]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
nspawn: add support for owneridmap bind option
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <getopt.h>
5 #include <linux/loop.h>
6 #if HAVE_SELINUX
7 #include <selinux/selinux.h>
8 #endif
9 #include <stdlib.h>
10 #include <sys/file.h>
11 #include <sys/ioctl.h>
12 #include <sys/mount.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
15 #include <sys/types.h>
16 #include <sys/wait.h>
17 #include <termios.h>
18 #include <unistd.h>
19
20 #include <linux/fs.h> /* Must be included after <sys/mount.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 "ether-addr-util.h"
28 #include "barrier.h"
29 #include "base-filesystem.h"
30 #include "blkid-util.h"
31 #include "btrfs-util.h"
32 #include "build.h"
33 #include "bus-error.h"
34 #include "bus-locator.h"
35 #include "bus-util.h"
36 #include "cap-list.h"
37 #include "capability-util.h"
38 #include "cgroup-util.h"
39 #include "chase.h"
40 #include "common-signal.h"
41 #include "copy.h"
42 #include "cpu-set-util.h"
43 #include "creds-util.h"
44 #include "dev-setup.h"
45 #include "discover-image.h"
46 #include "dissect-image.h"
47 #include "env-util.h"
48 #include "escape.h"
49 #include "fd-util.h"
50 #include "fdset.h"
51 #include "fileio.h"
52 #include "format-util.h"
53 #include "fs-util.h"
54 #include "gpt.h"
55 #include "hexdecoct.h"
56 #include "hostname-setup.h"
57 #include "hostname-util.h"
58 #include "id128-util.h"
59 #include "io-util.h"
60 #include "log.h"
61 #include "loop-util.h"
62 #include "loopback-setup.h"
63 #include "machine-credential.h"
64 #include "macro.h"
65 #include "main-func.h"
66 #include "missing_sched.h"
67 #include "mkdir.h"
68 #include "mount-util.h"
69 #include "mountpoint-util.h"
70 #include "namespace-util.h"
71 #include "netlink-util.h"
72 #include "nspawn-bind-user.h"
73 #include "nspawn-cgroup.h"
74 #include "nspawn-def.h"
75 #include "nspawn-expose-ports.h"
76 #include "nspawn-mount.h"
77 #include "nspawn-network.h"
78 #include "nspawn-oci.h"
79 #include "nspawn-patch-uid.h"
80 #include "nspawn-register.h"
81 #include "nspawn-seccomp.h"
82 #include "nspawn-settings.h"
83 #include "nspawn-setuid.h"
84 #include "nspawn-stub-pid1.h"
85 #include "nspawn-util.h"
86 #include "nspawn.h"
87 #include "nulstr-util.h"
88 #include "os-util.h"
89 #include "pager.h"
90 #include "parse-argument.h"
91 #include "parse-util.h"
92 #include "pretty-print.h"
93 #include "process-util.h"
94 #include "ptyfwd.h"
95 #include "random-util.h"
96 #include "raw-clone.h"
97 #include "resolve-util.h"
98 #include "rlimit-util.h"
99 #include "rm-rf.h"
100 #include "seccomp-util.h"
101 #include "selinux-util.h"
102 #include "signal-util.h"
103 #include "socket-util.h"
104 #include "stat-util.h"
105 #include "stdio-util.h"
106 #include "string-table.h"
107 #include "string-util.h"
108 #include "strv.h"
109 #include "sysctl-util.h"
110 #include "terminal-util.h"
111 #include "tmpfile-util.h"
112 #include "umask-util.h"
113 #include "unit-name.h"
114 #include "user-util.h"
115 #include "vpick.h"
116
117 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
119 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
120
121 #define EXIT_FORCE_RESTART 133
122
123 typedef enum ContainerStatus {
124 CONTAINER_TERMINATED,
125 CONTAINER_REBOOTED,
126 } ContainerStatus;
127
128 static char *arg_directory = NULL;
129 static char *arg_template = NULL;
130 static char *arg_chdir = NULL;
131 static char *arg_pivot_root_new = NULL;
132 static char *arg_pivot_root_old = NULL;
133 static char *arg_user = NULL;
134 static uid_t arg_uid = UID_INVALID;
135 static gid_t arg_gid = GID_INVALID;
136 static gid_t* arg_supplementary_gids = NULL;
137 static size_t arg_n_supplementary_gids = 0;
138 static sd_id128_t arg_uuid = {};
139 static char *arg_machine = NULL; /* The name used by the host to refer to this */
140 static char *arg_hostname = NULL; /* The name the payload sees by default */
141 static const char *arg_selinux_context = NULL;
142 static const char *arg_selinux_apifs_context = NULL;
143 static char *arg_slice = NULL;
144 static bool arg_private_network = false;
145 static bool arg_read_only = false;
146 static StartMode arg_start_mode = START_PID1;
147 static bool arg_ephemeral = false;
148 static LinkJournal arg_link_journal = LINK_AUTO;
149 static bool arg_link_journal_try = false;
150 static uint64_t arg_caps_retain =
151 (1ULL << CAP_AUDIT_CONTROL) |
152 (1ULL << CAP_AUDIT_WRITE) |
153 (1ULL << CAP_CHOWN) |
154 (1ULL << CAP_DAC_OVERRIDE) |
155 (1ULL << CAP_DAC_READ_SEARCH) |
156 (1ULL << CAP_FOWNER) |
157 (1ULL << CAP_FSETID) |
158 (1ULL << CAP_IPC_OWNER) |
159 (1ULL << CAP_KILL) |
160 (1ULL << CAP_LEASE) |
161 (1ULL << CAP_LINUX_IMMUTABLE) |
162 (1ULL << CAP_MKNOD) |
163 (1ULL << CAP_NET_BIND_SERVICE) |
164 (1ULL << CAP_NET_BROADCAST) |
165 (1ULL << CAP_NET_RAW) |
166 (1ULL << CAP_SETFCAP) |
167 (1ULL << CAP_SETGID) |
168 (1ULL << CAP_SETPCAP) |
169 (1ULL << CAP_SETUID) |
170 (1ULL << CAP_SYS_ADMIN) |
171 (1ULL << CAP_SYS_BOOT) |
172 (1ULL << CAP_SYS_CHROOT) |
173 (1ULL << CAP_SYS_NICE) |
174 (1ULL << CAP_SYS_PTRACE) |
175 (1ULL << CAP_SYS_RESOURCE) |
176 (1ULL << CAP_SYS_TTY_CONFIG);
177 static uint64_t arg_caps_ambient = 0;
178 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
179 static CustomMount *arg_custom_mounts = NULL;
180 static size_t arg_n_custom_mounts = 0;
181 static char **arg_setenv = NULL;
182 static bool arg_quiet = false;
183 static bool arg_register = true;
184 static bool arg_keep_unit = false;
185 static char **arg_network_interfaces = NULL;
186 static char **arg_network_macvlan = NULL;
187 static char **arg_network_ipvlan = NULL;
188 static bool arg_network_veth = false;
189 static char **arg_network_veth_extra = NULL;
190 static char *arg_network_bridge = NULL;
191 static char *arg_network_zone = NULL;
192 static char *arg_network_namespace_path = NULL;
193 struct ether_addr arg_network_provided_mac = {};
194 static PagerFlags arg_pager_flags = 0;
195 static unsigned long arg_personality = PERSONALITY_INVALID;
196 static char *arg_image = NULL;
197 static char *arg_oci_bundle = NULL;
198 static VolatileMode arg_volatile_mode = VOLATILE_NO;
199 static ExposePort *arg_expose_ports = NULL;
200 static char **arg_property = NULL;
201 static sd_bus_message *arg_property_message = NULL;
202 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
203 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
204 static UserNamespaceOwnership arg_userns_ownership = _USER_NAMESPACE_OWNERSHIP_INVALID;
205 static int arg_kill_signal = 0;
206 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
207 static SettingsMask arg_settings_mask = 0;
208 static int arg_settings_trusted = -1;
209 static char **arg_parameters = NULL;
210 static const char *arg_container_service_name = "systemd-nspawn";
211 static bool arg_notify_ready = false;
212 static bool arg_use_cgns = true;
213 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
214 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
215 static VeritySettings arg_verity_settings = VERITY_SETTINGS_DEFAULT;
216 static char **arg_syscall_allow_list = NULL;
217 static char **arg_syscall_deny_list = NULL;
218 #if HAVE_SECCOMP
219 static scmp_filter_ctx arg_seccomp = NULL;
220 #endif
221 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
222 static bool arg_no_new_privileges = false;
223 static int arg_oom_score_adjust = 0;
224 static bool arg_oom_score_adjust_set = false;
225 static CPUSet arg_cpu_set = {};
226 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
227 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
228 static unsigned arg_console_width = UINT_MAX, arg_console_height = UINT_MAX;
229 static DeviceNode* arg_extra_nodes = NULL;
230 static size_t arg_n_extra_nodes = 0;
231 static char **arg_sysctl = NULL;
232 static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
233 static MachineCredentialContext arg_credentials = {};
234 static char **arg_bind_user = NULL;
235 static bool arg_suppress_sync = false;
236 static char *arg_settings_filename = NULL;
237 static Architecture arg_architecture = _ARCHITECTURE_INVALID;
238 static ImagePolicy *arg_image_policy = NULL;
239 static char *arg_background = NULL;
240
241 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
248 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
250 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
256 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
257 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
258 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
259 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
260 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
261 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
262 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
263 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
264 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings, verity_settings_done);
265 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list, strv_freep);
266 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list, strv_freep);
267 #if HAVE_SECCOMP
268 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
269 #endif
270 STATIC_DESTRUCTOR_REGISTER(arg_credentials, machine_credential_context_done);
271 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
272 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
273 STATIC_DESTRUCTOR_REGISTER(arg_bind_user, strv_freep);
274 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename, freep);
275 STATIC_DESTRUCTOR_REGISTER(arg_image_policy, image_policy_freep);
276 STATIC_DESTRUCTOR_REGISTER(arg_background, freep);
277
278 static int handle_arg_console(const char *arg) {
279 if (streq(arg, "help")) {
280 puts("autopipe\n"
281 "interactive\n"
282 "passive\n"
283 "pipe\n"
284 "read-only");
285 return 0;
286 }
287
288 if (streq(arg, "interactive"))
289 arg_console_mode = CONSOLE_INTERACTIVE;
290 else if (streq(arg, "read-only"))
291 arg_console_mode = CONSOLE_READ_ONLY;
292 else if (streq(arg, "passive"))
293 arg_console_mode = CONSOLE_PASSIVE;
294 else if (streq(arg, "pipe")) {
295 if (isatty(STDIN_FILENO) && isatty(STDOUT_FILENO))
296 log_full(arg_quiet ? LOG_DEBUG : LOG_NOTICE,
297 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
298 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
299 "Proceeding anyway.");
300
301 arg_console_mode = CONSOLE_PIPE;
302 } else if (streq(arg, "autopipe")) {
303 if (isatty(STDIN_FILENO) && isatty(STDOUT_FILENO))
304 arg_console_mode = CONSOLE_INTERACTIVE;
305 else
306 arg_console_mode = CONSOLE_PIPE;
307 } else
308 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
309
310 arg_settings_mask |= SETTING_CONSOLE_MODE;
311 return 1;
312 }
313
314 static int help(void) {
315 _cleanup_free_ char *link = NULL;
316 int r;
317
318 pager_open(arg_pager_flags);
319
320 r = terminal_urlify_man("systemd-nspawn", "1", &link);
321 if (r < 0)
322 return log_oom();
323
324 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
325 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
326 " -h --help Show this help\n"
327 " --version Print version string\n"
328 " -q --quiet Do not show status information\n"
329 " --no-pager Do not pipe output into a pager\n"
330 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
331 "\n%3$sImage:%4$s\n"
332 " -D --directory=PATH Root directory for the container\n"
333 " --template=PATH Initialize root directory from template directory,\n"
334 " if missing\n"
335 " -x --ephemeral Run container with snapshot of root directory, and\n"
336 " remove it after exit\n"
337 " -i --image=PATH Root file system disk image (or device node) for\n"
338 " the container\n"
339 " --image-policy=POLICY Specify disk image dissection policy\n"
340 " --oci-bundle=PATH OCI bundle directory\n"
341 " --read-only Mount the root directory read-only\n"
342 " --volatile[=MODE] Run the system in volatile mode\n"
343 " --root-hash=HASH Specify verity root hash for root disk image\n"
344 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
345 " as a DER encoded PKCS7, either as a path to a file\n"
346 " or as an ASCII base64 encoded string prefixed by\n"
347 " 'base64:'\n"
348 " --verity-data=PATH Specify hash device for verity\n"
349 " --pivot-root=PATH[:PATH]\n"
350 " Pivot root to given directory in the container\n"
351 "\n%3$sExecution:%4$s\n"
352 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
353 " -b --boot Boot up full system (i.e. invoke init)\n"
354 " --chdir=PATH Set working directory in the container\n"
355 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
356 " -u --user=USER Run the command under specified user or UID\n"
357 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
358 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
359 " --suppress-sync=BOOLEAN\n"
360 " Suppress any form of disk data synchronization\n"
361 "\n%3$sSystem Identity:%4$s\n"
362 " -M --machine=NAME Set the machine name for the container\n"
363 " --hostname=NAME Override the hostname for the container\n"
364 " --uuid=UUID Set a specific machine UUID for the container\n"
365 "\n%3$sProperties:%4$s\n"
366 " -S --slice=SLICE Place the container in the specified slice\n"
367 " --property=NAME=VALUE Set scope unit property\n"
368 " --register=BOOLEAN Register container as machine\n"
369 " --keep-unit Do not register a scope for the machine, reuse\n"
370 " the service unit nspawn is running in\n"
371 "\n%3$sUser Namespacing:%4$s\n"
372 " --private-users=no Run without user namespacing\n"
373 " --private-users=yes|pick|identity\n"
374 " Run within user namespace, autoselect UID/GID range\n"
375 " --private-users=UIDBASE[:NUIDS]\n"
376 " Similar, but with user configured UID/GID range\n"
377 " --private-users-ownership=MODE\n"
378 " Adjust ('chown') or map ('map') OS tree ownership\n"
379 " to private UID/GID range\n"
380 " -U Equivalent to --private-users=pick and\n"
381 " --private-users-ownership=auto\n"
382 "\n%3$sNetworking:%4$s\n"
383 " --private-network Disable network in container\n"
384 " --network-interface=HOSTIF[:CONTAINERIF]\n"
385 " Assign an existing network interface to the\n"
386 " container\n"
387 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
388 " Create a macvlan network interface based on an\n"
389 " existing network interface to the container\n"
390 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
391 " Create an ipvlan network interface based on an\n"
392 " existing network interface to the container\n"
393 " -n --network-veth Add a virtual Ethernet connection between host\n"
394 " and container\n"
395 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
396 " Add an additional virtual Ethernet link between\n"
397 " host and container\n"
398 " --network-bridge=INTERFACE\n"
399 " Add a virtual Ethernet connection to the container\n"
400 " and attach it to an existing bridge on the host\n"
401 " --network-zone=NAME Similar, but attach the new interface to an\n"
402 " an automatically managed bridge interface\n"
403 " --network-namespace-path=PATH\n"
404 " Set network namespace to the one represented by\n"
405 " the specified kernel namespace file node\n"
406 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
407 " Expose a container IP port on the host\n"
408 "\n%3$sSecurity:%4$s\n"
409 " --capability=CAP In addition to the default, retain specified\n"
410 " capability\n"
411 " --drop-capability=CAP Drop the specified capability from the default set\n"
412 " --ambient-capability=CAP\n"
413 " Sets the specified capability for the started\n"
414 " process. Not useful if booting a machine.\n"
415 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
416 " --system-call-filter=LIST|~LIST\n"
417 " Permit/prohibit specific system calls\n"
418 " -Z --selinux-context=SECLABEL\n"
419 " Set the SELinux security context to be used by\n"
420 " processes in the container\n"
421 " -L --selinux-apifs-context=SECLABEL\n"
422 " Set the SELinux security context to be used by\n"
423 " API/tmpfs file systems in the container\n"
424 "\n%3$sResources:%4$s\n"
425 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
426 " --oom-score-adjust=VALUE\n"
427 " Adjust the OOM score value for the payload\n"
428 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
429 " --personality=ARCH Pick personality for this container\n"
430 "\n%3$sIntegration:%4$s\n"
431 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
432 " --timezone=MODE Select mode of /etc/localtime initialization\n"
433 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
434 " host, try-guest, try-host\n"
435 " -j Equivalent to --link-journal=try-guest\n"
436 "\n%3$sMounts:%4$s\n"
437 " --bind=PATH[:PATH[:OPTIONS]]\n"
438 " Bind mount a file or directory from the host into\n"
439 " the container\n"
440 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
441 " Similar, but creates a read-only bind mount\n"
442 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
443 " it\n"
444 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
445 " --overlay=PATH[:PATH...]:PATH\n"
446 " Create an overlay mount from the host to \n"
447 " the container\n"
448 " --overlay-ro=PATH[:PATH...]:PATH\n"
449 " Similar, but creates a read-only overlay mount\n"
450 " --bind-user=NAME Bind user from host to container\n"
451 "\n%3$sInput/Output:%4$s\n"
452 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
453 " set up for the container.\n"
454 " -P --pipe Equivalent to --console=pipe\n"
455 " --background=COLOR Set ANSI color for background\n"
456 "\n%3$sCredentials:%4$s\n"
457 " --set-credential=ID:VALUE\n"
458 " Pass a credential with literal value to container.\n"
459 " --load-credential=ID:PATH\n"
460 " Load credential to pass to container from file or\n"
461 " AF_UNIX stream socket.\n"
462 "\nSee the %2$s for details.\n",
463 program_invocation_short_name,
464 link,
465 ansi_underline(),
466 ansi_normal(),
467 ansi_highlight(),
468 ansi_normal());
469
470 return 0;
471 }
472
473 static int custom_mount_check_all(void) {
474 size_t i;
475
476 for (i = 0; i < arg_n_custom_mounts; i++) {
477 CustomMount *m = &arg_custom_mounts[i];
478
479 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
480 if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_OFF)
481 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
482 "--private-users-ownership=own may not be combined with custom root mounts.");
483 if (arg_uid_shift == UID_INVALID)
484 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
485 "--private-users with automatic UID shift may not be combined with custom root mounts.");
486 }
487 }
488
489 return 0;
490 }
491
492 static int detect_unified_cgroup_hierarchy_from_environment(void) {
493 const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
494 int r;
495
496 /* Allow the user to control whether the unified hierarchy is used */
497
498 e = getenv(var);
499 if (!e) {
500 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
501 var = "UNIFIED_CGROUP_HIERARCHY";
502 e = getenv(var);
503 }
504
505 if (!isempty(e)) {
506 r = parse_boolean(e);
507 if (r < 0)
508 return log_error_errno(r, "Failed to parse $%s: %m", var);
509 if (r > 0)
510 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
511 else
512 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
513 }
514
515 return 0;
516 }
517
518 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
519 int r;
520
521 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
522 * in the image actually supports. */
523 r = cg_all_unified();
524 if (r < 0)
525 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
526 if (r > 0) {
527 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
528 * routine only detects 231, so we'll have a false negative here for 230. */
529 r = systemd_installation_has_version(directory, "230");
530 if (r < 0)
531 return log_error_errno(r, "Failed to determine systemd version in container: %m");
532 if (r > 0)
533 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
534 else
535 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
536 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
537 /* Mixed cgroup hierarchy support was added in 233 */
538 r = systemd_installation_has_version(directory, "233");
539 if (r < 0)
540 return log_error_errno(r, "Failed to determine systemd version in container: %m");
541 if (r > 0)
542 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
543 else
544 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
545 } else
546 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
547
548 log_debug("Using %s hierarchy for container.",
549 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
550 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
551
552 return 0;
553 }
554
555 static int parse_capability_spec(const char *spec, uint64_t *ret_mask) {
556 uint64_t mask = 0;
557 int r;
558
559 for (;;) {
560 _cleanup_free_ char *t = NULL;
561
562 r = extract_first_word(&spec, &t, ",", 0);
563 if (r < 0)
564 return log_error_errno(r, "Failed to parse capability %s.", t);
565 if (r == 0)
566 break;
567
568 if (streq(t, "help")) {
569 for (int i = 0; i < capability_list_length(); i++) {
570 const char *name;
571
572 name = capability_to_name(i);
573 if (name)
574 puts(name);
575 }
576
577 return 0; /* quit */
578 }
579
580 if (streq(t, "all"))
581 mask = UINT64_MAX;
582 else {
583 r = capability_from_name(t);
584 if (r < 0)
585 return log_error_errno(r, "Failed to parse capability %s.", t);
586
587 mask |= 1ULL << r;
588 }
589 }
590
591 *ret_mask = mask;
592 return 1; /* continue */
593 }
594
595 static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
596 int r;
597
598 r = getenv_bool(name);
599 if (r == -ENXIO)
600 return 0;
601 if (r < 0)
602 return log_error_errno(r, "Failed to parse $%s: %m", name);
603
604 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
605 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
606 return 0;
607 }
608
609 static int parse_mount_settings_env(void) {
610 const char *e;
611 int r;
612
613 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
614 if (r < 0 && r != -ENXIO)
615 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
616 if (r >= 0)
617 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
618
619 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
620 if (streq_ptr(e, "network"))
621 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
622
623 else if (e) {
624 r = parse_boolean(e);
625 if (r < 0)
626 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
627
628 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
629 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
630 }
631
632 return 0;
633 }
634
635 static int parse_environment(void) {
636 const char *e;
637 int r;
638
639 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
640 if (r < 0)
641 return r;
642 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
643 if (r < 0)
644 return r;
645 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
646 if (r < 0)
647 return r;
648 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
649 if (r < 0)
650 return r;
651
652 r = parse_mount_settings_env();
653 if (r < 0)
654 return r;
655
656 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
657 * even if it is supported. If not supported, it has no effect. */
658 if (!cg_ns_supported())
659 arg_use_cgns = false;
660 else {
661 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
662 if (r < 0) {
663 if (r != -ENXIO)
664 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
665
666 arg_use_cgns = true;
667 } else {
668 arg_use_cgns = r > 0;
669 arg_settings_mask |= SETTING_USE_CGNS;
670 }
671 }
672
673 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
674 if (e)
675 arg_container_service_name = e;
676
677 e = getenv("SYSTEMD_NSPAWN_NETWORK_MAC");
678 if (e) {
679 r = parse_ether_addr(e, &arg_network_provided_mac);
680 if (r < 0)
681 return log_error_errno(r, "Failed to parse provided MAC address via environment variable");
682 }
683
684 r = getenv_bool("SYSTEMD_SUPPRESS_SYNC");
685 if (r >= 0)
686 arg_suppress_sync = r;
687 else if (r != -ENXIO)
688 log_debug_errno(r, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
689
690 return detect_unified_cgroup_hierarchy_from_environment();
691 }
692
693 static int parse_argv(int argc, char *argv[]) {
694 enum {
695 ARG_VERSION = 0x100,
696 ARG_PRIVATE_NETWORK,
697 ARG_UUID,
698 ARG_READ_ONLY,
699 ARG_CAPABILITY,
700 ARG_AMBIENT_CAPABILITY,
701 ARG_DROP_CAPABILITY,
702 ARG_LINK_JOURNAL,
703 ARG_BIND,
704 ARG_BIND_RO,
705 ARG_TMPFS,
706 ARG_OVERLAY,
707 ARG_OVERLAY_RO,
708 ARG_INACCESSIBLE,
709 ARG_SHARE_SYSTEM,
710 ARG_REGISTER,
711 ARG_KEEP_UNIT,
712 ARG_NETWORK_INTERFACE,
713 ARG_NETWORK_MACVLAN,
714 ARG_NETWORK_IPVLAN,
715 ARG_NETWORK_BRIDGE,
716 ARG_NETWORK_ZONE,
717 ARG_NETWORK_VETH_EXTRA,
718 ARG_NETWORK_NAMESPACE_PATH,
719 ARG_PERSONALITY,
720 ARG_VOLATILE,
721 ARG_TEMPLATE,
722 ARG_PROPERTY,
723 ARG_PRIVATE_USERS,
724 ARG_KILL_SIGNAL,
725 ARG_SETTINGS,
726 ARG_CHDIR,
727 ARG_PIVOT_ROOT,
728 ARG_PRIVATE_USERS_CHOWN,
729 ARG_PRIVATE_USERS_OWNERSHIP,
730 ARG_NOTIFY_READY,
731 ARG_ROOT_HASH,
732 ARG_ROOT_HASH_SIG,
733 ARG_VERITY_DATA,
734 ARG_SYSTEM_CALL_FILTER,
735 ARG_RLIMIT,
736 ARG_HOSTNAME,
737 ARG_NO_NEW_PRIVILEGES,
738 ARG_OOM_SCORE_ADJUST,
739 ARG_CPU_AFFINITY,
740 ARG_RESOLV_CONF,
741 ARG_TIMEZONE,
742 ARG_CONSOLE,
743 ARG_PIPE,
744 ARG_OCI_BUNDLE,
745 ARG_NO_PAGER,
746 ARG_SET_CREDENTIAL,
747 ARG_LOAD_CREDENTIAL,
748 ARG_BIND_USER,
749 ARG_SUPPRESS_SYNC,
750 ARG_IMAGE_POLICY,
751 ARG_BACKGROUND,
752 };
753
754 static const struct option options[] = {
755 { "help", no_argument, NULL, 'h' },
756 { "version", no_argument, NULL, ARG_VERSION },
757 { "directory", required_argument, NULL, 'D' },
758 { "template", required_argument, NULL, ARG_TEMPLATE },
759 { "ephemeral", no_argument, NULL, 'x' },
760 { "user", required_argument, NULL, 'u' },
761 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
762 { "as-pid2", no_argument, NULL, 'a' },
763 { "boot", no_argument, NULL, 'b' },
764 { "uuid", required_argument, NULL, ARG_UUID },
765 { "read-only", no_argument, NULL, ARG_READ_ONLY },
766 { "capability", required_argument, NULL, ARG_CAPABILITY },
767 { "ambient-capability", required_argument, NULL, ARG_AMBIENT_CAPABILITY },
768 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
769 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
770 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
771 { "bind", required_argument, NULL, ARG_BIND },
772 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
773 { "tmpfs", required_argument, NULL, ARG_TMPFS },
774 { "overlay", required_argument, NULL, ARG_OVERLAY },
775 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
776 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
777 { "machine", required_argument, NULL, 'M' },
778 { "hostname", required_argument, NULL, ARG_HOSTNAME },
779 { "slice", required_argument, NULL, 'S' },
780 { "setenv", required_argument, NULL, 'E' },
781 { "selinux-context", required_argument, NULL, 'Z' },
782 { "selinux-apifs-context", required_argument, NULL, 'L' },
783 { "quiet", no_argument, NULL, 'q' },
784 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
785 { "register", required_argument, NULL, ARG_REGISTER },
786 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
787 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
788 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
789 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
790 { "network-veth", no_argument, NULL, 'n' },
791 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
792 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
793 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
794 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
795 { "personality", required_argument, NULL, ARG_PERSONALITY },
796 { "image", required_argument, NULL, 'i' },
797 { "volatile", optional_argument, NULL, ARG_VOLATILE },
798 { "port", required_argument, NULL, 'p' },
799 { "property", required_argument, NULL, ARG_PROPERTY },
800 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
801 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN }, /* obsolete */
802 { "private-users-ownership",required_argument, NULL, ARG_PRIVATE_USERS_OWNERSHIP},
803 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
804 { "settings", required_argument, NULL, ARG_SETTINGS },
805 { "chdir", required_argument, NULL, ARG_CHDIR },
806 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
807 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
808 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
809 { "root-hash-sig", required_argument, NULL, ARG_ROOT_HASH_SIG },
810 { "verity-data", required_argument, NULL, ARG_VERITY_DATA },
811 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
812 { "rlimit", required_argument, NULL, ARG_RLIMIT },
813 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
814 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
815 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
816 { "timezone", required_argument, NULL, ARG_TIMEZONE },
817 { "console", required_argument, NULL, ARG_CONSOLE },
818 { "pipe", no_argument, NULL, ARG_PIPE },
819 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
820 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
821 { "set-credential", required_argument, NULL, ARG_SET_CREDENTIAL },
822 { "load-credential", required_argument, NULL, ARG_LOAD_CREDENTIAL },
823 { "bind-user", required_argument, NULL, ARG_BIND_USER },
824 { "suppress-sync", required_argument, NULL, ARG_SUPPRESS_SYNC },
825 { "image-policy", required_argument, NULL, ARG_IMAGE_POLICY },
826 { "background", required_argument, NULL, ARG_BACKGROUND },
827 {}
828 };
829
830 int c, r;
831 uint64_t plus = 0, minus = 0;
832 bool mask_all_settings = false, mask_no_settings = false;
833
834 assert(argc >= 0);
835 assert(argv);
836
837 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
838 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
839 optind = 0;
840 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
841 switch (c) {
842
843 case 'h':
844 return help();
845
846 case ARG_VERSION:
847 return version();
848
849 case 'D':
850 r = parse_path_argument(optarg, false, &arg_directory);
851 if (r < 0)
852 return r;
853
854 arg_settings_mask |= SETTING_DIRECTORY;
855 break;
856
857 case ARG_TEMPLATE:
858 r = parse_path_argument(optarg, false, &arg_template);
859 if (r < 0)
860 return r;
861
862 arg_settings_mask |= SETTING_DIRECTORY;
863 break;
864
865 case 'i':
866 r = parse_path_argument(optarg, false, &arg_image);
867 if (r < 0)
868 return r;
869
870 arg_settings_mask |= SETTING_DIRECTORY;
871 break;
872
873 case ARG_OCI_BUNDLE:
874 r = parse_path_argument(optarg, false, &arg_oci_bundle);
875 if (r < 0)
876 return r;
877
878 break;
879
880 case 'x':
881 arg_ephemeral = true;
882 arg_settings_mask |= SETTING_EPHEMERAL;
883 break;
884
885 case 'u':
886 r = free_and_strdup(&arg_user, optarg);
887 if (r < 0)
888 return log_oom();
889
890 arg_settings_mask |= SETTING_USER;
891 break;
892
893 case ARG_NETWORK_ZONE: {
894 _cleanup_free_ char *j = NULL;
895
896 j = strjoin("vz-", optarg);
897 if (!j)
898 return log_oom();
899
900 if (!ifname_valid(j))
901 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
902 "Network zone name not valid: %s", j);
903
904 free_and_replace(arg_network_zone, j);
905
906 arg_network_veth = true;
907 arg_private_network = true;
908 arg_settings_mask |= SETTING_NETWORK;
909 break;
910 }
911
912 case ARG_NETWORK_BRIDGE:
913
914 if (!ifname_valid(optarg))
915 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
916 "Bridge interface name not valid: %s", optarg);
917
918 r = free_and_strdup(&arg_network_bridge, optarg);
919 if (r < 0)
920 return log_oom();
921
922 _fallthrough_;
923 case 'n':
924 arg_network_veth = true;
925 arg_private_network = true;
926 arg_settings_mask |= SETTING_NETWORK;
927 break;
928
929 case ARG_NETWORK_VETH_EXTRA:
930 r = veth_extra_parse(&arg_network_veth_extra, optarg);
931 if (r < 0)
932 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
933
934 arg_private_network = true;
935 arg_settings_mask |= SETTING_NETWORK;
936 break;
937
938 case ARG_NETWORK_INTERFACE:
939 r = interface_pair_parse(&arg_network_interfaces, optarg);
940 if (r < 0)
941 return r;
942
943 arg_private_network = true;
944 arg_settings_mask |= SETTING_NETWORK;
945 break;
946
947 case ARG_NETWORK_MACVLAN:
948 r = macvlan_pair_parse(&arg_network_macvlan, optarg);
949 if (r < 0)
950 return r;
951
952 arg_private_network = true;
953 arg_settings_mask |= SETTING_NETWORK;
954 break;
955
956 case ARG_NETWORK_IPVLAN:
957 r = ipvlan_pair_parse(&arg_network_ipvlan, optarg);
958 if (r < 0)
959 return r;
960
961 _fallthrough_;
962 case ARG_PRIVATE_NETWORK:
963 arg_private_network = true;
964 arg_settings_mask |= SETTING_NETWORK;
965 break;
966
967 case ARG_NETWORK_NAMESPACE_PATH:
968 r = parse_path_argument(optarg, false, &arg_network_namespace_path);
969 if (r < 0)
970 return r;
971
972 arg_settings_mask |= SETTING_NETWORK;
973 break;
974
975 case 'b':
976 if (arg_start_mode == START_PID2)
977 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
978 "--boot and --as-pid2 may not be combined.");
979
980 arg_start_mode = START_BOOT;
981 arg_settings_mask |= SETTING_START_MODE;
982 break;
983
984 case 'a':
985 if (arg_start_mode == START_BOOT)
986 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
987 "--boot and --as-pid2 may not be combined.");
988
989 arg_start_mode = START_PID2;
990 arg_settings_mask |= SETTING_START_MODE;
991 break;
992
993 case ARG_UUID:
994 r = id128_from_string_nonzero(optarg, &arg_uuid);
995 if (r == -ENXIO)
996 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
997 "Machine UUID may not be all zeroes.");
998 if (r < 0)
999 return log_error_errno(r, "Invalid UUID: %s", optarg);
1000
1001 arg_settings_mask |= SETTING_MACHINE_ID;
1002 break;
1003
1004 case 'S': {
1005 _cleanup_free_ char *mangled = NULL;
1006
1007 r = unit_name_mangle_with_suffix(optarg, NULL, UNIT_NAME_MANGLE_WARN, ".slice", &mangled);
1008 if (r < 0)
1009 return log_oom();
1010
1011 free_and_replace(arg_slice, mangled);
1012 arg_settings_mask |= SETTING_SLICE;
1013 break;
1014 }
1015
1016 case 'M':
1017 if (isempty(optarg))
1018 arg_machine = mfree(arg_machine);
1019 else {
1020 if (!hostname_is_valid(optarg, 0))
1021 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1022 "Invalid machine name: %s", optarg);
1023
1024 r = free_and_strdup(&arg_machine, optarg);
1025 if (r < 0)
1026 return log_oom();
1027 }
1028 break;
1029
1030 case ARG_HOSTNAME:
1031 if (isempty(optarg))
1032 arg_hostname = mfree(arg_hostname);
1033 else {
1034 if (!hostname_is_valid(optarg, 0))
1035 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1036 "Invalid hostname: %s", optarg);
1037
1038 r = free_and_strdup(&arg_hostname, optarg);
1039 if (r < 0)
1040 return log_oom();
1041 }
1042
1043 arg_settings_mask |= SETTING_HOSTNAME;
1044 break;
1045
1046 case 'Z':
1047 arg_selinux_context = optarg;
1048 break;
1049
1050 case 'L':
1051 arg_selinux_apifs_context = optarg;
1052 break;
1053
1054 case ARG_READ_ONLY:
1055 arg_read_only = true;
1056 arg_settings_mask |= SETTING_READ_ONLY;
1057 break;
1058
1059 case ARG_AMBIENT_CAPABILITY: {
1060 uint64_t m;
1061 r = parse_capability_spec(optarg, &m);
1062 if (r <= 0)
1063 return r;
1064 arg_caps_ambient |= m;
1065 arg_settings_mask |= SETTING_CAPABILITY;
1066 break;
1067 }
1068 case ARG_CAPABILITY:
1069 case ARG_DROP_CAPABILITY: {
1070 uint64_t m;
1071 r = parse_capability_spec(optarg, &m);
1072 if (r <= 0)
1073 return r;
1074
1075 if (c == ARG_CAPABILITY)
1076 plus |= m;
1077 else
1078 minus |= m;
1079 arg_settings_mask |= SETTING_CAPABILITY;
1080 break;
1081 }
1082 case ARG_NO_NEW_PRIVILEGES:
1083 r = parse_boolean(optarg);
1084 if (r < 0)
1085 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
1086
1087 arg_no_new_privileges = r;
1088 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
1089 break;
1090
1091 case 'j':
1092 arg_link_journal = LINK_GUEST;
1093 arg_link_journal_try = true;
1094 arg_settings_mask |= SETTING_LINK_JOURNAL;
1095 break;
1096
1097 case ARG_LINK_JOURNAL:
1098 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
1099 if (r < 0)
1100 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
1101
1102 arg_settings_mask |= SETTING_LINK_JOURNAL;
1103 break;
1104
1105 case ARG_BIND:
1106 case ARG_BIND_RO:
1107 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
1108 if (r < 0)
1109 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
1110
1111 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1112 break;
1113
1114 case ARG_TMPFS:
1115 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1116 if (r < 0)
1117 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
1118
1119 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1120 break;
1121
1122 case ARG_OVERLAY:
1123 case ARG_OVERLAY_RO:
1124 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
1125 if (r == -EADDRNOTAVAIL)
1126 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1127 if (r < 0)
1128 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
1129
1130 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1131 break;
1132
1133 case ARG_INACCESSIBLE:
1134 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1135 if (r < 0)
1136 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
1137
1138 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1139 break;
1140
1141 case 'E':
1142 r = strv_env_replace_strdup_passthrough(&arg_setenv, optarg);
1143 if (r < 0)
1144 return log_error_errno(r, "Cannot assign environment variable %s: %m", optarg);
1145
1146 arg_settings_mask |= SETTING_ENVIRONMENT;
1147 break;
1148
1149 case 'q':
1150 arg_quiet = true;
1151 break;
1152
1153 case ARG_SHARE_SYSTEM:
1154 /* We don't officially support this anymore, except for compat reasons. People should use the
1155 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1156 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1157 arg_clone_ns_flags = 0;
1158 break;
1159
1160 case ARG_REGISTER:
1161 r = parse_boolean(optarg);
1162 if (r < 0) {
1163 log_error("Failed to parse --register= argument: %s", optarg);
1164 return r;
1165 }
1166
1167 arg_register = r;
1168 break;
1169
1170 case ARG_KEEP_UNIT:
1171 arg_keep_unit = true;
1172 break;
1173
1174 case ARG_PERSONALITY:
1175
1176 arg_personality = personality_from_string(optarg);
1177 if (arg_personality == PERSONALITY_INVALID)
1178 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1179 "Unknown or unsupported personality '%s'.", optarg);
1180
1181 arg_settings_mask |= SETTING_PERSONALITY;
1182 break;
1183
1184 case ARG_VOLATILE:
1185
1186 if (!optarg)
1187 arg_volatile_mode = VOLATILE_YES;
1188 else if (streq(optarg, "help")) {
1189 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1190 return 0;
1191 } else {
1192 VolatileMode m;
1193
1194 m = volatile_mode_from_string(optarg);
1195 if (m < 0)
1196 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1197 "Failed to parse --volatile= argument: %s", optarg);
1198 else
1199 arg_volatile_mode = m;
1200 }
1201
1202 arg_settings_mask |= SETTING_VOLATILE_MODE;
1203 break;
1204
1205 case 'p':
1206 r = expose_port_parse(&arg_expose_ports, optarg);
1207 if (r == -EEXIST)
1208 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1209 if (r < 0)
1210 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1211
1212 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1213 break;
1214
1215 case ARG_PROPERTY:
1216 if (strv_extend(&arg_property, optarg) < 0)
1217 return log_oom();
1218
1219 break;
1220
1221 case ARG_PRIVATE_USERS: {
1222 int boolean;
1223
1224 if (!optarg)
1225 boolean = true;
1226 else if (!in_charset(optarg, DIGITS))
1227 /* do *not* parse numbers as booleans */
1228 boolean = parse_boolean(optarg);
1229 else
1230 boolean = -1;
1231
1232 if (boolean == 0) {
1233 /* no: User namespacing off */
1234 arg_userns_mode = USER_NAMESPACE_NO;
1235 arg_uid_shift = UID_INVALID;
1236 arg_uid_range = UINT32_C(0x10000);
1237 } else if (boolean > 0) {
1238 /* yes: User namespacing on, UID range is read from root dir */
1239 arg_userns_mode = USER_NAMESPACE_FIXED;
1240 arg_uid_shift = UID_INVALID;
1241 arg_uid_range = UINT32_C(0x10000);
1242 } else if (streq(optarg, "pick")) {
1243 /* pick: User namespacing on, UID range is picked randomly */
1244 arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
1245 * implied by USER_NAMESPACE_PICK
1246 * further down. */
1247 arg_uid_shift = UID_INVALID;
1248 arg_uid_range = UINT32_C(0x10000);
1249
1250 } else if (streq(optarg, "identity")) {
1251 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1252 * itself, i.e. we don't actually map anything, but do take benefit of
1253 * isolation of capability sets. */
1254 arg_userns_mode = USER_NAMESPACE_FIXED;
1255 arg_uid_shift = 0;
1256 arg_uid_range = UINT32_C(0x10000);
1257 } else {
1258 /* anything else: User namespacing on, UID range is explicitly configured */
1259 r = parse_userns_uid_range(optarg, &arg_uid_shift, &arg_uid_range);
1260 if (r < 0)
1261 return r;
1262 arg_userns_mode = USER_NAMESPACE_FIXED;
1263 }
1264
1265 arg_settings_mask |= SETTING_USERNS;
1266 break;
1267 }
1268
1269 case 'U':
1270 if (userns_supported()) {
1271 arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
1272 * implied by USER_NAMESPACE_PICK
1273 * further down. */
1274 arg_uid_shift = UID_INVALID;
1275 arg_uid_range = UINT32_C(0x10000);
1276
1277 arg_settings_mask |= SETTING_USERNS;
1278 }
1279
1280 break;
1281
1282 case ARG_PRIVATE_USERS_CHOWN:
1283 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
1284
1285 arg_settings_mask |= SETTING_USERNS;
1286 break;
1287
1288 case ARG_PRIVATE_USERS_OWNERSHIP:
1289 if (streq(optarg, "help")) {
1290 DUMP_STRING_TABLE(user_namespace_ownership, UserNamespaceOwnership, _USER_NAMESPACE_OWNERSHIP_MAX);
1291 return 0;
1292 }
1293
1294 arg_userns_ownership = user_namespace_ownership_from_string(optarg);
1295 if (arg_userns_ownership < 0)
1296 return log_error_errno(arg_userns_ownership, "Cannot parse --user-namespace-ownership= value: %s", optarg);
1297
1298 arg_settings_mask |= SETTING_USERNS;
1299 break;
1300
1301 case ARG_KILL_SIGNAL:
1302 if (streq(optarg, "help")) {
1303 DUMP_STRING_TABLE(signal, int, _NSIG);
1304 return 0;
1305 }
1306
1307 arg_kill_signal = signal_from_string(optarg);
1308 if (arg_kill_signal < 0)
1309 return log_error_errno(arg_kill_signal, "Cannot parse signal: %s", optarg);
1310
1311 arg_settings_mask |= SETTING_KILL_SIGNAL;
1312 break;
1313
1314 case ARG_SETTINGS:
1315
1316 /* no → do not read files
1317 * yes → read files, do not override cmdline, trust only subset
1318 * override → read files, override cmdline, trust only subset
1319 * trusted → read files, do not override cmdline, trust all
1320 */
1321
1322 r = parse_boolean(optarg);
1323 if (r < 0) {
1324 if (streq(optarg, "trusted")) {
1325 mask_all_settings = false;
1326 mask_no_settings = false;
1327 arg_settings_trusted = true;
1328
1329 } else if (streq(optarg, "override")) {
1330 mask_all_settings = false;
1331 mask_no_settings = true;
1332 arg_settings_trusted = -1;
1333 } else
1334 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1335 } else if (r > 0) {
1336 /* yes */
1337 mask_all_settings = false;
1338 mask_no_settings = false;
1339 arg_settings_trusted = -1;
1340 } else {
1341 /* no */
1342 mask_all_settings = true;
1343 mask_no_settings = false;
1344 arg_settings_trusted = false;
1345 }
1346
1347 break;
1348
1349 case ARG_CHDIR: {
1350 _cleanup_free_ char *wd = NULL;
1351
1352 if (!path_is_absolute(optarg))
1353 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1354 "Working directory %s is not an absolute path.", optarg);
1355
1356 r = path_simplify_alloc(optarg, &wd);
1357 if (r < 0)
1358 return log_error_errno(r, "Failed to simplify path %s: %m", optarg);
1359
1360 if (!path_is_normalized(wd))
1361 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Working directory path is not normalized: %s", wd);
1362
1363 if (path_below_api_vfs(wd))
1364 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Working directory is below API VFS, refusing: %s", wd);
1365
1366 free_and_replace(arg_chdir, wd);
1367 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1368 break;
1369 }
1370
1371 case ARG_PIVOT_ROOT:
1372 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1373 if (r < 0)
1374 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1375
1376 arg_settings_mask |= SETTING_PIVOT_ROOT;
1377 break;
1378
1379 case ARG_NOTIFY_READY:
1380 r = parse_boolean(optarg);
1381 if (r < 0)
1382 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1383 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1384 arg_notify_ready = r;
1385 arg_settings_mask |= SETTING_NOTIFY_READY;
1386 break;
1387
1388 case ARG_ROOT_HASH: {
1389 _cleanup_free_ void *k = NULL;
1390 size_t l;
1391
1392 r = unhexmem(optarg, &k, &l);
1393 if (r < 0)
1394 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1395 if (l < sizeof(sd_id128_t))
1396 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128-bit long: %s", optarg);
1397
1398 free_and_replace(arg_verity_settings.root_hash, k);
1399 arg_verity_settings.root_hash_size = l;
1400 break;
1401 }
1402
1403 case ARG_ROOT_HASH_SIG: {
1404 char *value;
1405 size_t l;
1406 void *p;
1407
1408 if ((value = startswith(optarg, "base64:"))) {
1409 r = unbase64mem(value, &p, &l);
1410 if (r < 0)
1411 return log_error_errno(r, "Failed to parse root hash signature '%s': %m", optarg);
1412
1413 } else {
1414 r = read_full_file(optarg, (char**) &p, &l);
1415 if (r < 0)
1416 return log_error_errno(r, "Failed parse root hash signature file '%s': %m", optarg);
1417 }
1418
1419 free_and_replace(arg_verity_settings.root_hash_sig, p);
1420 arg_verity_settings.root_hash_sig_size = l;
1421 break;
1422 }
1423
1424 case ARG_VERITY_DATA:
1425 r = parse_path_argument(optarg, false, &arg_verity_settings.data_path);
1426 if (r < 0)
1427 return r;
1428 break;
1429
1430 case ARG_SYSTEM_CALL_FILTER: {
1431 bool negative;
1432 const char *items;
1433
1434 negative = optarg[0] == '~';
1435 items = negative ? optarg + 1 : optarg;
1436
1437 for (;;) {
1438 _cleanup_free_ char *word = NULL;
1439
1440 r = extract_first_word(&items, &word, NULL, 0);
1441 if (r == 0)
1442 break;
1443 if (r == -ENOMEM)
1444 return log_oom();
1445 if (r < 0)
1446 return log_error_errno(r, "Failed to parse system call filter: %m");
1447
1448 if (negative)
1449 r = strv_extend(&arg_syscall_deny_list, word);
1450 else
1451 r = strv_extend(&arg_syscall_allow_list, word);
1452 if (r < 0)
1453 return log_oom();
1454 }
1455
1456 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1457 break;
1458 }
1459
1460 case ARG_RLIMIT: {
1461 const char *eq;
1462 _cleanup_free_ char *name = NULL;
1463 int rl;
1464
1465 if (streq(optarg, "help")) {
1466 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1467 return 0;
1468 }
1469
1470 eq = strchr(optarg, '=');
1471 if (!eq)
1472 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1473 "--rlimit= expects an '=' assignment.");
1474
1475 name = strndup(optarg, eq - optarg);
1476 if (!name)
1477 return log_oom();
1478
1479 rl = rlimit_from_string_harder(name);
1480 if (rl < 0)
1481 return log_error_errno(rl, "Unknown resource limit: %s", name);
1482
1483 if (!arg_rlimit[rl]) {
1484 arg_rlimit[rl] = new0(struct rlimit, 1);
1485 if (!arg_rlimit[rl])
1486 return log_oom();
1487 }
1488
1489 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1490 if (r < 0)
1491 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1492
1493 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1494 break;
1495 }
1496
1497 case ARG_OOM_SCORE_ADJUST:
1498 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1499 if (r < 0)
1500 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1501
1502 arg_oom_score_adjust_set = true;
1503 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1504 break;
1505
1506 case ARG_CPU_AFFINITY: {
1507 CPUSet cpuset;
1508
1509 r = parse_cpu_set(optarg, &cpuset);
1510 if (r < 0)
1511 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
1512
1513 cpu_set_reset(&arg_cpu_set);
1514 arg_cpu_set = cpuset;
1515 arg_settings_mask |= SETTING_CPU_AFFINITY;
1516 break;
1517 }
1518
1519 case ARG_RESOLV_CONF:
1520 if (streq(optarg, "help")) {
1521 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1522 return 0;
1523 }
1524
1525 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1526 if (arg_resolv_conf < 0)
1527 return log_error_errno(arg_resolv_conf,
1528 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1529
1530 arg_settings_mask |= SETTING_RESOLV_CONF;
1531 break;
1532
1533 case ARG_TIMEZONE:
1534 if (streq(optarg, "help")) {
1535 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1536 return 0;
1537 }
1538
1539 arg_timezone = timezone_mode_from_string(optarg);
1540 if (arg_timezone < 0)
1541 return log_error_errno(arg_timezone,
1542 "Failed to parse /etc/localtime mode: %s", optarg);
1543
1544 arg_settings_mask |= SETTING_TIMEZONE;
1545 break;
1546
1547 case ARG_CONSOLE:
1548 r = handle_arg_console(optarg);
1549 if (r <= 0)
1550 return r;
1551 break;
1552
1553 case 'P':
1554 case ARG_PIPE:
1555 r = handle_arg_console("pipe");
1556 if (r <= 0)
1557 return r;
1558 break;
1559
1560 case ARG_NO_PAGER:
1561 arg_pager_flags |= PAGER_DISABLE;
1562 break;
1563
1564 case ARG_SET_CREDENTIAL:
1565 r = machine_credential_set(&arg_credentials, optarg);
1566 if (r < 0)
1567 return r;
1568
1569 arg_settings_mask |= SETTING_CREDENTIALS;
1570 break;
1571
1572 case ARG_LOAD_CREDENTIAL:
1573 r = machine_credential_load(&arg_credentials, optarg);
1574 if (r < 0)
1575 return r;
1576
1577 arg_settings_mask |= SETTING_CREDENTIALS;
1578 break;
1579
1580 case ARG_BIND_USER:
1581 if (!valid_user_group_name(optarg, 0))
1582 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid user name to bind: %s", optarg);
1583
1584 if (strv_extend(&arg_bind_user, optarg) < 0)
1585 return log_oom();
1586
1587 arg_settings_mask |= SETTING_BIND_USER;
1588 break;
1589
1590 case ARG_SUPPRESS_SYNC:
1591 r = parse_boolean_argument("--suppress-sync=", optarg, &arg_suppress_sync);
1592 if (r < 0)
1593 return r;
1594
1595 arg_settings_mask |= SETTING_SUPPRESS_SYNC;
1596 break;
1597
1598 case ARG_IMAGE_POLICY:
1599 r = parse_image_policy_argument(optarg, &arg_image_policy);
1600 if (r < 0)
1601 return r;
1602 break;
1603
1604 case ARG_BACKGROUND:
1605 r = free_and_strdup_warn(&arg_background, optarg);
1606 if (r < 0)
1607 return r;
1608 break;
1609
1610 case '?':
1611 return -EINVAL;
1612
1613 default:
1614 assert_not_reached();
1615 }
1616
1617 if (argc > optind) {
1618 strv_free(arg_parameters);
1619 arg_parameters = strv_copy(argv + optind);
1620 if (!arg_parameters)
1621 return log_oom();
1622
1623 arg_settings_mask |= SETTING_START_MODE;
1624 }
1625
1626 if (arg_ephemeral && arg_template && !arg_directory)
1627 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1628 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1629 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1630 * --directory=". */
1631 arg_directory = TAKE_PTR(arg_template);
1632
1633 arg_caps_retain |= plus;
1634 arg_caps_retain |= arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0;
1635 arg_caps_retain &= ~minus;
1636
1637 /* Make sure to parse environment before we reset the settings mask below */
1638 r = parse_environment();
1639 if (r < 0)
1640 return r;
1641
1642 /* Load all settings from .nspawn files */
1643 if (mask_no_settings)
1644 arg_settings_mask = 0;
1645
1646 /* Don't load any settings from .nspawn files */
1647 if (mask_all_settings)
1648 arg_settings_mask = _SETTINGS_MASK_ALL;
1649
1650 return 1;
1651 }
1652
1653 static int verify_arguments(void) {
1654 int r;
1655
1656 if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
1657 /* If we are running the stub init in the container, we don't need to look at what the init
1658 * in the container supports, because we are not using it. Let's immediately pick the right
1659 * setting based on the host system configuration.
1660 *
1661 * We only do this, if the user didn't use an environment variable to override the detection.
1662 */
1663
1664 r = cg_all_unified();
1665 if (r < 0)
1666 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
1667 if (r > 0)
1668 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
1669 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
1670 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
1671 else
1672 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
1673 }
1674
1675 if (arg_userns_mode != USER_NAMESPACE_NO)
1676 arg_mount_settings |= MOUNT_USE_USERNS;
1677
1678 if (arg_private_network)
1679 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1680
1681 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1682 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1683 arg_register = false;
1684 if (arg_start_mode != START_PID1)
1685 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1686 }
1687
1688 if (arg_userns_ownership < 0)
1689 arg_userns_ownership =
1690 arg_userns_mode == USER_NAMESPACE_PICK ? USER_NAMESPACE_OWNERSHIP_AUTO :
1691 USER_NAMESPACE_OWNERSHIP_OFF;
1692
1693 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1694 arg_kill_signal = SIGRTMIN+3;
1695
1696 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 */
1697 arg_read_only = true;
1698
1699 if (has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts))
1700 arg_read_only = true;
1701
1702 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1703 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1704 * The latter is not technically a user session, but we don't need to labour the point. */
1705 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1706
1707 if (arg_directory && arg_image)
1708 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1709
1710 if (arg_template && arg_image)
1711 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1712
1713 if (arg_template && !(arg_directory || arg_machine))
1714 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1715
1716 if (arg_ephemeral && arg_template)
1717 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1718
1719 /* Permit --ephemeral with --link-journal=try-* to satisfy principle of the least astonishment
1720 * (by common sense, "try" means "do not fail if not possible") */
1721 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO) && !arg_link_journal_try)
1722 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal={host,guest} may not be combined.");
1723
1724 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1725 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1726
1727 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_read_only)
1728 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1729 "--read-only and --private-users-ownership=chown may not be combined.");
1730
1731 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1732 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1733 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1734 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_volatile_mode != VOLATILE_NO)
1735 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-ownership=chown may not be combined.");
1736
1737 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1738 * we need to error out, to avoid conflicts between different network options. */
1739 if (arg_network_namespace_path &&
1740 (arg_network_interfaces || arg_network_macvlan ||
1741 arg_network_ipvlan || arg_network_veth_extra ||
1742 arg_network_bridge || arg_network_zone ||
1743 arg_network_veth))
1744 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1745
1746 if (arg_network_bridge && arg_network_zone)
1747 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1748 "--network-bridge= and --network-zone= may not be combined.");
1749
1750 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1751 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1752
1753 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1754 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1755
1756 if (arg_expose_ports && !arg_private_network)
1757 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1758
1759 if (arg_caps_ambient) {
1760 if (arg_caps_ambient == UINT64_MAX)
1761 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= does not support the value all.");
1762
1763 if ((arg_caps_ambient & arg_caps_retain) != arg_caps_ambient)
1764 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not fully covered by Capability= setting.");
1765
1766 if (arg_start_mode == START_BOOT)
1767 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not useful for boot mode.");
1768 }
1769
1770 if (arg_userns_mode == USER_NAMESPACE_NO && !strv_isempty(arg_bind_user))
1771 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--bind-user= requires --private-users");
1772
1773 /* Drop duplicate --bind-user= entries */
1774 strv_uniq(arg_bind_user);
1775
1776 r = custom_mount_check_all();
1777 if (r < 0)
1778 return r;
1779
1780 return 0;
1781 }
1782
1783 static int verify_network_interfaces_initialized(void) {
1784 int r;
1785 r = test_network_interfaces_initialized(arg_network_interfaces);
1786 if (r < 0)
1787 return r;
1788
1789 r = test_network_interfaces_initialized(arg_network_macvlan);
1790 if (r < 0)
1791 return r;
1792
1793 r = test_network_interfaces_initialized(arg_network_ipvlan);
1794 if (r < 0)
1795 return r;
1796
1797 return 0;
1798 }
1799
1800 int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1801 assert(p);
1802
1803 if (arg_userns_mode == USER_NAMESPACE_NO)
1804 return 0;
1805
1806 if (uid == UID_INVALID && gid == GID_INVALID)
1807 return 0;
1808
1809 if (uid != UID_INVALID) {
1810 uid += arg_uid_shift;
1811
1812 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1813 return -EOVERFLOW;
1814 }
1815
1816 if (gid != GID_INVALID) {
1817 gid += (gid_t) arg_uid_shift;
1818
1819 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1820 return -EOVERFLOW;
1821 }
1822
1823 return RET_NERRNO(lchown(p, uid, gid));
1824 }
1825
1826 int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1827 const char *q;
1828 int r;
1829
1830 q = prefix_roota(root, path);
1831 r = RET_NERRNO(mkdir(q, mode));
1832 if (r == -EEXIST)
1833 return 0;
1834 if (r < 0)
1835 return r;
1836
1837 return userns_lchown(q, uid, gid);
1838 }
1839
1840 static const char *timezone_from_path(const char *path) {
1841 return PATH_STARTSWITH_SET(
1842 path,
1843 "../usr/share/zoneinfo/",
1844 "/usr/share/zoneinfo/");
1845 }
1846
1847 static bool etc_writable(void) {
1848 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1849 }
1850
1851 static int setup_timezone(const char *dest) {
1852 _cleanup_free_ char *p = NULL, *etc = NULL;
1853 const char *where, *check;
1854 TimezoneMode m;
1855 int r;
1856
1857 assert(dest);
1858
1859 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1860 r = readlink_malloc("/etc/localtime", &p);
1861 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1862 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1863 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1864 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1865 else if (r < 0) {
1866 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1867 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1868 * file.
1869 *
1870 * Example:
1871 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1872 */
1873 return 0;
1874 } else if (arg_timezone == TIMEZONE_AUTO)
1875 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1876 else
1877 m = arg_timezone;
1878 } else
1879 m = arg_timezone;
1880
1881 if (m == TIMEZONE_OFF)
1882 return 0;
1883
1884 r = chase("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1885 if (r < 0) {
1886 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1887 return 0;
1888 }
1889
1890 where = strjoina(etc, "/localtime");
1891
1892 switch (m) {
1893
1894 case TIMEZONE_DELETE:
1895 if (unlink(where) < 0)
1896 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1897
1898 return 0;
1899
1900 case TIMEZONE_SYMLINK: {
1901 _cleanup_free_ char *q = NULL;
1902 const char *z, *what;
1903
1904 z = timezone_from_path(p);
1905 if (!z) {
1906 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1907 return 0;
1908 }
1909
1910 r = readlink_malloc(where, &q);
1911 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1912 return 0; /* Already pointing to the right place? Then do nothing .. */
1913
1914 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1915 r = chase(check, dest, 0, NULL, NULL);
1916 if (r < 0)
1917 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1918 else {
1919 if (unlink(where) < 0 && errno != ENOENT) {
1920 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1921 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1922 return 0;
1923 }
1924
1925 what = strjoina("../usr/share/zoneinfo/", z);
1926 if (symlink(what, where) < 0) {
1927 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1928 errno, "Failed to correct timezone of container, ignoring: %m");
1929 return 0;
1930 }
1931
1932 break;
1933 }
1934
1935 _fallthrough_;
1936 }
1937
1938 case TIMEZONE_BIND: {
1939 _cleanup_free_ char *resolved = NULL;
1940 int found;
1941
1942 found = chase(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
1943 if (found < 0) {
1944 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1945 return 0;
1946 }
1947
1948 if (found == 0) /* missing? */
1949 (void) touch(resolved);
1950
1951 r = mount_nofollow_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1952 if (r >= 0)
1953 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1954
1955 _fallthrough_;
1956 }
1957
1958 case TIMEZONE_COPY:
1959 /* If mounting failed, try to copy */
1960 r = copy_file_atomic("/etc/localtime", where, 0644, COPY_REFLINK|COPY_REPLACE);
1961 if (r < 0) {
1962 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1963 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
1964 return 0;
1965 }
1966
1967 break;
1968
1969 default:
1970 assert_not_reached();
1971 }
1972
1973 /* Fix permissions of the symlink or file copy we just created */
1974 r = userns_lchown(where, 0, 0);
1975 if (r < 0)
1976 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
1977
1978 return 0;
1979 }
1980
1981 static int have_resolv_conf(const char *path) {
1982 assert(path);
1983
1984 if (access(path, F_OK) < 0) {
1985 if (errno == ENOENT)
1986 return 0;
1987
1988 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1989 }
1990
1991 return 1;
1992 }
1993
1994 static int resolved_listening(void) {
1995 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1996 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1997 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1998 int r;
1999
2000 /* Check if resolved is listening */
2001
2002 r = sd_bus_open_system(&bus);
2003 if (r < 0)
2004 return log_debug_errno(r, "Failed to open system bus: %m");
2005
2006 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
2007 if (r < 0)
2008 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2009 if (r == 0)
2010 return 0;
2011
2012 r = bus_get_property_string(bus, bus_resolve_mgr, "DNSStubListener", &error, &dns_stub_listener_mode);
2013 if (r < 0)
2014 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
2015
2016 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
2017 }
2018
2019 static int setup_resolv_conf(const char *dest) {
2020 _cleanup_free_ char *etc = NULL;
2021 const char *where, *what;
2022 ResolvConfMode m;
2023 int r;
2024
2025 assert(dest);
2026
2027 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
2028 if (arg_private_network)
2029 m = RESOLV_CONF_OFF;
2030 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF) > 0 && resolved_listening() > 0)
2031 m = etc_writable() ? RESOLV_CONF_COPY_STUB : RESOLV_CONF_BIND_STUB;
2032 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2033 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
2034 else
2035 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
2036
2037 } else
2038 m = arg_resolv_conf;
2039
2040 if (m == RESOLV_CONF_OFF)
2041 return 0;
2042
2043 r = chase("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
2044 if (r < 0) {
2045 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
2046 return 0;
2047 }
2048
2049 where = strjoina(etc, "/resolv.conf");
2050
2051 if (m == RESOLV_CONF_DELETE) {
2052 if (unlink(where) < 0)
2053 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
2054
2055 return 0;
2056 }
2057
2058 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_COPY_STATIC))
2059 what = PRIVATE_STATIC_RESOLV_CONF;
2060 else if (IN_SET(m, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_COPY_UPLINK))
2061 what = PRIVATE_UPLINK_RESOLV_CONF;
2062 else if (IN_SET(m, RESOLV_CONF_BIND_STUB, RESOLV_CONF_REPLACE_STUB, RESOLV_CONF_COPY_STUB))
2063 what = PRIVATE_STUB_RESOLV_CONF;
2064 else
2065 what = "/etc/resolv.conf";
2066
2067 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_BIND_STUB)) {
2068 _cleanup_free_ char *resolved = NULL;
2069 int found;
2070
2071 found = chase(where, dest, CHASE_NONEXISTENT|CHASE_NOFOLLOW, &resolved, NULL);
2072 if (found < 0) {
2073 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2074 return 0;
2075 }
2076
2077 if (found == 0) /* missing? */
2078 (void) touch(resolved);
2079
2080 r = mount_nofollow_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
2081 if (r >= 0)
2082 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
2083
2084 /* If that didn't work, let's copy the file */
2085 }
2086
2087 if (IN_SET(m, RESOLV_CONF_REPLACE_HOST, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_REPLACE_STUB))
2088 r = copy_file_atomic(what, where, 0644, COPY_REFLINK|COPY_REPLACE);
2089 else
2090 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, COPY_REFLINK);
2091 if (r < 0) {
2092 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2093 * resolved or something similar runs inside and the symlink points there.
2094 *
2095 * If the disk image is read-only, there's also no point in complaining.
2096 */
2097 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC, RESOLV_CONF_COPY_UPLINK, RESOLV_CONF_COPY_STUB) &&
2098 IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
2099 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
2100 return 0;
2101 }
2102
2103 r = userns_lchown(where, 0, 0);
2104 if (r < 0)
2105 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
2106
2107 return 0;
2108 }
2109
2110 static int setup_boot_id(void) {
2111 _cleanup_(unlink_and_freep) char *from = NULL;
2112 _cleanup_free_ char *path = NULL;
2113 sd_id128_t rnd = SD_ID128_NULL;
2114 const char *to;
2115 int r;
2116
2117 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2118
2119 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
2120 if (r < 0)
2121 return log_error_errno(r, "Failed to generate random boot ID path: %m");
2122
2123 r = sd_id128_randomize(&rnd);
2124 if (r < 0)
2125 return log_error_errno(r, "Failed to generate random boot id: %m");
2126
2127 r = id128_write(path, ID128_FORMAT_UUID, rnd);
2128 if (r < 0)
2129 return log_error_errno(r, "Failed to write boot id: %m");
2130
2131 from = TAKE_PTR(path);
2132 to = "/proc/sys/kernel/random/boot_id";
2133
2134 r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
2135 if (r < 0)
2136 return r;
2137
2138 return mount_nofollow_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2139 }
2140
2141 static int copy_devnodes(const char *dest) {
2142 static const char devnodes[] =
2143 "null\0"
2144 "zero\0"
2145 "full\0"
2146 "random\0"
2147 "urandom\0"
2148 "tty\0"
2149 "net/tun\0";
2150
2151 int r = 0;
2152
2153 assert(dest);
2154
2155 BLOCK_WITH_UMASK(0000);
2156
2157 /* Create /dev/net, so that we can create /dev/net/tun in it */
2158 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
2159 return log_error_errno(r, "Failed to create /dev/net directory: %m");
2160
2161 NULSTR_FOREACH(d, devnodes) {
2162 _cleanup_free_ char *from = NULL, *to = NULL;
2163 struct stat st;
2164
2165 from = path_join("/dev/", d);
2166 if (!from)
2167 return log_oom();
2168
2169 to = path_join(dest, from);
2170 if (!to)
2171 return log_oom();
2172
2173 if (stat(from, &st) < 0) {
2174
2175 if (errno != ENOENT)
2176 return log_error_errno(errno, "Failed to stat %s: %m", from);
2177
2178 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
2179 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2180 "%s is not a char or block device, cannot copy.", from);
2181 else {
2182 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
2183
2184 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
2185 /* Explicitly warn the user when /dev is already populated. */
2186 if (errno == EEXIST)
2187 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);
2188 if (errno != EPERM)
2189 return log_error_errno(errno, "mknod(%s) failed: %m", to);
2190
2191 /* Some systems abusively restrict mknod but allow bind mounts. */
2192 r = touch(to);
2193 if (r < 0)
2194 return log_error_errno(r, "touch (%s) failed: %m", to);
2195 r = mount_nofollow_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
2196 if (r < 0)
2197 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
2198 }
2199
2200 r = userns_lchown(to, 0, 0);
2201 if (r < 0)
2202 return log_error_errno(r, "chown() of device node %s failed: %m", to);
2203
2204 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
2205 if (!dn)
2206 return log_oom();
2207
2208 r = userns_mkdir(dest, dn, 0755, 0, 0);
2209 if (r < 0)
2210 return log_error_errno(r, "Failed to create '%s': %m", dn);
2211
2212 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
2213 return log_oom();
2214
2215 prefixed = path_join(dest, sl);
2216 if (!prefixed)
2217 return log_oom();
2218
2219 t = path_join("..", d);
2220 if (!t)
2221 return log_oom();
2222
2223 if (symlink(t, prefixed) < 0)
2224 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
2225 }
2226 }
2227
2228 return r;
2229 }
2230
2231 static int make_extra_nodes(const char *dest) {
2232 size_t i;
2233 int r;
2234
2235 BLOCK_WITH_UMASK(0000);
2236
2237 for (i = 0; i < arg_n_extra_nodes; i++) {
2238 _cleanup_free_ char *path = NULL;
2239 DeviceNode *n = arg_extra_nodes + i;
2240
2241 path = path_join(dest, n->path);
2242 if (!path)
2243 return log_oom();
2244
2245 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
2246 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
2247
2248 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
2249 if (r < 0)
2250 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
2251 }
2252
2253 return 0;
2254 }
2255
2256 static int setup_pts(const char *dest) {
2257 _cleanup_free_ char *options = NULL;
2258 const char *p;
2259 int r;
2260
2261 #if HAVE_SELINUX
2262 if (arg_selinux_apifs_context)
2263 (void) asprintf(&options,
2264 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
2265 arg_uid_shift + TTY_GID,
2266 arg_selinux_apifs_context);
2267 else
2268 #endif
2269 (void) asprintf(&options,
2270 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
2271 arg_uid_shift + TTY_GID);
2272
2273 if (!options)
2274 return log_oom();
2275
2276 /* Mount /dev/pts itself */
2277 p = prefix_roota(dest, "/dev/pts");
2278 r = RET_NERRNO(mkdir(p, 0755));
2279 if (r < 0)
2280 return log_error_errno(r, "Failed to create /dev/pts: %m");
2281
2282 r = mount_nofollow_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2283 if (r < 0)
2284 return r;
2285 r = userns_lchown(p, 0, 0);
2286 if (r < 0)
2287 return log_error_errno(r, "Failed to chown /dev/pts: %m");
2288
2289 /* Create /dev/ptmx symlink */
2290 p = prefix_roota(dest, "/dev/ptmx");
2291 if (symlink("pts/ptmx", p) < 0)
2292 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
2293 r = userns_lchown(p, 0, 0);
2294 if (r < 0)
2295 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
2296
2297 /* And fix /dev/pts/ptmx ownership */
2298 p = prefix_roota(dest, "/dev/pts/ptmx");
2299 r = userns_lchown(p, 0, 0);
2300 if (r < 0)
2301 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
2302
2303 return 0;
2304 }
2305
2306 static int setup_stdio_as_dev_console(void) {
2307 _cleanup_close_ int terminal = -EBADF;
2308 int r;
2309
2310 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2311 * explicitly, if we are configured to. */
2312 terminal = open_terminal("/dev/console", O_RDWR|O_NOCTTY);
2313 if (terminal < 0)
2314 return log_error_errno(terminal, "Failed to open console: %m");
2315
2316 /* Make sure we can continue logging to the original stderr, even if
2317 * stderr points elsewhere now */
2318 r = log_dup_console();
2319 if (r < 0)
2320 return log_error_errno(r, "Failed to duplicate stderr: %m");
2321
2322 /* invalidates 'terminal' on success and failure */
2323 r = rearrange_stdio(terminal, terminal, terminal);
2324 TAKE_FD(terminal);
2325 if (r < 0)
2326 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2327
2328 return 0;
2329 }
2330
2331 static int setup_dev_console(const char *console) {
2332 _cleanup_free_ char *p = NULL;
2333 int r;
2334
2335 /* Create /dev/console symlink */
2336 r = path_make_relative("/dev", console, &p);
2337 if (r < 0)
2338 return log_error_errno(r, "Failed to create relative path: %m");
2339
2340 if (symlink(p, "/dev/console") < 0)
2341 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
2342
2343 return 0;
2344 }
2345
2346 static int setup_keyring(void) {
2347 key_serial_t keyring;
2348
2349 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2350 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2351 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2352 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2353 * into the container. */
2354
2355 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2356 if (keyring == -1) {
2357 if (errno == ENOSYS)
2358 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2359 else if (ERRNO_IS_PRIVILEGE(errno))
2360 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2361 else
2362 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2363 }
2364
2365 return 0;
2366 }
2367
2368 int make_run_host(const char *root) {
2369 int r;
2370
2371 assert(root);
2372
2373 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2374 if (r < 0)
2375 return log_error_errno(r, "Failed to create /run/host/: %m");
2376
2377 return 0;
2378 }
2379
2380 static int setup_credentials(const char *root) {
2381 const char *q;
2382 int r;
2383
2384 if (arg_credentials.n_credentials == 0)
2385 return 0;
2386
2387 r = make_run_host(root);
2388 if (r < 0)
2389 return r;
2390
2391 r = userns_mkdir(root, "/run/host/credentials", 0700, 0, 0);
2392 if (r < 0)
2393 return log_error_errno(r, "Failed to create /run/host/credentials: %m");
2394
2395 q = prefix_roota(root, "/run/host/credentials");
2396 r = mount_nofollow_verbose(LOG_ERR, NULL, q, "ramfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0700");
2397 if (r < 0)
2398 return r;
2399
2400 FOREACH_ARRAY(cred, arg_credentials.credentials, arg_credentials.n_credentials) {
2401 _cleanup_free_ char *j = NULL;
2402 _cleanup_close_ int fd = -EBADF;
2403
2404 j = path_join(q, cred->id);
2405 if (!j)
2406 return log_oom();
2407
2408 fd = open(j, O_CREAT|O_EXCL|O_WRONLY|O_CLOEXEC|O_NOFOLLOW, 0600);
2409 if (fd < 0)
2410 return log_error_errno(errno, "Failed to create credential file %s: %m", j);
2411
2412 r = loop_write(fd, cred->data, cred->size);
2413 if (r < 0)
2414 return log_error_errno(r, "Failed to write credential to file %s: %m", j);
2415
2416 if (fchmod(fd, 0400) < 0)
2417 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", j);
2418
2419 if (arg_userns_mode != USER_NAMESPACE_NO) {
2420 if (fchown(fd, arg_uid_shift, arg_uid_shift) < 0)
2421 return log_error_errno(errno, "Failed to adjust ownership of %s: %m", j);
2422 }
2423 }
2424
2425 if (chmod(q, 0500) < 0)
2426 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", q);
2427
2428 r = userns_lchown(q, 0, 0);
2429 if (r < 0)
2430 return r;
2431
2432 /* Make both mount and superblock read-only now */
2433 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2434 if (r < 0)
2435 return r;
2436
2437 return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0500");
2438 }
2439
2440 static int setup_kmsg(int fd_inner_socket) {
2441 _cleanup_(unlink_and_freep) char *from = NULL;
2442 _cleanup_free_ char *fifo = NULL;
2443 _cleanup_close_ int fd = -EBADF;
2444 int r;
2445
2446 assert(fd_inner_socket >= 0);
2447
2448 BLOCK_WITH_UMASK(0000);
2449
2450 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2451 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2452 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2453 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2454
2455 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2456 if (r < 0)
2457 return log_error_errno(r, "Failed to generate kmsg path: %m");
2458
2459 if (mkfifo(fifo, 0600) < 0)
2460 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2461
2462 from = TAKE_PTR(fifo);
2463
2464 r = mount_nofollow_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2465 if (r < 0)
2466 return r;
2467
2468 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2469 if (fd < 0)
2470 return log_error_errno(errno, "Failed to open fifo: %m");
2471
2472 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2473 r = send_one_fd(fd_inner_socket, fd, 0);
2474 if (r < 0)
2475 return log_error_errno(r, "Failed to send FIFO fd: %m");
2476
2477 return 0;
2478 }
2479
2480 struct ExposeArgs {
2481 union in_addr_union address4;
2482 union in_addr_union address6;
2483 struct FirewallContext *fw_ctx;
2484 };
2485
2486 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2487 struct ExposeArgs *args = ASSERT_PTR(userdata);
2488
2489 assert(rtnl);
2490 assert(m);
2491
2492 (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET, &args->address4);
2493 (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET6, &args->address6);
2494 return 0;
2495 }
2496
2497 static int setup_hostname(void) {
2498 int r;
2499
2500 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2501 return 0;
2502
2503 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2504 if (r < 0)
2505 return log_error_errno(r, "Failed to set hostname: %m");
2506
2507 return 0;
2508 }
2509
2510 static int setup_journal(const char *directory) {
2511 _cleanup_free_ char *d = NULL;
2512 const char *p, *q;
2513 sd_id128_t this_id;
2514 bool try;
2515 int r;
2516
2517 /* Don't link journals in ephemeral mode */
2518 if (arg_ephemeral)
2519 return 0;
2520
2521 if (arg_link_journal == LINK_NO)
2522 return 0;
2523
2524 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2525
2526 r = sd_id128_get_machine(&this_id);
2527 if (r < 0)
2528 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2529
2530 if (sd_id128_equal(arg_uuid, this_id)) {
2531 log_full(try ? LOG_WARNING : LOG_ERR,
2532 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid));
2533 if (try)
2534 return 0;
2535 return -EEXIST;
2536 }
2537
2538 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2539 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2540 if (r < 0) {
2541 bool ignore = r == -EROFS && try;
2542 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2543 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2544 return ignore ? 0 : r;
2545 }
2546 }
2547
2548 p = strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid));
2549 q = prefix_roota(directory, p);
2550
2551 if (path_is_mount_point(p) > 0) {
2552 if (try)
2553 return 0;
2554
2555 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2556 "%s: already a mount point, refusing to use for journal", p);
2557 }
2558
2559 if (path_is_mount_point(q) > 0) {
2560 if (try)
2561 return 0;
2562
2563 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2564 "%s: already a mount point, refusing to use for journal", q);
2565 }
2566
2567 r = readlink_and_make_absolute(p, &d);
2568 if (r >= 0) {
2569 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2570 path_equal(d, q)) {
2571
2572 r = userns_mkdir(directory, p, 0755, 0, 0);
2573 if (r < 0)
2574 log_warning_errno(r, "Failed to create directory %s: %m", q);
2575 return 0;
2576 }
2577
2578 if (unlink(p) < 0)
2579 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2580 } else if (r == -EINVAL) {
2581
2582 if (arg_link_journal == LINK_GUEST &&
2583 rmdir(p) < 0) {
2584
2585 if (errno == ENOTDIR) {
2586 log_error("%s already exists and is neither a symlink nor a directory", p);
2587 return r;
2588 } else
2589 return log_error_errno(errno, "Failed to remove %s: %m", p);
2590 }
2591 } else if (r != -ENOENT)
2592 return log_error_errno(r, "readlink(%s) failed: %m", p);
2593
2594 if (arg_link_journal == LINK_GUEST) {
2595
2596 if (symlink(q, p) < 0) {
2597 if (try) {
2598 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2599 return 0;
2600 } else
2601 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2602 }
2603
2604 r = userns_mkdir(directory, p, 0755, 0, 0);
2605 if (r < 0)
2606 log_warning_errno(r, "Failed to create directory %s: %m", q);
2607 return 0;
2608 }
2609
2610 if (arg_link_journal == LINK_HOST) {
2611 /* don't create parents here — if the host doesn't have
2612 * permanent journal set up, don't force it here */
2613
2614 r = RET_NERRNO(mkdir(p, 0755));
2615 if (r < 0 && r != -EEXIST) {
2616 if (try) {
2617 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2618 return 0;
2619 } else
2620 return log_error_errno(r, "Failed to create %s: %m", p);
2621 }
2622
2623 } else if (access(p, F_OK) < 0)
2624 return 0;
2625
2626 if (dir_is_empty(q, /* ignore_hidden_or_backup= */ false) == 0)
2627 log_warning("%s is not empty, proceeding anyway.", q);
2628
2629 r = userns_mkdir(directory, p, 0755, 0, 0);
2630 if (r < 0)
2631 return log_error_errno(r, "Failed to create %s: %m", q);
2632
2633 r = mount_nofollow_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2634 if (r < 0)
2635 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2636
2637 return 0;
2638 }
2639
2640 static int drop_capabilities(uid_t uid) {
2641 CapabilityQuintet q;
2642
2643 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2644 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2645 * arg_caps_retain. */
2646
2647 if (capability_quintet_is_set(&arg_full_capabilities)) {
2648 q = arg_full_capabilities;
2649
2650 if (q.bounding == UINT64_MAX)
2651 q.bounding = uid == 0 ? arg_caps_retain : 0;
2652
2653 if (q.effective == UINT64_MAX)
2654 q.effective = uid == 0 ? q.bounding : 0;
2655
2656 if (q.inheritable == UINT64_MAX)
2657 q.inheritable = uid == 0 ? q.bounding : arg_caps_ambient;
2658
2659 if (q.permitted == UINT64_MAX)
2660 q.permitted = uid == 0 ? q.bounding : arg_caps_ambient;
2661
2662 if (q.ambient == UINT64_MAX && ambient_capabilities_supported())
2663 q.ambient = arg_caps_ambient;
2664
2665 if (capability_quintet_mangle(&q))
2666 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2667
2668 } else {
2669 q = (CapabilityQuintet) {
2670 .bounding = arg_caps_retain,
2671 .effective = uid == 0 ? arg_caps_retain : 0,
2672 .inheritable = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2673 .permitted = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2674 .ambient = ambient_capabilities_supported() ? arg_caps_ambient : UINT64_MAX,
2675 };
2676
2677 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2678 * in order to maintain the same behavior as systemd < 242. */
2679 if (capability_quintet_mangle(&q))
2680 log_full(arg_quiet ? LOG_DEBUG : LOG_WARNING,
2681 "Some capabilities will not be set because they are not in the current bounding set.");
2682
2683 }
2684
2685 return capability_quintet_enforce(&q);
2686 }
2687
2688 static int reset_audit_loginuid(void) {
2689 _cleanup_free_ char *p = NULL;
2690 int r;
2691
2692 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2693 return 0;
2694
2695 r = read_one_line_file("/proc/self/loginuid", &p);
2696 if (r == -ENOENT)
2697 return 0;
2698 if (r < 0)
2699 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2700
2701 /* Already reset? */
2702 if (streq(p, "4294967295"))
2703 return 0;
2704
2705 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2706 if (r < 0) {
2707 log_error_errno(r,
2708 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2709 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2710 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2711 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2712 "using systemd-nspawn. Sleeping for 5s... (%m)");
2713
2714 sleep(5);
2715 }
2716
2717 return 0;
2718 }
2719
2720 static int mount_tunnel_dig(const char *root) {
2721 const char *p, *q;
2722 int r;
2723
2724 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2725 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2726 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2727 (void) mkdir_p(p, 0600);
2728
2729 r = make_run_host(root);
2730 if (r < 0)
2731 return r;
2732
2733 r = userns_mkdir(root, NSPAWN_MOUNT_TUNNEL, 0600, 0, 0);
2734 if (r < 0)
2735 return log_error_errno(r, "Failed to create "NSPAWN_MOUNT_TUNNEL": %m");
2736
2737 q = prefix_roota(root, NSPAWN_MOUNT_TUNNEL);
2738 r = mount_nofollow_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2739 if (r < 0)
2740 return r;
2741
2742 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2743 if (r < 0)
2744 return r;
2745
2746 return 0;
2747 }
2748
2749 static int mount_tunnel_open(void) {
2750 int r;
2751
2752 r = mount_follow_verbose(LOG_ERR, NULL, NSPAWN_MOUNT_TUNNEL, NULL, MS_SLAVE, NULL);
2753 if (r < 0)
2754 return r;
2755
2756 return 0;
2757 }
2758
2759 static int setup_machine_id(const char *directory) {
2760 int r;
2761
2762 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2763 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2764 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2765 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2766 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2767 * container behaves nicely). */
2768
2769 r = id128_get_machine(directory, &arg_uuid);
2770 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r)) {
2771 /* If the file is missing, empty, or uninitialized, we don't mind */
2772 if (sd_id128_is_null(arg_uuid)) {
2773 r = sd_id128_randomize(&arg_uuid);
2774 if (r < 0)
2775 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2776 }
2777 } else if (r < 0)
2778 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2779
2780 return 0;
2781 }
2782
2783 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2784 int r;
2785
2786 assert(directory);
2787
2788 if (arg_userns_mode == USER_NAMESPACE_NO || arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_CHOWN)
2789 return 0;
2790
2791 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2792 if (r == -EOPNOTSUPP)
2793 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.");
2794 if (r == -EBADE)
2795 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2796 if (r < 0)
2797 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2798 if (r == 0)
2799 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2800 else
2801 log_debug("Patched directory tree to match UID/GID range.");
2802
2803 return r;
2804 }
2805
2806 /*
2807 * Return values:
2808 * < 0 : wait_for_terminate() failed to get the state of the
2809 * container, the container was terminated by a signal, or
2810 * failed for an unknown reason. No change is made to the
2811 * container argument.
2812 * > 0 : The program executed in the container terminated with an
2813 * error. The exit code of the program executed in the
2814 * container is returned. The container argument has been set
2815 * to CONTAINER_TERMINATED.
2816 * 0 : The container is being rebooted, has been shut down or exited
2817 * successfully. The container argument has been set to either
2818 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2819 *
2820 * That is, success is indicated by a return value of zero, and an
2821 * error is indicated by a non-zero value.
2822 */
2823 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2824 siginfo_t status;
2825 int r;
2826
2827 r = wait_for_terminate(pid, &status);
2828 if (r < 0)
2829 return log_warning_errno(r, "Failed to wait for container: %m");
2830
2831 switch (status.si_code) {
2832
2833 case CLD_EXITED:
2834 if (status.si_status == 0)
2835 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2836 else
2837 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2838
2839 *container = CONTAINER_TERMINATED;
2840 return status.si_status;
2841
2842 case CLD_KILLED:
2843 if (status.si_status == SIGINT) {
2844 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2845 *container = CONTAINER_TERMINATED;
2846 return 0;
2847
2848 } else if (status.si_status == SIGHUP) {
2849 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2850 *container = CONTAINER_REBOOTED;
2851 return 0;
2852 }
2853
2854 _fallthrough_;
2855 case CLD_DUMPED:
2856 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2857 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2858
2859 default:
2860 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2861 "Container %s failed due to unknown reason.", arg_machine);
2862 }
2863 }
2864
2865 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2866 pid_t pid;
2867
2868 pid = PTR_TO_PID(userdata);
2869 if (pid > 0) {
2870 if (kill(pid, arg_kill_signal) >= 0) {
2871 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2872 sd_event_source_set_userdata(s, NULL);
2873 return 0;
2874 }
2875 }
2876
2877 sd_event_exit(sd_event_source_get_event(s), 0);
2878 return 0;
2879 }
2880
2881 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2882 pid_t pid;
2883
2884 assert(s);
2885 assert(ssi);
2886
2887 pid = PTR_TO_PID(userdata);
2888
2889 for (;;) {
2890 siginfo_t si = {};
2891
2892 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2893 return log_error_errno(errno, "Failed to waitid(): %m");
2894 if (si.si_pid == 0) /* No pending children. */
2895 break;
2896 if (si.si_pid == pid) {
2897 /* The main process we care for has exited. Return from
2898 * signal handler but leave the zombie. */
2899 sd_event_exit(sd_event_source_get_event(s), 0);
2900 break;
2901 }
2902
2903 /* Reap all other children. */
2904 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2905 }
2906
2907 return 0;
2908 }
2909
2910 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2911 pid_t pid;
2912
2913 assert(m);
2914
2915 pid = PTR_TO_PID(userdata);
2916
2917 if (arg_kill_signal > 0) {
2918 log_info("Container termination requested. Attempting to halt container.");
2919 (void) kill(pid, arg_kill_signal);
2920 } else {
2921 log_info("Container termination requested. Exiting.");
2922 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2923 }
2924
2925 return 0;
2926 }
2927
2928 static int pick_paths(void) {
2929 int r;
2930
2931 if (arg_directory) {
2932 _cleanup_(pick_result_done) PickResult result = PICK_RESULT_NULL;
2933 PickFilter filter = pick_filter_image_dir;
2934
2935 filter.architecture = arg_architecture;
2936
2937 r = path_pick_update_warn(
2938 &arg_directory,
2939 &filter,
2940 PICK_ARCHITECTURE|PICK_TRIES,
2941 &result);
2942 if (r < 0) {
2943 /* Accept ENOENT here so that the --template= logic can work */
2944 if (r != -ENOENT)
2945 return r;
2946 } else
2947 arg_architecture = result.architecture;
2948 }
2949
2950 if (arg_image) {
2951 _cleanup_(pick_result_done) PickResult result = PICK_RESULT_NULL;
2952 PickFilter filter = pick_filter_image_raw;
2953
2954 filter.architecture = arg_architecture;
2955
2956 r = path_pick_update_warn(
2957 &arg_image,
2958 &filter,
2959 PICK_ARCHITECTURE|PICK_TRIES,
2960 &result);
2961 if (r < 0)
2962 return r;
2963
2964 arg_architecture = result.architecture;
2965 }
2966
2967 if (arg_template) {
2968 _cleanup_(pick_result_done) PickResult result = PICK_RESULT_NULL;
2969 PickFilter filter = pick_filter_image_dir;
2970
2971 filter.architecture = arg_architecture;
2972
2973 r = path_pick_update_warn(
2974 &arg_template,
2975 &filter,
2976 PICK_ARCHITECTURE,
2977 &result);
2978 if (r < 0)
2979 return r;
2980
2981 arg_architecture = result.architecture;
2982 }
2983
2984 return 0;
2985 }
2986
2987 static int determine_names(void) {
2988 int r;
2989
2990 if (arg_template && !arg_directory && arg_machine) {
2991
2992 /* If --template= was specified then we should not search for a machine, but instead create a
2993 * new one in /var/lib/machine. */
2994
2995 arg_directory = path_join("/var/lib/machines", arg_machine);
2996 if (!arg_directory)
2997 return log_oom();
2998 }
2999
3000 if (!arg_image && !arg_directory) {
3001 if (arg_machine) {
3002 _cleanup_(image_unrefp) Image *i = NULL;
3003
3004 r = image_find(IMAGE_MACHINE, arg_machine, NULL, &i);
3005 if (r == -ENOENT)
3006 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
3007 if (r < 0)
3008 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
3009
3010 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
3011 r = free_and_strdup(&arg_image, i->path);
3012 else
3013 r = free_and_strdup(&arg_directory, i->path);
3014 if (r < 0)
3015 return log_oom();
3016
3017 if (!arg_ephemeral)
3018 arg_read_only = arg_read_only || i->read_only;
3019 } else {
3020 r = safe_getcwd(&arg_directory);
3021 if (r < 0)
3022 return log_error_errno(r, "Failed to determine current directory: %m");
3023 }
3024
3025 if (!arg_directory && !arg_image)
3026 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
3027 }
3028
3029 if (!arg_machine) {
3030 if (arg_directory && path_equal(arg_directory, "/")) {
3031 arg_machine = gethostname_malloc();
3032 if (!arg_machine)
3033 return log_oom();
3034 } else if (arg_image) {
3035 char *e;
3036
3037 r = path_extract_filename(arg_image, &arg_machine);
3038 if (r < 0)
3039 return log_error_errno(r, "Failed to extract file name from '%s': %m", arg_image);
3040
3041 /* Truncate suffix if there is one */
3042 e = endswith(arg_machine, ".raw");
3043 if (e)
3044 *e = 0;
3045 } else {
3046 r = path_extract_filename(arg_directory, &arg_machine);
3047 if (r < 0)
3048 return log_error_errno(r, "Failed to extract file name from '%s': %m", arg_directory);
3049 }
3050
3051 hostname_cleanup(arg_machine);
3052 if (!hostname_is_valid(arg_machine, 0))
3053 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
3054
3055 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3056 * to match fixed config file names. */
3057 arg_settings_filename = strjoin(arg_machine, ".nspawn");
3058 if (!arg_settings_filename)
3059 return log_oom();
3060
3061 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3062 * instances at once without manually having to specify -M each time. */
3063 if (arg_ephemeral)
3064 if (strextendf(&arg_machine, "-%016" PRIx64, random_u64()) < 0)
3065 return log_oom();
3066 } else {
3067 arg_settings_filename = strjoin(arg_machine, ".nspawn");
3068 if (!arg_settings_filename)
3069 return log_oom();
3070 }
3071
3072 return 0;
3073 }
3074
3075 static int chase_and_update(char **p, unsigned flags) {
3076 char *chased;
3077 int r;
3078
3079 assert(p);
3080
3081 if (!*p)
3082 return 0;
3083
3084 r = chase(*p, NULL, flags, &chased, NULL);
3085 if (r < 0)
3086 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
3087
3088 return free_and_replace(*p, chased);
3089 }
3090
3091 static int determine_uid_shift(const char *directory) {
3092
3093 if (arg_userns_mode == USER_NAMESPACE_NO) {
3094 arg_uid_shift = 0;
3095 return 0;
3096 }
3097
3098 if (arg_uid_shift == UID_INVALID) {
3099 struct stat st;
3100
3101 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3102
3103 if (stat(directory, &st) < 0)
3104 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
3105
3106 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
3107
3108 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
3109 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3110 "UID and GID base of %s don't match.", directory);
3111
3112 arg_uid_range = UINT32_C(0x10000);
3113
3114 if (arg_uid_shift != 0) {
3115 /* If the image is shifted already, then we'll fall back to classic chowning, for
3116 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3117
3118 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_AUTO) {
3119 log_debug("UID base of %s is non-zero, not using UID mapping.", directory);
3120 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
3121 } else if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_MAP)
3122 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3123 "UID base of %s is not zero, UID mapping not supported.", directory);
3124 }
3125 }
3126
3127 if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
3128 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID base too high for UID range.");
3129
3130 return 0;
3131 }
3132
3133 static unsigned long effective_clone_ns_flags(void) {
3134 unsigned long flags = arg_clone_ns_flags;
3135
3136 if (arg_private_network)
3137 flags |= CLONE_NEWNET;
3138 if (arg_use_cgns)
3139 flags |= CLONE_NEWCGROUP;
3140 if (arg_userns_mode != USER_NAMESPACE_NO)
3141 flags |= CLONE_NEWUSER;
3142
3143 return flags;
3144 }
3145
3146 static int patch_sysctl(void) {
3147
3148 /* This table is inspired by runc's sysctl() function */
3149 static const struct {
3150 const char *key;
3151 bool prefix;
3152 unsigned long clone_flags;
3153 } safe_sysctl[] = {
3154 { "kernel.hostname", false, CLONE_NEWUTS },
3155 { "kernel.domainname", false, CLONE_NEWUTS },
3156 { "kernel.msgmax", false, CLONE_NEWIPC },
3157 { "kernel.msgmnb", false, CLONE_NEWIPC },
3158 { "kernel.msgmni", false, CLONE_NEWIPC },
3159 { "kernel.sem", false, CLONE_NEWIPC },
3160 { "kernel.shmall", false, CLONE_NEWIPC },
3161 { "kernel.shmmax", false, CLONE_NEWIPC },
3162 { "kernel.shmmni", false, CLONE_NEWIPC },
3163 { "fs.mqueue.", true, CLONE_NEWIPC },
3164 { "net.", true, CLONE_NEWNET },
3165 };
3166
3167 unsigned long flags;
3168 int r;
3169
3170 flags = effective_clone_ns_flags();
3171
3172 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
3173 bool good = false;
3174 size_t i;
3175
3176 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
3177
3178 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
3179 continue;
3180
3181 if (safe_sysctl[i].prefix)
3182 good = startswith(*k, safe_sysctl[i].key);
3183 else
3184 good = streq(*k, safe_sysctl[i].key);
3185
3186 if (good)
3187 break;
3188 }
3189
3190 if (!good)
3191 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
3192
3193 r = sysctl_write(*k, *v);
3194 if (r < 0)
3195 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
3196 }
3197
3198 return 0;
3199 }
3200
3201 static int inner_child(
3202 Barrier *barrier,
3203 int fd_inner_socket,
3204 FDSet *fds,
3205 char **os_release_pairs) {
3206
3207 _cleanup_free_ char *home = NULL;
3208 size_t n_env = 1;
3209 char *envp[] = {
3210 (char*) "PATH=" DEFAULT_PATH_COMPAT,
3211 NULL, /* container */
3212 NULL, /* TERM */
3213 NULL, /* HOME */
3214 NULL, /* USER */
3215 NULL, /* LOGNAME */
3216 NULL, /* container_uuid */
3217 NULL, /* LISTEN_FDS */
3218 NULL, /* LISTEN_PID */
3219 NULL, /* NOTIFY_SOCKET */
3220 NULL, /* CREDENTIALS_DIRECTORY */
3221 NULL, /* LANG */
3222 NULL
3223 };
3224 const char *exec_target;
3225 _cleanup_strv_free_ char **env_use = NULL;
3226 int r, which_failed;
3227
3228 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3229 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3230 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3231 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3232 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3233 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3234 * namespace.
3235 *
3236 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3237 * unshare(). See below. */
3238
3239 assert(barrier);
3240 assert(fd_inner_socket >= 0);
3241
3242 log_debug("Inner child is initializing.");
3243
3244 if (arg_userns_mode != USER_NAMESPACE_NO) {
3245 /* Tell the parent, that it now can write the UID map. */
3246 (void) barrier_place(barrier); /* #1 */
3247
3248 /* Wait until the parent wrote the UID map */
3249 if (!barrier_place_and_sync(barrier)) /* #2 */
3250 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3251
3252 /* Become the new root user inside our namespace */
3253 r = reset_uid_gid();
3254 if (r < 0)
3255 return log_error_errno(r, "Couldn't become new root: %m");
3256
3257 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3258 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3259 * propagation, but simply create new peer groups for all our mounts). */
3260 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SHARED|MS_REC, NULL);
3261 if (r < 0)
3262 return r;
3263 }
3264
3265 r = mount_all(NULL,
3266 arg_mount_settings | MOUNT_IN_USERNS,
3267 arg_uid_shift,
3268 arg_selinux_apifs_context);
3269 if (r < 0)
3270 return r;
3271
3272 if (!arg_network_namespace_path && arg_private_network) {
3273 r = unshare(CLONE_NEWNET);
3274 if (r < 0)
3275 return log_error_errno(errno, "Failed to unshare network namespace: %m");
3276
3277 /* Tell the parent that it can setup network interfaces. */
3278 (void) barrier_place(barrier); /* #3 */
3279 }
3280
3281 r = mount_sysfs(NULL, arg_mount_settings);
3282 if (r < 0)
3283 return r;
3284
3285 /* Wait until we are cgroup-ified, so that we
3286 * can mount the right cgroup path writable */
3287 if (!barrier_place_and_sync(barrier)) /* #4 */
3288 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3289 "Parent died too early");
3290
3291 if (arg_use_cgns) {
3292 r = unshare(CLONE_NEWCGROUP);
3293 if (r < 0)
3294 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
3295 r = mount_cgroups(
3296 "",
3297 arg_unified_cgroup_hierarchy,
3298 arg_userns_mode != USER_NAMESPACE_NO,
3299 arg_uid_shift,
3300 arg_uid_range,
3301 arg_selinux_apifs_context,
3302 true);
3303 } else
3304 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
3305 if (r < 0)
3306 return r;
3307
3308 r = setup_boot_id();
3309 if (r < 0)
3310 return r;
3311
3312 r = setup_kmsg(fd_inner_socket);
3313 if (r < 0)
3314 return r;
3315
3316 r = mount_custom(
3317 "/",
3318 arg_custom_mounts,
3319 arg_n_custom_mounts,
3320 0,
3321 0,
3322 arg_selinux_apifs_context,
3323 MOUNT_NON_ROOT_ONLY | MOUNT_IN_USERNS);
3324 if (r < 0)
3325 return r;
3326
3327 if (setsid() < 0)
3328 return log_error_errno(errno, "setsid() failed: %m");
3329
3330 if (arg_private_network)
3331 (void) loopback_setup();
3332
3333 if (arg_expose_ports) {
3334 r = expose_port_send_rtnl(fd_inner_socket);
3335 if (r < 0)
3336 return r;
3337 }
3338
3339 if (arg_console_mode != CONSOLE_PIPE) {
3340 _cleanup_close_ int master = -EBADF;
3341 _cleanup_free_ char *console = NULL;
3342
3343 /* Allocate a pty and make it available as /dev/console. */
3344 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
3345 if (master < 0)
3346 return log_error_errno(master, "Failed to allocate a pty: %m");
3347
3348 r = setup_dev_console(console);
3349 if (r < 0)
3350 return log_error_errno(r, "Failed to set up /dev/console: %m");
3351
3352 r = send_one_fd(fd_inner_socket, master, 0);
3353 if (r < 0)
3354 return log_error_errno(r, "Failed to send master fd: %m");
3355
3356 r = setup_stdio_as_dev_console();
3357 if (r < 0)
3358 return r;
3359 }
3360
3361 r = patch_sysctl();
3362 if (r < 0)
3363 return r;
3364
3365 if (arg_oom_score_adjust_set) {
3366 r = set_oom_score_adjust(arg_oom_score_adjust);
3367 if (r < 0)
3368 return log_error_errno(r, "Failed to adjust OOM score: %m");
3369 }
3370
3371 if (arg_cpu_set.set)
3372 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
3373 return log_error_errno(errno, "Failed to set CPU affinity: %m");
3374
3375 (void) setup_hostname();
3376
3377 if (arg_personality != PERSONALITY_INVALID) {
3378 r = safe_personality(arg_personality);
3379 if (r < 0)
3380 return log_error_errno(r, "personality() failed: %m");
3381 #ifdef ARCHITECTURE_SECONDARY
3382 } else if (arg_architecture == ARCHITECTURE_SECONDARY) {
3383 r = safe_personality(PER_LINUX32);
3384 if (r < 0)
3385 return log_error_errno(r, "personality() failed: %m");
3386 #endif
3387 } else if (!arg_quiet && arg_architecture >= 0 && arg_architecture != native_architecture())
3388 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3389 "invocation with qemu userspace emulator (or equivalent) in effect.",
3390 architecture_to_string(arg_architecture));
3391
3392 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3393 if (r < 0)
3394 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3395
3396 #if HAVE_SECCOMP
3397 if (arg_seccomp) {
3398
3399 if (is_seccomp_available()) {
3400 r = seccomp_load(arg_seccomp);
3401 if (ERRNO_IS_NEG_SECCOMP_FATAL(r))
3402 return log_error_errno(r, "Failed to install seccomp filter: %m");
3403 if (r < 0)
3404 log_debug_errno(r, "Failed to install seccomp filter: %m");
3405 }
3406 } else
3407 #endif
3408 {
3409 r = setup_seccomp(arg_caps_retain, arg_syscall_allow_list, arg_syscall_deny_list);
3410 if (r < 0)
3411 return r;
3412 }
3413
3414 if (arg_suppress_sync) {
3415 #if HAVE_SECCOMP
3416 r = seccomp_suppress_sync();
3417 if (r < 0)
3418 log_debug_errno(r, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3419 #else
3420 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3421 #endif
3422 }
3423
3424 #if HAVE_SELINUX
3425 if (arg_selinux_context)
3426 if (setexeccon(arg_selinux_context) < 0)
3427 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3428 #endif
3429
3430 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3431 * if we need to later on. */
3432 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3433 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3434
3435 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3436 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids, arg_console_mode != CONSOLE_PIPE);
3437 else
3438 r = change_uid_gid(arg_user, arg_console_mode != CONSOLE_PIPE, &home);
3439 if (r < 0)
3440 return r;
3441
3442 r = drop_capabilities(getuid());
3443 if (r < 0)
3444 return log_error_errno(r, "Dropping capabilities failed: %m");
3445
3446 if (arg_no_new_privileges)
3447 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3448 return log_error_errno(errno, "Failed to disable new privileges: %m");
3449
3450 /* LXC sets container=lxc, so follow the scheme here */
3451 envp[n_env++] = strjoina("container=", arg_container_service_name);
3452
3453 envp[n_env] = strv_find_prefix(environ, "TERM=");
3454 if (envp[n_env])
3455 n_env++;
3456
3457 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3458 if (asprintf(envp + n_env++, "HOME=%s", home ?: "/root") < 0)
3459 return log_oom();
3460
3461 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3462 if (asprintf(envp + n_env++, "USER=%s", arg_user ?: "root") < 0 ||
3463 asprintf(envp + n_env++, "LOGNAME=%s", arg_user ?: "root") < 0)
3464 return log_oom();
3465
3466 assert(!sd_id128_is_null(arg_uuid));
3467
3468 if (asprintf(envp + n_env++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid)) < 0)
3469 return log_oom();
3470
3471 if (!fdset_isempty(fds)) {
3472 r = fdset_cloexec(fds, false);
3473 if (r < 0)
3474 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3475
3476 if ((asprintf(envp + n_env++, "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3477 (asprintf(envp + n_env++, "LISTEN_PID=1") < 0))
3478 return log_oom();
3479 }
3480 if (asprintf(envp + n_env++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3481 return log_oom();
3482
3483 if (arg_credentials.n_credentials > 0) {
3484 envp[n_env] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3485 if (!envp[n_env])
3486 return log_oom();
3487 n_env++;
3488 }
3489
3490 if (arg_start_mode != START_BOOT) {
3491 envp[n_env] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE);
3492 if (!envp[n_env])
3493 return log_oom();
3494 n_env++;
3495 }
3496
3497 env_use = strv_env_merge(envp, os_release_pairs, arg_setenv);
3498 if (!env_use)
3499 return log_oom();
3500
3501 /* Let the parent know that we are ready and wait until the parent is ready with the setup, too... */
3502 if (!barrier_place_and_sync(barrier)) /* #5 */
3503 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3504
3505 /* Note, this should be done this late (💣 and not moved earlier! 💣), so that all namespacing
3506 * changes are already in effect by now, so that any resolved paths here definitely reference
3507 * resources inside the container, and not outside of them. */
3508 if (arg_chdir)
3509 if (chdir(arg_chdir) < 0)
3510 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3511
3512 if (arg_start_mode == START_PID2) {
3513 r = stub_pid1(arg_uuid);
3514 if (r < 0)
3515 return r;
3516 }
3517
3518 if (arg_console_mode != CONSOLE_PIPE) {
3519 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3520 * are configured for that. Acquire it as controlling tty. */
3521 if (ioctl(STDIN_FILENO, TIOCSCTTY) < 0)
3522 return log_error_errno(errno, "Failed to acquire controlling TTY: %m");
3523 }
3524
3525 log_debug("Inner child completed, invoking payload.");
3526
3527 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3528 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3529 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3530 log_close();
3531 log_set_open_when_needed(true);
3532 log_settle_target();
3533
3534 (void) fdset_close_others(fds);
3535
3536 if (arg_start_mode == START_BOOT) {
3537 char **a;
3538 size_t m;
3539
3540 /* Automatically search for the init system */
3541
3542 m = strv_length(arg_parameters);
3543 a = newa(char*, m + 2);
3544 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3545 a[1 + m] = NULL;
3546
3547 FOREACH_STRING(init,
3548 "/usr/lib/systemd/systemd",
3549 "/lib/systemd/systemd",
3550 "/sbin/init") {
3551 a[0] = (char*) init;
3552 execve(a[0], a, env_use);
3553 }
3554
3555 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3556 } else if (!strv_isempty(arg_parameters)) {
3557 const char *dollar_path;
3558
3559 exec_target = arg_parameters[0];
3560
3561 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3562 * binary. */
3563 dollar_path = strv_env_get(env_use, "PATH");
3564 if (dollar_path) {
3565 if (setenv("PATH", dollar_path, 1) < 0)
3566 return log_error_errno(errno, "Failed to update $PATH: %m");
3567 }
3568
3569 execvpe(arg_parameters[0], arg_parameters, env_use);
3570 } else {
3571 if (!arg_chdir)
3572 /* If we cannot change the directory, we'll end up in /, that is expected. */
3573 (void) chdir(home ?: "/root");
3574
3575 execle(DEFAULT_USER_SHELL, "-" DEFAULT_USER_SHELL_NAME, NULL, env_use);
3576 if (!streq(DEFAULT_USER_SHELL, "/bin/bash"))
3577 execle("/bin/bash", "-bash", NULL, env_use);
3578 if (!streq(DEFAULT_USER_SHELL, "/bin/sh"))
3579 execle("/bin/sh", "-sh", NULL, env_use);
3580
3581 exec_target = DEFAULT_USER_SHELL ", /bin/bash, /bin/sh";
3582 }
3583
3584 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3585 }
3586
3587 static int setup_notify_child(void) {
3588 _cleanup_close_ int fd = -EBADF;
3589 static const union sockaddr_union sa = {
3590 .un.sun_family = AF_UNIX,
3591 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3592 };
3593 int r;
3594
3595 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3596 if (fd < 0)
3597 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3598
3599 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3600 (void) sockaddr_un_unlink(&sa.un);
3601
3602 WITH_UMASK(0577) { /* only set "w" bit, which is all that's necessary for connecting from the container */
3603 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3604 if (r < 0)
3605 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3606 }
3607
3608 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3609 if (r < 0)
3610 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3611
3612 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3613 if (r < 0)
3614 return log_error_errno(r, "SO_PASSCRED failed: %m");
3615
3616 return TAKE_FD(fd);
3617 }
3618
3619 static int setup_unix_export_dir_outside(char **ret) {
3620 int r;
3621
3622 assert(ret);
3623
3624 _cleanup_free_ char *p = NULL;
3625 p = path_join("/run/systemd/nspawn/unix-export", arg_machine);
3626 if (!p)
3627 return log_oom();
3628
3629 r = path_is_mount_point(p);
3630 if (r > 0)
3631 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Mount point '%s' exists already, refusing.", p);
3632 if (r < 0 && r != -ENOENT)
3633 return log_error_errno(r, "Failed to detect if '%s' is a mount point: %m", p);
3634
3635 r = mkdir_p(p, 0755);
3636 if (r < 0)
3637 return log_error_errno(r, "Failed to create '%s': %m", p);
3638
3639 _cleanup_(rmdir_and_freep) char *q = TAKE_PTR(p);
3640
3641 /* Mount the "unix export" directory really tiny, just 64 inodes. We mark the superblock writable
3642 * (since the container shall bind sockets into it). */
3643 r = mount_nofollow_verbose(
3644 LOG_ERR,
3645 "tmpfs",
3646 q,
3647 "tmpfs",
3648 MS_NODEV|MS_NOEXEC|MS_NOSUID|ms_nosymfollow_supported(),
3649 "size=4M,nr_inodes=64,mode=0755");
3650 if (r < 0)
3651 return r;
3652
3653 _cleanup_(umount_and_rmdir_and_freep) char *w = TAKE_PTR(q);
3654
3655 /* After creating the superblock we change the bind mount to be read-only. This means that the fs
3656 * itself is writable, but not through the mount accessible from the host. */
3657 r = mount_nofollow_verbose(
3658 LOG_ERR,
3659 /* source= */ NULL,
3660 w,
3661 /* fstype= */ NULL,
3662 MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NODEV|MS_NOEXEC|MS_NOSUID|ms_nosymfollow_supported(),
3663 /* options= */ NULL);
3664 if (r < 0)
3665 return r;
3666
3667 *ret = TAKE_PTR(w);
3668 return 0;
3669 }
3670
3671 static int setup_unix_export_host_inside(const char *directory, const char *unix_export_path) {
3672 int r;
3673
3674 assert(directory);
3675 assert(unix_export_path);
3676
3677 r = make_run_host(directory);
3678 if (r < 0)
3679 return r;
3680
3681 _cleanup_free_ char *p = path_join(directory, "run/host/unix-export");
3682 if (!p)
3683 return log_oom();
3684
3685 if (mkdir(p, 0755) < 0)
3686 return log_error_errno(errno, "Failed to create '%s': %m", p);
3687
3688 r = mount_nofollow_verbose(
3689 LOG_ERR,
3690 unix_export_path,
3691 p,
3692 /* fstype= */ NULL,
3693 MS_BIND,
3694 /* options= */ NULL);
3695 if (r < 0)
3696 return r;
3697
3698 r = mount_nofollow_verbose(
3699 LOG_ERR,
3700 /* source= */ NULL,
3701 p,
3702 /* fstype= */ NULL,
3703 MS_BIND|MS_REMOUNT|MS_NODEV|MS_NOEXEC|MS_NOSUID|ms_nosymfollow_supported(),
3704 /* options= */ NULL);
3705 if (r < 0)
3706 return r;
3707
3708 r = userns_lchown(p, 0, 0);
3709 if (r < 0)
3710 return log_error_errno(r, "Failed to chown '%s': %m", p);
3711
3712 return 0;
3713 }
3714
3715 static int outer_child(
3716 Barrier *barrier,
3717 const char *directory,
3718 DissectedImage *dissected_image,
3719 int fd_outer_socket,
3720 int fd_inner_socket,
3721 FDSet *fds,
3722 int netns_fd,
3723 const char *unix_export_path) {
3724
3725 _cleanup_(bind_user_context_freep) BindUserContext *bind_user_context = NULL;
3726 _cleanup_strv_free_ char **os_release_pairs = NULL;
3727 _cleanup_close_ int fd = -EBADF, mntns_fd = -EBADF;
3728 bool idmap = false;
3729 const char *p;
3730 pid_t pid;
3731 ssize_t l;
3732 int r;
3733
3734 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3735 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3736 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3737 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3738 * forked off it, and it exits. */
3739
3740 assert(barrier);
3741 assert(directory);
3742 assert(fd_outer_socket >= 0);
3743 assert(fd_inner_socket >= 0);
3744
3745 log_debug("Outer child is initializing.");
3746
3747 r = load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs);
3748 if (r < 0)
3749 log_debug_errno(r, "Failed to read os-release from host for container, ignoring: %m");
3750
3751 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3752 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3753
3754 r = reset_audit_loginuid();
3755 if (r < 0)
3756 return r;
3757
3758 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3759 * mounts to the real root. */
3760 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3761 if (r < 0)
3762 return r;
3763
3764 if (dissected_image) {
3765 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3766 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3767 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3768 * right place right away. This makes sure ESP partitions and userns are compatible. */
3769
3770 r = dissected_image_mount_and_warn(
3771 dissected_image,
3772 directory,
3773 arg_uid_shift,
3774 arg_uid_range,
3775 /* userns_fd= */ -EBADF,
3776 DISSECT_IMAGE_MOUNT_ROOT_ONLY|
3777 DISSECT_IMAGE_DISCARD_ON_LOOP|
3778 DISSECT_IMAGE_USR_NO_ROOT|
3779 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
3780 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3781 if (r < 0)
3782 return r;
3783 }
3784
3785 r = determine_uid_shift(directory);
3786 if (r < 0)
3787 return r;
3788
3789 if (arg_userns_mode != USER_NAMESPACE_NO) {
3790 r = namespace_open(0,
3791 /* ret_pidns_fd = */ NULL,
3792 &mntns_fd,
3793 /* ret_netns_fd = */ NULL,
3794 /* ret_userns_fd = */ NULL,
3795 /* ret_root_fd = */ NULL);
3796 if (r < 0)
3797 return log_error_errno(r, "Failed to pin outer mount namespace: %m");
3798
3799 l = send_one_fd(fd_outer_socket, mntns_fd, 0);
3800 if (l < 0)
3801 return log_error_errno(l, "Failed to send outer mount namespace fd: %m");
3802 mntns_fd = safe_close(mntns_fd);
3803
3804 /* Let the parent know which UID shift we read from the image */
3805 l = send(fd_outer_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3806 if (l < 0)
3807 return log_error_errno(errno, "Failed to send UID shift: %m");
3808 if (l != sizeof(arg_uid_shift))
3809 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3810 "Short write while sending UID shift.");
3811
3812 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3813 /* When we are supposed to pick the UID shift, the parent will check now whether the
3814 * UID shift we just read from the image is available. If yes, it will send the UID
3815 * shift back to us, if not it will pick a different one, and send it back to us. */
3816
3817 l = recv(fd_outer_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3818 if (l < 0)
3819 return log_error_errno(errno, "Failed to recv UID shift: %m");
3820 if (l != sizeof(arg_uid_shift))
3821 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3822 "Short read while receiving UID shift.");
3823 }
3824
3825 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3826 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3827 }
3828
3829 if (path_equal(directory, "/")) {
3830 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3831 * place, so that we can make changes to its mount structure (for example, to implement
3832 * --volatile=) without this interfering with our ability to access files such as
3833 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3834 * (instead of a temporary directory, since we are living in our own mount namespace here
3835 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3836 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3837
3838 r = mount_nofollow_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
3839 if (r < 0)
3840 return r;
3841
3842 directory = "/run/systemd/nspawn-root";
3843 }
3844
3845 /* Make sure we always have a mount that we can move to root later on. */
3846 r = make_mount_point(directory);
3847 if (r < 0)
3848 return r;
3849
3850 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3851 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3852 * we'll live in our own little world from now on, and propagation from the host may only happen via
3853 * the mount tunnel dir, or not at all. */
3854 r = mount_follow_verbose(LOG_ERR, NULL, directory, NULL, MS_PRIVATE|MS_REC, NULL);
3855 if (r < 0)
3856 return r;
3857
3858 r = setup_pivot_root(
3859 directory,
3860 arg_pivot_root_new,
3861 arg_pivot_root_old);
3862 if (r < 0)
3863 return r;
3864
3865 r = setup_volatile_mode(
3866 directory,
3867 arg_volatile_mode,
3868 arg_uid_shift,
3869 arg_selinux_apifs_context);
3870 if (r < 0)
3871 return r;
3872
3873 r = bind_user_prepare(
3874 directory,
3875 arg_bind_user,
3876 arg_uid_shift,
3877 arg_uid_range,
3878 &arg_custom_mounts, &arg_n_custom_mounts,
3879 &bind_user_context);
3880 if (r < 0)
3881 return r;
3882
3883 if (arg_userns_mode != USER_NAMESPACE_NO && bind_user_context) {
3884 /* Send the user maps we determined to the parent, so that it installs it in our user
3885 * namespace UID map table */
3886
3887 for (size_t i = 0; i < bind_user_context->n_data; i++) {
3888 uid_t map[] = {
3889 bind_user_context->data[i].payload_user->uid,
3890 bind_user_context->data[i].host_user->uid,
3891 (uid_t) bind_user_context->data[i].payload_group->gid,
3892 (uid_t) bind_user_context->data[i].host_group->gid,
3893 };
3894
3895 l = send(fd_outer_socket, map, sizeof(map), MSG_NOSIGNAL);
3896 if (l < 0)
3897 return log_error_errno(errno, "Failed to send user UID map: %m");
3898 if (l != sizeof(map))
3899 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3900 "Short write while sending user UID map.");
3901 }
3902 }
3903
3904 r = mount_custom(
3905 directory,
3906 arg_custom_mounts,
3907 arg_n_custom_mounts,
3908 arg_uid_shift,
3909 arg_uid_range,
3910 arg_selinux_apifs_context,
3911 MOUNT_ROOT_ONLY);
3912 if (r < 0)
3913 return r;
3914
3915 if (arg_userns_mode != USER_NAMESPACE_NO &&
3916 IN_SET(arg_userns_ownership, USER_NAMESPACE_OWNERSHIP_MAP, USER_NAMESPACE_OWNERSHIP_AUTO) &&
3917 arg_uid_shift != 0) {
3918 _cleanup_free_ char *usr_subtree = NULL;
3919 char *dirs[3];
3920 size_t i = 0;
3921
3922 dirs[i++] = (char*) directory;
3923
3924 if (dissected_image && dissected_image->partitions[PARTITION_USR].found) {
3925 usr_subtree = path_join(directory, "/usr");
3926 if (!usr_subtree)
3927 return log_oom();
3928
3929 dirs[i++] = usr_subtree;
3930 }
3931
3932 dirs[i] = NULL;
3933
3934 r = remount_idmap(dirs, arg_uid_shift, arg_uid_range, UID_INVALID, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
3935 if (r == -EINVAL || ERRNO_IS_NEG_NOT_SUPPORTED(r)) {
3936 /* This might fail because the kernel or file system doesn't support idmapping. We
3937 * can't really distinguish this nicely, nor do we have any guarantees about the
3938 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3939 if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_AUTO)
3940 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
3941 "ID mapped mounts are apparently not available, sorry.");
3942
3943 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3944 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
3945 } else if (r < 0)
3946 return log_error_errno(r, "Failed to set up ID mapped mounts: %m");
3947 else {
3948 log_debug("ID mapped mounts available, making use of them.");
3949 idmap = true;
3950 }
3951 }
3952
3953 if (dissected_image) {
3954 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3955 r = dissected_image_mount(
3956 dissected_image,
3957 directory,
3958 arg_uid_shift,
3959 arg_uid_range,
3960 /* userns_fd= */ -EBADF,
3961 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|
3962 DISSECT_IMAGE_DISCARD_ON_LOOP|
3963 DISSECT_IMAGE_USR_NO_ROOT|
3964 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
3965 (idmap ? DISSECT_IMAGE_MOUNT_IDMAPPED : 0));
3966 if (r == -EUCLEAN)
3967 return log_error_errno(r, "File system check for image failed: %m");
3968 if (r < 0)
3969 return log_error_errno(r, "Failed to mount image file system: %m");
3970 }
3971
3972 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3973 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3974
3975 r = detect_unified_cgroup_hierarchy_from_image(directory);
3976 if (r < 0)
3977 return r;
3978
3979 l = send(fd_outer_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3980 if (l < 0)
3981 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3982 if (l != sizeof(arg_unified_cgroup_hierarchy))
3983 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3984 "Short write while sending cgroup mode.");
3985 }
3986
3987 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3988 if (r < 0)
3989 return r;
3990
3991 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3992 if (r < 0)
3993 return r;
3994
3995 if (arg_read_only && arg_volatile_mode == VOLATILE_NO &&
3996 !has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts)) {
3997 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3998 if (r < 0)
3999 return log_error_errno(r, "Failed to make tree read-only: %m");
4000 }
4001
4002 r = mount_all(directory,
4003 arg_mount_settings,
4004 arg_uid_shift,
4005 arg_selinux_apifs_context);
4006 if (r < 0)
4007 return r;
4008
4009 r = copy_devnodes(directory);
4010 if (r < 0)
4011 return r;
4012
4013 r = make_extra_nodes(directory);
4014 if (r < 0)
4015 return r;
4016
4017 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
4018
4019 p = prefix_roota(directory, "/run/host");
4020 (void) make_inaccessible_nodes(p, arg_uid_shift, arg_uid_shift);
4021
4022 r = setup_unix_export_host_inside(directory, unix_export_path);
4023 if (r < 0)
4024 return r;
4025
4026 r = setup_pts(directory);
4027 if (r < 0)
4028 return r;
4029
4030 r = mount_tunnel_dig(directory);
4031 if (r < 0)
4032 return r;
4033
4034 r = setup_keyring();
4035 if (r < 0)
4036 return r;
4037
4038 r = setup_credentials(directory);
4039 if (r < 0)
4040 return r;
4041
4042 r = bind_user_setup(bind_user_context, directory);
4043 if (r < 0)
4044 return r;
4045
4046 r = mount_custom(
4047 directory,
4048 arg_custom_mounts,
4049 arg_n_custom_mounts,
4050 arg_uid_shift,
4051 arg_uid_range,
4052 arg_selinux_apifs_context,
4053 MOUNT_NON_ROOT_ONLY);
4054 if (r < 0)
4055 return r;
4056
4057 r = setup_timezone(directory);
4058 if (r < 0)
4059 return r;
4060
4061 r = setup_resolv_conf(directory);
4062 if (r < 0)
4063 return r;
4064
4065 r = setup_machine_id(directory);
4066 if (r < 0)
4067 return r;
4068
4069 r = setup_journal(directory);
4070 if (r < 0)
4071 return r;
4072
4073 /* The same stuff as the $container env var, but nicely readable for the entire payload */
4074 p = prefix_roota(directory, "/run/host/container-manager");
4075 (void) write_string_file(p, arg_container_service_name, WRITE_STRING_FILE_CREATE|WRITE_STRING_FILE_MODE_0444);
4076
4077 /* The same stuff as the $container_uuid env var */
4078 p = prefix_roota(directory, "/run/host/container-uuid");
4079 (void) write_string_filef(p, WRITE_STRING_FILE_CREATE|WRITE_STRING_FILE_MODE_0444, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid));
4080
4081 if (!arg_use_cgns) {
4082 r = mount_cgroups(
4083 directory,
4084 arg_unified_cgroup_hierarchy,
4085 arg_userns_mode != USER_NAMESPACE_NO,
4086 arg_uid_shift,
4087 arg_uid_range,
4088 arg_selinux_apifs_context,
4089 false);
4090 if (r < 0)
4091 return r;
4092 }
4093
4094 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
4095 * mounts available in systemd services inside the container that create a new mount namespace. See
4096 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
4097 * will inherit the shared propagation mode.
4098 *
4099 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
4100 * directory mount to root later on.
4101 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
4102 */
4103 r = mount_switch_root(directory, MS_SHARED);
4104 if (r < 0)
4105 return log_error_errno(r, "Failed to move root directory: %m");
4106
4107 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
4108 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
4109 * the container. */
4110 r = mount_tunnel_open();
4111 if (r < 0)
4112 return r;
4113
4114 if (arg_userns_mode != USER_NAMESPACE_NO) {
4115 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4116 * requires that a fully visible instance is already present in the target mount
4117 * namespace. Mount one here so the inner child can mount its own instances. Later
4118 * we umount the temporary instances created here before we actually exec the
4119 * payload. Since the rootfs is shared the umount will propagate into the container.
4120 * Note, the inner child wouldn't be able to unmount the instances on its own since
4121 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4122 * this. */
4123 r = pin_fully_visible_fs();
4124 if (r < 0)
4125 return r;
4126 }
4127
4128 fd = setup_notify_child();
4129 if (fd < 0)
4130 return fd;
4131
4132 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
4133 arg_clone_ns_flags |
4134 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
4135 if (pid < 0)
4136 return log_error_errno(errno, "Failed to fork inner child: %m");
4137 if (pid == 0) {
4138 fd_outer_socket = safe_close(fd_outer_socket);
4139
4140 /* The inner child has all namespaces that are requested, so that we all are owned by the
4141 * user if user namespaces are turned on. */
4142
4143 if (arg_network_namespace_path) {
4144 r = namespace_enter(/* pidns_fd = */ -EBADF,
4145 /* mntns_fd = */ -EBADF,
4146 netns_fd,
4147 /* userns_fd = */ -EBADF,
4148 /* root_fd = */ -EBADF);
4149 if (r < 0)
4150 return log_error_errno(r, "Failed to join network namespace: %m");
4151 }
4152
4153 r = inner_child(barrier, fd_inner_socket, fds, os_release_pairs);
4154 if (r < 0)
4155 _exit(EXIT_FAILURE);
4156
4157 _exit(EXIT_SUCCESS);
4158 }
4159
4160 l = send(fd_outer_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
4161 if (l < 0)
4162 return log_error_errno(errno, "Failed to send PID: %m");
4163 if (l != sizeof(pid))
4164 return log_error_errno(SYNTHETIC_ERRNO(EIO),
4165 "Short write while sending PID.");
4166
4167 l = send(fd_outer_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
4168 if (l < 0)
4169 return log_error_errno(errno, "Failed to send machine ID: %m");
4170 if (l != sizeof(arg_uuid))
4171 return log_error_errno(SYNTHETIC_ERRNO(EIO),
4172 "Short write while sending machine ID.");
4173
4174 l = send_one_fd(fd_outer_socket, fd, 0);
4175 if (l < 0)
4176 return log_error_errno(l, "Failed to send notify fd: %m");
4177
4178 fd_outer_socket = safe_close(fd_outer_socket);
4179 fd_inner_socket = safe_close(fd_inner_socket);
4180 netns_fd = safe_close(netns_fd);
4181
4182 return 0;
4183 }
4184
4185 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
4186 bool tried_hashed = false;
4187 unsigned n_tries = 100;
4188 uid_t candidate;
4189 int r;
4190
4191 assert(shift);
4192 assert(ret_lock_file);
4193 assert(arg_userns_mode == USER_NAMESPACE_PICK);
4194 assert(arg_uid_range == 0x10000U);
4195
4196 candidate = *shift;
4197
4198 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4199
4200 for (;;) {
4201 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
4202 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
4203
4204 if (--n_tries <= 0)
4205 return -EBUSY;
4206
4207 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
4208 goto next;
4209 if ((candidate & UINT32_C(0xFFFF)) != 0)
4210 goto next;
4211
4212 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
4213 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
4214 if (r == -EBUSY) /* Range already taken by another nspawn instance */
4215 goto next;
4216 if (r < 0)
4217 return r;
4218
4219 /* Make some superficial checks whether the range is currently known in the user database */
4220 if (getpwuid_malloc(candidate, /* ret= */ NULL) >= 0)
4221 goto next;
4222 if (getpwuid_malloc(candidate + UINT32_C(0xFFFE), /* ret= */ NULL) >= 0)
4223 goto next;
4224 if (getgrgid_malloc(candidate, /* ret= */ NULL) >= 0)
4225 goto next;
4226 if (getgrgid_malloc(candidate + UINT32_C(0xFFFE), /* ret= */ NULL) >= 0)
4227 goto next;
4228
4229 *ret_lock_file = lf;
4230 lf = (struct LockFile) LOCK_FILE_INIT;
4231 *shift = candidate;
4232 return 0;
4233
4234 next:
4235 if (arg_machine && !tried_hashed) {
4236 /* Try to hash the base from the container name */
4237
4238 static const uint8_t hash_key[] = {
4239 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4240 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4241 };
4242
4243 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
4244
4245 tried_hashed = true;
4246 } else
4247 random_bytes(&candidate, sizeof(candidate));
4248
4249 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
4250 candidate &= (uid_t) UINT32_C(0xFFFF0000);
4251 }
4252 }
4253
4254 static int add_one_uid_map(
4255 char **p,
4256 uid_t container_uid,
4257 uid_t host_uid,
4258 uid_t range) {
4259
4260 return strextendf(p,
4261 UID_FMT " " UID_FMT " " UID_FMT "\n",
4262 container_uid, host_uid, range);
4263 }
4264
4265 static int make_uid_map_string(
4266 const uid_t bind_user_uid[],
4267 size_t n_bind_user_uid,
4268 size_t offset,
4269 char **ret) {
4270
4271 _cleanup_free_ char *s = NULL;
4272 uid_t previous_uid = 0;
4273 int r;
4274
4275 assert(n_bind_user_uid == 0 || bind_user_uid);
4276 assert(IN_SET(offset, 0, 2)); /* used to switch between UID and GID map */
4277 assert(ret);
4278
4279 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4280 * quadruplet, consisting of host and container UID + GID. */
4281
4282 for (size_t i = 0; i < n_bind_user_uid; i++) {
4283 uid_t payload_uid = bind_user_uid[i*4+offset],
4284 host_uid = bind_user_uid[i*4+offset+1];
4285
4286 assert(previous_uid <= payload_uid);
4287 assert(payload_uid < arg_uid_range);
4288
4289 /* Add a range to close the gap to previous entry */
4290 if (payload_uid > previous_uid) {
4291 r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, payload_uid - previous_uid);
4292 if (r < 0)
4293 return r;
4294 }
4295
4296 /* Map this specific user */
4297 r = add_one_uid_map(&s, payload_uid, host_uid, 1);
4298 if (r < 0)
4299 return r;
4300
4301 previous_uid = payload_uid + 1;
4302 }
4303
4304 /* And add a range to close the gap to finish the range */
4305 if (arg_uid_range > previous_uid) {
4306 r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, arg_uid_range - previous_uid);
4307 if (r < 0)
4308 return r;
4309 }
4310
4311 assert(s);
4312
4313 *ret = TAKE_PTR(s);
4314 return 0;
4315 }
4316
4317 static int setup_uid_map(
4318 pid_t pid,
4319 const uid_t bind_user_uid[],
4320 size_t n_bind_user_uid) {
4321
4322 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1];
4323 _cleanup_free_ char *s = NULL;
4324 int r;
4325
4326 assert(pid > 1);
4327
4328 /* Build the UID map string */
4329 if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 0, &s) < 0) /* offset=0 contains the UID pair */
4330 return log_oom();
4331
4332 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
4333 r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
4334 if (r < 0)
4335 return log_error_errno(r, "Failed to write UID map: %m");
4336
4337 /* And now build the GID map string */
4338 s = mfree(s);
4339 if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 2, &s) < 0) /* offset=2 contains the GID pair */
4340 return log_oom();
4341
4342 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
4343 r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
4344 if (r < 0)
4345 return log_error_errno(r, "Failed to write GID map: %m");
4346
4347 return 0;
4348 }
4349
4350 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4351 char buf[NOTIFY_BUFFER_MAX+1];
4352 char *p = NULL;
4353 struct iovec iovec = {
4354 .iov_base = buf,
4355 .iov_len = sizeof(buf)-1,
4356 };
4357 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
4358 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
4359 struct msghdr msghdr = {
4360 .msg_iov = &iovec,
4361 .msg_iovlen = 1,
4362 .msg_control = &control,
4363 .msg_controllen = sizeof(control),
4364 };
4365 struct ucred *ucred;
4366 ssize_t n;
4367 pid_t inner_child_pid;
4368 _cleanup_strv_free_ char **tags = NULL;
4369 int r;
4370
4371 assert(userdata);
4372
4373 inner_child_pid = PTR_TO_PID(userdata);
4374
4375 if (revents != EPOLLIN) {
4376 log_warning("Got unexpected poll event for notify fd.");
4377 return 0;
4378 }
4379
4380 n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
4381 if (ERRNO_IS_NEG_TRANSIENT(n))
4382 return 0;
4383 else if (n == -EXFULL) {
4384 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4385 return 0;
4386 } else if (n < 0)
4387 return log_warning_errno(n, "Couldn't read notification socket: %m");
4388
4389 cmsg_close_all(&msghdr);
4390
4391 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
4392 if (!ucred || ucred->pid != inner_child_pid) {
4393 log_debug("Received notify message without valid credentials. Ignoring.");
4394 return 0;
4395 }
4396
4397 if ((size_t) n >= sizeof(buf)) {
4398 log_warning("Received notify message exceeded maximum size. Ignoring.");
4399 return 0;
4400 }
4401
4402 buf[n] = 0;
4403 tags = strv_split(buf, "\n\r");
4404 if (!tags)
4405 return log_oom();
4406
4407 if (strv_contains(tags, "READY=1")) {
4408 r = sd_notify(false, "READY=1\n");
4409 if (r < 0)
4410 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
4411 }
4412
4413 p = strv_find_startswith(tags, "STATUS=");
4414 if (p)
4415 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
4416
4417 return 0;
4418 }
4419
4420 static int setup_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
4421 int r;
4422
4423 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
4424 if (r < 0)
4425 return log_error_errno(r, "Failed to allocate notify event source: %m");
4426
4427 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
4428
4429 return 0;
4430 }
4431
4432 static void set_window_title(PTYForward *f) {
4433 _cleanup_free_ char *hn = NULL, *dot = NULL;
4434
4435 assert(f);
4436
4437 (void) gethostname_strict(&hn);
4438
4439 if (emoji_enabled())
4440 dot = strjoin(special_glyph(SPECIAL_GLYPH_BLUE_CIRCLE), " ");
4441
4442 if (hn)
4443 (void) pty_forward_set_titlef(f, "%sContainer %s on %s", strempty(dot), arg_machine, hn);
4444 else
4445 (void) pty_forward_set_titlef(f, "%sContainer %s", strempty(dot), arg_machine);
4446 }
4447
4448 static int merge_settings(Settings *settings, const char *path) {
4449 int rl;
4450
4451 assert(settings);
4452 assert(path);
4453
4454 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4455 * that this steals the fields of the Settings* structure, and hence modifies it. */
4456
4457 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
4458 settings->start_mode >= 0) {
4459 arg_start_mode = settings->start_mode;
4460 strv_free_and_replace(arg_parameters, settings->parameters);
4461 }
4462
4463 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0 &&
4464 settings->ephemeral >= 0)
4465 arg_ephemeral = settings->ephemeral;
4466
4467 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
4468 settings->root) {
4469
4470 if (!arg_settings_trusted)
4471 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
4472 else
4473 free_and_replace(arg_directory, settings->root);
4474 }
4475
4476 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
4477 settings->pivot_root_new) {
4478 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
4479 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
4480 }
4481
4482 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
4483 settings->working_directory)
4484 free_and_replace(arg_chdir, settings->working_directory);
4485
4486 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
4487 settings->environment)
4488 strv_free_and_replace(arg_setenv, settings->environment);
4489
4490 if ((arg_settings_mask & SETTING_USER) == 0) {
4491
4492 if (settings->user)
4493 free_and_replace(arg_user, settings->user);
4494
4495 if (uid_is_valid(settings->uid))
4496 arg_uid = settings->uid;
4497 if (gid_is_valid(settings->gid))
4498 arg_gid = settings->gid;
4499 if (settings->n_supplementary_gids > 0) {
4500 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
4501 arg_n_supplementary_gids = settings->n_supplementary_gids;
4502 }
4503 }
4504
4505 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
4506 uint64_t plus, minus;
4507 uint64_t network_minus = 0;
4508 uint64_t ambient;
4509
4510 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4511 * Settings structure */
4512
4513 plus = settings->capability;
4514 minus = settings->drop_capability;
4515
4516 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4517 settings_network_configured(settings)) {
4518 if (settings_private_network(settings))
4519 plus |= UINT64_C(1) << CAP_NET_ADMIN;
4520 else
4521 network_minus |= UINT64_C(1) << CAP_NET_ADMIN;
4522 }
4523
4524 if (!arg_settings_trusted && plus != 0) {
4525 if (settings->capability != 0)
4526 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
4527 } else {
4528 arg_caps_retain &= ~network_minus;
4529 arg_caps_retain |= plus;
4530 }
4531
4532 arg_caps_retain &= ~minus;
4533
4534 /* Copy the full capabilities over too */
4535 if (capability_quintet_is_set(&settings->full_capabilities)) {
4536 if (!arg_settings_trusted)
4537 log_warning("Ignoring capability settings, file %s is not trusted.", path);
4538 else
4539 arg_full_capabilities = settings->full_capabilities;
4540 }
4541
4542 ambient = settings->ambient_capability;
4543 if (!arg_settings_trusted && ambient != 0)
4544 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path);
4545 else
4546 arg_caps_ambient |= ambient;
4547 }
4548
4549 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
4550 settings->kill_signal > 0)
4551 arg_kill_signal = settings->kill_signal;
4552
4553 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
4554 settings->personality != PERSONALITY_INVALID)
4555 arg_personality = settings->personality;
4556
4557 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
4558 !sd_id128_is_null(settings->machine_id)) {
4559
4560 if (!arg_settings_trusted)
4561 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
4562 else
4563 arg_uuid = settings->machine_id;
4564 }
4565
4566 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
4567 settings->read_only >= 0)
4568 arg_read_only = settings->read_only;
4569
4570 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
4571 settings->volatile_mode != _VOLATILE_MODE_INVALID)
4572 arg_volatile_mode = settings->volatile_mode;
4573
4574 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
4575 settings->n_custom_mounts > 0) {
4576
4577 if (!arg_settings_trusted)
4578 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
4579 else {
4580 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4581 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
4582 arg_n_custom_mounts = settings->n_custom_mounts;
4583 settings->n_custom_mounts = 0;
4584 }
4585 }
4586
4587 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4588 settings_network_configured(settings)) {
4589
4590 if (!arg_settings_trusted)
4591 log_warning("Ignoring network settings, file %s is not trusted.", path);
4592 else {
4593 arg_network_veth = settings_network_veth(settings);
4594 arg_private_network = settings_private_network(settings);
4595
4596 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
4597 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
4598 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
4599 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
4600
4601 free_and_replace(arg_network_bridge, settings->network_bridge);
4602 free_and_replace(arg_network_zone, settings->network_zone);
4603
4604 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
4605 }
4606 }
4607
4608 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
4609 settings->expose_ports) {
4610
4611 if (!arg_settings_trusted)
4612 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
4613 else {
4614 expose_port_free_all(arg_expose_ports);
4615 arg_expose_ports = TAKE_PTR(settings->expose_ports);
4616 }
4617 }
4618
4619 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
4620 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
4621
4622 if (!arg_settings_trusted)
4623 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
4624 else {
4625 arg_userns_mode = settings->userns_mode;
4626 arg_uid_shift = settings->uid_shift;
4627 arg_uid_range = settings->uid_range;
4628 arg_userns_ownership = settings->userns_ownership;
4629 }
4630 }
4631
4632 if ((arg_settings_mask & SETTING_BIND_USER) == 0 &&
4633 !strv_isempty(settings->bind_user))
4634 strv_free_and_replace(arg_bind_user, settings->bind_user);
4635
4636 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0 &&
4637 settings->notify_ready >= 0)
4638 arg_notify_ready = settings->notify_ready;
4639
4640 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
4641
4642 if (!strv_isempty(settings->syscall_allow_list) || !strv_isempty(settings->syscall_deny_list)) {
4643 if (!arg_settings_trusted && !strv_isempty(settings->syscall_allow_list))
4644 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
4645 else {
4646 strv_free_and_replace(arg_syscall_allow_list, settings->syscall_allow_list);
4647 strv_free_and_replace(arg_syscall_deny_list, settings->syscall_deny_list);
4648 }
4649 }
4650
4651 #if HAVE_SECCOMP
4652 if (settings->seccomp) {
4653 if (!arg_settings_trusted)
4654 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
4655 else {
4656 seccomp_release(arg_seccomp);
4657 arg_seccomp = TAKE_PTR(settings->seccomp);
4658 }
4659 }
4660 #endif
4661 }
4662
4663 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
4664 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
4665 continue;
4666
4667 if (!settings->rlimit[rl])
4668 continue;
4669
4670 if (!arg_settings_trusted) {
4671 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
4672 continue;
4673 }
4674
4675 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
4676 }
4677
4678 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
4679 settings->hostname)
4680 free_and_replace(arg_hostname, settings->hostname);
4681
4682 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
4683 settings->no_new_privileges >= 0)
4684 arg_no_new_privileges = settings->no_new_privileges;
4685
4686 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
4687 settings->oom_score_adjust_set) {
4688
4689 if (!arg_settings_trusted)
4690 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
4691 else {
4692 arg_oom_score_adjust = settings->oom_score_adjust;
4693 arg_oom_score_adjust_set = true;
4694 }
4695 }
4696
4697 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
4698 settings->cpu_set.set) {
4699
4700 if (!arg_settings_trusted)
4701 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
4702 else {
4703 cpu_set_reset(&arg_cpu_set);
4704 arg_cpu_set = TAKE_STRUCT(settings->cpu_set);
4705 }
4706 }
4707
4708 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
4709 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
4710 arg_resolv_conf = settings->resolv_conf;
4711
4712 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
4713 settings->link_journal != _LINK_JOURNAL_INVALID) {
4714
4715 if (!arg_settings_trusted)
4716 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
4717 else {
4718 arg_link_journal = settings->link_journal;
4719 arg_link_journal_try = settings->link_journal_try;
4720 }
4721 }
4722
4723 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
4724 settings->timezone != _TIMEZONE_MODE_INVALID)
4725 arg_timezone = settings->timezone;
4726
4727 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
4728 settings->slice) {
4729
4730 if (!arg_settings_trusted)
4731 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
4732 else
4733 free_and_replace(arg_slice, settings->slice);
4734 }
4735
4736 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
4737 settings->use_cgns >= 0) {
4738
4739 if (!arg_settings_trusted)
4740 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
4741 else
4742 arg_use_cgns = settings->use_cgns;
4743 }
4744
4745 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
4746 settings->clone_ns_flags != ULONG_MAX) {
4747
4748 if (!arg_settings_trusted)
4749 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
4750 else
4751 arg_clone_ns_flags = settings->clone_ns_flags;
4752 }
4753
4754 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
4755 settings->console_mode >= 0) {
4756
4757 if (!arg_settings_trusted)
4758 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
4759 else
4760 arg_console_mode = settings->console_mode;
4761 }
4762
4763 if ((arg_settings_mask & SETTING_SUPPRESS_SYNC) == 0 &&
4764 settings->suppress_sync >= 0)
4765 arg_suppress_sync = settings->suppress_sync;
4766
4767 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4768 * don't consult arg_settings_mask for them. */
4769
4770 sd_bus_message_unref(arg_property_message);
4771 arg_property_message = TAKE_PTR(settings->properties);
4772
4773 arg_console_width = settings->console_width;
4774 arg_console_height = settings->console_height;
4775
4776 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
4777 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
4778 arg_n_extra_nodes = settings->n_extra_nodes;
4779 settings->n_extra_nodes = 0;
4780
4781 return 0;
4782 }
4783
4784 static int load_settings(void) {
4785 _cleanup_(settings_freep) Settings *settings = NULL;
4786 _cleanup_fclose_ FILE *f = NULL;
4787 _cleanup_free_ char *p = NULL;
4788 int r;
4789
4790 if (arg_oci_bundle)
4791 return 0;
4792
4793 /* If all settings are masked, there's no point in looking for
4794 * the settings file */
4795 if (FLAGS_SET(arg_settings_mask, _SETTINGS_MASK_ALL))
4796 return 0;
4797
4798 /* We first look in the admin's directories in /etc and /run */
4799 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4800 _cleanup_free_ char *j = NULL;
4801
4802 j = path_join(i, arg_settings_filename);
4803 if (!j)
4804 return log_oom();
4805
4806 f = fopen(j, "re");
4807 if (f) {
4808 p = TAKE_PTR(j);
4809
4810 /* By default, we trust configuration from /etc and /run */
4811 if (arg_settings_trusted < 0)
4812 arg_settings_trusted = true;
4813
4814 break;
4815 }
4816
4817 if (errno != ENOENT)
4818 return log_error_errno(errno, "Failed to open %s: %m", j);
4819 }
4820
4821 if (!f) {
4822 /* After that, let's look for a file next to the
4823 * actual image we shall boot. */
4824
4825 if (arg_image) {
4826 r = file_in_same_dir(arg_image, arg_settings_filename, &p);
4827 if (r < 0)
4828 return log_error_errno(r, "Failed to generate settings path from image path: %m");
4829 } else if (arg_directory) {
4830 r = file_in_same_dir(arg_directory, arg_settings_filename, &p);
4831 if (r < 0 && r != -EADDRNOTAVAIL) /* if directory is root fs, don't complain */
4832 return log_error_errno(r, "Failed to generate settings path from directory path: %m");
4833 }
4834
4835 if (p) {
4836 f = fopen(p, "re");
4837 if (!f && errno != ENOENT)
4838 return log_error_errno(errno, "Failed to open %s: %m", p);
4839
4840 /* By default, we do not trust configuration from /var/lib/machines */
4841 if (arg_settings_trusted < 0)
4842 arg_settings_trusted = false;
4843 }
4844 }
4845
4846 if (!f)
4847 return 0;
4848
4849 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4850
4851 r = settings_load(f, p, &settings);
4852 if (r < 0)
4853 return r;
4854
4855 return merge_settings(settings, p);
4856 }
4857
4858 static int load_oci_bundle(void) {
4859 _cleanup_(settings_freep) Settings *settings = NULL;
4860 int r;
4861
4862 if (!arg_oci_bundle)
4863 return 0;
4864
4865 /* By default let's trust OCI bundles */
4866 if (arg_settings_trusted < 0)
4867 arg_settings_trusted = true;
4868
4869 r = oci_load(NULL, arg_oci_bundle, &settings);
4870 if (r < 0)
4871 return r;
4872
4873 return merge_settings(settings, arg_oci_bundle);
4874 }
4875
4876 static int run_container(
4877 DissectedImage *dissected_image,
4878 FDSet *fds,
4879 char veth_name[IFNAMSIZ], bool *veth_created,
4880 struct ExposeArgs *expose_args,
4881 int *master, pid_t *pid, int *ret) {
4882
4883 static const struct sigaction sa = {
4884 .sa_handler = nop_signal_handler,
4885 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4886 };
4887
4888 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4889 _cleanup_close_ int etc_passwd_lock = -EBADF;
4890 _cleanup_close_pair_ int
4891 fd_inner_socket_pair[2] = EBADF_PAIR,
4892 fd_outer_socket_pair[2] = EBADF_PAIR;
4893
4894 _cleanup_close_ int notify_socket = -EBADF, mntns_fd = -EBADF, fd_kmsg_fifo = -EBADF;
4895 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4896 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4897 _cleanup_(umount_and_rmdir_and_freep) char *unix_export_host_dir = NULL;
4898 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4899 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4900 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4901 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4902 _cleanup_free_ uid_t *bind_user_uid = NULL;
4903 size_t n_bind_user_uid = 0;
4904 ContainerStatus container_status = 0;
4905 int ifi = 0, r;
4906 ssize_t l;
4907 sigset_t mask_chld;
4908 _cleanup_close_ int child_netns_fd = -EBADF;
4909
4910 assert_se(sigemptyset(&mask_chld) == 0);
4911 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4912
4913 /* Set up the unix export host directory on the host first */
4914 r = setup_unix_export_dir_outside(&unix_export_host_dir);
4915 if (r < 0)
4916 return r;
4917
4918 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4919 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4920 * check with getpwuid() if the specific user already exists. Note that /etc might be
4921 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4922 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4923 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4924 * really ours. */
4925
4926 etc_passwd_lock = take_etc_passwd_lock(NULL);
4927 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4928 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4929 }
4930
4931 r = barrier_create(&barrier);
4932 if (r < 0)
4933 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4934
4935 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, fd_inner_socket_pair) < 0)
4936 return log_error_errno(errno, "Failed to create inner socket pair: %m");
4937
4938 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, fd_outer_socket_pair) < 0)
4939 return log_error_errno(errno, "Failed to create outer socket pair: %m");
4940
4941 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4942 * parent's blocking calls and give it a chance to call wait() and terminate. */
4943 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4944 if (r < 0)
4945 return log_error_errno(errno, "Failed to change the signal mask: %m");
4946
4947 r = sigaction(SIGCHLD, &sa, NULL);
4948 if (r < 0)
4949 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4950
4951 if (arg_network_namespace_path) {
4952 child_netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4953 if (child_netns_fd < 0)
4954 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4955
4956 r = fd_is_ns(child_netns_fd, CLONE_NEWNET);
4957 if (r == -EUCLEAN)
4958 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);
4959 else if (r < 0)
4960 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4961 else if (r == 0)
4962 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4963 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4964 }
4965
4966 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4967 if (*pid < 0)
4968 return log_error_errno(errno, "clone() failed%s: %m",
4969 errno == EINVAL ?
4970 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4971
4972 if (*pid == 0) {
4973 /* The outer child only has a file system namespace. */
4974 barrier_set_role(&barrier, BARRIER_CHILD);
4975
4976 fd_inner_socket_pair[0] = safe_close(fd_inner_socket_pair[0]);
4977 fd_outer_socket_pair[0] = safe_close(fd_outer_socket_pair[0]);
4978
4979 (void) reset_all_signal_handlers();
4980 (void) reset_signal_mask();
4981
4982 r = outer_child(&barrier,
4983 arg_directory,
4984 dissected_image,
4985 fd_outer_socket_pair[1],
4986 fd_inner_socket_pair[1],
4987 fds,
4988 child_netns_fd,
4989 unix_export_host_dir);
4990 if (r < 0)
4991 _exit(EXIT_FAILURE);
4992
4993 _exit(EXIT_SUCCESS);
4994 }
4995
4996 barrier_set_role(&barrier, BARRIER_PARENT);
4997
4998 fdset_close(fds);
4999
5000 fd_inner_socket_pair[1] = safe_close(fd_inner_socket_pair[1]);
5001 fd_outer_socket_pair[1] = safe_close(fd_outer_socket_pair[1]);
5002
5003 if (arg_userns_mode != USER_NAMESPACE_NO) {
5004 mntns_fd = receive_one_fd(fd_outer_socket_pair[0], 0);
5005 if (mntns_fd < 0)
5006 return log_error_errno(mntns_fd, "Failed to receive mount namespace fd from outer child: %m");
5007
5008 /* The child just let us know the UID shift it might have read from the image. */
5009 l = recv(fd_outer_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
5010 if (l < 0)
5011 return log_error_errno(errno, "Failed to read UID shift: %m");
5012 if (l != sizeof arg_uid_shift)
5013 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
5014
5015 if (arg_userns_mode == USER_NAMESPACE_PICK) {
5016 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
5017 * image, but if that's already in use, pick a new one, and report back to the child,
5018 * which one we now picked. */
5019
5020 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
5021 if (r < 0)
5022 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
5023
5024 l = send(fd_outer_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
5025 if (l < 0)
5026 return log_error_errno(errno, "Failed to send UID shift: %m");
5027 if (l != sizeof arg_uid_shift)
5028 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
5029 }
5030
5031 n_bind_user_uid = strv_length(arg_bind_user);
5032 if (n_bind_user_uid > 0) {
5033 /* Right after the UID shift, we'll receive the list of UID mappings for the
5034 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
5035
5036 bind_user_uid = new(uid_t, n_bind_user_uid*4);
5037 if (!bind_user_uid)
5038 return log_oom();
5039
5040 for (size_t i = 0; i < n_bind_user_uid; i++) {
5041 l = recv(fd_outer_socket_pair[0], bind_user_uid + i*4, sizeof(uid_t)*4, 0);
5042 if (l < 0)
5043 return log_error_errno(errno, "Failed to read user UID map pair: %m");
5044 if (l != sizeof(uid_t)*4)
5045 return log_full_errno(l == 0 ? LOG_DEBUG : LOG_WARNING,
5046 SYNTHETIC_ERRNO(EIO),
5047 "Short read while reading bind user UID pairs.");
5048 }
5049 }
5050 }
5051
5052 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
5053 /* The child let us know the support cgroup mode it might have read from the image. */
5054 l = recv(fd_outer_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
5055 if (l < 0)
5056 return log_error_errno(errno, "Failed to read cgroup mode: %m");
5057 if (l != sizeof(arg_unified_cgroup_hierarchy))
5058 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zi bytes).%s",
5059 l, l == 0 ? " The child is most likely dead." : "");
5060 }
5061
5062 /* Wait for the outer child. */
5063 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
5064 if (r < 0)
5065 return r;
5066 if (r != EXIT_SUCCESS)
5067 return -EIO;
5068
5069 /* And now retrieve the PID of the inner child. */
5070 l = recv(fd_outer_socket_pair[0], pid, sizeof *pid, 0);
5071 if (l < 0)
5072 return log_error_errno(errno, "Failed to read inner child PID: %m");
5073 if (l != sizeof *pid)
5074 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
5075
5076 /* We also retrieve container UUID in case it was generated by outer child */
5077 l = recv(fd_outer_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
5078 if (l < 0)
5079 return log_error_errno(errno, "Failed to read container machine ID: %m");
5080 if (l != sizeof(arg_uuid))
5081 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
5082
5083 /* We also retrieve the socket used for notifications generated by outer child */
5084 notify_socket = receive_one_fd(fd_outer_socket_pair[0], 0);
5085 if (notify_socket < 0)
5086 return log_error_errno(notify_socket,
5087 "Failed to receive notification socket from the outer child: %m");
5088
5089 log_debug("Init process invoked as PID "PID_FMT, *pid);
5090
5091 if (arg_userns_mode != USER_NAMESPACE_NO) {
5092 if (!barrier_place_and_sync(&barrier)) /* #1 */
5093 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
5094
5095 r = setup_uid_map(*pid, bind_user_uid, n_bind_user_uid);
5096 if (r < 0)
5097 return r;
5098
5099 (void) barrier_place(&barrier); /* #2 */
5100 }
5101
5102 if (arg_private_network) {
5103 if (!arg_network_namespace_path) {
5104 /* Wait until the child has unshared its network namespace. */
5105 if (!barrier_place_and_sync(&barrier)) /* #3 */
5106 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
5107 }
5108
5109 if (child_netns_fd < 0) {
5110 /* Make sure we have an open file descriptor to the child's network
5111 * namespace so it stays alive even if the child exits. */
5112 r = namespace_open(*pid,
5113 /* ret_pidns_fd = */ NULL,
5114 /* ret_mntns_fd = */ NULL,
5115 &child_netns_fd,
5116 /* ret_userns_fd = */ NULL,
5117 /* ret_root_fd = */ NULL);
5118 if (r < 0)
5119 return log_error_errno(r, "Failed to open child network namespace: %m");
5120 }
5121
5122 r = move_network_interfaces(child_netns_fd, arg_network_interfaces);
5123 if (r < 0)
5124 return r;
5125
5126 if (arg_network_veth) {
5127 r = setup_veth(arg_machine, *pid, veth_name,
5128 arg_network_bridge || arg_network_zone, &arg_network_provided_mac);
5129 if (r < 0)
5130 return r;
5131 else if (r > 0)
5132 ifi = r;
5133
5134 if (arg_network_bridge) {
5135 /* Add the interface to a bridge */
5136 r = setup_bridge(veth_name, arg_network_bridge, false);
5137 if (r < 0)
5138 return r;
5139 if (r > 0)
5140 ifi = r;
5141 } else if (arg_network_zone) {
5142 /* Add the interface to a bridge, possibly creating it */
5143 r = setup_bridge(veth_name, arg_network_zone, true);
5144 if (r < 0)
5145 return r;
5146 if (r > 0)
5147 ifi = r;
5148 }
5149 }
5150
5151 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
5152 if (r < 0)
5153 return r;
5154
5155 /* We created the primary and extra veth links now; let's remember this, so that we know to
5156 remove them later on. Note that we don't bother with removing veth links that were created
5157 here when their setup failed half-way, because in that case the kernel should be able to
5158 remove them on its own, since they cannot be referenced by anything yet. */
5159 *veth_created = true;
5160
5161 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
5162 if (r < 0)
5163 return r;
5164
5165 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
5166 if (r < 0)
5167 return r;
5168 }
5169
5170 if (arg_register || !arg_keep_unit) {
5171 r = sd_bus_default_system(&bus);
5172 if (r < 0)
5173 return log_error_errno(r, "Failed to open system bus: %m");
5174
5175 r = sd_bus_set_close_on_exit(bus, false);
5176 if (r < 0)
5177 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
5178 }
5179
5180 if (!arg_keep_unit) {
5181 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5182 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5183 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5184
5185 r = sd_bus_match_signal_async(
5186 bus,
5187 NULL,
5188 "org.freedesktop.systemd1",
5189 NULL,
5190 "org.freedesktop.systemd1.Scope",
5191 "RequestStop",
5192 on_request_stop, NULL, PID_TO_PTR(*pid));
5193 if (r < 0)
5194 return log_error_errno(r, "Failed to request RequestStop match: %m");
5195 }
5196
5197 if (arg_register) {
5198 r = register_machine(
5199 bus,
5200 arg_machine,
5201 *pid,
5202 arg_directory,
5203 arg_uuid,
5204 ifi,
5205 arg_slice,
5206 arg_custom_mounts, arg_n_custom_mounts,
5207 arg_kill_signal,
5208 arg_property,
5209 arg_property_message,
5210 arg_keep_unit,
5211 arg_container_service_name,
5212 arg_start_mode);
5213 if (r < 0)
5214 return r;
5215
5216 } else if (!arg_keep_unit) {
5217 r = allocate_scope(
5218 bus,
5219 arg_machine,
5220 *pid,
5221 arg_slice,
5222 arg_custom_mounts, arg_n_custom_mounts,
5223 arg_kill_signal,
5224 arg_property,
5225 arg_property_message,
5226 /* allow_pidfds= */ true,
5227 arg_start_mode);
5228 if (r < 0)
5229 return r;
5230
5231 } else if (arg_slice || arg_property)
5232 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5233
5234 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
5235 if (r < 0)
5236 return r;
5237
5238 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
5239 if (r < 0)
5240 return r;
5241
5242 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
5243 if (r < 0)
5244 return r;
5245
5246 /* Notify the child that the parent is ready with all
5247 * its setup (including cgroup-ification), and that
5248 * the child can now hand over control to the code to
5249 * run inside the container. */
5250 (void) barrier_place(&barrier); /* #4 */
5251
5252 /* Block SIGCHLD here, before notifying child.
5253 * process_pty() will handle it with the other signals. */
5254 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
5255
5256 /* Reset signal to default */
5257 r = default_signals(SIGCHLD);
5258 if (r < 0)
5259 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
5260
5261 r = sd_event_new(&event);
5262 if (r < 0)
5263 return log_error_errno(r, "Failed to get default event source: %m");
5264
5265 (void) sd_event_set_watchdog(event, true);
5266
5267 if (bus) {
5268 r = sd_bus_attach_event(bus, event, 0);
5269 if (r < 0)
5270 return log_error_errno(r, "Failed to attach bus to event loop: %m");
5271 }
5272
5273 r = setup_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
5274 if (r < 0)
5275 return r;
5276
5277 /* Wait that the child is completely ready now, and has mounted their own copies of procfs and so on,
5278 * before we take the fully visible instances away. */
5279 if (!barrier_sync(&barrier)) /* #5.1 */
5280 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
5281
5282 if (arg_userns_mode != USER_NAMESPACE_NO) {
5283 r = wipe_fully_visible_fs(mntns_fd);
5284 if (r < 0)
5285 return r;
5286 mntns_fd = safe_close(mntns_fd);
5287 }
5288
5289 /* And now let the child know that we completed removing the procfs instances, and it can start the
5290 * payload. */
5291 if (!barrier_place(&barrier)) /* #5.2 */
5292 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
5293
5294 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5295 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5296 etc_passwd_lock = safe_close(etc_passwd_lock);
5297
5298 (void) sd_notifyf(false,
5299 "STATUS=Container running.\n"
5300 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
5301 if (!arg_notify_ready) {
5302 r = sd_notify(false, "READY=1\n");
5303 if (r < 0)
5304 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
5305 }
5306
5307 if (arg_kill_signal > 0) {
5308 /* Try to kill the init system on SIGINT or SIGTERM */
5309 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
5310 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
5311 } else {
5312 /* Immediately exit */
5313 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
5314 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
5315 }
5316
5317 (void) sd_event_add_signal(event, NULL, SIGRTMIN+18, sigrtmin18_handler, NULL);
5318
5319 r = sd_event_add_memory_pressure(event, NULL, NULL, NULL);
5320 if (r < 0)
5321 log_debug_errno(r, "Failed allocate memory pressure event source, ignoring: %m");
5322
5323 /* Exit when the child exits */
5324 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
5325
5326 /* Retrieve the kmsg fifo allocated by inner child */
5327 fd_kmsg_fifo = receive_one_fd(fd_inner_socket_pair[0], 0);
5328 if (fd_kmsg_fifo < 0)
5329 return log_error_errno(fd_kmsg_fifo, "Failed to receive kmsg fifo from inner child: %m");
5330
5331 if (arg_expose_ports) {
5332 r = expose_port_watch_rtnl(event, fd_inner_socket_pair[0], on_address_change, expose_args, &rtnl);
5333 if (r < 0)
5334 return r;
5335
5336 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
5337 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
5338 }
5339
5340 if (arg_console_mode != CONSOLE_PIPE) {
5341 _cleanup_close_ int fd = -EBADF;
5342 PTYForwardFlags flags = 0;
5343
5344 /* Retrieve the master pty allocated by inner child */
5345 fd = receive_one_fd(fd_inner_socket_pair[0], 0);
5346 if (fd < 0)
5347 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
5348
5349 switch (arg_console_mode) {
5350
5351 case CONSOLE_READ_ONLY:
5352 flags |= PTY_FORWARD_READ_ONLY;
5353
5354 _fallthrough_;
5355
5356 case CONSOLE_INTERACTIVE:
5357 flags |= PTY_FORWARD_IGNORE_VHANGUP;
5358
5359 r = pty_forward_new(event, fd, flags, &forward);
5360 if (r < 0)
5361 return log_error_errno(r, "Failed to create PTY forwarder: %m");
5362
5363 if (arg_console_width != UINT_MAX || arg_console_height != UINT_MAX)
5364 (void) pty_forward_set_width_height(
5365 forward,
5366 arg_console_width,
5367 arg_console_height);
5368
5369 if (!arg_background) {
5370 _cleanup_free_ char *bg = NULL;
5371
5372 r = terminal_tint_color(220 /* blue */, &bg);
5373 if (r < 0)
5374 log_debug_errno(r, "Failed to determine terminal background color, not tinting.");
5375 else
5376 (void) pty_forward_set_background_color(forward, bg);
5377 } else if (!isempty(arg_background))
5378 (void) pty_forward_set_background_color(forward, arg_background);
5379
5380 set_window_title(forward);
5381 break;
5382
5383 default:
5384 assert(arg_console_mode == CONSOLE_PASSIVE);
5385 }
5386
5387 *master = TAKE_FD(fd);
5388 }
5389
5390 fd_inner_socket_pair[0] = safe_close(fd_inner_socket_pair[0]);
5391
5392 r = sd_event_loop(event);
5393 if (r < 0)
5394 return log_error_errno(r, "Failed to run event loop: %m");
5395
5396 if (forward) {
5397 char last_char = 0;
5398
5399 (void) pty_forward_get_last_char(forward, &last_char);
5400 forward = pty_forward_free(forward);
5401
5402 if (!arg_quiet && last_char != '\n')
5403 putc('\n', stdout);
5404 }
5405
5406 /* Kill if it is not dead yet anyway */
5407 if (!arg_register && !arg_keep_unit && bus)
5408 terminate_scope(bus, arg_machine);
5409
5410 /* Normally redundant, but better safe than sorry */
5411 (void) kill(*pid, SIGKILL);
5412
5413 fd_kmsg_fifo = safe_close(fd_kmsg_fifo);
5414
5415 if (arg_private_network) {
5416 r = move_back_network_interfaces(child_netns_fd, arg_network_interfaces);
5417 if (r < 0)
5418 return r;
5419 }
5420
5421 r = wait_for_container(TAKE_PID(*pid), &container_status);
5422
5423 /* Tell machined that we are gone. */
5424 if (bus)
5425 (void) unregister_machine(bus, arg_machine);
5426
5427 if (r < 0)
5428 /* We failed to wait for the container, or the container exited abnormally. */
5429 return r;
5430 if (r > 0 || container_status == CONTAINER_TERMINATED) {
5431 /* r > 0 → The container exited with a non-zero status.
5432 * As a special case, we need to replace 133 with a different value,
5433 * because 133 is special-cased in the service file to reboot the container.
5434 * otherwise → The container exited with zero status and a reboot was not requested.
5435 */
5436 if (r == EXIT_FORCE_RESTART)
5437 r = EXIT_FAILURE; /* replace 133 with the general failure code */
5438 *ret = r;
5439 return 0; /* finito */
5440 }
5441
5442 /* CONTAINER_REBOOTED, loop again */
5443
5444 if (arg_keep_unit) {
5445 /* Special handling if we are running as a service: instead of simply
5446 * restarting the machine we want to restart the entire service, so let's
5447 * inform systemd about this with the special exit code 133. The service
5448 * file uses RestartForceExitStatus=133 so that this results in a full
5449 * nspawn restart. This is necessary since we might have cgroup parameters
5450 * set we want to have flushed out. */
5451 *ret = EXIT_FORCE_RESTART;
5452 return 0; /* finito */
5453 }
5454
5455 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
5456 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
5457
5458 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5459 *veth_created = false;
5460 return 1; /* loop again */
5461 }
5462
5463 static int initialize_rlimits(void) {
5464 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5465 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5466 * container execution environments. */
5467
5468 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
5469 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
5470 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
5471 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
5472 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
5473 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
5474 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
5475 [RLIMIT_MEMLOCK] = { DEFAULT_RLIMIT_MEMLOCK, DEFAULT_RLIMIT_MEMLOCK },
5476 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
5477 [RLIMIT_NICE] = { 0, 0 },
5478 [RLIMIT_NOFILE] = { 1024, 4096 },
5479 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
5480 [RLIMIT_RTPRIO] = { 0, 0 },
5481 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
5482 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
5483
5484 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5485 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5486 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5487 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5488 * that PID 1 changes a number of other resource limits during early initialization which is why we
5489 * don't read the other limits from PID 1 but prefer the static table above. */
5490 };
5491
5492 int rl, r;
5493
5494 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
5495 /* Let's only fill in what the user hasn't explicitly configured anyway */
5496 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
5497 const struct rlimit *v;
5498 struct rlimit buffer;
5499
5500 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
5501 /* For these two let's read the limits off PID 1. See above for an explanation. */
5502
5503 r = pid_getrlimit(1, rl, &buffer);
5504 if (r < 0)
5505 return log_error_errno(r, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
5506
5507 v = &buffer;
5508 } else if (rl == RLIMIT_NOFILE) {
5509 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5510 * userspace. Given that nspawn containers are often run without our PID 1,
5511 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5512 * so that container userspace gets similar resources as host userspace
5513 * gets. */
5514 buffer = kernel_defaults[rl];
5515 buffer.rlim_max = MIN((rlim_t) read_nr_open(), (rlim_t) HIGH_RLIMIT_NOFILE);
5516 v = &buffer;
5517 } else
5518 v = kernel_defaults + rl;
5519
5520 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
5521 if (!arg_rlimit[rl])
5522 return log_oom();
5523 }
5524
5525 if (DEBUG_LOGGING) {
5526 _cleanup_free_ char *k = NULL;
5527
5528 (void) rlimit_format(arg_rlimit[rl], &k);
5529 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
5530 }
5531 }
5532
5533 return 0;
5534 }
5535
5536 static int cant_be_in_netns(void) {
5537 _cleanup_close_ int fd = -EBADF;
5538 struct ucred ucred;
5539 int r;
5540
5541 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5542 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5543 * nice message. */
5544
5545 if (!arg_image) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5546 return 0;
5547
5548 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_NONBLOCK|SOCK_CLOEXEC, 0);
5549 if (fd < 0)
5550 return log_error_errno(errno, "Failed to allocate udev control socket: %m");
5551
5552 r = connect_unix_path(fd, AT_FDCWD, "/run/udev/control");
5553 if (r == -ENOENT || ERRNO_IS_NEG_DISCONNECT(r))
5554 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5555 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5556 if (r < 0)
5557 return log_error_errno(r, "Failed to connect socket to udev control socket: %m");
5558
5559 r = getpeercred(fd, &ucred);
5560 if (r < 0)
5561 return log_error_errno(r, "Failed to determine peer of udev control socket: %m");
5562
5563 r = in_same_namespace(ucred.pid, 0, NAMESPACE_NET);
5564 if (r < 0)
5565 return log_error_errno(r, "Failed to determine network namespace of udev: %m");
5566 if (r == 0)
5567 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5568 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5569 return 0;
5570 }
5571
5572 static int run(int argc, char *argv[]) {
5573 bool remove_directory = false, remove_image = false, veth_created = false, remove_tmprootdir = false;
5574 _cleanup_close_ int master = -EBADF;
5575 _cleanup_fdset_free_ FDSet *fds = NULL;
5576 int r, n_fd_passed, ret = EXIT_SUCCESS;
5577 char veth_name[IFNAMSIZ] = "";
5578 struct ExposeArgs expose_args = {};
5579 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
5580 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
5581 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
5582 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
5583 _cleanup_(fw_ctx_freep) FirewallContext *fw_ctx = NULL;
5584 pid_t pid = 0;
5585
5586 log_parse_environment();
5587 log_open();
5588
5589 r = parse_argv(argc, argv);
5590 if (r <= 0)
5591 goto finish;
5592
5593 if (geteuid() != 0) {
5594 r = log_warning_errno(SYNTHETIC_ERRNO(EPERM),
5595 argc >= 2 ? "Need to be root." :
5596 "Need to be root (and some arguments are usually required).\nHint: try --help");
5597 goto finish;
5598 }
5599
5600 r = cant_be_in_netns();
5601 if (r < 0)
5602 goto finish;
5603
5604 r = initialize_rlimits();
5605 if (r < 0)
5606 goto finish;
5607
5608 r = load_oci_bundle();
5609 if (r < 0)
5610 goto finish;
5611
5612 r = pick_paths();
5613 if (r < 0)
5614 goto finish;
5615
5616 r = determine_names();
5617 if (r < 0)
5618 goto finish;
5619
5620 r = load_settings();
5621 if (r < 0)
5622 goto finish;
5623
5624 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
5625 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
5626 * indicate that. */
5627 if (!arg_private_network && arg_userns_mode != USER_NAMESPACE_NO && arg_uid_shift > 0)
5628 arg_caps_retain &= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE);
5629
5630 r = cg_unified();
5631 if (r < 0) {
5632 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5633 goto finish;
5634 }
5635
5636 r = verify_arguments();
5637 if (r < 0)
5638 goto finish;
5639
5640 r = resolve_network_interface_names(arg_network_interfaces);
5641 if (r < 0)
5642 goto finish;
5643
5644 r = verify_network_interfaces_initialized();
5645 if (r < 0)
5646 goto finish;
5647
5648 /* Reapply environment settings. */
5649 (void) detect_unified_cgroup_hierarchy_from_environment();
5650
5651 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5652 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5653 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5654 (void) ignore_signals(SIGPIPE);
5655
5656 n_fd_passed = sd_listen_fds(false);
5657 if (n_fd_passed > 0) {
5658 r = fdset_new_listen_fds(&fds, false);
5659 if (r < 0) {
5660 log_error_errno(r, "Failed to collect file descriptors: %m");
5661 goto finish;
5662 }
5663 }
5664
5665 /* The "default" umask. This is appropriate for most file and directory
5666 * operations performed by nspawn, and is the umask that will be used for
5667 * the child. Functions like copy_devnodes() change the umask temporarily. */
5668 umask(0022);
5669
5670 if (arg_directory) {
5671 assert(!arg_image);
5672
5673 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5674 * /var from the host will propagate into container dynamically (because bad things happen if
5675 * two systems write to the same /var). Let's allow it for the special cases where /var is
5676 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5677 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
5678 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5679 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5680 goto finish;
5681 }
5682
5683 if (arg_ephemeral) {
5684 _cleanup_free_ char *np = NULL;
5685
5686 r = chase_and_update(&arg_directory, 0);
5687 if (r < 0)
5688 goto finish;
5689
5690 /* If the specified path is a mount point we generate the new snapshot immediately
5691 * inside it under a random name. However if the specified is not a mount point we
5692 * create the new snapshot in the parent directory, just next to it. */
5693 r = path_is_mount_point(arg_directory);
5694 if (r < 0) {
5695 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
5696 goto finish;
5697 }
5698 if (r > 0)
5699 r = tempfn_random_child(arg_directory, "machine.", &np);
5700 else
5701 r = tempfn_random(arg_directory, "machine.", &np);
5702 if (r < 0) {
5703 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
5704 goto finish;
5705 }
5706
5707 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5708 * only owned by us and no one else. */
5709 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5710 if (r < 0) {
5711 log_error_errno(r, "Failed to lock %s: %m", np);
5712 goto finish;
5713 }
5714
5715 {
5716 BLOCK_SIGNALS(SIGINT);
5717 r = btrfs_subvol_snapshot_at(AT_FDCWD, arg_directory, AT_FDCWD, np,
5718 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5719 BTRFS_SNAPSHOT_FALLBACK_COPY |
5720 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5721 BTRFS_SNAPSHOT_RECURSIVE |
5722 BTRFS_SNAPSHOT_QUOTA |
5723 BTRFS_SNAPSHOT_SIGINT);
5724 }
5725 if (r == -EINTR) {
5726 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
5727 goto finish;
5728 }
5729 if (r < 0) {
5730 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
5731 goto finish;
5732 }
5733
5734 free_and_replace(arg_directory, np);
5735 remove_directory = true;
5736 } else {
5737 r = chase_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
5738 if (r < 0)
5739 goto finish;
5740
5741 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5742 if (r == -EBUSY) {
5743 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
5744 goto finish;
5745 }
5746 if (r < 0) {
5747 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
5748 goto finish;
5749 }
5750
5751 if (arg_template) {
5752 r = chase_and_update(&arg_template, 0);
5753 if (r < 0)
5754 goto finish;
5755
5756 {
5757 BLOCK_SIGNALS(SIGINT);
5758 r = btrfs_subvol_snapshot_at(AT_FDCWD, arg_template, AT_FDCWD, arg_directory,
5759 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5760 BTRFS_SNAPSHOT_FALLBACK_COPY |
5761 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5762 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
5763 BTRFS_SNAPSHOT_RECURSIVE |
5764 BTRFS_SNAPSHOT_QUOTA |
5765 BTRFS_SNAPSHOT_SIGINT);
5766 }
5767 if (r == -EEXIST)
5768 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5769 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
5770 else if (r == -EINTR) {
5771 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
5772 goto finish;
5773 } else if (r < 0) {
5774 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
5775 goto finish;
5776 } else
5777 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5778 "Populated %s from template %s.", arg_directory, arg_template);
5779 }
5780 }
5781
5782 if (arg_start_mode == START_BOOT) {
5783 _cleanup_free_ char *b = NULL;
5784 const char *p;
5785 int check_os_release, is_os_tree;
5786
5787 if (arg_pivot_root_new) {
5788 b = path_join(arg_directory, arg_pivot_root_new);
5789 if (!b) {
5790 r = log_oom();
5791 goto finish;
5792 }
5793
5794 p = b;
5795 } else
5796 p = arg_directory;
5797
5798 check_os_release = getenv_bool("SYSTEMD_NSPAWN_CHECK_OS_RELEASE");
5799 if (check_os_release < 0 && check_os_release != -ENXIO) {
5800 r = log_error_errno(check_os_release, "Failed to parse $SYSTEMD_NSPAWN_CHECK_OS_RELEASE: %m");
5801 goto finish;
5802 }
5803
5804 is_os_tree = path_is_os_tree(p);
5805 if (is_os_tree == 0 && check_os_release == 0)
5806 log_debug("Directory %s is missing an os-release file, continuing anyway.", p);
5807 else if (is_os_tree <= 0) {
5808 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5809 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
5810 goto finish;
5811 }
5812 } else {
5813 _cleanup_free_ char *p = NULL;
5814
5815 if (arg_pivot_root_new)
5816 p = path_join(arg_directory, arg_pivot_root_new, "/usr/");
5817 else
5818 p = path_join(arg_directory, "/usr/");
5819 if (!p) {
5820 r = log_oom();
5821 goto finish;
5822 }
5823
5824 if (laccess(p, F_OK) < 0) {
5825 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5826 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory);
5827 goto finish;
5828 }
5829 }
5830
5831 } else {
5832 DissectImageFlags dissect_image_flags =
5833 DISSECT_IMAGE_GENERIC_ROOT |
5834 DISSECT_IMAGE_REQUIRE_ROOT |
5835 DISSECT_IMAGE_RELAX_VAR_CHECK |
5836 DISSECT_IMAGE_USR_NO_ROOT |
5837 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
5838 DISSECT_IMAGE_PIN_PARTITION_DEVICES;
5839 assert(arg_image);
5840 assert(!arg_template);
5841
5842 r = chase_and_update(&arg_image, 0);
5843 if (r < 0)
5844 goto finish;
5845
5846 if (arg_ephemeral) {
5847 _cleanup_free_ char *np = NULL;
5848
5849 r = tempfn_random(arg_image, "machine.", &np);
5850 if (r < 0) {
5851 log_error_errno(r, "Failed to generate name for image snapshot: %m");
5852 goto finish;
5853 }
5854
5855 /* Always take an exclusive lock on our own ephemeral copy. */
5856 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5857 if (r < 0) {
5858 log_error_errno(r, "Failed to create image lock: %m");
5859 goto finish;
5860 }
5861
5862 {
5863 BLOCK_SIGNALS(SIGINT);
5864 r = copy_file_full(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600,
5865 FS_NOCOW_FL, FS_NOCOW_FL,
5866 COPY_REFLINK|COPY_CRTIME|COPY_SIGINT,
5867 NULL, NULL);
5868 }
5869 if (r == -EINTR) {
5870 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
5871 goto finish;
5872 }
5873 if (r < 0) {
5874 r = log_error_errno(r, "Failed to copy image file: %m");
5875 goto finish;
5876 }
5877
5878 free_and_replace(arg_image, np);
5879 remove_image = true;
5880 } else {
5881 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5882 if (r == -EBUSY) {
5883 log_error_errno(r, "Disk image %s is currently busy.", arg_image);
5884 goto finish;
5885 }
5886 if (r < 0) {
5887 log_error_errno(r, "Failed to create image lock: %m");
5888 goto finish;
5889 }
5890
5891 r = verity_settings_load(
5892 &arg_verity_settings,
5893 arg_image, NULL, NULL);
5894 if (r < 0) {
5895 log_error_errno(r, "Failed to read verity artefacts for %s: %m", arg_image);
5896 goto finish;
5897 }
5898
5899 if (arg_verity_settings.data_path)
5900 dissect_image_flags |= DISSECT_IMAGE_NO_PARTITION_TABLE;
5901 }
5902
5903 if (!mkdtemp(tmprootdir)) {
5904 r = log_error_errno(errno, "Failed to create temporary directory: %m");
5905 goto finish;
5906 }
5907
5908 remove_tmprootdir = true;
5909
5910 arg_directory = strdup(tmprootdir);
5911 if (!arg_directory) {
5912 r = log_oom();
5913 goto finish;
5914 }
5915
5916 r = loop_device_make_by_path(
5917 arg_image,
5918 arg_read_only ? O_RDONLY : O_RDWR,
5919 /* sector_size= */ UINT32_MAX,
5920 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
5921 LOCK_SH,
5922 &loop);
5923 if (r < 0) {
5924 log_error_errno(r, "Failed to set up loopback block device: %m");
5925 goto finish;
5926 }
5927
5928 r = dissect_loop_device_and_warn(
5929 loop,
5930 &arg_verity_settings,
5931 /* mount_options=*/ NULL,
5932 arg_image_policy ?: &image_policy_container,
5933 dissect_image_flags,
5934 &dissected_image);
5935 if (r == -ENOPKG) {
5936 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5937 log_notice("Note that the disk image needs to\n"
5938 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5939 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5940 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5941 " d) or contain a file system without a partition table\n"
5942 "in order to be bootable with systemd-nspawn.");
5943 goto finish;
5944 }
5945 if (r < 0)
5946 goto finish;
5947
5948 r = dissected_image_load_verity_sig_partition(
5949 dissected_image,
5950 loop->fd,
5951 &arg_verity_settings);
5952 if (r < 0)
5953 goto finish;
5954
5955 if (dissected_image->has_verity && !arg_verity_settings.root_hash && !dissected_image->has_verity_sig)
5956 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5957 "root hash signature found! Proceeding without integrity checking.", arg_image);
5958
5959 r = dissected_image_decrypt_interactively(
5960 dissected_image,
5961 NULL,
5962 &arg_verity_settings,
5963 0);
5964 if (r < 0)
5965 goto finish;
5966
5967 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5968 if (remove_image && unlink(arg_image) >= 0)
5969 remove_image = false;
5970
5971 if (arg_architecture < 0)
5972 arg_architecture = dissected_image_architecture(dissected_image);
5973 }
5974
5975 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5976 if (r < 0)
5977 goto finish;
5978
5979 if (arg_console_mode < 0)
5980 arg_console_mode = isatty(STDIN_FILENO) && isatty(STDOUT_FILENO) ?
5981 CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
5982
5983 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5984 arg_quiet = true;
5985
5986 if (!arg_quiet) {
5987 const char *t = arg_image ?: arg_directory;
5988 _cleanup_free_ char *u = NULL;
5989 (void) terminal_urlify_path(t, t, &u);
5990
5991 log_info("%s %sSpawning container %s on %s.%s\n"
5992 "%s %sPress %sCtrl-]%s three times within 1s to kill container.%s",
5993 special_glyph(SPECIAL_GLYPH_LIGHT_SHADE), ansi_grey(), arg_machine, u ?: t, ansi_normal(),
5994 special_glyph(SPECIAL_GLYPH_LIGHT_SHADE), ansi_grey(), ansi_highlight(), ansi_grey(), ansi_normal());
5995 }
5996
5997 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, SIGRTMIN+18, -1) >= 0);
5998
5999 r = make_reaper_process(true);
6000 if (r < 0) {
6001 log_error_errno(r, "Failed to become subreaper: %m");
6002 goto finish;
6003 }
6004
6005 if (arg_expose_ports) {
6006 r = fw_ctx_new(&fw_ctx);
6007 if (r < 0) {
6008 log_error_errno(r, "Cannot expose configured ports, firewall initialization failed: %m");
6009 goto finish;
6010 }
6011 expose_args.fw_ctx = fw_ctx;
6012 }
6013 for (;;) {
6014 r = run_container(dissected_image,
6015 fds,
6016 veth_name, &veth_created,
6017 &expose_args, &master,
6018 &pid, &ret);
6019 if (r <= 0)
6020 break;
6021 }
6022
6023 finish:
6024 (void) sd_notify(false,
6025 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
6026 "STOPPING=1\nSTATUS=Terminating...");
6027
6028 if (pid > 0)
6029 (void) kill(pid, SIGKILL);
6030
6031 /* Try to flush whatever is still queued in the pty */
6032 if (master >= 0) {
6033 (void) copy_bytes(master, STDOUT_FILENO, UINT64_MAX, 0);
6034 master = safe_close(master);
6035 }
6036
6037 if (pid > 0)
6038 (void) wait_for_terminate(pid, NULL);
6039
6040 pager_close();
6041
6042 if (remove_directory && arg_directory) {
6043 int k;
6044
6045 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
6046 if (k < 0)
6047 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
6048 }
6049
6050 if (remove_image && arg_image) {
6051 if (unlink(arg_image) < 0)
6052 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
6053 }
6054
6055 if (remove_tmprootdir) {
6056 if (rmdir(tmprootdir) < 0)
6057 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
6058 }
6059
6060 if (arg_machine) {
6061 const char *p;
6062
6063 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
6064 (void) rm_rf(p, REMOVE_ROOT);
6065
6066 p = strjoina("/run/systemd/nspawn/unix-export/", arg_machine);
6067 (void) umount2(p, MNT_DETACH|UMOUNT_NOFOLLOW);
6068 (void) rmdir(p);
6069 }
6070
6071 expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET, &expose_args.address4);
6072 expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET6, &expose_args.address6);
6073
6074 if (veth_created)
6075 (void) remove_veth_links(veth_name, arg_network_veth_extra);
6076 (void) remove_bridge(arg_network_zone);
6077
6078 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
6079 expose_port_free_all(arg_expose_ports);
6080 rlimit_free_all(arg_rlimit);
6081 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
6082
6083 if (r < 0)
6084 return r;
6085
6086 return ret;
6087 }
6088
6089 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.