calls\n"
402 " -Z --selinux-context=SECLABEL\n"
403 " Set the SELinux security context to be used by\n"
404 " processes in the container\n"
405 " -L --selinux-apifs-context=SECLABEL\n"
406 " Set the SELinux security context to be used by\n"
407 " API/tmpfs file systems in the container\n\n"
408 "%3$sResources:%4$s\n"
409 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
410 " --oom-score-adjust=VALUE\n"
411 " Adjust the OOM score value for the payload\n"
412 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
413 " --personality=ARCH Pick personality for this container\n\n"
414 "%3$sIntegration:%4$s\n"
415 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
416 " --timezone=MODE Select mode of /etc/localtime initialization\n"
417 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
418 " host, try-guest, try-host\n"
419 " -j Equivalent to --link-journal=try-guest\n\n"
420 "%3$sMounts:%4$s\n"
421 " --bind=PATH[:PATH[:OPTIONS]]\n"
422 " Bind mount a file or directory from the host into\n"
423 " the container\n"
424 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
425 " Similar, but creates a read-only bind mount\n"
426 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
427 " it\n"
428 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
429 " --overlay=PATH[:PATH...]:PATH\n"
430 " Create an overlay mount from the host to \n"
431 " the container\n"
432 " --overlay-ro=PATH[:PATH...]:PATH\n"
433 " Similar, but creates a read-only overlay mount\n"
434 " --bind-user=NAME Bind user from host to container\n\n"
435 "%3$sInput/Output:%4$s\n"
436 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
437 " set up for the container.\n"
438 " -P --pipe Equivalent to --console=pipe\n\n"
439 "%3$sCredentials:%4$s\n"
440 " --set-credential=ID:VALUE\n"
441 " Pass a credential with literal value to container.\n"
442 " --load-credential=ID:PATH\n"
443 " Load credential to pass to container from file or\n"
444 " AF_UNIX stream socket.\n"
445 "\nSee the %2$s for details.\n",
446 program_invocation_short_name,
447 link,
448 ansi_underline(),
449 ansi_normal(),
450 ansi_highlight(),
451 ansi_normal());
452
453 return 0;
454 }
455
456 static int custom_mount_check_all(void) {
457 size_t i;
458
459 for (i = 0; i < arg_n_custom_mounts; i++) {
460 CustomMount *m = &arg_custom_mounts[i];
461
462 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
463 if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_OFF)
464 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
465 "--private-users-ownership=own may not be combined with custom root mounts.");
466 if (arg_uid_shift == UID_INVALID)
467 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
468 "--private-users with automatic UID shift may not be combined with custom root mounts.");
469 }
470 }
471
472 return 0;
473 }
474
475 static int detect_unified_cgroup_hierarchy_from_environment(void) {
476 const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
477 int r;
478
479 /* Allow the user to control whether the unified hierarchy is used */
480
481 e = getenv(var);
482 if (!e) {
483 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
484 var = "UNIFIED_CGROUP_HIERARCHY";
485 e = getenv(var);
486 }
487
488 if (!isempty(e)) {
489 r = parse_boolean(e);
490 if (r < 0)
491 return log_error_errno(r, "Failed to parse $%s: %m", var);
492 if (r > 0)
493 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
494 else
495 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
496 }
497
498 return 0;
499 }
500
501 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
502 int r;
503
504 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
505 * in the image actually supports. */
506 r = cg_all_unified();
507 if (r < 0)
508 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
509 if (r > 0) {
510 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
511 * routine only detects 231, so we'll have a false negative here for 230. */
512 r = systemd_installation_has_version(directory, 230);
513 if (r < 0)
514 return log_error_errno(r, "Failed to determine systemd version in container: %m");
515 if (r > 0)
516 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
517 else
518 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
519 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
520 /* Mixed cgroup hierarchy support was added in 233 */
521 r = systemd_installation_has_version(directory, 233);
522 if (r < 0)
523 return log_error_errno(r, "Failed to determine systemd version in container: %m");
524 if (r > 0)
525 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
526 else
527 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
528 } else
529 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
530
531 log_debug("Using %s hierarchy for container.",
532 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
533 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
534
535 return 0;
536 }
537
538 static int parse_capability_spec(const char *spec, uint64_t *ret_mask) {
539 uint64_t mask = 0;
540 int r;
541
542 for (;;) {
543 _cleanup_free_ char *t = NULL;
544
545 r = extract_first_word(&spec, &t, ",", 0);
546 if (r < 0)
547 return log_error_errno(r, "Failed to parse capability %s.", t);
548 if (r == 0)
549 break;
550
551 if (streq(t, "help")) {
552 for (int i = 0; i < capability_list_length(); i++) {
553 const char *name;
554
555 name = capability_to_name(i);
556 if (name)
557 puts(name);
558 }
559
560 return 0; /* quit */
561 }
562
563 if (streq(t, "all"))
564 mask = UINT64_MAX;
565 else {
566 r = capability_from_name(t);
567 if (r < 0)
568 return log_error_errno(r, "Failed to parse capability %s.", t);
569
570 mask |= 1ULL << r;
571 }
572 }
573
574 *ret_mask = mask;
575 return 1; /* continue */
576 }
577
578 static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
579 int r;
580
581 r = getenv_bool(name);
582 if (r == -ENXIO)
583 return 0;
584 if (r < 0)
585 return log_error_errno(r, "Failed to parse $%s: %m", name);
586
587 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
588 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
589 return 0;
590 }
591
592 static int parse_mount_settings_env(void) {
593 const char *e;
594 int r;
595
596 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
597 if (r < 0 && r != -ENXIO)
598 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
599 if (r >= 0)
600 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
601
602 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
603 if (streq_ptr(e, "network"))
604 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
605
606 else if (e) {
607 r = parse_boolean(e);
608 if (r < 0)
609 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
610
611 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
612 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
613 }
614
615 return 0;
616 }
617
618 static int parse_environment(void) {
619 const char *e;
620 int r;
621
622 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
623 if (r < 0)
624 return r;
625 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
626 if (r < 0)
627 return r;
628 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
629 if (r < 0)
630 return r;
631 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
632 if (r < 0)
633 return r;
634
635 r = parse_mount_settings_env();
636 if (r < 0)
637 return r;
638
639 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
640 * even if it is supported. If not supported, it has no effect. */
641 if (!cg_ns_supported())
642 arg_use_cgns = false;
643 else {
644 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
645 if (r < 0) {
646 if (r != -ENXIO)
647 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
648
649 arg_use_cgns = true;
650 } else {
651 arg_use_cgns = r > 0;
652 arg_settings_mask |= SETTING_USE_CGNS;
653 }
654 }
655
656 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
657 if (e)
658 arg_container_service_name = e;
659
660 r = getenv_bool("SYSTEMD_SUPPRESS_SYNC");
661 if (r >= 0)
662 arg_suppress_sync = r;
663 else if (r != -ENXIO)
664 log_debug_errno(r, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
665
666 return detect_unified_cgroup_hierarchy_from_environment();
667 }
668
669 static int parse_argv(int argc, char *argv[]) {
670 enum {
671 ARG_VERSION = 0x100,
672 ARG_PRIVATE_NETWORK,
673 ARG_UUID,
674 ARG_READ_ONLY,
675 ARG_CAPABILITY,
676 ARG_AMBIENT_CAPABILITY,
677 ARG_DROP_CAPABILITY,
678 ARG_LINK_JOURNAL,
679 ARG_BIND,
680 ARG_BIND_RO,
681 ARG_TMPFS,
682 ARG_OVERLAY,
683 ARG_OVERLAY_RO,
684 ARG_INACCESSIBLE,
685 ARG_SHARE_SYSTEM,
686 ARG_REGISTER,
687 ARG_KEEP_UNIT,
688 ARG_NETWORK_INTERFACE,
689 ARG_NETWORK_MACVLAN,
690 ARG_NETWORK_IPVLAN,
691 ARG_NETWORK_BRIDGE,
692 ARG_NETWORK_ZONE,
693 ARG_NETWORK_VETH_EXTRA,
694 ARG_NETWORK_NAMESPACE_PATH,
695 ARG_PERSONALITY,
696 ARG_VOLATILE,
697 ARG_TEMPLATE,
698 ARG_PROPERTY,
699 ARG_PRIVATE_USERS,
700 ARG_KILL_SIGNAL,
701 ARG_SETTINGS,
702 ARG_CHDIR,
703 ARG_PIVOT_ROOT,
704 ARG_PRIVATE_USERS_CHOWN,
705 ARG_PRIVATE_USERS_OWNERSHIP,
706 ARG_NOTIFY_READY,
707 ARG_ROOT_HASH,
708 ARG_ROOT_HASH_SIG,
709 ARG_VERITY_DATA,
710 ARG_SYSTEM_CALL_FILTER,
711 ARG_RLIMIT,
712 ARG_HOSTNAME,
713 ARG_NO_NEW_PRIVILEGES,
714 ARG_OOM_SCORE_ADJUST,
715 ARG_CPU_AFFINITY,
716 ARG_RESOLV_CONF,
717 ARG_TIMEZONE,
718 ARG_CONSOLE,
719 ARG_PIPE,
720 ARG_OCI_BUNDLE,
721 ARG_NO_PAGER,
722 ARG_SET_CREDENTIAL,
723 ARG_LOAD_CREDENTIAL,
724 ARG_BIND_USER,
725 ARG_SUPPRESS_SYNC,
726 };
727
728 static const struct option options[] = {
729 { "help", no_argument, NULL, 'h' },
730 { "version", no_argument, NULL, ARG_VERSION },
731 { "directory", required_argument, NULL, 'D' },
732 { "template", required_argument, NULL, ARG_TEMPLATE },
733 { "ephemeral", no_argument, NULL, 'x' },
734 { "user", required_argument, NULL, 'u' },
735 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
736 { "as-pid2", no_argument, NULL, 'a' },
737 { "boot", no_argument, NULL, 'b' },
738 { "uuid", required_argument, NULL, ARG_UUID },
739 { "read-only", no_argument, NULL, ARG_READ_ONLY },
740 { "capability", required_argument, NULL, ARG_CAPABILITY },
741 { "ambient-capability", required_argument, NULL, ARG_AMBIENT_CAPABILITY },
742 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
743 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
744 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
745 { "bind", required_argument, NULL, ARG_BIND },
746 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
747 { "tmpfs", required_argument, NULL, ARG_TMPFS },
748 { "overlay", required_argument, NULL, ARG_OVERLAY },
749 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
750 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
751 { "machine", required_argument, NULL, 'M' },
752 { "hostname", required_argument, NULL, ARG_HOSTNAME },
753 { "slice", required_argument, NULL, 'S' },
754 { "setenv", required_argument, NULL, 'E' },
755 { "selinux-context", required_argument, NULL, 'Z' },
756 { "selinux-apifs-context", required_argument, NULL, 'L' },
757 { "quiet", no_argument, NULL, 'q' },
758 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
759 { "register", required_argument, NULL, ARG_REGISTER },
760 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
761 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
762 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
763 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
764 { "network-veth", no_argument, NULL, 'n' },
765 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
766 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
767 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
768 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
769 { "personality", required_argument, NULL, ARG_PERSONALITY },
770 { "image", required_argument, NULL, 'i' },
771 { "volatile", optional_argument, NULL, ARG_VOLATILE },
772 { "port", required_argument, NULL, 'p' },
773 { "property", required_argument, NULL, ARG_PROPERTY },
774 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
775 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN }, /* obsolete */
776 { "private-users-ownership",required_argument, NULL, ARG_PRIVATE_USERS_OWNERSHIP},
777 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
778 { "settings", required_argument, NULL, ARG_SETTINGS },
779 { "chdir", required_argument, NULL, ARG_CHDIR },
780 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
781 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
782 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
783 { "root-hash-sig", required_argument, NULL, ARG_ROOT_HASH_SIG },
784 { "verity-data", required_argument, NULL, ARG_VERITY_DATA },
785 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
786 { "rlimit", required_argument, NULL, ARG_RLIMIT },
787 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
788 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
789 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
790 { "timezone", required_argument, NULL, ARG_TIMEZONE },
791 { "console", required_argument, NULL, ARG_CONSOLE },
792 { "pipe", no_argument, NULL, ARG_PIPE },
793 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
794 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
795 { "set-credential", required_argument, NULL, ARG_SET_CREDENTIAL },
796 { "load-credential", required_argument, NULL, ARG_LOAD_CREDENTIAL },
797 { "bind-user", required_argument, NULL, ARG_BIND_USER },
798 { "suppress-sync", required_argument, NULL, ARG_SUPPRESS_SYNC },
799 {}
800 };
801
802 int c, r;
803 uint64_t plus = 0, minus = 0;
804 bool mask_all_settings = false, mask_no_settings = false;
805
806 assert(argc >= 0);
807 assert(argv);
808
809 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
810 switch (c) {
811
812 case 'h':
813 return help();
814
815 case ARG_VERSION:
816 return version();
817
818 case 'D':
819 r = parse_path_argument(optarg, false, &arg_directory);
820 if (r < 0)
821 return r;
822
823 arg_settings_mask |= SETTING_DIRECTORY;
824 break;
825
826 case ARG_TEMPLATE:
827 r = parse_path_argument(optarg, false, &arg_template);
828 if (r < 0)
829 return r;
830
831 arg_settings_mask |= SETTING_DIRECTORY;
832 break;
833
834 case 'i':
835 r = parse_path_argument(optarg, false, &arg_image);
836 if (r < 0)
837 return r;
838
839 arg_settings_mask |= SETTING_DIRECTORY;
840 break;
841
842 case ARG_OCI_BUNDLE:
843 r = parse_path_argument(optarg, false, &arg_oci_bundle);
844 if (r < 0)
845 return r;
846
847 break;
848
849 case 'x':
850 arg_ephemeral = true;
851 arg_settings_mask |= SETTING_EPHEMERAL;
852 break;
853
854 case 'u':
855 r = free_and_strdup(&arg_user, optarg);
856 if (r < 0)
857 return log_oom();
858
859 arg_settings_mask |= SETTING_USER;
860 break;
861
862 case ARG_NETWORK_ZONE: {
863 char *j;
864
865 j = strjoin("vz-", optarg);
866 if (!j)
867 return log_oom();
868
869 if (!ifname_valid(j)) {
870 log_error("Network zone name not valid: %s", j);
871 free(j);
872 return -EINVAL;
873 }
874
875 free_and_replace(arg_network_zone, j);
876
877 arg_network_veth = true;
878 arg_private_network = true;
879 arg_settings_mask |= SETTING_NETWORK;
880 break;
881 }
882
883 case ARG_NETWORK_BRIDGE:
884
885 if (!ifname_valid(optarg))
886 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
887 "Bridge interface name not valid: %s", optarg);
888
889 r = free_and_strdup(&arg_network_bridge, optarg);
890 if (r < 0)
891 return log_oom();
892
893 _fallthrough_;
894 case 'n':
895 arg_network_veth = true;
896 arg_private_network = true;
897 arg_settings_mask |= SETTING_NETWORK;
898 break;
899
900 case ARG_NETWORK_VETH_EXTRA:
901 r = veth_extra_parse(&arg_network_veth_extra, optarg);
902 if (r < 0)
903 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
904
905 arg_private_network = true;
906 arg_settings_mask |= SETTING_NETWORK;
907 break;
908
909 case ARG_NETWORK_INTERFACE:
910 if (!ifname_valid(optarg))
911 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
912 "Network interface name not valid: %s", optarg);
913
914 r = test_network_interface_initialized(optarg);
915 if (r < 0)
916 return r;
917
918 if (strv_extend(&arg_network_interfaces, optarg) < 0)
919 return log_oom();
920
921 arg_private_network = true;
922 arg_settings_mask |= SETTING_NETWORK;
923 break;
924
925 case ARG_NETWORK_MACVLAN:
926
927 if (!ifname_valid(optarg))
928 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
929 "MACVLAN network interface name not valid: %s", optarg);
930
931 r = test_network_interface_initialized(optarg);
932 if (r < 0)
933 return r;
934
935 if (strv_extend(&arg_network_macvlan, optarg) < 0)
936 return log_oom();
937
938 arg_private_network = true;
939 arg_settings_mask |= SETTING_NETWORK;
940 break;
941
942 case ARG_NETWORK_IPVLAN:
943
944 if (!ifname_valid(optarg))
945 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
946 "IPVLAN network interface name not valid: %s", optarg);
947
948 r = test_network_interface_initialized(optarg);
949 if (r < 0)
950 return r;
951
952 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
953 return log_oom();
954
955 _fallthrough_;
956 case ARG_PRIVATE_NETWORK:
957 arg_private_network = true;
958 arg_settings_mask |= SETTING_NETWORK;
959 break;
960
961 case ARG_NETWORK_NAMESPACE_PATH:
962 r = parse_path_argument(optarg, false, &arg_network_namespace_path);
963 if (r < 0)
964 return r;
965
966 arg_settings_mask |= SETTING_NETWORK;
967 break;
968
969 case 'b':
970 if (arg_start_mode == START_PID2)
971 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
972 "--boot and --as-pid2 may not be combined.");
973
974 arg_start_mode = START_BOOT;
975 arg_settings_mask |= SETTING_START_MODE;
976 break;
977
978 case 'a':
979 if (arg_start_mode == START_BOOT)
980 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
981 "--boot and --as-pid2 may not be combined.");
982
983 arg_start_mode = START_PID2;
984 arg_settings_mask |= SETTING_START_MODE;
985 break;
986
987 case ARG_UUID:
988 r = sd_id128_from_string(optarg, &arg_uuid);
989 if (r < 0)
990 return log_error_errno(r, "Invalid UUID: %s", optarg);
991
992 if (sd_id128_is_null(arg_uuid))
993 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
994 "Machine UUID may not be all zeroes.");
995
996 arg_settings_mask |= SETTING_MACHINE_ID;
997 break;
998
999 case 'S': {
1000 _cleanup_free_ char *mangled = NULL;
1001
1002 r = unit_name_mangle_with_suffix(optarg, NULL, UNIT_NAME_MANGLE_WARN, ".slice", &mangled);
1003 if (r < 0)
1004 return log_oom();
1005
1006 free_and_replace(arg_slice, mangled);
1007 arg_settings_mask |= SETTING_SLICE;
1008 break;
1009 }
1010
1011 case 'M':
1012 if (isempty(optarg))
1013 arg_machine = mfree(arg_machine);
1014 else {
1015 if (!hostname_is_valid(optarg, 0))
1016 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1017 "Invalid machine name: %s", optarg);
1018
1019 r = free_and_strdup(&arg_machine, optarg);
1020 if (r < 0)
1021 return log_oom();
1022 }
1023 break;
1024
1025 case ARG_HOSTNAME:
1026 if (isempty(optarg))
1027 arg_hostname = mfree(arg_hostname);
1028 else {
1029 if (!hostname_is_valid(optarg, 0))
1030 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1031 "Invalid hostname: %s", optarg);
1032
1033 r = free_and_strdup(&arg_hostname, optarg);
1034 if (r < 0)
1035 return log_oom();
1036 }
1037
1038 arg_settings_mask |= SETTING_HOSTNAME;
1039 break;
1040
1041 case 'Z':
1042 arg_selinux_context = optarg;
1043 break;
1044
1045 case 'L':
1046 arg_selinux_apifs_context = optarg;
1047 break;
1048
1049 case ARG_READ_ONLY:
1050 arg_read_only = true;
1051 arg_settings_mask |= SETTING_READ_ONLY;
1052 break;
1053
1054 case ARG_AMBIENT_CAPABILITY: {
1055 uint64_t m;
1056 r = parse_capability_spec(optarg, &m);
1057 if (r <= 0)
1058 return r;
1059 arg_caps_ambient |= m;
1060 arg_settings_mask |= SETTING_CAPABILITY;
1061 break;
1062 }
1063 case ARG_CAPABILITY:
1064 case ARG_DROP_CAPABILITY: {
1065 uint64_t m;
1066 r = parse_capability_spec(optarg, &m);
1067 if (r <= 0)
1068 return r;
1069
1070 if (c == ARG_CAPABILITY)
1071 plus |= m;
1072 else
1073 minus |= m;
1074 arg_settings_mask |= SETTING_CAPABILITY;
1075 break;
1076 }
1077 case ARG_NO_NEW_PRIVILEGES:
1078 r = parse_boolean(optarg);
1079 if (r < 0)
1080 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
1081
1082 arg_no_new_privileges = r;
1083 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
1084 break;
1085
1086 case 'j':
1087 arg_link_journal = LINK_GUEST;
1088 arg_link_journal_try = true;
1089 arg_settings_mask |= SETTING_LINK_JOURNAL;
1090 break;
1091
1092 case ARG_LINK_JOURNAL:
1093 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
1094 if (r < 0)
1095 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
1096
1097 arg_settings_mask |= SETTING_LINK_JOURNAL;
1098 break;
1099
1100 case ARG_BIND:
1101 case ARG_BIND_RO:
1102 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
1103 if (r < 0)
1104 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
1105
1106 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1107 break;
1108
1109 case ARG_TMPFS:
1110 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1111 if (r < 0)
1112 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
1113
1114 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1115 break;
1116
1117 case ARG_OVERLAY:
1118 case ARG_OVERLAY_RO:
1119 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
1120 if (r == -EADDRNOTAVAIL)
1121 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1122 if (r < 0)
1123 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
1124
1125 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1126 break;
1127
1128 case ARG_INACCESSIBLE:
1129 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1130 if (r < 0)
1131 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
1132
1133 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1134 break;
1135
1136 case 'E':
1137 r = strv_env_replace_strdup_passthrough(&arg_setenv, optarg);
1138 if (r < 0)
1139 return log_error_errno(r, "Cannot assign environment variable %s: %m", optarg);
1140
1141 arg_settings_mask |= SETTING_ENVIRONMENT;
1142 break;
1143
1144 case 'q':
1145 arg_quiet = true;
1146 break;
1147
1148 case ARG_SHARE_SYSTEM:
1149 /* We don't officially support this anymore, except for compat reasons. People should use the
1150 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1151 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1152 arg_clone_ns_flags = 0;
1153 break;
1154
1155 case ARG_REGISTER:
1156 r = parse_boolean(optarg);
1157 if (r < 0) {
1158 log_error("Failed to parse --register= argument: %s", optarg);
1159 return r;
1160 }
1161
1162 arg_register = r;
1163 break;
1164
1165 case ARG_KEEP_UNIT:
1166 arg_keep_unit = true;
1167 break;
1168
1169 case ARG_PERSONALITY:
1170
1171 arg_personality = personality_from_string(optarg);
1172 if (arg_personality == PERSONALITY_INVALID)
1173 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1174 "Unknown or unsupported personality '%s'.", optarg);
1175
1176 arg_settings_mask |= SETTING_PERSONALITY;
1177 break;
1178
1179 case ARG_VOLATILE:
1180
1181 if (!optarg)
1182 arg_volatile_mode = VOLATILE_YES;
1183 else if (streq(optarg, "help")) {
1184 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1185 return 0;
1186 } else {
1187 VolatileMode m;
1188
1189 m = volatile_mode_from_string(optarg);
1190 if (m < 0)
1191 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1192 "Failed to parse --volatile= argument: %s", optarg);
1193 else
1194 arg_volatile_mode = m;
1195 }
1196
1197 arg_settings_mask |= SETTING_VOLATILE_MODE;
1198 break;
1199
1200 case 'p':
1201 r = expose_port_parse(&arg_expose_ports, optarg);
1202 if (r == -EEXIST)
1203 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1204 if (r < 0)
1205 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1206
1207 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1208 break;
1209
1210 case ARG_PROPERTY:
1211 if (strv_extend(&arg_property, optarg) < 0)
1212 return log_oom();
1213
1214 break;
1215
1216 case ARG_PRIVATE_USERS: {
1217 int boolean;
1218
1219 if (!optarg)
1220 boolean = true;
1221 else if (!in_charset(optarg, DIGITS))
1222 /* do *not* parse numbers as booleans */
1223 boolean = parse_boolean(optarg);
1224 else
1225 boolean = -1;
1226
1227 if (boolean == 0) {
1228 /* no: User namespacing off */
1229 arg_userns_mode = USER_NAMESPACE_NO;
1230 arg_uid_shift = UID_INVALID;
1231 arg_uid_range = UINT32_C(0x10000);
1232 } else if (boolean > 0) {
1233 /* yes: User namespacing on, UID range is read from root dir */
1234 arg_userns_mode = USER_NAMESPACE_FIXED;
1235 arg_uid_shift = UID_INVALID;
1236 arg_uid_range = UINT32_C(0x10000);
1237 } else if (streq(optarg, "pick")) {
1238 /* pick: User namespacing on, UID range is picked randomly */
1239 arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
1240 * implied by USER_NAMESPACE_PICK
1241 * further down. */
1242 arg_uid_shift = UID_INVALID;
1243 arg_uid_range = UINT32_C(0x10000);
1244
1245 } else if (streq(optarg, "identity")) {
1246 /* identitiy: User namespaces on, UID range is map the 0…0xFFFF range to
1247 * itself, i.e. we don't actually map anything, but do take benefit of
1248 * isolation of capability sets. */
1249 arg_userns_mode = USER_NAMESPACE_FIXED;
1250 arg_uid_shift = 0;
1251 arg_uid_range = UINT32_C(0x10000);
1252 } else {
1253 _cleanup_free_ char *buffer = NULL;
1254 const char *range, *shift;
1255
1256 /* anything else: User namespacing on, UID range is explicitly configured */
1257
1258 range = strchr(optarg, ':');
1259 if (range) {
1260 buffer = strndup(optarg, range - optarg);
1261 if (!buffer)
1262 return log_oom();
1263 shift = buffer;
1264
1265 range++;
1266 r = safe_atou32(range, &arg_uid_range);
1267 if (r < 0)
1268 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
1269 } else
1270 shift = optarg;
1271
1272 r = parse_uid(shift, &arg_uid_shift);
1273 if (r < 0)
1274 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
1275
1276 arg_userns_mode = USER_NAMESPACE_FIXED;
1277
1278 if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
1279 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID range cannot be empty or go beyond " UID_FMT ".", UID_INVALID);
1280 }
1281
1282 arg_settings_mask |= SETTING_USERNS;
1283 break;
1284 }
1285
1286 case 'U':
1287 if (userns_supported()) {
1288 arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
1289 * implied by USER_NAMESPACE_PICK
1290 * further down. */
1291 arg_uid_shift = UID_INVALID;
1292 arg_uid_range = UINT32_C(0x10000);
1293
1294 arg_settings_mask |= SETTING_USERNS;
1295 }
1296
1297 break;
1298
1299 case ARG_PRIVATE_USERS_CHOWN:
1300 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
1301
1302 arg_settings_mask |= SETTING_USERNS;
1303 break;
1304
1305 case ARG_PRIVATE_USERS_OWNERSHIP:
1306 if (streq(optarg, "help")) {
1307 DUMP_STRING_TABLE(user_namespace_ownership, UserNamespaceOwnership, _USER_NAMESPACE_OWNERSHIP_MAX);
1308 return 0;
1309 }
1310
1311 arg_userns_ownership = user_namespace_ownership_from_string(optarg);
1312 if (arg_userns_ownership < 0)
1313 return log_error_errno(arg_userns_ownership, "Cannot parse --user-namespace-ownership= value: %s", optarg);
1314
1315 arg_settings_mask |= SETTING_USERNS;
1316 break;
1317
1318 case ARG_KILL_SIGNAL:
1319 if (streq(optarg, "help")) {
1320 DUMP_STRING_TABLE(signal, int, _NSIG);
1321 return 0;
1322 }
1323
1324 arg_kill_signal = signal_from_string(optarg);
1325 if (arg_kill_signal < 0)
1326 return log_error_errno(arg_kill_signal, "Cannot parse signal: %s", optarg);
1327
1328 arg_settings_mask |= SETTING_KILL_SIGNAL;
1329 break;
1330
1331 case ARG_SETTINGS:
1332
1333 /* no → do not read files
1334 * yes → read files, do not override cmdline, trust only subset
1335 * override → read files, override cmdline, trust only subset
1336 * trusted → read files, do not override cmdline, trust all
1337 */
1338
1339 r = parse_boolean(optarg);
1340 if (r < 0) {
1341 if (streq(optarg, "trusted")) {
1342 mask_all_settings = false;
1343 mask_no_settings = false;
1344 arg_settings_trusted = true;
1345
1346 } else if (streq(optarg, "override")) {
1347 mask_all_settings = false;
1348 mask_no_settings = true;
1349 arg_settings_trusted = -1;
1350 } else
1351 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1352 } else if (r > 0) {
1353 /* yes */
1354 mask_all_settings = false;
1355 mask_no_settings = false;
1356 arg_settings_trusted = -1;
1357 } else {
1358 /* no */
1359 mask_all_settings = true;
1360 mask_no_settings = false;
1361 arg_settings_trusted = false;
1362 }
1363
1364 break;
1365
1366 case ARG_CHDIR:
1367 if (!path_is_absolute(optarg))
1368 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1369 "Working directory %s is not an absolute path.", optarg);
1370
1371 r = free_and_strdup(&arg_chdir, optarg);
1372 if (r < 0)
1373 return log_oom();
1374
1375 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1376 break;
1377
1378 case ARG_PIVOT_ROOT:
1379 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1380 if (r < 0)
1381 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1382
1383 arg_settings_mask |= SETTING_PIVOT_ROOT;
1384 break;
1385
1386 case ARG_NOTIFY_READY:
1387 r = parse_boolean(optarg);
1388 if (r < 0)
1389 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1390 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1391 arg_notify_ready = r;
1392 arg_settings_mask |= SETTING_NOTIFY_READY;
1393 break;
1394
1395 case ARG_ROOT_HASH: {
1396 _cleanup_free_ void *k = NULL;
1397 size_t l;
1398
1399 r = unhexmem(optarg, strlen(optarg), &k, &l);
1400 if (r < 0)
1401 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1402 if (l < sizeof(sd_id128_t))
1403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128bit long: %s", optarg);
1404
1405 free_and_replace(arg_verity_settings.root_hash, k);
1406 arg_verity_settings.root_hash_size = l;
1407 break;
1408 }
1409
1410 case ARG_ROOT_HASH_SIG: {
1411 char *value;
1412 size_t l;
1413 void *p;
1414
1415 if ((value = startswith(optarg, "base64:"))) {
1416 r = unbase64mem(value, strlen(value), &p, &l);
1417 if (r < 0)
1418 return log_error_errno(r, "Failed to parse root hash signature '%s': %m", optarg);
1419
1420 } else {
1421 r = read_full_file(optarg, (char**) &p, &l);
1422 if (r < 0)
1423 return log_error_errno(r, "Failed parse root hash signature file '%s': %m", optarg);
1424 }
1425
1426 free_and_replace(arg_verity_settings.root_hash_sig, p);
1427 arg_verity_settings.root_hash_sig_size = l;
1428 break;
1429 }
1430
1431 case ARG_VERITY_DATA:
1432 r = parse_path_argument(optarg, false, &arg_verity_settings.data_path);
1433 if (r < 0)
1434 return r;
1435 break;
1436
1437 case ARG_SYSTEM_CALL_FILTER: {
1438 bool negative;
1439 const char *items;
1440
1441 negative = optarg[0] == '~';
1442 items = negative ? optarg + 1 : optarg;
1443
1444 for (;;) {
1445 _cleanup_free_ char *word = NULL;
1446
1447 r = extract_first_word(&items, &word, NULL, 0);
1448 if (r == 0)
1449 break;
1450 if (r == -ENOMEM)
1451 return log_oom();
1452 if (r < 0)
1453 return log_error_errno(r, "Failed to parse system call filter: %m");
1454
1455 if (negative)
1456 r = strv_extend(&arg_syscall_deny_list, word);
1457 else
1458 r = strv_extend(&arg_syscall_allow_list, word);
1459 if (r < 0)
1460 return log_oom();
1461 }
1462
1463 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1464 break;
1465 }
1466
1467 case ARG_RLIMIT: {
1468 const char *eq;
1469 _cleanup_free_ char *name = NULL;
1470 int rl;
1471
1472 if (streq(optarg, "help")) {
1473 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1474 return 0;
1475 }
1476
1477 eq = strchr(optarg, '=');
1478 if (!eq)
1479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1480 "--rlimit= expects an '=' assignment.");
1481
1482 name = strndup(optarg, eq - optarg);
1483 if (!name)
1484 return log_oom();
1485
1486 rl = rlimit_from_string_harder(name);
1487 if (rl < 0)
1488 return log_error_errno(rl, "Unknown resource limit: %s", name);
1489
1490 if (!arg_rlimit[rl]) {
1491 arg_rlimit[rl] = new0(struct rlimit, 1);
1492 if (!arg_rlimit[rl])
1493 return log_oom();
1494 }
1495
1496 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1497 if (r < 0)
1498 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1499
1500 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1501 break;
1502 }
1503
1504 case ARG_OOM_SCORE_ADJUST:
1505 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1506 if (r < 0)
1507 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1508
1509 arg_oom_score_adjust_set = true;
1510 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1511 break;
1512
1513 case ARG_CPU_AFFINITY: {
1514 CPUSet cpuset;
1515
1516 r = parse_cpu_set(optarg, &cpuset);
1517 if (r < 0)
1518 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
1519
1520 cpu_set_reset(&arg_cpu_set);
1521 arg_cpu_set = cpuset;
1522 arg_settings_mask |= SETTING_CPU_AFFINITY;
1523 break;
1524 }
1525
1526 case ARG_RESOLV_CONF:
1527 if (streq(optarg, "help")) {
1528 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1529 return 0;
1530 }
1531
1532 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1533 if (arg_resolv_conf < 0)
1534 return log_error_errno(arg_resolv_conf,
1535 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1536
1537 arg_settings_mask |= SETTING_RESOLV_CONF;
1538 break;
1539
1540 case ARG_TIMEZONE:
1541 if (streq(optarg, "help")) {
1542 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1543 return 0;
1544 }
1545
1546 arg_timezone = timezone_mode_from_string(optarg);
1547 if (arg_timezone < 0)
1548 return log_error_errno(arg_timezone,
1549 "Failed to parse /etc/localtime mode: %s", optarg);
1550
1551 arg_settings_mask |= SETTING_TIMEZONE;
1552 break;
1553
1554 case ARG_CONSOLE:
1555 r = handle_arg_console(optarg);
1556 if (r <= 0)
1557 return r;
1558 break;
1559
1560 case 'P':
1561 case ARG_PIPE:
1562 r = handle_arg_console("pipe");
1563 if (r <= 0)
1564 return r;
1565 break;
1566
1567 case ARG_NO_PAGER:
1568 arg_pager_flags |= PAGER_DISABLE;
1569 break;
1570
1571 case ARG_SET_CREDENTIAL: {
1572 _cleanup_free_ char *word = NULL, *data = NULL;
1573 const char *p = optarg;
1574 Credential *a;
1575 ssize_t l;
1576
1577 r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1578 if (r == -ENOMEM)
1579 return log_oom();
1580 if (r < 0)
1581 return log_error_errno(r, "Failed to parse --set-credential= parameter: %m");
1582 if (r == 0 || !p)
1583 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --set-credential=: %s", optarg);
1584
1585 if (!credential_name_valid(word))
1586 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
1587
1588 for (size_t i = 0; i < arg_n_credentials; i++)
1589 if (streq(arg_credentials[i].id, word))
1590 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
1591
1592 l = cunescape(p, UNESCAPE_ACCEPT_NUL, &data);
1593 if (l < 0)
1594 return log_error_errno(l, "Failed to unescape credential data: %s", p);
1595
1596 a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
1597 if (!a)
1598 return log_oom();
1599
1600 a[arg_n_credentials++] = (Credential) {
1601 .id = TAKE_PTR(word),
1602 .data = TAKE_PTR(data),
1603 .size = l,
1604 };
1605
1606 arg_credentials = a;
1607
1608 arg_settings_mask |= SETTING_CREDENTIALS;
1609 break;
1610 }
1611
1612 case ARG_LOAD_CREDENTIAL: {
1613 ReadFullFileFlags flags = READ_FULL_FILE_SECURE;
1614 _cleanup_(erase_and_freep) char *data = NULL;
1615 _cleanup_free_ char *word = NULL, *j = NULL;
1616 const char *p = optarg;
1617 Credential *a;
1618 size_t size, i;
1619
1620 r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1621 if (r == -ENOMEM)
1622 return log_oom();
1623 if (r < 0)
1624 return log_error_errno(r, "Failed to parse --set-credential= parameter: %m");
1625 if (r == 0 || !p)
1626 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --set-credential=: %s", optarg);
1627
1628 if (!credential_name_valid(word))
1629 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
1630
1631 for (i = 0; i < arg_n_credentials; i++)
1632 if (streq(arg_credentials[i].id, word))
1633 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
1634
1635 if (path_is_absolute(p))
1636 flags |= READ_FULL_FILE_CONNECT_SOCKET;
1637 else {
1638 const char *e;
1639
1640 r = get_credentials_dir(&e);
1641 if (r < 0)
1642 return log_error_errno(r, "Credential not available (no credentials passed at all): %s", word);
1643
1644 j = path_join(e, p);
1645 if (!j)
1646 return log_oom();
1647 }
1648
1649 r = read_full_file_full(AT_FDCWD, j ?: p, UINT64_MAX, SIZE_MAX,
1650 flags,
1651 NULL,
1652 &data, &size);
1653 if (r < 0)
1654 return log_error_errno(r, "Failed to read credential '%s': %m", j ?: p);
1655
1656 a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
1657 if (!a)
1658 return log_oom();
1659
1660 a[arg_n_credentials++] = (Credential) {
1661 .id = TAKE_PTR(word),
1662 .data = TAKE_PTR(data),
1663 .size = size,
1664 };
1665
1666 arg_credentials = a;
1667
1668 arg_settings_mask |= SETTING_CREDENTIALS;
1669 break;
1670 }
1671
1672 case ARG_BIND_USER:
1673 if (!valid_user_group_name(optarg, 0))
1674 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid user name to bind: %s", optarg);
1675
1676 if (strv_extend(&arg_bind_user, optarg) < 0)
1677 return log_oom();
1678
1679 arg_settings_mask |= SETTING_BIND_USER;
1680 break;
1681
1682 case ARG_SUPPRESS_SYNC:
1683 r = parse_boolean_argument("--suppress-sync=", optarg, &arg_suppress_sync);
1684 if (r < 0)
1685 return r;
1686
1687 arg_settings_mask |= SETTING_SUPPRESS_SYNC;
1688 break;
1689
1690 case '?':
1691 return -EINVAL;
1692
1693 default:
1694 assert_not_reached();
1695 }
1696
1697 if (argc > optind) {
1698 strv_free(arg_parameters);
1699 arg_parameters = strv_copy(argv + optind);
1700 if (!arg_parameters)
1701 return log_oom();
1702
1703 arg_settings_mask |= SETTING_START_MODE;
1704 }
1705
1706 if (arg_ephemeral && arg_template && !arg_directory)
1707 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1708 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1709 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1710 * --directory=". */
1711 arg_directory = TAKE_PTR(arg_template);
1712
1713 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0)) & ~minus;
1714
1715 /* Make sure to parse environment before we reset the settings mask below */
1716 r = parse_environment();
1717 if (r < 0)
1718 return r;
1719
1720 /* Load all settings from .nspawn files */
1721 if (mask_no_settings)
1722 arg_settings_mask = 0;
1723
1724 /* Don't load any settings from .nspawn files */
1725 if (mask_all_settings)
1726 arg_settings_mask = _SETTINGS_MASK_ALL;
1727
1728 return 1;
1729 }
1730
1731 static int verify_arguments(void) {
1732 int r;
1733
1734 if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
1735 /* If we are running the stub init in the container, we don't need to look at what the init
1736 * in the container supports, because we are not using it. Let's immediately pick the right
1737 * setting based on the host system configuration.
1738 *
1739 * We only do this, if the user didn't use an environment variable to override the detection.
1740 */
1741
1742 r = cg_all_unified();
1743 if (r < 0)
1744 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
1745 if (r > 0)
1746 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
1747 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
1748 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
1749 else
1750 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
1751 }
1752
1753 if (arg_userns_mode != USER_NAMESPACE_NO)
1754 arg_mount_settings |= MOUNT_USE_USERNS;
1755
1756 if (arg_private_network)
1757 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1758
1759 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1760 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1761 arg_register = false;
1762 if (arg_start_mode != START_PID1)
1763 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1764 }
1765
1766 if (arg_userns_ownership < 0)
1767 arg_userns_ownership =
1768 arg_userns_mode == USER_NAMESPACE_PICK ? USER_NAMESPACE_OWNERSHIP_AUTO :
1769 USER_NAMESPACE_OWNERSHIP_OFF;
1770
1771 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1772 arg_kill_signal = SIGRTMIN+3;
1773
1774 if (arg_volatile_mode != VOLATILE_NO) /* Make sure all file systems contained in the image are mounted read-only if we are in volatile mode */
1775 arg_read_only = true;
1776
1777 if (has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts))
1778 arg_read_only = true;
1779
1780 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1781 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1782 * The latter is not technically a user session, but we don't need to labour the point. */
1783 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1784
1785 if (arg_directory && arg_image)
1786 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1787
1788 if (arg_template && arg_image)
1789 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1790
1791 if (arg_template && !(arg_directory || arg_machine))
1792 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1793
1794 if (arg_ephemeral && arg_template)
1795 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1796
1797 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
1798 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
1799
1800 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1801 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1802
1803 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_read_only)
1804 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1805 "--read-only and --private-users-ownership=chown may not be combined.");
1806
1807 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1808 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1809 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1810 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_volatile_mode != VOLATILE_NO)
1811 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-ownership=chown may not be combined.");
1812
1813 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1814 * we need to error out, to avoid conflicts between different network options. */
1815 if (arg_network_namespace_path &&
1816 (arg_network_interfaces || arg_network_macvlan ||
1817 arg_network_ipvlan || arg_network_veth_extra ||
1818 arg_network_bridge || arg_network_zone ||
1819 arg_network_veth))
1820 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1821
1822 if (arg_network_bridge && arg_network_zone)
1823 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1824 "--network-bridge= and --network-zone= may not be combined.");
1825
1826 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1827 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1828
1829 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1830 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1831
1832 if (arg_expose_ports && !arg_private_network)
1833 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1834
1835 if (arg_caps_ambient) {
1836 if (arg_caps_ambient == UINT64_MAX)
1837 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= does not support the value all.");
1838
1839 if ((arg_caps_ambient & arg_caps_retain) != arg_caps_ambient)
1840 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not fully covered by Capability= setting.");
1841
1842 if (arg_start_mode == START_BOOT)
1843 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not useful for boot mode.");
1844 }
1845
1846 if (arg_userns_mode == USER_NAMESPACE_NO && !strv_isempty(arg_bind_user))
1847 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--bind-user= requires --private-users");
1848
1849 /* Drop duplicate --bind-user= entries */
1850 strv_uniq(arg_bind_user);
1851
1852 r = custom_mount_check_all();
1853 if (r < 0)
1854 return r;
1855
1856 return 0;
1857 }
1858
1859 int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1860 assert(p);
1861
1862 if (arg_userns_mode == USER_NAMESPACE_NO)
1863 return 0;
1864
1865 if (uid == UID_INVALID && gid == GID_INVALID)
1866 return 0;
1867
1868 if (uid != UID_INVALID) {
1869 uid += arg_uid_shift;
1870
1871 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1872 return -EOVERFLOW;
1873 }
1874
1875 if (gid != GID_INVALID) {
1876 gid += (gid_t) arg_uid_shift;
1877
1878 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1879 return -EOVERFLOW;
1880 }
1881
1882 return RET_NERRNO(lchown(p, uid, gid));
1883 }
1884
1885 int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1886 const char *q;
1887 int r;
1888
1889 q = prefix_roota(root, path);
1890 r = RET_NERRNO(mkdir(q, mode));
1891 if (r == -EEXIST)
1892 return 0;
1893 if (r < 0)
1894 return r;
1895
1896 return userns_lchown(q, uid, gid);
1897 }
1898
1899 static const char *timezone_from_path(const char *path) {
1900 return PATH_STARTSWITH_SET(
1901 path,
1902 "../usr/share/zoneinfo/",
1903 "/usr/share/zoneinfo/");
1904 }
1905
1906 static bool etc_writable(void) {
1907 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1908 }
1909
1910 static int setup_timezone(const char *dest) {
1911 _cleanup_free_ char *p = NULL, *etc = NULL;
1912 const char *where, *check;
1913 TimezoneMode m;
1914 int r;
1915
1916 assert(dest);
1917
1918 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1919 r = readlink_malloc("/etc/localtime", &p);
1920 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1921 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1922 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1923 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1924 else if (r < 0) {
1925 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1926 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1927 * file.
1928 *
1929 * Example:
1930 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1931 */
1932 return 0;
1933 } else if (arg_timezone == TIMEZONE_AUTO)
1934 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1935 else
1936 m = arg_timezone;
1937 } else
1938 m = arg_timezone;
1939
1940 if (m == TIMEZONE_OFF)
1941 return 0;
1942
1943 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1944 if (r < 0) {
1945 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1946 return 0;
1947 }
1948
1949 where = strjoina(etc, "/localtime");
1950
1951 switch (m) {
1952
1953 case TIMEZONE_DELETE:
1954 if (unlink(where) < 0)
1955 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1956
1957 return 0;
1958
1959 case TIMEZONE_SYMLINK: {
1960 _cleanup_free_ char *q = NULL;
1961 const char *z, *what;
1962
1963 z = timezone_from_path(p);
1964 if (!z) {
1965 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1966 return 0;
1967 }
1968
1969 r = readlink_malloc(where, &q);
1970 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1971 return 0; /* Already pointing to the right place? Then do nothing .. */
1972
1973 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1974 r = chase_symlinks(check, dest, 0, NULL, NULL);
1975 if (r < 0)
1976 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1977 else {
1978 if (unlink(where) < 0 && errno != ENOENT) {
1979 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1980 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1981 return 0;
1982 }
1983
1984 what = strjoina("../usr/share/zoneinfo/", z);
1985 if (symlink(what, where) < 0) {
1986 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1987 errno, "Failed to correct timezone of container, ignoring: %m");
1988 return 0;
1989 }
1990
1991 break;
1992 }
1993
1994 _fallthrough_;
1995 }
1996
1997 case TIMEZONE_BIND: {
1998 _cleanup_free_ char *resolved = NULL;
1999 int found;
2000
2001 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
2002 if (found < 0) {
2003 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
2004 return 0;
2005 }
2006
2007 if (found == 0) /* missing? */
2008 (void) touch(resolved);
2009
2010 r = mount_nofollow_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
2011 if (r >= 0)
2012 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
2013
2014 _fallthrough_;
2015 }
2016
2017 case TIMEZONE_COPY:
2018 /* If mounting failed, try to copy */
2019 r = copy_file_atomic("/etc/localtime", where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
2020 if (r < 0) {
2021 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
2022 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
2023 return 0;
2024 }
2025
2026 break;
2027
2028 default:
2029 assert_not_reached();
2030 }
2031
2032 /* Fix permissions of the symlink or file copy we just created */
2033 r = userns_lchown(where, 0, 0);
2034 if (r < 0)
2035 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
2036
2037 return 0;
2038 }
2039
2040 static int have_resolv_conf(const char *path) {
2041 assert(path);
2042
2043 if (access(path, F_OK) < 0) {
2044 if (errno == ENOENT)
2045 return 0;
2046
2047 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
2048 }
2049
2050 return 1;
2051 }
2052
2053 static int resolved_listening(void) {
2054 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2055 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
2056 _cleanup_free_ char *dns_stub_listener_mode = NULL;
2057 int r;
2058
2059 /* Check if resolved is listening */
2060
2061 r = sd_bus_open_system(&bus);
2062 if (r < 0)
2063 return log_debug_errno(r, "Failed to open system bus: %m");
2064
2065 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
2066 if (r < 0)
2067 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2068 if (r == 0)
2069 return 0;
2070
2071 r = sd_bus_get_property_string(bus,
2072 "org.freedesktop.resolve1",
2073 "/org/freedesktop/resolve1",
2074 "org.freedesktop.resolve1.Manager",
2075 "DNSStubListener",
2076 &error,
2077 &dns_stub_listener_mode);
2078 if (r < 0)
2079 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
2080
2081 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
2082 }
2083
2084 static int setup_resolv_conf(const char *dest) {
2085 _cleanup_free_ char *etc = NULL;
2086 const char *where, *what;
2087 ResolvConfMode m;
2088 int r;
2089
2090 assert(dest);
2091
2092 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
2093 if (arg_private_network)
2094 m = RESOLV_CONF_OFF;
2095 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF) > 0 && resolved_listening() > 0)
2096 m = etc_writable() ? RESOLV_CONF_COPY_STUB : RESOLV_CONF_BIND_STUB;
2097 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2098 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
2099 else
2100 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
2101
2102 } else
2103 m = arg_resolv_conf;
2104
2105 if (m == RESOLV_CONF_OFF)
2106 return 0;
2107
2108 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
2109 if (r < 0) {
2110 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
2111 return 0;
2112 }
2113
2114 where = strjoina(etc, "/resolv.conf");
2115
2116 if (m == RESOLV_CONF_DELETE) {
2117 if (unlink(where) < 0)
2118 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
2119
2120 return 0;
2121 }
2122
2123 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_COPY_STATIC))
2124 what = PRIVATE_STATIC_RESOLV_CONF;
2125 else if (IN_SET(m, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_COPY_UPLINK))
2126 what = PRIVATE_UPLINK_RESOLV_CONF;
2127 else if (IN_SET(m, RESOLV_CONF_BIND_STUB, RESOLV_CONF_REPLACE_STUB, RESOLV_CONF_COPY_STUB))
2128 what = PRIVATE_STUB_RESOLV_CONF;
2129 else
2130 what = "/etc/resolv.conf";
2131
2132 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_BIND_STUB)) {
2133 _cleanup_free_ char *resolved = NULL;
2134 int found;
2135
2136 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
2137 if (found < 0) {
2138 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2139 return 0;
2140 }
2141
2142 if (found == 0) /* missing? */
2143 (void) touch(resolved);
2144
2145 r = mount_nofollow_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
2146 if (r >= 0)
2147 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
2148
2149 /* If that didn't work, let's copy the file */
2150 }
2151
2152 if (IN_SET(m, RESOLV_CONF_REPLACE_HOST, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_REPLACE_STUB))
2153 r = copy_file_atomic(what, where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
2154 else
2155 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, 0, COPY_REFLINK);
2156 if (r < 0) {
2157 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2158 * resolved or something similar runs inside and the symlink points there.
2159 *
2160 * If the disk image is read-only, there's also no point in complaining.
2161 */
2162 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC, RESOLV_CONF_COPY_UPLINK, RESOLV_CONF_COPY_STUB) &&
2163 IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
2164 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
2165 return 0;
2166 }
2167
2168 r = userns_lchown(where, 0, 0);
2169 if (r < 0)
2170 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
2171
2172 return 0;
2173 }
2174
2175 static int setup_boot_id(void) {
2176 _cleanup_(unlink_and_freep) char *from = NULL;
2177 _cleanup_free_ char *path = NULL;
2178 sd_id128_t rnd = SD_ID128_NULL;
2179 const char *to;
2180 int r;
2181
2182 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2183
2184 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
2185 if (r < 0)
2186 return log_error_errno(r, "Failed to generate random boot ID path: %m");
2187
2188 r = sd_id128_randomize(&rnd);
2189 if (r < 0)
2190 return log_error_errno(r, "Failed to generate random boot id: %m");
2191
2192 r = id128_write(path, ID128_UUID, rnd, false);
2193 if (r < 0)
2194 return log_error_errno(r, "Failed to write boot id: %m");
2195
2196 from = TAKE_PTR(path);
2197 to = "/proc/sys/kernel/random/boot_id";
2198
2199 r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
2200 if (r < 0)
2201 return r;
2202
2203 return mount_nofollow_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2204 }
2205
2206 static int copy_devnodes(const char *dest) {
2207 static const char devnodes[] =
2208 "null\0"
2209 "zero\0"
2210 "full\0"
2211 "random\0"
2212 "urandom\0"
2213 "tty\0"
2214 "net/tun\0";
2215
2216 const char *d;
2217 int r = 0;
2218
2219 assert(dest);
2220
2221 BLOCK_WITH_UMASK(0000);
2222
2223 /* Create /dev/net, so that we can create /dev/net/tun in it */
2224 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
2225 return log_error_errno(r, "Failed to create /dev/net directory: %m");
2226
2227 NULSTR_FOREACH(d, devnodes) {
2228 _cleanup_free_ char *from = NULL, *to = NULL;
2229 struct stat st;
2230
2231 from = path_join("/dev/", d);
2232 if (!from)
2233 return log_oom();
2234
2235 to = path_join(dest, from);
2236 if (!to)
2237 return log_oom();
2238
2239 if (stat(from, &st) < 0) {
2240
2241 if (errno != ENOENT)
2242 return log_error_errno(errno, "Failed to stat %s: %m", from);
2243
2244 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
2245 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2246 "%s is not a char or block device, cannot copy.", from);
2247 else {
2248 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
2249
2250 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
2251 /* Explicitly warn the user when /dev is already populated. */
2252 if (errno == EEXIST)
2253 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);
2254 if (errno != EPERM)
2255 return log_error_errno(errno, "mknod(%s) failed: %m", to);
2256
2257 /* Some systems abusively restrict mknod but allow bind mounts. */
2258 r = touch(to);
2259 if (r < 0)
2260 return log_error_errno(r, "touch (%s) failed: %m", to);
2261 r = mount_nofollow_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
2262 if (r < 0)
2263 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
2264 }
2265
2266 r = userns_lchown(to, 0, 0);
2267 if (r < 0)
2268 return log_error_errno(r, "chown() of device node %s failed: %m", to);
2269
2270 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
2271 if (!dn)
2272 return log_oom();
2273
2274 r = userns_mkdir(dest, dn, 0755, 0, 0);
2275 if (r < 0)
2276 return log_error_errno(r, "Failed to create '%s': %m", dn);
2277
2278 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
2279 return log_oom();
2280
2281 prefixed = path_join(dest, sl);
2282 if (!prefixed)
2283 return log_oom();
2284
2285 t = path_join("..", d);
2286 if (!t)
2287 return log_oom();
2288
2289 if (symlink(t, prefixed) < 0)
2290 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
2291 }
2292 }
2293
2294 return r;
2295 }
2296
2297 static int make_extra_nodes(const char *dest) {
2298 size_t i;
2299 int r;
2300
2301 BLOCK_WITH_UMASK(0000);
2302
2303 for (i = 0; i < arg_n_extra_nodes; i++) {
2304 _cleanup_free_ char *path = NULL;
2305 DeviceNode *n = arg_extra_nodes + i;
2306
2307 path = path_join(dest, n->path);
2308 if (!path)
2309 return log_oom();
2310
2311 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
2312 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
2313
2314 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
2315 if (r < 0)
2316 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
2317 }
2318
2319 return 0;
2320 }
2321
2322 static int setup_pts(const char *dest) {
2323 _cleanup_free_ char *options = NULL;
2324 const char *p;
2325 int r;
2326
2327 #if HAVE_SELINUX
2328 if (arg_selinux_apifs_context)
2329 (void) asprintf(&options,
2330 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
2331 arg_uid_shift + TTY_GID,
2332 arg_selinux_apifs_context);
2333 else
2334 #endif
2335 (void) asprintf(&options,
2336 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
2337 arg_uid_shift + TTY_GID);
2338
2339 if (!options)
2340 return log_oom();
2341
2342 /* Mount /dev/pts itself */
2343 p = prefix_roota(dest, "/dev/pts");
2344 r = RET_NERRNO(mkdir(p, 0755));
2345 if (r < 0)
2346 return log_error_errno(r, "Failed to create /dev/pts: %m");
2347
2348 r = mount_nofollow_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2349 if (r < 0)
2350 return r;
2351 r = userns_lchown(p, 0, 0);
2352 if (r < 0)
2353 return log_error_errno(r, "Failed to chown /dev/pts: %m");
2354
2355 /* Create /dev/ptmx symlink */
2356 p = prefix_roota(dest, "/dev/ptmx");
2357 if (symlink("pts/ptmx", p) < 0)
2358 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
2359 r = userns_lchown(p, 0, 0);
2360 if (r < 0)
2361 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
2362
2363 /* And fix /dev/pts/ptmx ownership */
2364 p = prefix_roota(dest, "/dev/pts/ptmx");
2365 r = userns_lchown(p, 0, 0);
2366 if (r < 0)
2367 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
2368
2369 return 0;
2370 }
2371
2372 static int setup_stdio_as_dev_console(void) {
2373 _cleanup_close_ int terminal = -1;
2374 int r;
2375
2376 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2377 * explicitly, if we are configured to. */
2378 terminal = open_terminal("/dev/console", O_RDWR|O_NOCTTY);
2379 if (terminal < 0)
2380 return log_error_errno(terminal, "Failed to open console: %m");
2381
2382 /* Make sure we can continue logging to the original stderr, even if
2383 * stderr points elsewhere now */
2384 r = log_dup_console();
2385 if (r < 0)
2386 return log_error_errno(r, "Failed to duplicate stderr: %m");
2387
2388 /* invalidates 'terminal' on success and failure */
2389 r = rearrange_stdio(terminal, terminal, terminal);
2390 TAKE_FD(terminal);
2391 if (r < 0)
2392 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2393
2394 return 0;
2395 }
2396
2397 static int setup_dev_console(const char *console) {
2398 _cleanup_free_ char *p = NULL;
2399 int r;
2400
2401 /* Create /dev/console symlink */
2402 r = path_make_relative("/dev", console, &p);
2403 if (r < 0)
2404 return log_error_errno(r, "Failed to create relative path: %m");
2405
2406 if (symlink(p, "/dev/console") < 0)
2407 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
2408
2409 return 0;
2410 }
2411
2412 static int setup_keyring(void) {
2413 key_serial_t keyring;
2414
2415 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2416 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2417 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2418 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2419 * into the container. */
2420
2421 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2422 if (keyring == -1) {
2423 if (errno == ENOSYS)
2424 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2425 else if (ERRNO_IS_PRIVILEGE(errno))
2426 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2427 else
2428 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2429 }
2430
2431 return 0;
2432 }
2433
2434 static int setup_credentials(const char *root) {
2435 const char *q;
2436 int r;
2437
2438 if (arg_n_credentials <= 0)
2439 return 0;
2440
2441 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2442 if (r < 0)
2443 return log_error_errno(r, "Failed to create /run/host: %m");
2444
2445 r = userns_mkdir(root, "/run/host/credentials", 0700, 0, 0);
2446 if (r < 0)
2447 return log_error_errno(r, "Failed to create /run/host/credentials: %m");
2448
2449 q = prefix_roota(root, "/run/host/credentials");
2450 r = mount_nofollow_verbose(LOG_ERR, NULL, q, "ramfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0700");
2451 if (r < 0)
2452 return r;
2453
2454 for (size_t i = 0; i < arg_n_credentials; i++) {
2455 _cleanup_free_ char *j = NULL;
2456 _cleanup_close_ int fd = -1;
2457
2458 j = path_join(q, arg_credentials[i].id);
2459 if (!j)
2460 return log_oom();
2461
2462 fd = open(j, O_CREAT|O_EXCL|O_WRONLY|O_CLOEXEC|O_NOFOLLOW, 0600);
2463 if (fd < 0)
2464 return log_error_errno(errno, "Failed to create credential file %s: %m", j);
2465
2466 r = loop_write(fd, arg_credentials[i].data, arg_credentials[i].size, /* do_poll= */ false);
2467 if (r < 0)
2468 return log_error_errno(r, "Failed to write credential to file %s: %m", j);
2469
2470 if (fchmod(fd, 0400) < 0)
2471 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", j);
2472
2473 if (arg_userns_mode != USER_NAMESPACE_NO) {
2474 if (fchown(fd, arg_uid_shift, arg_uid_shift) < 0)
2475 return log_error_errno(errno, "Failed to adjust ownership of %s: %m", j);
2476 }
2477 }
2478
2479 if (chmod(q, 0500) < 0)
2480 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", q);
2481
2482 r = userns_lchown(q, 0, 0);
2483 if (r < 0)
2484 return r;
2485
2486 /* Make both mount and superblock read-only now */
2487 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2488 if (r < 0)
2489 return r;
2490
2491 return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0500");
2492 }
2493
2494 static int setup_kmsg(int kmsg_socket) {
2495 _cleanup_(unlink_and_freep) char *from = NULL;
2496 _cleanup_free_ char *fifo = NULL;
2497 _cleanup_close_ int fd = -1;
2498 int r;
2499
2500 assert(kmsg_socket >= 0);
2501
2502 BLOCK_WITH_UMASK(0000);
2503
2504 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2505 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2506 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2507 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2508
2509 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2510 if (r < 0)
2511 return log_error_errno(r, "Failed to generate kmsg path: %m");
2512
2513 if (mkfifo(fifo, 0600) < 0)
2514 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2515
2516 from = TAKE_PTR(fifo);
2517
2518 r = mount_nofollow_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2519 if (r < 0)
2520 return r;
2521
2522 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2523 if (fd < 0)
2524 return log_error_errno(errno, "Failed to open fifo: %m");
2525
2526 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2527 r = send_one_fd(kmsg_socket, fd, 0);
2528 if (r < 0)
2529 return log_error_errno(r, "Failed to send FIFO fd: %m");
2530
2531 return 0;
2532 }
2533
2534 struct ExposeArgs {
2535 union in_addr_union address4;
2536 union in_addr_union address6;
2537 struct FirewallContext *fw_ctx;
2538 };
2539
2540 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2541 struct ExposeArgs *args = userdata;
2542
2543 assert(rtnl);
2544 assert(m);
2545 assert(args);
2546
2547 (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET, &args->address4);
2548 (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET6, &args->address6);
2549 return 0;
2550 }
2551
2552 static int setup_hostname(void) {
2553 int r;
2554
2555 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2556 return 0;
2557
2558 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2559 if (r < 0)
2560 return log_error_errno(r, "Failed to set hostname: %m");
2561
2562 return 0;
2563 }
2564
2565 static int setup_journal(const char *directory) {
2566 _cleanup_free_ char *d = NULL;
2567 const char *dirname, *p, *q;
2568 sd_id128_t this_id;
2569 bool try;
2570 int r;
2571
2572 /* Don't link journals in ephemeral mode */
2573 if (arg_ephemeral)
2574 return 0;
2575
2576 if (arg_link_journal == LINK_NO)
2577 return 0;
2578
2579 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2580
2581 r = sd_id128_get_machine(&this_id);
2582 if (r < 0)
2583 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2584
2585 if (sd_id128_equal(arg_uuid, this_id)) {
2586 log_full(try ? LOG_WARNING : LOG_ERR,
2587 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid));
2588 if (try)
2589 return 0;
2590 return -EEXIST;
2591 }
2592
2593 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2594 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2595 if (r < 0) {
2596 bool ignore = r == -EROFS && try;
2597 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2598 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2599 return ignore ? 0 : r;
2600 }
2601 }
2602
2603 p = strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid));
2604 q = prefix_roota(directory, p);
2605
2606 if (path_is_mount_point(p, NULL, 0) > 0) {
2607 if (try)
2608 return 0;
2609
2610 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2611 "%s: already a mount point, refusing to use for journal", p);
2612 }
2613
2614 if (path_is_mount_point(q, NULL, 0) > 0) {
2615 if (try)
2616 return 0;
2617
2618 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2619 "%s: already a mount point, refusing to use for journal", q);
2620 }
2621
2622 r = readlink_and_make_absolute(p, &d);
2623 if (r >= 0) {
2624 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2625 path_equal(d, q)) {
2626
2627 r = userns_mkdir(directory, p, 0755, 0, 0);
2628 if (r < 0)
2629 log_warning_errno(r, "Failed to create directory %s: %m", q);
2630 return 0;
2631 }
2632
2633 if (unlink(p) < 0)
2634 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2635 } else if (r == -EINVAL) {
2636
2637 if (arg_link_journal == LINK_GUEST &&
2638 rmdir(p) < 0) {
2639
2640 if (errno == ENOTDIR) {
2641 log_error("%s already exists and is neither a symlink nor a directory", p);
2642 return r;
2643 } else
2644 return log_error_errno(errno, "Failed to remove %s: %m", p);
2645 }
2646 } else if (r != -ENOENT)
2647 return log_error_errno(r, "readlink(%s) failed: %m", p);
2648
2649 if (arg_link_journal == LINK_GUEST) {
2650
2651 if (symlink(q, p) < 0) {
2652 if (try) {
2653 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2654 return 0;
2655 } else
2656 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2657 }
2658
2659 r = userns_mkdir(directory, p, 0755, 0, 0);
2660 if (r < 0)
2661 log_warning_errno(r, "Failed to create directory %s: %m", q);
2662 return 0;
2663 }
2664
2665 if (arg_link_journal == LINK_HOST) {
2666 /* don't create parents here — if the host doesn't have
2667 * permanent journal set up, don't force it here */
2668
2669 r = RET_NERRNO(mkdir(p, 0755));
2670 if (r < 0 && r != -EEXIST) {
2671 if (try) {
2672 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2673 return 0;
2674 } else
2675 return log_error_errno(r, "Failed to create %s: %m", p);
2676 }
2677
2678 } else if (access(p, F_OK) < 0)
2679 return 0;
2680
2681 if (dir_is_empty(q) == 0)
2682 log_warning("%s is not empty, proceeding anyway.", q);
2683
2684 r = userns_mkdir(directory, p, 0755, 0, 0);
2685 if (r < 0)
2686 return log_error_errno(r, "Failed to create %s: %m", q);
2687
2688 r = mount_nofollow_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2689 if (r < 0)
2690 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2691
2692 return 0;
2693 }
2694
2695 static int drop_capabilities(uid_t uid) {
2696 CapabilityQuintet q;
2697
2698 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2699 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2700 * arg_caps_retain. */
2701
2702 if (capability_quintet_is_set(&arg_full_capabilities)) {
2703 q = arg_full_capabilities;
2704
2705 if (q.bounding == UINT64_MAX)
2706 q.bounding = uid == 0 ? arg_caps_retain : 0;
2707
2708 if (q.effective == UINT64_MAX)
2709 q.effective = uid == 0 ? q.bounding : 0;
2710
2711 if (q.inheritable == UINT64_MAX)
2712 q.inheritable = uid == 0 ? q.bounding : arg_caps_ambient;
2713
2714 if (q.permitted == UINT64_MAX)
2715 q.permitted = uid == 0 ? q.bounding : arg_caps_ambient;
2716
2717 if (q.ambient == UINT64_MAX && ambient_capabilities_supported())
2718 q.ambient = arg_caps_ambient;
2719
2720 if (capability_quintet_mangle(&q))
2721 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2722
2723 } else {
2724 q = (CapabilityQuintet) {
2725 .bounding = arg_caps_retain,
2726 .effective = uid == 0 ? arg_caps_retain : 0,
2727 .inheritable = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2728 .permitted = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2729 .ambient = ambient_capabilities_supported() ? arg_caps_ambient : UINT64_MAX,
2730 };
2731
2732 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2733 * in order to maintain the same behavior as systemd < 242. */
2734 if (capability_quintet_mangle(&q))
2735 log_full(arg_quiet ? LOG_DEBUG : LOG_WARNING,
2736 "Some capabilities will not be set because they are not in the current bounding set.");
2737
2738 }
2739
2740 return capability_quintet_enforce(&q);
2741 }
2742
2743 static int reset_audit_loginuid(void) {
2744 _cleanup_free_ char *p = NULL;
2745 int r;
2746
2747 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2748 return 0;
2749
2750 r = read_one_line_file("/proc/self/loginuid", &p);
2751 if (r == -ENOENT)
2752 return 0;
2753 if (r < 0)
2754 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2755
2756 /* Already reset? */
2757 if (streq(p, "4294967295"))
2758 return 0;
2759
2760 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2761 if (r < 0) {
2762 log_error_errno(r,
2763 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2764 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2765 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2766 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2767 "using systemd-nspawn. Sleeping for 5s... (%m)");
2768
2769 sleep(5);
2770 }
2771
2772 return 0;
2773 }
2774
2775 static int setup_propagate(const char *root) {
2776 const char *p, *q;
2777 int r;
2778
2779 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2780 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2781 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2782 (void) mkdir_p(p, 0600);
2783
2784 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2785 if (r < 0)
2786 return log_error_errno(r, "Failed to create /run/host: %m");
2787
2788 r = userns_mkdir(root, "/run/host/incoming", 0600, 0, 0);
2789 if (r < 0)
2790 return log_error_errno(r, "Failed to create /run/host/incoming: %m");
2791
2792 q = prefix_roota(root, "/run/host/incoming");
2793 r = mount_nofollow_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2794 if (r < 0)
2795 return r;
2796
2797 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2798 if (r < 0)
2799 return r;
2800
2801 /* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
2802 return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2803 }
2804
2805 static int setup_machine_id(const char *directory) {
2806 const char *etc_machine_id;
2807 sd_id128_t id;
2808 int r;
2809
2810 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2811 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2812 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2813 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2814 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2815 * container behaves nicely). */
2816
2817 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2818
2819 r = id128_read(etc_machine_id, ID128_PLAIN_OR_UNINIT, &id);
2820 if (r < 0) {
2821 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2822 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2823
2824 if (sd_id128_is_null(arg_uuid)) {
2825 r = sd_id128_randomize(&arg_uuid);
2826 if (r < 0)
2827 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2828 }
2829 } else {
2830 if (sd_id128_is_null(id))
2831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2832 "Machine ID in container image is zero, refusing.");
2833
2834 arg_uuid = id;
2835 }
2836
2837 return 0;
2838 }
2839
2840 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2841 int r;
2842
2843 assert(directory);
2844
2845 if (arg_userns_mode == USER_NAMESPACE_NO || arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_CHOWN)
2846 return 0;
2847
2848 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2849 if (r == -EOPNOTSUPP)
2850 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.");
2851 if (r == -EBADE)
2852 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2853 if (r < 0)
2854 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2855 if (r == 0)
2856 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2857 else
2858 log_debug("Patched directory tree to match UID/GID range.");
2859
2860 return r;
2861 }
2862
2863 /*
2864 * Return values:
2865 * < 0 : wait_for_terminate() failed to get the state of the
2866 * container, the container was terminated by a signal, or
2867 * failed for an unknown reason. No change is made to the
2868 * container argument.
2869 * > 0 : The program executed in the container terminated with an
2870 * error. The exit code of the program executed in the
2871 * container is returned. The container argument has been set
2872 * to CONTAINER_TERMINATED.
2873 * 0 : The container is being rebooted, has been shut down or exited
2874 * successfully. The container argument has been set to either
2875 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2876 *
2877 * That is, success is indicated by a return value of zero, and an
2878 * error is indicated by a non-zero value.
2879 */
2880 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2881 siginfo_t status;
2882 int r;
2883
2884 r = wait_for_terminate(pid, &status);
2885 if (r < 0)
2886 return log_warning_errno(r, "Failed to wait for container: %m");
2887
2888 switch (status.si_code) {
2889
2890 case CLD_EXITED:
2891 if (status.si_status == 0)
2892 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2893 else
2894 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2895
2896 *container = CONTAINER_TERMINATED;
2897 return status.si_status;
2898
2899 case CLD_KILLED:
2900 if (status.si_status == SIGINT) {
2901 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2902 *container = CONTAINER_TERMINATED;
2903 return 0;
2904
2905 } else if (status.si_status == SIGHUP) {
2906 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2907 *container = CONTAINER_REBOOTED;
2908 return 0;
2909 }
2910
2911 _fallthrough_;
2912 case CLD_DUMPED:
2913 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2914 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2915
2916 default:
2917 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2918 "Container %s failed due to unknown reason.", arg_machine);
2919 }
2920 }
2921
2922 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2923 pid_t pid;
2924
2925 pid = PTR_TO_PID(userdata);
2926 if (pid > 0) {
2927 if (kill(pid, arg_kill_signal) >= 0) {
2928 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2929 sd_event_source_set_userdata(s, NULL);
2930 return 0;
2931 }
2932 }
2933
2934 sd_event_exit(sd_event_source_get_event(s), 0);
2935 return 0;
2936 }
2937
2938 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2939 pid_t pid;
2940
2941 assert(s);
2942 assert(ssi);
2943
2944 pid = PTR_TO_PID(userdata);
2945
2946 for (;;) {
2947 siginfo_t si = {};
2948
2949 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2950 return log_error_errno(errno, "Failed to waitid(): %m");
2951 if (si.si_pid == 0) /* No pending children. */
2952 break;
2953 if (si.si_pid == pid) {
2954 /* The main process we care for has exited. Return from
2955 * signal handler but leave the zombie. */
2956 sd_event_exit(sd_event_source_get_event(s), 0);
2957 break;
2958 }
2959
2960 /* Reap all other children. */
2961 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2962 }
2963
2964 return 0;
2965 }
2966
2967 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2968 pid_t pid;
2969
2970 assert(m);
2971
2972 pid = PTR_TO_PID(userdata);
2973
2974 if (arg_kill_signal > 0) {
2975 log_info("Container termination requested. Attempting to halt container.");
2976 (void) kill(pid, arg_kill_signal);
2977 } else {
2978 log_info("Container termination requested. Exiting.");
2979 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2980 }
2981
2982 return 0;
2983 }
2984
2985 static int determine_names(void) {
2986 int r;
2987
2988 if (arg_template && !arg_directory && arg_machine) {
2989
2990 /* If --template= was specified then we should not
2991 * search for a machine, but instead create a new one
2992 * in /var/lib/machine. */
2993
2994 arg_directory = path_join("/var/lib/machines", arg_machine);
2995 if (!arg_directory)
2996 return log_oom();
2997 }
2998
2999 if (!arg_image && !arg_directory) {
3000 if (arg_machine) {
3001 _cleanup_(image_unrefp) Image *i = NULL;
3002
3003 r = image_find(IMAGE_MACHINE, arg_machine, NULL, &i);
3004 if (r == -ENOENT)
3005 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
3006 if (r < 0)
3007 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
3008
3009 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
3010 r = free_and_strdup(&arg_image, i->path);
3011 else
3012 r = free_and_strdup(&arg_directory, i->path);
3013 if (r < 0)
3014 return log_oom();
3015
3016 if (!arg_ephemeral)
3017 arg_read_only = arg_read_only || i->read_only;
3018 } else {
3019 r = safe_getcwd(&arg_directory);
3020 if (r < 0)
3021 return log_error_errno(r, "Failed to determine current directory: %m");
3022 }
3023
3024 if (!arg_directory && !arg_image)
3025 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
3026 }
3027
3028 if (!arg_machine) {
3029 if (arg_directory && path_equal(arg_directory, "/"))
3030 arg_machine = gethostname_malloc();
3031 else if (arg_image) {
3032 char *e;
3033
3034 arg_machine = strdup(basename(arg_image));
3035
3036 /* Truncate suffix if there is one */
3037 e = endswith(arg_machine, ".raw");
3038 if (e)
3039 *e = 0;
3040 } else
3041 arg_machine = strdup(basename(arg_directory));
3042 if (!arg_machine)
3043 return log_oom();
3044
3045 hostname_cleanup(arg_machine);
3046 if (!hostname_is_valid(arg_machine, 0))
3047 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
3048
3049 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3050 * instances at once without manually having to specify -M each time. */
3051 if (arg_ephemeral)
3052 if (strextendf(&arg_machine, "-%016" PRIx64, random_u64()) < 0)
3053 return log_oom();
3054 }
3055
3056 return 0;
3057 }
3058
3059 static int chase_symlinks_and_update(char **p, unsigned flags) {
3060 char *chased;
3061 int r;
3062
3063 assert(p);
3064
3065 if (!*p)
3066 return 0;
3067
3068 r = chase_symlinks(*p, NULL, flags, &chased, NULL);
3069 if (r < 0)
3070 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
3071
3072 return free_and_replace(*p, chased);
3073 }
3074
3075 static int determine_uid_shift(const char *directory) {
3076
3077 if (arg_userns_mode == USER_NAMESPACE_NO) {
3078 arg_uid_shift = 0;
3079 return 0;
3080 }
3081
3082 if (arg_uid_shift == UID_INVALID) {
3083 struct stat st;
3084
3085 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3086
3087 if (stat(directory, &st) < 0)
3088 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
3089
3090 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
3091
3092 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
3093 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3094 "UID and GID base of %s don't match.", directory);
3095
3096 arg_uid_range = UINT32_C(0x10000);
3097
3098 if (arg_uid_shift != 0) {
3099 /* If the image is shifted already, then we'll fall back to classic chowning, for
3100 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3101
3102 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_AUTO) {
3103 log_debug("UID base of %s is non-zero, not using UID mapping.", directory);
3104 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
3105 } else if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_MAP)
3106 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3107 "UID base of %s is not zero, UID mapping not supported.", directory);
3108 }
3109 }
3110
3111 if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
3112 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID base too high for UID range.");
3113
3114 return 0;
3115 }
3116
3117 static unsigned long effective_clone_ns_flags(void) {
3118 unsigned long flags = arg_clone_ns_flags;
3119
3120 if (arg_private_network)
3121 flags |= CLONE_NEWNET;
3122 if (arg_use_cgns)
3123 flags |= CLONE_NEWCGROUP;
3124 if (arg_userns_mode != USER_NAMESPACE_NO)
3125 flags |= CLONE_NEWUSER;
3126
3127 return flags;
3128 }
3129
3130 static int patch_sysctl(void) {
3131
3132 /* This table is inspired by runc's sysctl() function */
3133 static const struct {
3134 const char *key;
3135 bool prefix;
3136 unsigned long clone_flags;
3137 } safe_sysctl[] = {
3138 { "kernel.hostname", false, CLONE_NEWUTS },
3139 { "kernel.domainname", false, CLONE_NEWUTS },
3140 { "kernel.msgmax", false, CLONE_NEWIPC },
3141 { "kernel.msgmnb", false, CLONE_NEWIPC },
3142 { "kernel.msgmni", false, CLONE_NEWIPC },
3143 { "kernel.sem", false, CLONE_NEWIPC },
3144 { "kernel.shmall", false, CLONE_NEWIPC },
3145 { "kernel.shmmax", false, CLONE_NEWIPC },
3146 { "kernel.shmmni", false, CLONE_NEWIPC },
3147 { "fs.mqueue.", true, CLONE_NEWIPC },
3148 { "net.", true, CLONE_NEWNET },
3149 };
3150
3151 unsigned long flags;
3152 char **k, **v;
3153 int r;
3154
3155 flags = effective_clone_ns_flags();
3156
3157 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
3158 bool good = false;
3159 size_t i;
3160
3161 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
3162
3163 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
3164 continue;
3165
3166 if (safe_sysctl[i].prefix)
3167 good = startswith(*k, safe_sysctl[i].key);
3168 else
3169 good = streq(*k, safe_sysctl[i].key);
3170
3171 if (good)
3172 break;
3173 }
3174
3175 if (!good)
3176 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
3177
3178 r = sysctl_write(*k, *v);
3179 if (r < 0)
3180 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
3181 }
3182
3183 return 0;
3184 }
3185
3186 static int inner_child(
3187 Barrier *barrier,
3188 const char *directory,
3189 bool secondary,
3190 int kmsg_socket,
3191 int rtnl_socket,
3192 int master_pty_socket,
3193 FDSet *fds,
3194 char **os_release_pairs) {
3195
3196 _cleanup_free_ char *home = NULL;
3197 size_t n_env = 1;
3198 char *envp[] = {
3199 (char*) "PATH=" DEFAULT_PATH_COMPAT,
3200 NULL, /* container */
3201 NULL, /* TERM */
3202 NULL, /* HOME */
3203 NULL, /* USER */
3204 NULL, /* LOGNAME */
3205 NULL, /* container_uuid */
3206 NULL, /* LISTEN_FDS */
3207 NULL, /* LISTEN_PID */
3208 NULL, /* NOTIFY_SOCKET */
3209 NULL, /* CREDENTIALS_DIRECTORY */
3210 NULL
3211 };
3212 const char *exec_target;
3213 _cleanup_strv_free_ char **env_use = NULL;
3214 int r, which_failed;
3215
3216 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3217 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3218 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3219 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3220 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3221 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3222 * namespace.
3223 *
3224 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3225 * unshare(). See below. */
3226
3227 assert(barrier);
3228 assert(directory);
3229 assert(kmsg_socket >= 0);
3230
3231 log_debug("Inner child is initializing.");
3232
3233 if (arg_userns_mode != USER_NAMESPACE_NO) {
3234 /* Tell the parent, that it now can write the UID map. */
3235 (void) barrier_place(barrier); /* #1 */
3236
3237 /* Wait until the parent wrote the UID map */
3238 if (!barrier_place_and_sync(barrier)) /* #2 */
3239 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3240
3241 /* Become the new root user inside our namespace */
3242 r = reset_uid_gid();
3243 if (r < 0)
3244 return log_error_errno(r, "Couldn't become new root: %m");
3245
3246 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3247 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3248 * propagation, but simply create new peer groups for all our mounts). */
3249 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SHARED|MS_REC, NULL);
3250 if (r < 0)
3251 return r;
3252 }
3253
3254 r = mount_all(NULL,
3255 arg_mount_settings | MOUNT_IN_USERNS,
3256 arg_uid_shift,
3257 arg_selinux_apifs_context);
3258 if (r < 0)
3259 return r;
3260
3261 if (!arg_network_namespace_path && arg_private_network) {
3262 r = unshare(CLONE_NEWNET);
3263 if (r < 0)
3264 return log_error_errno(errno, "Failed to unshare network namespace: %m");
3265
3266 /* Tell the parent that it can setup network interfaces. */
3267 (void) barrier_place(barrier); /* #3 */
3268 }
3269
3270 r = mount_sysfs(NULL, arg_mount_settings);
3271 if (r < 0)
3272 return r;
3273
3274 /* Wait until we are cgroup-ified, so that we
3275 * can mount the right cgroup path writable */
3276 if (!barrier_place_and_sync(barrier)) /* #4 */
3277 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3278 "Parent died too early");
3279
3280 if (arg_use_cgns) {
3281 r = unshare(CLONE_NEWCGROUP);
3282 if (r < 0)
3283 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
3284 r = mount_cgroups(
3285 "",
3286 arg_unified_cgroup_hierarchy,
3287 arg_userns_mode != USER_NAMESPACE_NO,
3288 arg_uid_shift,
3289 arg_uid_range,
3290 arg_selinux_apifs_context,
3291 true);
3292 } else
3293 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
3294 if (r < 0)
3295 return r;
3296
3297 r = setup_boot_id();
3298 if (r < 0)
3299 return r;
3300
3301 r = setup_kmsg(kmsg_socket);
3302 if (r < 0)
3303 return r;
3304 kmsg_socket = safe_close(kmsg_socket);
3305
3306 r = mount_custom(
3307 "/",
3308 arg_custom_mounts,
3309 arg_n_custom_mounts,
3310 0,
3311 0,
3312 arg_selinux_apifs_context,
3313 MOUNT_NON_ROOT_ONLY | MOUNT_IN_USERNS);
3314 if (r < 0)
3315 return r;
3316
3317 if (setsid() < 0)
3318 return log_error_errno(errno, "setsid() failed: %m");
3319
3320 if (arg_private_network)
3321 (void) loopback_setup();
3322
3323 if (arg_expose_ports) {
3324 r = expose_port_send_rtnl(rtnl_socket);
3325 if (r < 0)
3326 return r;
3327 rtnl_socket = safe_close(rtnl_socket);
3328 }
3329
3330 if (arg_console_mode != CONSOLE_PIPE) {
3331 _cleanup_close_ int master = -1;
3332 _cleanup_free_ char *console = NULL;
3333
3334 /* Allocate a pty and make it available as /dev/console. */
3335 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
3336 if (master < 0)
3337 return log_error_errno(master, "Failed to allocate a pty: %m");
3338
3339 r = setup_dev_console(console);
3340 if (r < 0)
3341 return log_error_errno(r, "Failed to set up /dev/console: %m");
3342
3343 r = send_one_fd(master_pty_socket, master, 0);
3344 if (r < 0)
3345 return log_error_errno(r, "Failed to send master fd: %m");
3346 master_pty_socket = safe_close(master_pty_socket);
3347
3348 r = setup_stdio_as_dev_console();
3349 if (r < 0)
3350 return r;
3351 }
3352
3353 r = patch_sysctl();
3354 if (r < 0)
3355 return r;
3356
3357 if (arg_oom_score_adjust_set) {
3358 r = set_oom_score_adjust(arg_oom_score_adjust);
3359 if (r < 0)
3360 return log_error_errno(r, "Failed to adjust OOM score: %m");
3361 }
3362
3363 if (arg_cpu_set.set)
3364 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
3365 return log_error_errno(errno, "Failed to set CPU affinity: %m");
3366
3367 (void) setup_hostname();
3368
3369 if (arg_personality != PERSONALITY_INVALID) {
3370 r = safe_personality(arg_personality);
3371 if (r < 0)
3372 return log_error_errno(r, "personality() failed: %m");
3373 } else if (secondary) {
3374 r = safe_personality(PER_LINUX32);
3375 if (r < 0)
3376 return log_error_errno(r, "personality() failed: %m");
3377 }
3378
3379 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3380 if (r < 0)
3381 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3382
3383 #if HAVE_SECCOMP
3384 if (arg_seccomp) {
3385
3386 if (is_seccomp_available()) {
3387
3388 r = seccomp_load(arg_seccomp);
3389 if (ERRNO_IS_SECCOMP_FATAL(r))
3390 return log_error_errno(r, "Failed to install seccomp filter: %m");
3391 if (r < 0)
3392 log_debug_errno(r, "Failed to install seccomp filter: %m");
3393 }
3394 } else
3395 #endif
3396 {
3397 r = setup_seccomp(arg_caps_retain, arg_syscall_allow_list, arg_syscall_deny_list);
3398 if (r < 0)
3399 return r;
3400 }
3401
3402 if (arg_suppress_sync) {
3403 #if HAVE_SECCOMP
3404 r = seccomp_suppress_sync();
3405 if (r < 0)
3406 log_debug_errno(r, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3407 #else
3408 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3409 #endif
3410 }
3411
3412 #if HAVE_SELINUX
3413 if (arg_selinux_context)
3414 if (setexeccon(arg_selinux_context) < 0)
3415 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3416 #endif
3417
3418 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3419 * if we need to later on. */
3420 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3421 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3422
3423 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3424 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids, arg_console_mode != CONSOLE_PIPE);
3425 else
3426 r = change_uid_gid(arg_user, arg_console_mode != CONSOLE_PIPE, &home);
3427 if (r < 0)
3428 return r;
3429
3430 r = drop_capabilities(getuid());
3431 if (r < 0)
3432 return log_error_errno(r, "Dropping capabilities failed: %m");
3433
3434 if (arg_no_new_privileges)
3435 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3436 return log_error_errno(errno, "Failed to disable new privileges: %m");
3437
3438 /* LXC sets container=lxc, so follow the scheme here */
3439 envp[n_env++] = strjoina("container=", arg_container_service_name);
3440
3441 envp[n_env] = strv_find_prefix(environ, "TERM=");
3442 if (envp[n_env])
3443 n_env++;
3444
3445 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3446 if (asprintf(envp + n_env++, "HOME=%s", home ?: "/root") < 0)
3447 return log_oom();
3448
3449 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3450 if (asprintf(envp + n_env++, "USER=%s", arg_user ?: "root") < 0 ||
3451 asprintf(envp + n_env++, "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3452 return log_oom();
3453
3454 assert(!sd_id128_is_null(arg_uuid));
3455
3456 if (asprintf(envp + n_env++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid)) < 0)
3457 return log_oom();
3458
3459 if (fdset_size(fds) > 0) {
3460 r = fdset_cloexec(fds, false);
3461 if (r < 0)
3462 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3463
3464 if ((asprintf(envp + n_env++, "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3465 (asprintf(envp + n_env++, "LISTEN_PID=1") < 0))
3466 return log_oom();
3467 }
3468 if (asprintf(envp + n_env++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3469 return log_oom();
3470
3471 if (arg_n_credentials > 0) {
3472 envp[n_env] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3473 if (!envp[n_env])
3474 return log_oom();
3475 n_env++;
3476 }
3477
3478 env_use = strv_env_merge(envp, os_release_pairs, arg_setenv);
3479 if (!env_use)
3480 return log_oom();
3481
3482 /* Let the parent know that we are ready and
3483 * wait until the parent is ready with the
3484 * setup, too... */
3485 if (!barrier_place_and_sync(barrier)) /* #5 */
3486 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3487
3488 if (arg_chdir)
3489 if (chdir(arg_chdir) < 0)
3490 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3491
3492 if (arg_start_mode == START_PID2) {
3493 r = stub_pid1(arg_uuid);
3494 if (r < 0)
3495 return r;
3496 }
3497
3498 if (arg_console_mode != CONSOLE_PIPE) {
3499 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3500 * are configured for that. Acquire it as controlling tty. */
3501 if (ioctl(STDIN_FILENO, TIOCSCTTY) < 0)
3502 return log_error_errno(errno, "Failed to acquire controlling TTY: %m");
3503 }
3504
3505 log_debug("Inner child completed, invoking payload.");
3506
3507 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3508 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3509 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3510 log_close();
3511 log_set_open_when_needed(true);
3512
3513 (void) fdset_close_others(fds);
3514
3515 if (arg_start_mode == START_BOOT) {
3516 const char *init;
3517 char **a;
3518 size_t m;
3519
3520 /* Automatically search for the init system */
3521
3522 m = strv_length(arg_parameters);
3523 a = newa(char*, m + 2);
3524 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3525 a[1 + m] = NULL;
3526
3527 FOREACH_STRING(init,
3528 "/usr/lib/systemd/systemd",
3529 "/lib/systemd/systemd",
3530 "/sbin/init") {
3531 a[0] = (char*) init;
3532 execve(a[0], a, env_use);
3533 }
3534
3535 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3536 } else if (!strv_isempty(arg_parameters)) {
3537 const char *dollar_path;
3538
3539 exec_target = arg_parameters[0];
3540
3541 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3542 * binary. */
3543 dollar_path = strv_env_get(env_use, "PATH");
3544 if (dollar_path) {
3545 if (setenv("PATH", dollar_path, 1) < 0)
3546 return log_error_errno(errno, "Failed to update $PATH: %m");
3547 }
3548
3549 execvpe(arg_parameters[0], arg_parameters, env_use);
3550 } else {
3551 if (!arg_chdir)
3552 /* If we cannot change the directory, we'll end up in /, that is expected. */
3553 (void) chdir(home ?: "/root");
3554
3555 execle("/bin/bash", "-bash", NULL, env_use);
3556 execle("/bin/sh", "-sh", NULL, env_use);
3557
3558 exec_target = "/bin/bash, /bin/sh";
3559 }
3560
3561 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3562 }
3563
3564 static int setup_notify_child(void) {
3565 _cleanup_close_ int fd = -1;
3566 static const union sockaddr_union sa = {
3567 .un.sun_family = AF_UNIX,
3568 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3569 };
3570 int r;
3571
3572 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3573 if (fd < 0)
3574 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3575
3576 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3577 (void) sockaddr_un_unlink(&sa.un);
3578
3579 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3580 if (r < 0)
3581 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3582
3583 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3584 if (r < 0)
3585 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3586
3587 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3588 if (r < 0)
3589 return log_error_errno(r, "SO_PASSCRED failed: %m");
3590
3591 return TAKE_FD(fd);
3592 }
3593
3594 static int outer_child(
3595 Barrier *barrier,
3596 const char *directory,
3597 DissectedImage *dissected_image,
3598 bool secondary,
3599 int pid_socket,
3600 int uuid_socket,
3601 int notify_socket,
3602 int kmsg_socket,
3603 int rtnl_socket,
3604 int uid_shift_socket,
3605 int master_pty_socket,
3606 int unified_cgroup_hierarchy_socket,
3607 FDSet *fds,
3608 int netns_fd) {
3609
3610 _cleanup_(bind_user_context_freep) BindUserContext *bind_user_context = NULL;
3611 _cleanup_strv_free_ char **os_release_pairs = NULL;
3612 _cleanup_close_ int fd = -1;
3613 bool idmap = false;
3614 const char *p;
3615 pid_t pid;
3616 ssize_t l;
3617 int r;
3618
3619 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3620 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3621 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3622 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3623 * forked off it, and it exits. */
3624
3625 assert(barrier);
3626 assert(directory);
3627 assert(pid_socket >= 0);
3628 assert(uuid_socket >= 0);
3629 assert(notify_socket >= 0);
3630 assert(master_pty_socket >= 0);
3631 assert(kmsg_socket >= 0);
3632
3633 log_debug("Outer child is initializing.");
3634
3635 r = load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs);
3636 if (r < 0)
3637 log_debug_errno(r, "Failed to read os-release from host for container, ignoring: %m");
3638
3639 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3640 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3641
3642 r = reset_audit_loginuid();
3643 if (r < 0)
3644 return r;
3645
3646 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3647 * mounts to the real root. */
3648 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3649 if (r < 0)
3650 return r;
3651
3652 if (dissected_image) {
3653 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3654 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3655 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3656 * right place right away. This makes sure ESP partitions and userns are compatible. */
3657
3658 r = dissected_image_mount_and_warn(
3659 dissected_image,
3660 directory,
3661 arg_uid_shift,
3662 arg_uid_range,
3663 DISSECT_IMAGE_MOUNT_ROOT_ONLY|
3664 DISSECT_IMAGE_DISCARD_ON_LOOP|
3665 DISSECT_IMAGE_USR_NO_ROOT|
3666 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
3667 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3668 if (r < 0)
3669 return r;
3670 }
3671
3672 r = determine_uid_shift(directory);
3673 if (r < 0)
3674 return r;
3675
3676 if (arg_userns_mode != USER_NAMESPACE_NO) {
3677 /* Let the parent know which UID shift we read from the image */
3678 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3679 if (l < 0)
3680 return log_error_errno(errno, "Failed to send UID shift: %m");
3681 if (l != sizeof(arg_uid_shift))
3682 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3683 "Short write while sending UID shift.");
3684
3685 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3686 /* When we are supposed to pick the UID shift, the parent will check now whether the
3687 * UID shift we just read from the image is available. If yes, it will send the UID
3688 * shift back to us, if not it will pick a different one, and send it back to us. */
3689
3690 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3691 if (l < 0)
3692 return log_error_errno(errno, "Failed to recv UID shift: %m");
3693 if (l != sizeof(arg_uid_shift))
3694 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3695 "Short read while receiving UID shift.");
3696 }
3697
3698 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3699 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3700 }
3701
3702 if (path_equal(directory, "/")) {
3703 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3704 * place, so that we can make changes to its mount structure (for example, to implement
3705 * --volatile=) without this interfering with our ability to access files such as
3706 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3707 * (instead of a temporary directory, since we are living in our own mount namspace here
3708 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3709 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3710
3711 r = mount_nofollow_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
3712 if (r < 0)
3713 return r;
3714
3715 directory = "/run/systemd/nspawn-root";
3716 }
3717
3718 r = setup_pivot_root(
3719 directory,
3720 arg_pivot_root_new,
3721 arg_pivot_root_old);
3722 if (r < 0)
3723 return r;
3724
3725 r = setup_volatile_mode(
3726 directory,
3727 arg_volatile_mode,
3728 arg_uid_shift,
3729 arg_selinux_apifs_context);
3730 if (r < 0)
3731 return r;
3732
3733 r = bind_user_prepare(
3734 directory,
3735 arg_bind_user,
3736 arg_uid_shift,
3737 arg_uid_range,
3738 &arg_custom_mounts, &arg_n_custom_mounts,
3739 &bind_user_context);
3740 if (r < 0)
3741 return r;
3742
3743 if (arg_userns_mode != USER_NAMESPACE_NO && bind_user_context) {
3744 /* Send the user maps we determined to the parent, so that it installs it in our user
3745 * namespace UID map table */
3746
3747 for (size_t i = 0; i < bind_user_context->n_data; i++) {
3748 uid_t map[] = {
3749 bind_user_context->data[i].payload_user->uid,
3750 bind_user_context->data[i].host_user->uid,
3751 (uid_t) bind_user_context->data[i].payload_group->gid,
3752 (uid_t) bind_user_context->data[i].host_group->gid,
3753 };
3754
3755 l = send(uid_shift_socket, map, sizeof(map), MSG_NOSIGNAL);
3756 if (l < 0)
3757 return log_error_errno(errno, "Failed to send user UID map: %m");
3758 if (l != sizeof(map))
3759 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3760 "Short write while sending user UID map.");
3761 }
3762 }
3763
3764 r = mount_custom(
3765 directory,
3766 arg_custom_mounts,
3767 arg_n_custom_mounts,
3768 arg_uid_shift,
3769 arg_uid_range,
3770 arg_selinux_apifs_context,
3771 MOUNT_ROOT_ONLY);
3772 if (r < 0)
3773 return r;
3774
3775 /* Make sure we always have a mount that we can move to root later on. */
3776 r = make_mount_point(directory);
3777 if (r < 0)
3778 return r;
3779
3780 if (arg_userns_mode != USER_NAMESPACE_NO &&
3781 IN_SET(arg_userns_ownership, USER_NAMESPACE_OWNERSHIP_MAP, USER_NAMESPACE_OWNERSHIP_AUTO) &&
3782 arg_uid_shift != 0) {
3783
3784 r = remount_idmap(directory, arg_uid_shift, arg_uid_range, REMOUNT_IDMAP_HOST_ROOT);
3785 if (r == -EINVAL || ERRNO_IS_NOT_SUPPORTED(r)) {
3786 /* This might fail because the kernel or file system doesn't support idmapping. We
3787 * can't really distinguish this nicely, nor do we have any guarantees about the
3788 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3789 if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_AUTO)
3790 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
3791 "ID mapped mounts are apparently not available, sorry.");
3792
3793 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3794 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
3795 } else if (r < 0)
3796 return log_error_errno(r, "Failed to set up ID mapped mounts: %m");
3797 else {
3798 log_debug("ID mapped mounts available, making use of them.");
3799 idmap = true;
3800 }
3801 }
3802
3803 if (dissected_image) {
3804 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3805 r = dissected_image_mount(
3806 dissected_image,
3807 directory,
3808 arg_uid_shift,
3809 arg_uid_range,
3810 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|
3811 DISSECT_IMAGE_DISCARD_ON_LOOP|
3812 DISSECT_IMAGE_USR_NO_ROOT|
3813 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
3814 (idmap ? DISSECT_IMAGE_MOUNT_IDMAPPED : 0));
3815 if (r == -EUCLEAN)
3816 return log_error_errno(r, "File system check for image failed: %m");
3817 if (r < 0)
3818 return log_error_errno(r, "Failed to mount image file system: %m");
3819 }
3820
3821 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3822 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3823
3824 r = detect_unified_cgroup_hierarchy_from_image(directory);
3825 if (r < 0)
3826 return r;
3827
3828 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3829 if (l < 0)
3830 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3831 if (l != sizeof(arg_unified_cgroup_hierarchy))
3832 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3833 "Short write while sending cgroup mode.");
3834
3835 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3836 }
3837
3838 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3839 * mounts available in systemd services inside the container that create a new mount namespace. See
3840 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3841 * will inherit the shared propagation mode.
3842 *
3843 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3844 * directory mount to root later on.
3845 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3846 */
3847 r = mount_nofollow_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3848 if (r < 0)
3849 return r;
3850
3851 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3852 if (r < 0)
3853 return r;
3854
3855 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3856 if (r < 0)
3857 return r;
3858
3859 if (arg_read_only && arg_volatile_mode == VOLATILE_NO &&
3860 !has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts)) {
3861 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3862 if (r < 0)
3863 return log_error_errno(r, "Failed to make tree read-only: %m");
3864 }
3865
3866 r = mount_all(directory,
3867 arg_mount_settings,
3868 arg_uid_shift,
3869 arg_selinux_apifs_context);
3870 if (r < 0)
3871 return r;
3872
3873 r = copy_devnodes(directory);
3874 if (r < 0)
3875 return r;
3876
3877 r = make_extra_nodes(directory);
3878 if (r < 0)
3879 return r;
3880
3881 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3882
3883 p = prefix_roota(directory, "/run/host");
3884 (void) make_inaccessible_nodes(p, arg_uid_shift, arg_uid_shift);
3885
3886 r = setup_pts(directory);
3887 if (r < 0)
3888 return r;
3889
3890 r = setup_propagate(directory);
3891 if (r < 0)
3892 return r;
3893
3894 r = setup_keyring();
3895 if (r < 0)
3896 return r;
3897
3898 r = setup_credentials(directory);
3899 if (r < 0)
3900 return r;
3901
3902 r = bind_user_setup(bind_user_context, directory);
3903 if (r < 0)
3904 return r;
3905
3906 r = mount_custom(
3907 directory,
3908 arg_custom_mounts,
3909 arg_n_custom_mounts,
3910 arg_uid_shift,
3911 arg_uid_range,
3912 arg_selinux_apifs_context,
3913 MOUNT_NON_ROOT_ONLY);
3914 if (r < 0)
3915 return r;
3916
3917 r = setup_timezone(directory);
3918 if (r < 0)
3919 return r;
3920
3921 r = setup_resolv_conf(directory);
3922 if (r < 0)
3923 return r;
3924
3925 r = setup_machine_id(directory);
3926 if (r < 0)
3927 return r;
3928
3929 r = setup_journal(directory);
3930 if (r < 0)
3931 return r;
3932
3933 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3934 p = prefix_roota(directory, "/run/host/container-manager");
3935 (void) write_string_file(p, arg_container_service_name, WRITE_STRING_FILE_CREATE);
3936
3937 /* The same stuff as the $container_uuid env var */
3938 p = prefix_roota(directory, "/run/host/container-uuid");
3939 (void) write_string_filef(p, WRITE_STRING_FILE_CREATE, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid));
3940
3941 if (!arg_use_cgns) {
3942 r = mount_cgroups(
3943 directory,
3944 arg_unified_cgroup_hierarchy,
3945 arg_userns_mode != USER_NAMESPACE_NO,
3946 arg_uid_shift,
3947 arg_uid_range,
3948 arg_selinux_apifs_context,
3949 false);
3950 if (r < 0)
3951 return r;
3952 }
3953
3954 r = mount_move_root(directory);
3955 if (r < 0)
3956 return log_error_errno(r, "Failed to move root directory: %m");
3957
3958 fd = setup_notify_child();
3959 if (fd < 0)
3960 return fd;
3961
3962 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3963 arg_clone_ns_flags |
3964 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3965 if (pid < 0)
3966 return log_error_errno(errno, "Failed to fork inner child: %m");
3967 if (pid == 0) {
3968 pid_socket = safe_close(pid_socket);
3969 uuid_socket = safe_close(uuid_socket);
3970 notify_socket = safe_close(notify_socket);
3971 uid_shift_socket = safe_close(uid_shift_socket);
3972
3973 /* The inner child has all namespaces that are requested, so that we all are owned by the
3974 * user if user namespaces are turned on. */
3975
3976 if (arg_network_namespace_path) {
3977 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3978 if (r < 0)
3979 return log_error_errno(r, "Failed to join network namespace: %m");
3980 }
3981
3982 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds, os_release_pairs);
3983 if (r < 0)
3984 _exit(EXIT_FAILURE);
3985
3986 _exit(EXIT_SUCCESS);
3987 }
3988
3989 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3990 if (l < 0)
3991 return log_error_errno(errno, "Failed to send PID: %m");
3992 if (l != sizeof(pid))
3993 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3994 "Short write while sending PID.");
3995
3996 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3997 if (l < 0)
3998 return log_error_errno(errno, "Failed to send machine ID: %m");
3999 if (l != sizeof(arg_uuid))
4000 return log_error_errno(SYNTHETIC_ERRNO(EIO),
4001 "Short write while sending machine ID.");
4002
4003 l = send_one_fd(notify_socket, fd, 0);
4004 if (l < 0)
4005 return log_error_errno(l, "Failed to send notify fd: %m");
4006
4007 pid_socket = safe_close(pid_socket);
4008 uuid_socket = safe_close(uuid_socket);
4009 notify_socket = safe_close(notify_socket);
4010 master_pty_socket = safe_close(master_pty_socket);
4011 kmsg_socket = safe_close(kmsg_socket);
4012 rtnl_socket = safe_close(rtnl_socket);
4013 netns_fd = safe_close(netns_fd);
4014
4015 return 0;
4016 }
4017
4018 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
4019 bool tried_hashed = false;
4020 unsigned n_tries = 100;
4021 uid_t candidate;
4022 int r;
4023
4024 assert(shift);
4025 assert(ret_lock_file);
4026 assert(arg_userns_mode == USER_NAMESPACE_PICK);
4027 assert(arg_uid_range == 0x10000U);
4028
4029 candidate = *shift;
4030
4031 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4032
4033 for (;;) {
4034 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
4035 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
4036
4037 if (--n_tries <= 0)
4038 return -EBUSY;
4039
4040 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
4041 goto next;
4042 if ((candidate & UINT32_C(0xFFFF)) != 0)
4043 goto next;
4044
4045 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
4046 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
4047 if (r == -EBUSY) /* Range already taken by another nspawn instance */
4048 goto next;
4049 if (r < 0)
4050 return r;
4051
4052 /* Make some superficial checks whether the range is currently known in the user database */
4053 if (getpwuid(candidate))
4054 goto next;
4055 if (getpwuid(candidate + UINT32_C(0xFFFE)))
4056 goto next;
4057 if (getgrgid(candidate))
4058 goto next;
4059 if (getgrgid(candidate + UINT32_C(0xFFFE)))
4060 goto next;
4061
4062 *ret_lock_file = lf;
4063 lf = (struct LockFile) LOCK_FILE_INIT;
4064 *shift = candidate;
4065 return 0;
4066
4067 next:
4068 if (arg_machine && !tried_hashed) {
4069 /* Try to hash the base from the container name */
4070
4071 static const uint8_t hash_key[] = {
4072 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4073 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4074 };
4075
4076 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
4077
4078 tried_hashed = true;
4079 } else
4080 random_bytes(&candidate, sizeof(candidate));
4081
4082 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
4083 candidate &= (uid_t) UINT32_C(0xFFFF0000);
4084 }
4085 }
4086
4087 static int add_one_uid_map(
4088 char **p,
4089 uid_t container_uid,
4090 uid_t host_uid,
4091 uid_t range) {
4092
4093 return strextendf(p,
4094 UID_FMT " " UID_FMT " " UID_FMT "\n",
4095 container_uid, host_uid, range);
4096 }
4097
4098 static int make_uid_map_string(
4099 const uid_t bind_user_uid[],
4100 size_t n_bind_user_uid,
4101 size_t offset,
4102 char **ret) {
4103
4104 _cleanup_free_ char *s = NULL;
4105 uid_t previous_uid = 0;
4106 int r;
4107
4108 assert(n_bind_user_uid == 0 || bind_user_uid);
4109 assert(IN_SET(offset, 0, 2)); /* used to switch between UID and GID map */
4110 assert(ret);
4111
4112 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4113 * quadruplet, consisting of host and container UID + GID. */
4114
4115 for (size_t i = 0; i < n_bind_user_uid; i++) {
4116 uid_t payload_uid = bind_user_uid[i*2+offset],
4117 host_uid = bind_user_uid[i*2+offset+1];
4118
4119 assert(previous_uid <= payload_uid);
4120 assert(payload_uid < arg_uid_range);
4121
4122 /* Add a range to close the gap to previous entry */
4123 if (payload_uid > previous_uid) {
4124 r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, payload_uid - previous_uid);
4125 if (r < 0)
4126 return r;
4127 }
4128
4129 /* Map this specific user */
4130 r = add_one_uid_map(&s, payload_uid, host_uid, 1);
4131 if (r < 0)
4132 return r;
4133
4134 previous_uid = payload_uid + 1;
4135 }
4136
4137 /* And add a range to close the gap to finish the range */
4138 if (arg_uid_range > previous_uid) {
4139 r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, arg_uid_range - previous_uid);
4140 if (r < 0)
4141 return r;
4142 }
4143
4144 assert(s);
4145
4146 *ret = TAKE_PTR(s);
4147 return 0;
4148 }
4149
4150 static int setup_uid_map(
4151 pid_t pid,
4152 const uid_t bind_user_uid[],
4153 size_t n_bind_user_uid) {
4154
4155 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1];
4156 _cleanup_free_ char *s = NULL;
4157 int r;
4158
4159 assert(pid > 1);
4160
4161 /* Build the UID map string */
4162 if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 0, &s) < 0) /* offset=0 contains the UID pair */
4163 return log_oom();
4164
4165 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
4166 r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
4167 if (r < 0)
4168 return log_error_errno(r, "Failed to write UID map: %m");
4169
4170 /* And now build the GID map string */
4171 s = mfree(s);
4172 if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 2, &s) < 0) /* offset=2 contains the GID pair */
4173 return log_oom();
4174
4175 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
4176 r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
4177 if (r < 0)
4178 return log_error_errno(r, "Failed to write GID map: %m");
4179
4180 return 0;
4181 }
4182
4183 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4184 char buf[NOTIFY_BUFFER_MAX+1];
4185 char *p = NULL;
4186 struct iovec iovec = {
4187 .iov_base = buf,
4188 .iov_len = sizeof(buf)-1,
4189 };
4190 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
4191 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
4192 struct msghdr msghdr = {
4193 .msg_iov = &iovec,
4194 .msg_iovlen = 1,
4195 .msg_control = &control,
4196 .msg_controllen = sizeof(control),
4197 };
4198 struct ucred *ucred;
4199 ssize_t n;
4200 pid_t inner_child_pid;
4201 _cleanup_strv_free_ char **tags = NULL;
4202 int r;
4203
4204 assert(userdata);
4205
4206 inner_child_pid = PTR_TO_PID(userdata);
4207
4208 if (revents != EPOLLIN) {
4209 log_warning("Got unexpected poll event for notify fd.");
4210 return 0;
4211 }
4212
4213 n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
4214 if (n < 0) {
4215 if (ERRNO_IS_TRANSIENT(n))
4216 return 0;
4217 if (n == -EXFULL) {
4218 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4219 return 0;
4220 }
4221 return log_warning_errno(n, "Couldn't read notification socket: %m");
4222 }
4223
4224 cmsg_close_all(&msghdr);
4225
4226 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
4227 if (!ucred || ucred->pid != inner_child_pid) {
4228 log_debug("Received notify message without valid credentials. Ignoring.");
4229 return 0;
4230 }
4231
4232 if ((size_t) n >= sizeof(buf)) {
4233 log_warning("Received notify message exceeded maximum size. Ignoring.");
4234 return 0;
4235 }
4236
4237 buf[n] = 0;
4238 tags = strv_split(buf, "\n\r");
4239 if (!tags)
4240 return log_oom();
4241
4242 if (strv_contains(tags, "READY=1")) {
4243 r = sd_notify(false, "READY=1\n");
4244 if (r < 0)
4245 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
4246 }
4247
4248 p = strv_find_startswith(tags, "STATUS=");
4249 if (p)
4250 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
4251
4252 return 0;
4253 }
4254
4255 static int setup_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
4256 int r;
4257
4258 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
4259 if (r < 0)
4260 return log_error_errno(r, "Failed to allocate notify event source: %m");
4261
4262 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
4263
4264 return 0;
4265 }
4266
4267 static int merge_settings(Settings *settings, const char *path) {
4268 int rl;
4269
4270 assert(settings);
4271 assert(path);
4272
4273 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4274 * that this steals the fields of the Settings* structure, and hence modifies it. */
4275
4276 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
4277 settings->start_mode >= 0) {
4278 arg_start_mode = settings->start_mode;
4279 strv_free_and_replace(arg_parameters, settings->parameters);
4280 }
4281
4282 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0 &&
4283 settings->ephemeral >= 0)
4284 arg_ephemeral = settings->ephemeral;
4285
4286 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
4287 settings->root) {
4288
4289 if (!arg_settings_trusted)
4290 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
4291 else
4292 free_and_replace(arg_directory, settings->root);
4293 }
4294
4295 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
4296 settings->pivot_root_new) {
4297 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
4298 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
4299 }
4300
4301 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
4302 settings->working_directory)
4303 free_and_replace(arg_chdir, settings->working_directory);
4304
4305 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
4306 settings->environment)
4307 strv_free_and_replace(arg_setenv, settings->environment);
4308
4309 if ((arg_settings_mask & SETTING_USER) == 0) {
4310
4311 if (settings->user)
4312 free_and_replace(arg_user, settings->user);
4313
4314 if (uid_is_valid(settings->uid))
4315 arg_uid = settings->uid;
4316 if (gid_is_valid(settings->gid))
4317 arg_gid = settings->gid;
4318 if (settings->n_supplementary_gids > 0) {
4319 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
4320 arg_n_supplementary_gids = settings->n_supplementary_gids;
4321 }
4322 }
4323
4324 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
4325 uint64_t plus, minus;
4326 uint64_t network_minus = 0;
4327 uint64_t ambient;
4328
4329 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4330 * Settings structure */
4331
4332 plus = settings->capability;
4333 minus = settings->drop_capability;
4334
4335 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4336 settings_network_configured(settings)) {
4337 if (settings_private_network(settings))
4338 plus |= UINT64_C(1) << CAP_NET_ADMIN;
4339 else
4340 network_minus |= UINT64_C(1) << CAP_NET_ADMIN;
4341 }
4342
4343 if (!arg_settings_trusted && plus != 0) {
4344 if (settings->capability != 0)
4345 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
4346 } else {
4347 arg_caps_retain &= ~network_minus;
4348 arg_caps_retain |= plus;
4349 }
4350
4351 arg_caps_retain &= ~minus;
4352
4353 /* Copy the full capabilities over too */
4354 if (capability_quintet_is_set(&settings->full_capabilities)) {
4355 if (!arg_settings_trusted)
4356 log_warning("Ignoring capability settings, file %s is not trusted.", path);
4357 else
4358 arg_full_capabilities = settings->full_capabilities;
4359 }
4360
4361 ambient = settings->ambient_capability;
4362 if (!arg_settings_trusted && ambient != 0)
4363 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path);
4364 else
4365 arg_caps_ambient |= ambient;
4366 }
4367
4368 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
4369 settings->kill_signal > 0)
4370 arg_kill_signal = settings->kill_signal;
4371
4372 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
4373 settings->personality != PERSONALITY_INVALID)
4374 arg_personality = settings->personality;
4375
4376 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
4377 !sd_id128_is_null(settings->machine_id)) {
4378
4379 if (!arg_settings_trusted)
4380 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
4381 else
4382 arg_uuid = settings->machine_id;
4383 }
4384
4385 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
4386 settings->read_only >= 0)
4387 arg_read_only = settings->read_only;
4388
4389 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
4390 settings->volatile_mode != _VOLATILE_MODE_INVALID)
4391 arg_volatile_mode = settings->volatile_mode;
4392
4393 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
4394 settings->n_custom_mounts > 0) {
4395
4396 if (!arg_settings_trusted)
4397 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
4398 else {
4399 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4400 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
4401 arg_n_custom_mounts = settings->n_custom_mounts;
4402 settings->n_custom_mounts = 0;
4403 }
4404 }
4405
4406 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4407 settings_network_configured(settings)) {
4408
4409 if (!arg_settings_trusted)
4410 log_warning("Ignoring network settings, file %s is not trusted.", path);
4411 else {
4412 arg_network_veth = settings_network_veth(settings);
4413 arg_private_network = settings_private_network(settings);
4414
4415 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
4416 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
4417 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
4418 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
4419
4420 free_and_replace(arg_network_bridge, settings->network_bridge);
4421 free_and_replace(arg_network_zone, settings->network_zone);
4422
4423 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
4424 }
4425 }
4426
4427 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
4428 settings->expose_ports) {
4429
4430 if (!arg_settings_trusted)
4431 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
4432 else {
4433 expose_port_free_all(arg_expose_ports);
4434 arg_expose_ports = TAKE_PTR(settings->expose_ports);
4435 }
4436 }
4437
4438 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
4439 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
4440
4441 if (!arg_settings_trusted)
4442 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
4443 else {
4444 arg_userns_mode = settings->userns_mode;
4445 arg_uid_shift = settings->uid_shift;
4446 arg_uid_range = settings->uid_range;
4447 arg_userns_ownership = settings->userns_ownership;
4448 }
4449 }
4450
4451 if ((arg_settings_mask & SETTING_BIND_USER) == 0 &&
4452 !strv_isempty(settings->bind_user))
4453 strv_free_and_replace(arg_bind_user, settings->bind_user);
4454
4455 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0 &&
4456 settings->notify_ready >= 0)
4457 arg_notify_ready = settings->notify_ready;
4458
4459 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
4460
4461 if (!strv_isempty(settings->syscall_allow_list) || !strv_isempty(settings->syscall_deny_list)) {
4462 if (!arg_settings_trusted && !strv_isempty(settings->syscall_allow_list))
4463 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
4464 else {
4465 strv_free_and_replace(arg_syscall_allow_list, settings->syscall_allow_list);
4466 strv_free_and_replace(arg_syscall_deny_list, settings->syscall_deny_list);
4467 }
4468 }
4469
4470 #if HAVE_SECCOMP
4471 if (settings->seccomp) {
4472 if (!arg_settings_trusted)
4473 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
4474 else {
4475 seccomp_release(arg_seccomp);
4476 arg_seccomp = TAKE_PTR(settings->seccomp);
4477 }
4478 }
4479 #endif
4480 }
4481
4482 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
4483 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
4484 continue;
4485
4486 if (!settings->rlimit[rl])
4487 continue;
4488
4489 if (!arg_settings_trusted) {
4490 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
4491 continue;
4492 }
4493
4494 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
4495 }
4496
4497 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
4498 settings->hostname)
4499 free_and_replace(arg_hostname, settings->hostname);
4500
4501 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
4502 settings->no_new_privileges >= 0)
4503 arg_no_new_privileges = settings->no_new_privileges;
4504
4505 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
4506 settings->oom_score_adjust_set) {
4507
4508 if (!arg_settings_trusted)
4509 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
4510 else {
4511 arg_oom_score_adjust = settings->oom_score_adjust;
4512 arg_oom_score_adjust_set = true;
4513 }
4514 }
4515
4516 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
4517 settings->cpu_set.set) {
4518
4519 if (!arg_settings_trusted)
4520 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
4521 else {
4522 cpu_set_reset(&arg_cpu_set);
4523 arg_cpu_set = settings->cpu_set;
4524 settings->cpu_set = (CPUSet) {};
4525 }
4526 }
4527
4528 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
4529 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
4530 arg_resolv_conf = settings->resolv_conf;
4531
4532 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
4533 settings->link_journal != _LINK_JOURNAL_INVALID) {
4534
4535 if (!arg_settings_trusted)
4536 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
4537 else {
4538 arg_link_journal = settings->link_journal;
4539 arg_link_journal_try = settings->link_journal_try;
4540 }
4541 }
4542
4543 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
4544 settings->timezone != _TIMEZONE_MODE_INVALID)
4545 arg_timezone = settings->timezone;
4546
4547 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
4548 settings->slice) {
4549
4550 if (!arg_settings_trusted)
4551 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
4552 else
4553 free_and_replace(arg_slice, settings->slice);
4554 }
4555
4556 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
4557 settings->use_cgns >= 0) {
4558
4559 if (!arg_settings_trusted)
4560 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
4561 else
4562 arg_use_cgns = settings->use_cgns;
4563 }
4564
4565 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
4566 settings->clone_ns_flags != ULONG_MAX) {
4567
4568 if (!arg_settings_trusted)
4569 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
4570 else
4571 arg_clone_ns_flags = settings->clone_ns_flags;
4572 }
4573
4574 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
4575 settings->console_mode >= 0) {
4576
4577 if (!arg_settings_trusted)
4578 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
4579 else
4580 arg_console_mode = settings->console_mode;
4581 }
4582
4583 if ((arg_settings_mask & SETTING_SUPPRESS_SYNC) == 0 &&
4584 settings->suppress_sync >= 0)
4585 arg_suppress_sync = settings->suppress_sync;
4586
4587 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4588 * don't consult arg_settings_mask for them. */
4589
4590 sd_bus_message_unref(arg_property_message);
4591 arg_property_message = TAKE_PTR(settings->properties);
4592
4593 arg_console_width = settings->console_width;
4594 arg_console_height = settings->console_height;
4595
4596 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
4597 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
4598 arg_n_extra_nodes = settings->n_extra_nodes;
4599
4600 return 0;
4601 }
4602
4603 static int load_settings(void) {
4604 _cleanup_(settings_freep) Settings *settings = NULL;
4605 _cleanup_fclose_ FILE *f = NULL;
4606 _cleanup_free_ char *p = NULL;
4607 const char *fn, *i;
4608 int r;
4609
4610 if (arg_oci_bundle)
4611 return 0;
4612
4613 /* If all settings are masked, there's no point in looking for
4614 * the settings file */
4615 if (FLAGS_SET(arg_settings_mask, _SETTINGS_MASK_ALL))
4616 return 0;
4617
4618 fn = strjoina(arg_machine, ".nspawn");
4619
4620 /* We first look in the admin's directories in /etc and /run */
4621 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4622 _cleanup_free_ char *j = NULL;
4623
4624 j = path_join(i, fn);
4625 if (!j)
4626 return log_oom();
4627
4628 f = fopen(j, "re");
4629 if (f) {
4630 p = TAKE_PTR(j);
4631
4632 /* By default, we trust configuration from /etc and /run */
4633 if (arg_settings_trusted < 0)
4634 arg_settings_trusted = true;
4635
4636 break;
4637 }
4638
4639 if (errno != ENOENT)
4640 return log_error_errno(errno, "Failed to open %s: %m", j);
4641 }
4642
4643 if (!f) {
4644 /* After that, let's look for a file next to the
4645 * actual image we shall boot. */
4646
4647 if (arg_image) {
4648 p = file_in_same_dir(arg_image, fn);
4649 if (!p)
4650 return log_oom();
4651 } else if (arg_directory && !path_equal(arg_directory, "/")) {
4652 p = file_in_same_dir(arg_directory, fn);
4653 if (!p)
4654 return log_oom();
4655 }
4656
4657 if (p) {
4658 f = fopen(p, "re");
4659 if (!f && errno != ENOENT)
4660 return log_error_errno(errno, "Failed to open %s: %m", p);
4661
4662 /* By default, we do not trust configuration from /var/lib/machines */
4663 if (arg_settings_trusted < 0)
4664 arg_settings_trusted = false;
4665 }
4666 }
4667
4668 if (!f)
4669 return 0;
4670
4671 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4672
4673 r = settings_load(f, p, &settings);
4674 if (r < 0)
4675 return r;
4676
4677 return merge_settings(settings, p);
4678 }
4679
4680 static int load_oci_bundle(void) {
4681 _cleanup_(settings_freep) Settings *settings = NULL;
4682 int r;
4683
4684 if (!arg_oci_bundle)
4685 return 0;
4686
4687 /* By default let's trust OCI bundles */
4688 if (arg_settings_trusted < 0)
4689 arg_settings_trusted = true;
4690
4691 r = oci_load(NULL, arg_oci_bundle, &settings);
4692 if (r < 0)
4693 return r;
4694
4695 return merge_settings(settings, arg_oci_bundle);
4696 }
4697
4698 static int run_container(
4699 DissectedImage *dissected_image,
4700 bool secondary,
4701 FDSet *fds,
4702 char veth_name[IFNAMSIZ], bool *veth_created,
4703 struct ExposeArgs *expose_args,
4704 int *master, pid_t *pid, int *ret) {
4705
4706 static const struct sigaction sa = {
4707 .sa_handler = nop_signal_handler,
4708 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4709 };
4710
4711 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4712 _cleanup_close_ int etc_passwd_lock = -1;
4713 _cleanup_close_pair_ int
4714 kmsg_socket_pair[2] = { -1, -1 },
4715 rtnl_socket_pair[2] = { -1, -1 },
4716 pid_socket_pair[2] = { -1, -1 },
4717 uuid_socket_pair[2] = { -1, -1 },
4718 notify_socket_pair[2] = { -1, -1 },
4719 uid_shift_socket_pair[2] = { -1, -1 },
4720 master_pty_socket_pair[2] = { -1, -1 },
4721 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4722
4723 _cleanup_close_ int notify_socket = -1;
4724 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4725 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4726 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4727 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4728 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4729 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4730 _cleanup_free_ uid_t *bind_user_uid = NULL;
4731 size_t n_bind_user_uid = 0;
4732 ContainerStatus container_status = 0;
4733 int ifi = 0, r;
4734 ssize_t l;
4735 sigset_t mask_chld;
4736 _cleanup_close_ int child_netns_fd = -1;
4737
4738 assert_se(sigemptyset(&mask_chld) == 0);
4739 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4740
4741 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4742 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4743 * check with getpwuid() if the specific user already exists. Note that /etc might be
4744 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4745 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4746 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4747 * really ours. */
4748
4749 etc_passwd_lock = take_etc_passwd_lock(NULL);
4750 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4751 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4752 }
4753
4754 r = barrier_create(&barrier);
4755 if (r < 0)
4756 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4757
4758 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4759 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4760
4761 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4762 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4763
4764 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4765 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4766
4767 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4768 return log_error_errno(errno, "Failed to create id socket pair: %m");
4769
4770 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4771 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4772
4773 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
4774 return log_error_errno(errno, "Failed to create console socket pair: %m");
4775
4776 if (arg_userns_mode != USER_NAMESPACE_NO)
4777 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4778 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4779
4780 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4781 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4782 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4783
4784 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4785 * parent's blocking calls and give it a chance to call wait() and terminate. */
4786 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4787 if (r < 0)
4788 return log_error_errno(errno, "Failed to change the signal mask: %m");
4789
4790 r = sigaction(SIGCHLD, &sa, NULL);
4791 if (r < 0)
4792 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4793
4794 if (arg_network_namespace_path) {
4795 child_netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4796 if (child_netns_fd < 0)
4797 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4798
4799 r = fd_is_ns(child_netns_fd, CLONE_NEWNET);
4800 if (r == -EUCLEAN)
4801 log_debug_errno(r, "Cannot determine if passed network namespace path '%s' really refers to a network namespace, assuming it does.", arg_network_namespace_path);
4802 else if (r < 0)
4803 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4804 else if (r == 0)
4805 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4806 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4807 }
4808
4809 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4810 if (*pid < 0)
4811 return log_error_errno(errno, "clone() failed%s: %m",
4812 errno == EINVAL ?
4813 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4814
4815 if (*pid == 0) {
4816 /* The outer child only has a file system namespace. */
4817 barrier_set_role(&barrier, BARRIER_CHILD);
4818
4819 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4820 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4821 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4822 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4823 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
4824 master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
4825 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4826 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
4827
4828 (void) reset_all_signal_handlers();
4829 (void) reset_signal_mask();
4830
4831 r = outer_child(&barrier,
4832 arg_directory,
4833 dissected_image,
4834 secondary,
4835 pid_socket_pair[1],
4836 uuid_socket_pair[1],
4837 notify_socket_pair[1],
4838 kmsg_socket_pair[1],
4839 rtnl_socket_pair[1],
4840 uid_shift_socket_pair[1],
4841 master_pty_socket_pair[1],
4842 unified_cgroup_hierarchy_socket_pair[1],
4843 fds,
4844 child_netns_fd);
4845 if (r < 0)
4846 _exit(EXIT_FAILURE);
4847
4848 _exit(EXIT_SUCCESS);
4849 }
4850
4851 barrier_set_role(&barrier, BARRIER_PARENT);
4852
4853 fdset_close(fds);
4854
4855 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4856 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4857 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4858 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4859 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
4860 master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
4861 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
4862 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
4863
4864 if (arg_userns_mode != USER_NAMESPACE_NO) {
4865 /* The child just let us know the UID shift it might have read from the image. */
4866 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4867 if (l < 0)
4868 return log_error_errno(errno, "Failed to read UID shift: %m");
4869 if (l != sizeof arg_uid_shift)
4870 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
4871
4872 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4873 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4874 * image, but if that's already in use, pick a new one, and report back to the child,
4875 * which one we now picked. */
4876
4877 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4878 if (r < 0)
4879 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4880
4881 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4882 if (l < 0)
4883 return log_error_errno(errno, "Failed to send UID shift: %m");
4884 if (l != sizeof arg_uid_shift)
4885 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
4886 }
4887
4888 n_bind_user_uid = strv_length(arg_bind_user);
4889 if (n_bind_user_uid > 0) {
4890 /* Right after the UID shift, we'll receive the list of UID mappings for the
4891 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4892
4893 bind_user_uid = new(uid_t, n_bind_user_uid*4);
4894 if (!bind_user_uid)
4895 return log_oom();
4896
4897 for (size_t i = 0; i < n_bind_user_uid; i++) {
4898 l = recv(uid_shift_socket_pair[0], bind_user_uid + i*4, sizeof(uid_t)*4, 0);
4899 if (l < 0)
4900 return log_error_errno(errno, "Failed to read user UID map pair: %m");
4901 if (l != sizeof(uid_t)*4)
4902 return log_full_errno(l == 0 ? LOG_DEBUG : LOG_WARNING,
4903 SYNTHETIC_ERRNO(EIO),
4904 "Short read while reading bind user UID pairs.");
4905 }
4906 }
4907 }
4908
4909 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4910 /* The child let us know the support cgroup mode it might have read from the image. */
4911 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4912 if (l < 0)
4913 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4914 if (l != sizeof(arg_unified_cgroup_hierarchy))
4915 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zu bytes).%s",
4916 l, l == 0 ? " The child is most likely dead." : "");
4917 }
4918
4919 /* Wait for the outer child. */
4920 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4921 if (r < 0)
4922 return r;
4923 if (r != EXIT_SUCCESS)
4924 return -EIO;
4925
4926 /* And now retrieve the PID of the inner child. */
4927 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4928 if (l < 0)
4929 return log_error_errno(errno, "Failed to read inner child PID: %m");
4930 if (l != sizeof *pid)
4931 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
4932
4933 /* We also retrieve container UUID in case it was generated by outer child */
4934 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4935 if (l < 0)
4936 return log_error_errno(errno, "Failed to read container machine ID: %m");
4937 if (l != sizeof(arg_uuid))
4938 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
4939
4940 /* We also retrieve the socket used for notifications generated by outer child */
4941 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4942 if (notify_socket < 0)
4943 return log_error_errno(notify_socket,
4944 "Failed to receive notification socket from the outer child: %m");
4945
4946 log_debug("Init process invoked as PID "PID_FMT, *pid);
4947
4948 if (arg_userns_mode != USER_NAMESPACE_NO) {
4949 if (!barrier_place_and_sync(&barrier)) /* #1 */
4950 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4951
4952 r = setup_uid_map(*pid, bind_user_uid, n_bind_user_uid);
4953 if (r < 0)
4954 return r;
4955
4956 (void) barrier_place(&barrier); /* #2 */
4957 }
4958
4959 if (arg_private_network) {
4960 if (!arg_network_namespace_path) {
4961 /* Wait until the child has unshared its network namespace. */
4962 if (!barrier_place_and_sync(&barrier)) /* #3 */
4963 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
4964 }
4965
4966 if (child_netns_fd < 0) {
4967 /* Make sure we have an open file descriptor to the child's network
4968 * namespace so it stays alive even if the child exits. */
4969 r = namespace_open(*pid, NULL, NULL, &child_netns_fd, NULL, NULL);
4970 if (r < 0)
4971 return log_error_errno(r, "Failed to open child network namespace: %m");
4972 }
4973
4974 r = move_network_interfaces(child_netns_fd, arg_network_interfaces);
4975 if (r < 0)
4976 return r;
4977
4978 if (arg_network_veth) {
4979 r = setup_veth(arg_machine, *pid, veth_name,
4980 arg_network_bridge || arg_network_zone);
4981 if (r < 0)
4982 return r;
4983 else if (r > 0)
4984 ifi = r;
4985
4986 if (arg_network_bridge) {
4987 /* Add the interface to a bridge */
4988 r = setup_bridge(veth_name, arg_network_bridge, false);
4989 if (r < 0)
4990 return r;
4991 if (r > 0)
4992 ifi = r;
4993 } else if (arg_network_zone) {
4994 /* Add the interface to a bridge, possibly creating it */
4995 r = setup_bridge(veth_name, arg_network_zone, true);
4996 if (r < 0)
4997 return r;
4998 if (r > 0)
4999 ifi = r;
5000 }
5001 }
5002
5003 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
5004 if (r < 0)
5005 return r;
5006
5007 /* We created the primary and extra veth links now; let's remember this, so that we know to
5008 remove them later on. Note that we don't bother with removing veth links that were created
5009 here when their setup failed half-way, because in that case the kernel should be able to
5010 remove them on its own, since they cannot be referenced by anything yet. */
5011 *veth_created = true;
5012
5013 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
5014 if (r < 0)
5015 return r;
5016
5017 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
5018 if (r < 0)
5019 return r;
5020 }
5021
5022 if (arg_register || !arg_keep_unit) {
5023 r = sd_bus_default_system(&bus);
5024 if (r < 0)
5025 return log_error_errno(r, "Failed to open system bus: %m");
5026
5027 r = sd_bus_set_close_on_exit(bus, false);
5028 if (r < 0)
5029 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
5030 }
5031
5032 if (!arg_keep_unit) {
5033 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5034 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5035 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5036
5037 r = sd_bus_match_signal_async(
5038 bus,
5039 NULL,
5040 "org.freedesktop.systemd1",
5041 NULL,
5042 "org.freedesktop.systemd1.Scope",
5043 "RequestStop",
5044 on_request_stop, NULL, PID_TO_PTR(*pid));
5045 if (r < 0)
5046 return log_error_errno(r, "Failed to request RequestStop match: %m");
5047 }
5048
5049 if (arg_register) {
5050 r = register_machine(
5051 bus,
5052 arg_machine,
5053 *pid,
5054 arg_directory,
5055 arg_uuid,
5056 ifi,
5057 arg_slice,
5058 arg_custom_mounts, arg_n_custom_mounts,
5059 arg_kill_signal,
5060 arg_property,
5061 arg_property_message,
5062 arg_keep_unit,
5063 arg_container_service_name);
5064 if (r < 0)
5065 return r;
5066
5067 } else if (!arg_keep_unit) {
5068 r = allocate_scope(
5069 bus,
5070 arg_machine,
5071 *pid,
5072 arg_slice,
5073 arg_custom_mounts, arg_n_custom_mounts,
5074 arg_kill_signal,
5075 arg_property,
5076 arg_property_message);
5077 if (r < 0)
5078 return r;
5079
5080 } else if (arg_slice || arg_property)
5081 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5082
5083 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
5084 if (r < 0)
5085 return r;
5086
5087 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
5088 if (r < 0)
5089 return r;
5090
5091 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
5092 if (r < 0)
5093 return r;
5094
5095 /* Notify the child that the parent is ready with all
5096 * its setup (including cgroup-ification), and that
5097 * the child can now hand over control to the code to
5098 * run inside the container. */
5099 (void) barrier_place(&barrier); /* #4 */
5100
5101 /* Block SIGCHLD here, before notifying child.
5102 * process_pty() will handle it with the other signals. */
5103 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
5104
5105 /* Reset signal to default */
5106 r = default_signals(SIGCHLD);
5107 if (r < 0)
5108 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
5109
5110 r = sd_event_new(&event);
5111 if (r < 0)
5112 return log_error_errno(r, "Failed to get default event source: %m");
5113
5114 (void) sd_event_set_watchdog(event, true);
5115
5116 if (bus) {
5117 r = sd_bus_attach_event(bus, event, 0);
5118 if (r < 0)
5119 return log_error_errno(r, "Failed to attach bus to event loop: %m");
5120 }
5121
5122 r = setup_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
5123 if (r < 0)
5124 return r;
5125
5126 /* Let the child know that we are ready and wait that the child is completely ready now. */
5127 if (!barrier_place_and_sync(&barrier)) /* #5 */
5128 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
5129
5130 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5131 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5132 etc_passwd_lock = safe_close(etc_passwd_lock);
5133
5134 (void) sd_notifyf(false,
5135 "STATUS=Container running.\n"
5136 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
5137 if (!arg_notify_ready) {
5138 r = sd_notify(false, "READY=1\n");
5139 if (r < 0)
5140 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
5141 }
5142
5143 if (arg_kill_signal > 0) {
5144 /* Try to kill the init system on SIGINT or SIGTERM */
5145 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
5146 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
5147 } else {
5148 /* Immediately exit */
5149 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
5150 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
5151 }
5152
5153 /* Exit when the child exits */
5154 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
5155
5156 if (arg_expose_ports) {
5157 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, expose_args, &rtnl);
5158 if (r < 0)
5159 return r;
5160
5161 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
5162 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
5163 }
5164
5165 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
5166
5167 if (arg_console_mode != CONSOLE_PIPE) {
5168 _cleanup_close_ int fd = -1;
5169 PTYForwardFlags flags = 0;
5170
5171 /* Retrieve the master pty allocated by inner child */
5172 fd = receive_one_fd(master_pty_socket_pair[0], 0);
5173 if (fd < 0)
5174 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
5175
5176 switch (arg_console_mode) {
5177
5178 case CONSOLE_READ_ONLY:
5179 flags |= PTY_FORWARD_READ_ONLY;
5180
5181 _fallthrough_;
5182
5183 case CONSOLE_INTERACTIVE:
5184 flags |= PTY_FORWARD_IGNORE_VHANGUP;
5185
5186 r = pty_forward_new(event, fd, flags, &forward);
5187 if (r < 0)
5188 return log_error_errno(r, "Failed to create PTY forwarder: %m");
5189
5190 if (arg_console_width != UINT_MAX || arg_console_height != UINT_MAX)
5191 (void) pty_forward_set_width_height(forward,
5192 arg_console_width,
5193 arg_console_height);
5194 break;
5195
5196 default:
5197 assert(arg_console_mode == CONSOLE_PASSIVE);
5198 }
5199
5200 *master = TAKE_FD(fd);
5201 }
5202
5203 r = sd_event_loop(event);
5204 if (r < 0)
5205 return log_error_errno(r, "Failed to run event loop: %m");
5206
5207 if (forward) {
5208 char last_char = 0;
5209
5210 (void) pty_forward_get_last_char(forward, &last_char);
5211 forward = pty_forward_free(forward);
5212
5213 if (!arg_quiet && last_char != '\n')
5214 putc('\n', stdout);
5215 }
5216
5217 /* Kill if it is not dead yet anyway */
5218 if (!arg_register && !arg_keep_unit && bus)
5219 terminate_scope(bus, arg_machine);
5220
5221 /* Normally redundant, but better safe than sorry */
5222 (void) kill(*pid, SIGKILL);
5223
5224 if (arg_private_network) {
5225 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5226 * to avoid having to move the parent to the child network namespace. */
5227 r = safe_fork(NULL, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_WAIT|FORK_LOG, NULL);
5228 if (r < 0)
5229 return r;
5230
5231 if (r == 0) {
5232 _cleanup_close_ int parent_netns_fd = -1;
5233
5234 r = namespace_open(getpid(), NULL, NULL, &parent_netns_fd, NULL, NULL);
5235 if (r < 0) {
5236 log_error_errno(r, "Failed to open parent network namespace: %m");
5237 _exit(EXIT_FAILURE);
5238 }
5239
5240 r = namespace_enter(-1, -1, child_netns_fd, -1, -1);
5241 if (r < 0) {
5242 log_error_errno(r, "Failed to enter child network namespace: %m");
5243 _exit(EXIT_FAILURE);
5244 }
5245
5246 r = move_network_interfaces(parent_netns_fd, arg_network_interfaces);
5247 if (r < 0)
5248 log_error_errno(r, "Failed to move network interfaces back to parent network namespace: %m");
5249
5250 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
5251 }
5252 }
5253
5254 r = wait_for_container(TAKE_PID(*pid), &container_status);
5255
5256 /* Tell machined that we are gone. */
5257 if (bus)
5258 (void) unregister_machine(bus, arg_machine);
5259
5260 if (r < 0)
5261 /* We failed to wait for the container, or the container exited abnormally. */
5262 return r;
5263 if (r > 0 || container_status == CONTAINER_TERMINATED) {
5264 /* r > 0 → The container exited with a non-zero status.
5265 * As a special case, we need to replace 133 with a different value,
5266 * because 133 is special-cased in the service file to reboot the container.
5267 * otherwise → The container exited with zero status and a reboot was not requested.
5268 */
5269 if (r == EXIT_FORCE_RESTART)
5270 r = EXIT_FAILURE; /* replace 133 with the general failure code */
5271 *ret = r;
5272 return 0; /* finito */
5273 }
5274
5275 /* CONTAINER_REBOOTED, loop again */
5276
5277 if (arg_keep_unit) {
5278 /* Special handling if we are running as a service: instead of simply
5279 * restarting the machine we want to restart the entire service, so let's
5280 * inform systemd about this with the special exit code 133. The service
5281 * file uses RestartForceExitStatus=133 so that this results in a full
5282 * nspawn restart. This is necessary since we might have cgroup parameters
5283 * set we want to have flushed out. */
5284 *ret = EXIT_FORCE_RESTART;
5285 return 0; /* finito */
5286 }
5287
5288 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
5289 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
5290
5291 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5292 *veth_created = false;
5293 return 1; /* loop again */
5294 }
5295
5296 static int initialize_rlimits(void) {
5297 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5298 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5299 * container execution environments. */
5300
5301 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
5302 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
5303 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
5304 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
5305 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
5306 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
5307 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
5308 [RLIMIT_MEMLOCK] = { DEFAULT_RLIMIT_MEMLOCK, DEFAULT_RLIMIT_MEMLOCK },
5309 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
5310 [RLIMIT_NICE] = { 0, 0 },
5311 [RLIMIT_NOFILE] = { 1024, 4096 },
5312 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
5313 [RLIMIT_RTPRIO] = { 0, 0 },
5314 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
5315 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
5316
5317 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5318 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5319 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5320 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5321 * that PID 1 changes a number of other resource limits during early initialization which is why we
5322 * don't read the other limits from PID 1 but prefer the static table above. */
5323 };
5324
5325 int rl;
5326
5327 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
5328 /* Let's only fill in what the user hasn't explicitly configured anyway */
5329 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
5330 const struct rlimit *v;
5331 struct rlimit buffer;
5332
5333 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
5334 /* For these two let's read the limits off PID 1. See above for an explanation. */
5335
5336 if (prlimit(1, rl, NULL, &buffer) < 0)
5337 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
5338
5339 v = &buffer;
5340 } else if (rl == RLIMIT_NOFILE) {
5341 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5342 * userspace. Given that nspawn containers are often run without our PID 1,
5343 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5344 * so that container userspace gets similar resources as host userspace
5345 * gets. */
5346 buffer = kernel_defaults[rl];
5347 buffer.rlim_max = MIN((rlim_t) read_nr_open(), (rlim_t) HIGH_RLIMIT_NOFILE);
5348 v = &buffer;
5349 } else
5350 v = kernel_defaults + rl;
5351
5352 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
5353 if (!arg_rlimit[rl])
5354 return log_oom();
5355 }
5356
5357 if (DEBUG_LOGGING) {
5358 _cleanup_free_ char *k = NULL;
5359
5360 (void) rlimit_format(arg_rlimit[rl], &k);
5361 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
5362 }
5363 }
5364
5365 return 0;
5366 }
5367
5368 static int cant_be_in_netns(void) {
5369 union sockaddr_union sa = {
5370 .un = {
5371 .sun_family = AF_UNIX,
5372 .sun_path = "/run/udev/control",
5373 },
5374 };
5375 char udev_path[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t)];
5376 _cleanup_free_ char *udev_ns = NULL, *our_ns = NULL;
5377 _cleanup_close_ int fd = -1;
5378 struct ucred ucred;
5379 int r;
5380
5381 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5382 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5383 * nice message. */
5384
5385 if (!arg_image) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5386 return 0;
5387
5388 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_NONBLOCK|SOCK_CLOEXEC, 0);
5389 if (fd < 0)
5390 return log_error_errno(errno, "Failed to allocate udev control socket: %m");
5391
5392 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
5393
5394 if (errno == ENOENT || ERRNO_IS_DISCONNECT(errno))
5395 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5396 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5397
5398 return log_error_errno(errno, "Failed to connect socket to udev control socket: %m");
5399 }
5400
5401 r = getpeercred(fd, &ucred);
5402 if (r < 0)
5403 return log_error_errno(r, "Failed to determine peer of udev control socket: %m");
5404
5405 xsprintf(udev_path, "/proc/" PID_FMT "/ns/net", ucred.pid);
5406 r = readlink_malloc(udev_path, &udev_ns);
5407 if (r < 0)
5408 return log_error_errno(r, "Failed to read network namespace of udev: %m");
5409
5410 r = readlink_malloc("/proc/self/ns/net", &our_ns);
5411 if (r < 0)
5412 return log_error_errno(r, "Failed to read our own network namespace: %m");
5413
5414 if (!streq(our_ns, udev_ns))
5415 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5416 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5417 return 0;
5418 }
5419
5420 static int run(int argc, char *argv[]) {
5421 bool secondary = false, remove_directory = false, remove_image = false,
5422 veth_created = false, remove_tmprootdir = false;
5423 _cleanup_close_ int master = -1;
5424 _cleanup_fdset_free_ FDSet *fds = NULL;
5425 int r, n_fd_passed, ret = EXIT_SUCCESS;
5426 char veth_name[IFNAMSIZ] = "";
5427 struct ExposeArgs expose_args = {};
5428 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
5429 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
5430 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
5431 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
5432 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
5433 _cleanup_(fw_ctx_freep) FirewallContext *fw_ctx = NULL;
5434 pid_t pid = 0;
5435
5436 log_parse_environment();
5437 log_open();
5438
5439 r = parse_argv(argc, argv);
5440 if (r <= 0)
5441 goto finish;
5442
5443 if (geteuid() != 0) {
5444 r = log_warning_errno(SYNTHETIC_ERRNO(EPERM),
5445 argc >= 2 ? "Need to be root." :
5446 "Need to be root (and some arguments are usually required).\nHint: try --help");
5447 goto finish;
5448 }
5449
5450 r = cant_be_in_netns();
5451 if (r < 0)
5452 goto finish;
5453
5454 r = initialize_rlimits();
5455 if (r < 0)
5456 goto finish;
5457
5458 r = load_oci_bundle();
5459 if (r < 0)
5460 goto finish;
5461
5462 r = determine_names();
5463 if (r < 0)
5464 goto finish;
5465
5466 r = load_settings();
5467 if (r < 0)
5468 goto finish;
5469
5470 r = cg_unified();
5471 if (r < 0) {
5472 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5473 goto finish;
5474 }
5475
5476 r = verify_arguments();
5477 if (r < 0)
5478 goto finish;
5479
5480 /* Reapply environment settings. */
5481 (void) detect_unified_cgroup_hierarchy_from_environment();
5482
5483 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5484 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5485 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5486 (void) ignore_signals(SIGPIPE);
5487
5488 n_fd_passed = sd_listen_fds(false);
5489 if (n_fd_passed > 0) {
5490 r = fdset_new_listen_fds(&fds, false);
5491 if (r < 0) {
5492 log_error_errno(r, "Failed to collect file descriptors: %m");
5493 goto finish;
5494 }
5495 }
5496
5497 /* The "default" umask. This is appropriate for most file and directory
5498 * operations performed by nspawn, and is the umask that will be used for
5499 * the child. Functions like copy_devnodes() change the umask temporarily. */
5500 umask(0022);
5501
5502 if (arg_directory) {
5503 assert(!arg_image);
5504
5505 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5506 * /var from the host will propagate into container dynamically (because bad things happen if
5507 * two systems write to the same /var). Let's allow it for the special cases where /var is
5508 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5509 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
5510 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5511 r = -EINVAL;
5512 goto finish;
5513 }
5514
5515 if (arg_ephemeral) {
5516 _cleanup_free_ char *np = NULL;
5517
5518 r = chase_symlinks_and_update(&arg_directory, 0);
5519 if (r < 0)
5520 goto finish;
5521
5522 /* If the specified path is a mount point we generate the new snapshot immediately
5523 * inside it under a random name. However if the specified is not a mount point we
5524 * create the new snapshot in the parent directory, just next to it. */
5525 r = path_is_mount_point(arg_directory, NULL, 0);
5526 if (r < 0) {
5527 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
5528 goto finish;
5529 }
5530 if (r > 0)
5531 r = tempfn_random_child(arg_directory, "machine.", &np);
5532 else
5533 r = tempfn_random(arg_directory, "machine.", &np);
5534 if (r < 0) {
5535 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
5536 goto finish;
5537 }
5538
5539 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5540 * only owned by us and no one else. */
5541 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5542 if (r < 0) {
5543 log_error_errno(r, "Failed to lock %s: %m", np);
5544 goto finish;
5545 }
5546
5547 {
5548 BLOCK_SIGNALS(SIGINT);
5549 r = btrfs_subvol_snapshot(arg_directory, np,
5550 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5551 BTRFS_SNAPSHOT_FALLBACK_COPY |
5552 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5553 BTRFS_SNAPSHOT_RECURSIVE |
5554 BTRFS_SNAPSHOT_QUOTA |
5555 BTRFS_SNAPSHOT_SIGINT);
5556 }
5557 if (r == -EINTR) {
5558 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
5559 goto finish;
5560 }
5561 if (r < 0) {
5562 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
5563 goto finish;
5564 }
5565
5566 free_and_replace(arg_directory, np);
5567 remove_directory = true;
5568 } else {
5569 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
5570 if (r < 0)
5571 goto finish;
5572
5573 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5574 if (r == -EBUSY) {
5575 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
5576 goto finish;
5577 }
5578 if (r < 0) {
5579 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
5580 goto finish;
5581 }
5582
5583 if (arg_template) {
5584 r = chase_symlinks_and_update(&arg_template, 0);
5585 if (r < 0)
5586 goto finish;
5587
5588 {
5589 BLOCK_SIGNALS(SIGINT);
5590 r = btrfs_subvol_snapshot(arg_template, arg_directory,
5591 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5592 BTRFS_SNAPSHOT_FALLBACK_COPY |
5593 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5594 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
5595 BTRFS_SNAPSHOT_RECURSIVE |
5596 BTRFS_SNAPSHOT_QUOTA |
5597 BTRFS_SNAPSHOT_SIGINT);
5598 }
5599 if (r == -EEXIST)
5600 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5601 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
5602 else if (r == -EINTR) {
5603 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
5604 goto finish;
5605 } else if (r < 0) {
5606 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
5607 goto finish;
5608 } else
5609 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5610 "Populated %s from template %s.", arg_directory, arg_template);
5611 }
5612 }
5613
5614 if (arg_start_mode == START_BOOT) {
5615 _cleanup_free_ char *b = NULL;
5616 const char *p;
5617
5618 if (arg_pivot_root_new) {
5619 b = path_join(arg_directory, arg_pivot_root_new);
5620 if (!b)
5621 return log_oom();
5622
5623 p = b;
5624 } else
5625 p = arg_directory;
5626
5627 if (path_is_os_tree(p) <= 0) {
5628 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5629 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
5630 goto finish;
5631 }
5632 } else {
5633 _cleanup_free_ char *p = NULL;
5634
5635 if (arg_pivot_root_new)
5636 p = path_join(arg_directory, arg_pivot_root_new, "/usr/");
5637 else
5638 p = path_join(arg_directory, "/usr/");
5639 if (!p)
5640 return log_oom();
5641
5642 if (laccess(p, F_OK) < 0) {
5643 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5644 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory);
5645 goto finish;
5646 }
5647 }
5648
5649 } else {
5650 DissectImageFlags dissect_image_flags =
5651 DISSECT_IMAGE_GENERIC_ROOT |
5652 DISSECT_IMAGE_REQUIRE_ROOT |
5653 DISSECT_IMAGE_RELAX_VAR_CHECK |
5654 DISSECT_IMAGE_USR_NO_ROOT;
5655 assert(arg_image);
5656 assert(!arg_template);
5657
5658 r = chase_symlinks_and_update(&arg_image, 0);
5659 if (r < 0)
5660 goto finish;
5661
5662 if (arg_ephemeral) {
5663 _cleanup_free_ char *np = NULL;
5664
5665 r = tempfn_random(arg_image, "machine.", &np);
5666 if (r < 0) {
5667 log_error_errno(r, "Failed to generate name for image snapshot: %m");
5668 goto finish;
5669 }
5670
5671 /* Always take an exclusive lock on our own ephemeral copy. */
5672 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5673 if (r < 0) {
5674 r = log_error_errno(r, "Failed to create image lock: %m");
5675 goto finish;
5676 }
5677
5678 {
5679 BLOCK_SIGNALS(SIGINT);
5680 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME|COPY_SIGINT);
5681 }
5682 if (r == -EINTR) {
5683 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
5684 goto finish;
5685 }
5686 if (r < 0) {
5687 r = log_error_errno(r, "Failed to copy image file: %m");
5688 goto finish;
5689 }
5690
5691 free_and_replace(arg_image, np);
5692 remove_image = true;
5693 } else {
5694 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5695 if (r == -EBUSY) {
5696 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
5697 goto finish;
5698 }
5699 if (r < 0) {
5700 r = log_error_errno(r, "Failed to create image lock: %m");
5701 goto finish;
5702 }
5703
5704 r = verity_settings_load(
5705 &arg_verity_settings,
5706 arg_image, NULL, NULL);
5707 if (r < 0) {
5708 log_error_errno(r, "Failed to read verity artefacts for %s: %m", arg_image);
5709 goto finish;
5710 }
5711
5712 if (arg_verity_settings.data_path)
5713 dissect_image_flags |= DISSECT_IMAGE_NO_PARTITION_TABLE;
5714 }
5715
5716 if (!mkdtemp(tmprootdir)) {
5717 r = log_error_errno(errno, "Failed to create temporary directory: %m");
5718 goto finish;
5719 }
5720
5721 remove_tmprootdir = true;
5722
5723 arg_directory = strdup(tmprootdir);
5724 if (!arg_directory) {
5725 r = log_oom();
5726 goto finish;
5727 }
5728
5729 r = loop_device_make_by_path(
5730 arg_image,
5731 arg_read_only ? O_RDONLY : O_RDWR,
5732 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
5733 &loop);
5734 if (r < 0) {
5735 log_error_errno(r, "Failed to set up loopback block device: %m");
5736 goto finish;
5737 }
5738
5739 r = dissect_image_and_warn(
5740 loop->fd,
5741 arg_image,
5742 &arg_verity_settings,
5743 NULL,
5744 loop->diskseq,
5745 loop->uevent_seqnum_not_before,
5746 loop->timestamp_not_before,
5747 dissect_image_flags,
5748 &dissected_image);
5749 if (r == -ENOPKG) {
5750 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5751 log_notice("Note that the disk image needs to\n"
5752 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5753 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5754 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5755 " d) or contain a file system without a partition table\n"
5756 "in order to be bootable with systemd-nspawn.");
5757 goto finish;
5758 }
5759 if (r < 0)
5760 goto finish;
5761
5762 r = dissected_image_load_verity_sig_partition(
5763 dissected_image,
5764 loop->fd,
5765 &arg_verity_settings);
5766 if (r < 0)
5767 goto finish;
5768
5769 if (dissected_image->has_verity && !arg_verity_settings.root_hash && !dissected_image->has_verity_sig)
5770 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5771 "root hash signature found! Proceeding without integrity checking.", arg_image);
5772
5773 r = dissected_image_decrypt_interactively(
5774 dissected_image,
5775 NULL,
5776 &arg_verity_settings,
5777 0,
5778 &decrypted_image);
5779 if (r < 0)
5780 goto finish;
5781
5782 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5783 if (remove_image && unlink(arg_image) >= 0)
5784 remove_image = false;
5785 }
5786
5787 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5788 if (r < 0)
5789 goto finish;
5790
5791 if (arg_console_mode < 0)
5792 arg_console_mode =
5793 isatty(STDIN_FILENO) > 0 &&
5794 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
5795
5796 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5797 arg_quiet = true;
5798
5799 if (!arg_quiet)
5800 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5801 arg_machine, arg_image ?: arg_directory);
5802
5803 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
5804
5805 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
5806 r = log_error_errno(errno, "Failed to become subreaper: %m");
5807 goto finish;
5808 }
5809
5810 if (arg_expose_ports) {
5811 r = fw_ctx_new(&fw_ctx);
5812 if (r < 0) {
5813 log_error_errno(r, "Cannot expose configured ports, firewall initialization failed: %m");
5814 goto finish;
5815 }
5816 expose_args.fw_ctx = fw_ctx;
5817 }
5818 for (;;) {
5819 r = run_container(dissected_image,
5820 secondary,
5821 fds,
5822 veth_name, &veth_created,
5823 &expose_args, &master,
5824 &pid, &ret);
5825 if (r <= 0)
5826 break;
5827 }
5828
5829 finish:
5830 (void) sd_notify(false,
5831 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5832 "STOPPING=1\nSTATUS=Terminating...");
5833
5834 if (pid > 0)
5835 (void) kill(pid, SIGKILL);
5836
5837 /* Try to flush whatever is still queued in the pty */
5838 if (master >= 0) {
5839 (void) copy_bytes(master, STDOUT_FILENO, UINT64_MAX, 0);
5840 master = safe_close(master);
5841 }
5842
5843 if (pid > 0)
5844 (void) wait_for_terminate(pid, NULL);
5845
5846 pager_close();
5847
5848 if (remove_directory && arg_directory) {
5849 int k;
5850
5851 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5852 if (k < 0)
5853 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5854 }
5855
5856 if (remove_image && arg_image) {
5857 if (unlink(arg_image) < 0)
5858 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5859 }
5860
5861 if (remove_tmprootdir) {
5862 if (rmdir(tmprootdir) < 0)
5863 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5864 }
5865
5866 if (arg_machine) {
5867 const char *p;
5868
5869 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5870 (void) rm_rf(p, REMOVE_ROOT);
5871 }
5872
5873 expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET, &expose_args.address4);
5874 expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET6, &expose_args.address6);
5875
5876 if (veth_created)
5877 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5878 (void) remove_bridge(arg_network_zone);
5879
5880 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5881 expose_port_free_all(arg_expose_ports);
5882 rlimit_free_all(arg_rlimit);
5883 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
5884 credential_free_all(arg_credentials, arg_n_credentials);
5885
5886 if (r < 0)
5887 return r;
5888
5889 return ret;
5890 }
5891
5892 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.