]> 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
Merge pull request #18601 from keszybz/env-assign-cleanup
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #if HAVE_BLKID
4 #endif
5 #include <errno.h>
6 #include <getopt.h>
7 #include <linux/fs.h>
8 #include <linux/loop.h>
9 #if HAVE_SELINUX
10 #include <selinux/selinux.h>
11 #endif
12 #include <stdlib.h>
13 #include <sys/file.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.h>
18 #include <sys/wait.h>
19 #include <termios.h>
20 #include <unistd.h>
21
22 #include "sd-bus.h"
23 #include "sd-daemon.h"
24 #include "sd-id128.h"
25
26 #include "alloc-util.h"
27 #include "barrier.h"
28 #include "base-filesystem.h"
29 #include "blkid-util.h"
30 #include "btrfs-util.h"
31 #include "bus-error.h"
32 #include "bus-util.h"
33 #include "cap-list.h"
34 #include "capability-util.h"
35 #include "cgroup-util.h"
36 #include "copy.h"
37 #include "cpu-set-util.h"
38 #include "dev-setup.h"
39 #include "discover-image.h"
40 #include "dissect-image.h"
41 #include "env-util.h"
42 #include "escape.h"
43 #include "fd-util.h"
44 #include "fdset.h"
45 #include "fileio.h"
46 #include "format-util.h"
47 #include "fs-util.h"
48 #include "gpt.h"
49 #include "hexdecoct.h"
50 #include "hostname-setup.h"
51 #include "hostname-util.h"
52 #include "id128-util.h"
53 #include "io-util.h"
54 #include "log.h"
55 #include "loop-util.h"
56 #include "loopback-setup.h"
57 #include "macro.h"
58 #include "main-func.h"
59 #include "missing_sched.h"
60 #include "mkdir.h"
61 #include "mount-util.h"
62 #include "mountpoint-util.h"
63 #include "namespace-util.h"
64 #include "netlink-util.h"
65 #include "nspawn-cgroup.h"
66 #include "nspawn-creds.h"
67 #include "nspawn-def.h"
68 #include "nspawn-expose-ports.h"
69 #include "nspawn-mount.h"
70 #include "nspawn-network.h"
71 #include "nspawn-oci.h"
72 #include "nspawn-patch-uid.h"
73 #include "nspawn-register.h"
74 #include "nspawn-seccomp.h"
75 #include "nspawn-settings.h"
76 #include "nspawn-setuid.h"
77 #include "nspawn-stub-pid1.h"
78 #include "nulstr-util.h"
79 #include "os-util.h"
80 #include "pager.h"
81 #include "parse-argument.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "pretty-print.h"
85 #include "process-util.h"
86 #include "ptyfwd.h"
87 #include "random-util.h"
88 #include "raw-clone.h"
89 #include "resolve-util.h"
90 #include "rlimit-util.h"
91 #include "rm-rf.h"
92 #if HAVE_SECCOMP
93 #include "seccomp-util.h"
94 #endif
95 #include "selinux-util.h"
96 #include "signal-util.h"
97 #include "socket-util.h"
98 #include "stat-util.h"
99 #include "stdio-util.h"
100 #include "string-table.h"
101 #include "string-util.h"
102 #include "strv.h"
103 #include "sysctl-util.h"
104 #include "terminal-util.h"
105 #include "tmpfile-util.h"
106 #include "umask-util.h"
107 #include "unit-name.h"
108 #include "user-util.h"
109 #include "util.h"
110
111 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
112 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
113
114 #define EXIT_FORCE_RESTART 133
115
116 typedef enum ContainerStatus {
117 CONTAINER_TERMINATED,
118 CONTAINER_REBOOTED,
119 } ContainerStatus;
120
121 static char *arg_directory = NULL;
122 static char *arg_template = NULL;
123 static char *arg_chdir = NULL;
124 static char *arg_pivot_root_new = NULL;
125 static char *arg_pivot_root_old = NULL;
126 static char *arg_user = NULL;
127 static uid_t arg_uid = UID_INVALID;
128 static gid_t arg_gid = GID_INVALID;
129 static gid_t* arg_supplementary_gids = NULL;
130 static size_t arg_n_supplementary_gids = 0;
131 static sd_id128_t arg_uuid = {};
132 static char *arg_machine = NULL; /* The name used by the host to refer to this */
133 static char *arg_hostname = NULL; /* The name the payload sees by default */
134 static const char *arg_selinux_context = NULL;
135 static const char *arg_selinux_apifs_context = NULL;
136 static char *arg_slice = NULL;
137 static bool arg_private_network = false;
138 static bool arg_read_only = false;
139 static StartMode arg_start_mode = START_PID1;
140 static bool arg_ephemeral = false;
141 static LinkJournal arg_link_journal = LINK_AUTO;
142 static bool arg_link_journal_try = false;
143 static uint64_t arg_caps_retain =
144 (1ULL << CAP_AUDIT_CONTROL) |
145 (1ULL << CAP_AUDIT_WRITE) |
146 (1ULL << CAP_CHOWN) |
147 (1ULL << CAP_DAC_OVERRIDE) |
148 (1ULL << CAP_DAC_READ_SEARCH) |
149 (1ULL << CAP_FOWNER) |
150 (1ULL << CAP_FSETID) |
151 (1ULL << CAP_IPC_OWNER) |
152 (1ULL << CAP_KILL) |
153 (1ULL << CAP_LEASE) |
154 (1ULL << CAP_LINUX_IMMUTABLE) |
155 (1ULL << CAP_MKNOD) |
156 (1ULL << CAP_NET_BIND_SERVICE) |
157 (1ULL << CAP_NET_BROADCAST) |
158 (1ULL << CAP_NET_RAW) |
159 (1ULL << CAP_SETFCAP) |
160 (1ULL << CAP_SETGID) |
161 (1ULL << CAP_SETPCAP) |
162 (1ULL << CAP_SETUID) |
163 (1ULL << CAP_SYS_ADMIN) |
164 (1ULL << CAP_SYS_BOOT) |
165 (1ULL << CAP_SYS_CHROOT) |
166 (1ULL << CAP_SYS_NICE) |
167 (1ULL << CAP_SYS_PTRACE) |
168 (1ULL << CAP_SYS_RESOURCE) |
169 (1ULL << CAP_SYS_TTY_CONFIG);
170 static uint64_t arg_caps_ambient = 0;
171 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
172 static CustomMount *arg_custom_mounts = NULL;
173 static size_t arg_n_custom_mounts = 0;
174 static char **arg_setenv = NULL;
175 static bool arg_quiet = false;
176 static bool arg_register = true;
177 static bool arg_keep_unit = false;
178 static char **arg_network_interfaces = NULL;
179 static char **arg_network_macvlan = NULL;
180 static char **arg_network_ipvlan = NULL;
181 static bool arg_network_veth = false;
182 static char **arg_network_veth_extra = NULL;
183 static char *arg_network_bridge = NULL;
184 static char *arg_network_zone = NULL;
185 static char *arg_network_namespace_path = NULL;
186 static PagerFlags arg_pager_flags = 0;
187 static unsigned long arg_personality = PERSONALITY_INVALID;
188 static char *arg_image = NULL;
189 static char *arg_oci_bundle = NULL;
190 static VolatileMode arg_volatile_mode = VOLATILE_NO;
191 static ExposePort *arg_expose_ports = NULL;
192 static char **arg_property = NULL;
193 static sd_bus_message *arg_property_message = NULL;
194 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
195 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
196 static bool arg_userns_chown = false;
197 static int arg_kill_signal = 0;
198 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
199 static SettingsMask arg_settings_mask = 0;
200 static int arg_settings_trusted = -1;
201 static char **arg_parameters = NULL;
202 static const char *arg_container_service_name = "systemd-nspawn";
203 static bool arg_notify_ready = false;
204 static bool arg_use_cgns = true;
205 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
206 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
207 static VeritySettings arg_verity_settings = VERITY_SETTINGS_DEFAULT;
208 static char **arg_syscall_allow_list = NULL;
209 static char **arg_syscall_deny_list = NULL;
210 #if HAVE_SECCOMP
211 static scmp_filter_ctx arg_seccomp = NULL;
212 #endif
213 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
214 static bool arg_no_new_privileges = false;
215 static int arg_oom_score_adjust = 0;
216 static bool arg_oom_score_adjust_set = false;
217 static CPUSet arg_cpu_set = {};
218 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
219 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
220 static unsigned arg_console_width = (unsigned) -1, arg_console_height = (unsigned) -1;
221 static DeviceNode* arg_extra_nodes = NULL;
222 static size_t arg_n_extra_nodes = 0;
223 static char **arg_sysctl = NULL;
224 static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
225 static Credential *arg_credentials = NULL;
226 static size_t arg_n_credentials = 0;
227
228 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
229 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
230 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
231 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
232 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
233 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
234 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
235 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
236 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
237 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
238 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
239 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
240 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
241 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
248 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
250 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings, verity_settings_done);
252 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list, strv_freep);
253 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list, strv_freep);
254 #if HAVE_SECCOMP
255 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
256 #endif
257 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
258 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
259
260 static int handle_arg_console(const char *arg) {
261 if (streq(arg, "help")) {
262 puts("autopipe\n"
263 "interactive\n"
264 "passive\n"
265 "pipe\n"
266 "read-only");
267 return 0;
268 }
269
270 if (streq(arg, "interactive"))
271 arg_console_mode = CONSOLE_INTERACTIVE;
272 else if (streq(arg, "read-only"))
273 arg_console_mode = CONSOLE_READ_ONLY;
274 else if (streq(arg, "passive"))
275 arg_console_mode = CONSOLE_PASSIVE;
276 else if (streq(arg, "pipe")) {
277 if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
278 log_full(arg_quiet ? LOG_DEBUG : LOG_NOTICE,
279 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
280 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
281 "Proceeding anyway.");
282
283 arg_console_mode = CONSOLE_PIPE;
284 } else if (streq(arg, "autopipe")) {
285 if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
286 arg_console_mode = CONSOLE_INTERACTIVE;
287 else
288 arg_console_mode = CONSOLE_PIPE;
289 } else
290 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
291
292 arg_settings_mask |= SETTING_CONSOLE_MODE;
293 return 1;
294 }
295
296 static int help(void) {
297 _cleanup_free_ char *link = NULL;
298 int r;
299
300 (void) pager_open(arg_pager_flags);
301
302 r = terminal_urlify_man("systemd-nspawn", "1", &link);
303 if (r < 0)
304 return log_oom();
305
306 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
307 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
308 " -h --help Show this help\n"
309 " --version Print version string\n"
310 " -q --quiet Do not show status information\n"
311 " --no-pager Do not pipe output into a pager\n"
312 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
313 "%3$sImage:%4$s\n"
314 " -D --directory=PATH Root directory for the container\n"
315 " --template=PATH Initialize root directory from template directory,\n"
316 " if missing\n"
317 " -x --ephemeral Run container with snapshot of root directory, and\n"
318 " remove it after exit\n"
319 " -i --image=PATH Root file system disk image (or device node) for\n"
320 " the container\n"
321 " --oci-bundle=PATH OCI bundle directory\n"
322 " --read-only Mount the root directory read-only\n"
323 " --volatile[=MODE] Run the system in volatile mode\n"
324 " --root-hash=HASH Specify verity root hash for root disk image\n"
325 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
326 " as a DER encoded PKCS7, either as a path to a file\n"
327 " or as an ASCII base64 encoded string prefixed by\n"
328 " 'base64:'\n"
329 " --verity-data=PATH Specify hash device for verity\n"
330 " --pivot-root=PATH[:PATH]\n"
331 " Pivot root to given directory in the container\n\n"
332 "%3$sExecution:%4$s\n"
333 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
334 " -b --boot Boot up full system (i.e. invoke init)\n"
335 " --chdir=PATH Set working directory in the container\n"
336 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
337 " -u --user=USER Run the command under specified user or UID\n"
338 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
339 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
340 "%3$sSystem Identity:%4$s\n"
341 " -M --machine=NAME Set the machine name for the container\n"
342 " --hostname=NAME Override the hostname for the container\n"
343 " --uuid=UUID Set a specific machine UUID for the container\n\n"
344 "%3$sProperties:%4$s\n"
345 " -S --slice=SLICE Place the container in the specified slice\n"
346 " --property=NAME=VALUE Set scope unit property\n"
347 " --register=BOOLEAN Register container as machine\n"
348 " --keep-unit Do not register a scope for the machine, reuse\n"
349 " the service unit nspawn is running in\n\n"
350 "%3$sUser Namespacing:%4$s\n"
351 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
352 " --private-users[=UIDBASE[:NUIDS]]\n"
353 " Similar, but with user configured UID/GID range\n"
354 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
355 "%3$sNetworking:%4$s\n"
356 " --private-network Disable network in container\n"
357 " --network-interface=INTERFACE\n"
358 " Assign an existing network interface to the\n"
359 " container\n"
360 " --network-macvlan=INTERFACE\n"
361 " Create a macvlan network interface based on an\n"
362 " existing network interface to the container\n"
363 " --network-ipvlan=INTERFACE\n"
364 " Create a ipvlan network interface based on an\n"
365 " existing network interface to the container\n"
366 " -n --network-veth Add a virtual Ethernet connection between host\n"
367 " and container\n"
368 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
369 " Add an additional virtual Ethernet link between\n"
370 " host and container\n"
371 " --network-bridge=INTERFACE\n"
372 " Add a virtual Ethernet connection to the container\n"
373 " and attach it to an existing bridge on the host\n"
374 " --network-zone=NAME Similar, but attach the new interface to an\n"
375 " an automatically managed bridge interface\n"
376 " --network-namespace-path=PATH\n"
377 " Set network namespace to the one represented by\n"
378 " the specified kernel namespace file node\n"
379 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
380 " Expose a container IP port on the host\n\n"
381 "%3$sSecurity:%4$s\n"
382 " --capability=CAP In addition to the default, retain specified\n"
383 " capability\n"
384 " --drop-capability=CAP Drop the specified capability from the default set\n"
385 " --ambient-capability=CAP\n"
386 " Sets the specified capability for the started\n"
387 " process. Not useful if booting a machine.\n"
388 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
389 " --system-call-filter=LIST|~LIST\n"
390 " Permit/prohibit specific system calls\n"
391 " -Z --selinux-context=SECLABEL\n"
392 " Set the SELinux security context to be used by\n"
393 " processes in the container\n"
394 " -L --selinux-apifs-context=SECLABEL\n"
395 " Set the SELinux security context to be used by\n"
396 " API/tmpfs file systems in the container\n\n"
397 "%3$sResources:%4$s\n"
398 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
399 " --oom-score-adjust=VALUE\n"
400 " Adjust the OOM score value for the payload\n"
401 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
402 " --personality=ARCH Pick personality for this container\n\n"
403 "%3$sIntegration:%4$s\n"
404 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
405 " --timezone=MODE Select mode of /etc/localtime initialization\n"
406 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
407 " host, try-guest, try-host\n"
408 " -j Equivalent to --link-journal=try-guest\n\n"
409 "%3$sMounts:%4$s\n"
410 " --bind=PATH[:PATH[:OPTIONS]]\n"
411 " Bind mount a file or directory from the host into\n"
412 " the container\n"
413 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
414 " Similar, but creates a read-only bind mount\n"
415 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
416 " it\n"
417 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
418 " --overlay=PATH[:PATH...]:PATH\n"
419 " Create an overlay mount from the host to \n"
420 " the container\n"
421 " --overlay-ro=PATH[:PATH...]:PATH\n"
422 " Similar, but creates a read-only overlay mount\n\n"
423 "%3$sInput/Output:%4$s\n"
424 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
425 " set up for the container.\n"
426 " -P --pipe Equivalent to --console=pipe\n\n"
427 "%3$sCredentials:%4$s\n"
428 " --set-credential=ID:VALUE\n"
429 " Pass a credential with literal value to container.\n"
430 " --load-credential=ID:PATH\n"
431 " Load credential to pass to container from file or\n"
432 " AF_UNIX stream socket.\n"
433 "\nSee the %2$s for details.\n",
434 program_invocation_short_name,
435 link,
436 ansi_underline(),
437 ansi_normal(),
438 ansi_highlight(),
439 ansi_normal());
440
441 return 0;
442 }
443
444 static int custom_mount_check_all(void) {
445 size_t i;
446
447 for (i = 0; i < arg_n_custom_mounts; i++) {
448 CustomMount *m = &arg_custom_mounts[i];
449
450 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
451 if (arg_userns_chown)
452 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
453 "--private-users-chown may not be combined with custom root mounts.");
454 else if (arg_uid_shift == UID_INVALID)
455 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
456 "--private-users with automatic UID shift may not be combined with custom root mounts.");
457 }
458 }
459
460 return 0;
461 }
462
463 static int detect_unified_cgroup_hierarchy_from_environment(void) {
464 const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
465 int r;
466
467 /* Allow the user to control whether the unified hierarchy is used */
468
469 e = getenv(var);
470 if (!e) {
471 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
472 var = "UNIFIED_CGROUP_HIERARCHY";
473 e = getenv(var);
474 }
475
476 if (!isempty(e)) {
477 r = parse_boolean(e);
478 if (r < 0)
479 return log_error_errno(r, "Failed to parse $%s: %m", var);
480 if (r > 0)
481 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
482 else
483 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
484 }
485
486 return 0;
487 }
488
489 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
490 int r;
491
492 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
493 * in the image actually supports. */
494 r = cg_all_unified();
495 if (r < 0)
496 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
497 if (r > 0) {
498 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
499 * routine only detects 231, so we'll have a false negative here for 230. */
500 r = systemd_installation_has_version(directory, 230);
501 if (r < 0)
502 return log_error_errno(r, "Failed to determine systemd version in container: %m");
503 if (r > 0)
504 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
505 else
506 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
507 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
508 /* Mixed cgroup hierarchy support was added in 233 */
509 r = systemd_installation_has_version(directory, 233);
510 if (r < 0)
511 return log_error_errno(r, "Failed to determine systemd version in container: %m");
512 if (r > 0)
513 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
514 else
515 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
516 } else
517 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
518
519 log_debug("Using %s hierarchy for container.",
520 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
521 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
522
523 return 0;
524 }
525
526 static int parse_capability_spec(const char *spec, uint64_t *ret_mask) {
527 uint64_t mask = 0;
528 int r;
529
530 for (;;) {
531 _cleanup_free_ char *t = NULL;
532
533 r = extract_first_word(&spec, &t, ",", 0);
534 if (r < 0)
535 return log_error_errno(r, "Failed to parse capability %s.", t);
536 if (r == 0)
537 break;
538
539 if (streq(t, "help")) {
540 for (int i = 0; i < capability_list_length(); i++) {
541 const char *name;
542
543 name = capability_to_name(i);
544 if (name)
545 puts(name);
546 }
547
548 return 0; /* quit */
549 }
550
551 if (streq(t, "all"))
552 mask = (uint64_t) -1;
553 else {
554 r = capability_from_name(t);
555 if (r < 0)
556 return log_error_errno(r, "Failed to parse capability %s.", t);
557
558 mask |= 1ULL << r;
559 }
560 }
561
562 *ret_mask = mask;
563 return 1; /* continue */
564 }
565
566 static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
567 int r;
568
569 r = getenv_bool(name);
570 if (r == -ENXIO)
571 return 0;
572 if (r < 0)
573 return log_error_errno(r, "Failed to parse $%s: %m", name);
574
575 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
576 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
577 return 0;
578 }
579
580 static int parse_mount_settings_env(void) {
581 const char *e;
582 int r;
583
584 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
585 if (r < 0 && r != -ENXIO)
586 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
587 if (r >= 0)
588 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
589
590 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
591 if (streq_ptr(e, "network"))
592 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
593
594 else if (e) {
595 r = parse_boolean(e);
596 if (r < 0)
597 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
598
599 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
600 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
601 }
602
603 return 0;
604 }
605
606 static int parse_environment(void) {
607 const char *e;
608 int r;
609
610 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
611 if (r < 0)
612 return r;
613 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
614 if (r < 0)
615 return r;
616 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
617 if (r < 0)
618 return r;
619 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
620 if (r < 0)
621 return r;
622
623 r = parse_mount_settings_env();
624 if (r < 0)
625 return r;
626
627 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
628 * even if it is supported. If not supported, it has no effect. */
629 if (!cg_ns_supported())
630 arg_use_cgns = false;
631 else {
632 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
633 if (r < 0) {
634 if (r != -ENXIO)
635 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
636
637 arg_use_cgns = true;
638 } else {
639 arg_use_cgns = r > 0;
640 arg_settings_mask |= SETTING_USE_CGNS;
641 }
642 }
643
644 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
645 if (e)
646 arg_container_service_name = e;
647
648 return detect_unified_cgroup_hierarchy_from_environment();
649 }
650
651 static int parse_argv(int argc, char *argv[]) {
652 enum {
653 ARG_VERSION = 0x100,
654 ARG_PRIVATE_NETWORK,
655 ARG_UUID,
656 ARG_READ_ONLY,
657 ARG_CAPABILITY,
658 ARG_AMBIENT_CAPABILITY,
659 ARG_DROP_CAPABILITY,
660 ARG_LINK_JOURNAL,
661 ARG_BIND,
662 ARG_BIND_RO,
663 ARG_TMPFS,
664 ARG_OVERLAY,
665 ARG_OVERLAY_RO,
666 ARG_INACCESSIBLE,
667 ARG_SHARE_SYSTEM,
668 ARG_REGISTER,
669 ARG_KEEP_UNIT,
670 ARG_NETWORK_INTERFACE,
671 ARG_NETWORK_MACVLAN,
672 ARG_NETWORK_IPVLAN,
673 ARG_NETWORK_BRIDGE,
674 ARG_NETWORK_ZONE,
675 ARG_NETWORK_VETH_EXTRA,
676 ARG_NETWORK_NAMESPACE_PATH,
677 ARG_PERSONALITY,
678 ARG_VOLATILE,
679 ARG_TEMPLATE,
680 ARG_PROPERTY,
681 ARG_PRIVATE_USERS,
682 ARG_KILL_SIGNAL,
683 ARG_SETTINGS,
684 ARG_CHDIR,
685 ARG_PIVOT_ROOT,
686 ARG_PRIVATE_USERS_CHOWN,
687 ARG_NOTIFY_READY,
688 ARG_ROOT_HASH,
689 ARG_ROOT_HASH_SIG,
690 ARG_VERITY_DATA,
691 ARG_SYSTEM_CALL_FILTER,
692 ARG_RLIMIT,
693 ARG_HOSTNAME,
694 ARG_NO_NEW_PRIVILEGES,
695 ARG_OOM_SCORE_ADJUST,
696 ARG_CPU_AFFINITY,
697 ARG_RESOLV_CONF,
698 ARG_TIMEZONE,
699 ARG_CONSOLE,
700 ARG_PIPE,
701 ARG_OCI_BUNDLE,
702 ARG_NO_PAGER,
703 ARG_SET_CREDENTIAL,
704 ARG_LOAD_CREDENTIAL,
705 };
706
707 static const struct option options[] = {
708 { "help", no_argument, NULL, 'h' },
709 { "version", no_argument, NULL, ARG_VERSION },
710 { "directory", required_argument, NULL, 'D' },
711 { "template", required_argument, NULL, ARG_TEMPLATE },
712 { "ephemeral", no_argument, NULL, 'x' },
713 { "user", required_argument, NULL, 'u' },
714 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
715 { "as-pid2", no_argument, NULL, 'a' },
716 { "boot", no_argument, NULL, 'b' },
717 { "uuid", required_argument, NULL, ARG_UUID },
718 { "read-only", no_argument, NULL, ARG_READ_ONLY },
719 { "capability", required_argument, NULL, ARG_CAPABILITY },
720 { "ambient-capability", required_argument, NULL, ARG_AMBIENT_CAPABILITY },
721 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
722 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
723 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
724 { "bind", required_argument, NULL, ARG_BIND },
725 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
726 { "tmpfs", required_argument, NULL, ARG_TMPFS },
727 { "overlay", required_argument, NULL, ARG_OVERLAY },
728 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
729 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
730 { "machine", required_argument, NULL, 'M' },
731 { "hostname", required_argument, NULL, ARG_HOSTNAME },
732 { "slice", required_argument, NULL, 'S' },
733 { "setenv", required_argument, NULL, 'E' },
734 { "selinux-context", required_argument, NULL, 'Z' },
735 { "selinux-apifs-context", required_argument, NULL, 'L' },
736 { "quiet", no_argument, NULL, 'q' },
737 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
738 { "register", required_argument, NULL, ARG_REGISTER },
739 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
740 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
741 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
742 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
743 { "network-veth", no_argument, NULL, 'n' },
744 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
745 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
746 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
747 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
748 { "personality", required_argument, NULL, ARG_PERSONALITY },
749 { "image", required_argument, NULL, 'i' },
750 { "volatile", optional_argument, NULL, ARG_VOLATILE },
751 { "port", required_argument, NULL, 'p' },
752 { "property", required_argument, NULL, ARG_PROPERTY },
753 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
754 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
755 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
756 { "settings", required_argument, NULL, ARG_SETTINGS },
757 { "chdir", required_argument, NULL, ARG_CHDIR },
758 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
759 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
760 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
761 { "root-hash-sig", required_argument, NULL, ARG_ROOT_HASH_SIG },
762 { "verity-data", required_argument, NULL, ARG_VERITY_DATA },
763 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
764 { "rlimit", required_argument, NULL, ARG_RLIMIT },
765 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
766 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
767 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
768 { "timezone", required_argument, NULL, ARG_TIMEZONE },
769 { "console", required_argument, NULL, ARG_CONSOLE },
770 { "pipe", no_argument, NULL, ARG_PIPE },
771 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
772 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
773 { "set-credential", required_argument, NULL, ARG_SET_CREDENTIAL },
774 { "load-credential", required_argument, NULL, ARG_LOAD_CREDENTIAL },
775 {}
776 };
777
778 int c, r;
779 uint64_t plus = 0, minus = 0;
780 bool mask_all_settings = false, mask_no_settings = false;
781
782 assert(argc >= 0);
783 assert(argv);
784
785 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
786 switch (c) {
787
788 case 'h':
789 return help();
790
791 case ARG_VERSION:
792 return version();
793
794 case 'D':
795 r = parse_path_argument(optarg, false, &arg_directory);
796 if (r < 0)
797 return r;
798
799 arg_settings_mask |= SETTING_DIRECTORY;
800 break;
801
802 case ARG_TEMPLATE:
803 r = parse_path_argument(optarg, false, &arg_template);
804 if (r < 0)
805 return r;
806
807 arg_settings_mask |= SETTING_DIRECTORY;
808 break;
809
810 case 'i':
811 r = parse_path_argument(optarg, false, &arg_image);
812 if (r < 0)
813 return r;
814
815 arg_settings_mask |= SETTING_DIRECTORY;
816 break;
817
818 case ARG_OCI_BUNDLE:
819 r = parse_path_argument(optarg, false, &arg_oci_bundle);
820 if (r < 0)
821 return r;
822
823 break;
824
825 case 'x':
826 arg_ephemeral = true;
827 arg_settings_mask |= SETTING_EPHEMERAL;
828 break;
829
830 case 'u':
831 r = free_and_strdup(&arg_user, optarg);
832 if (r < 0)
833 return log_oom();
834
835 arg_settings_mask |= SETTING_USER;
836 break;
837
838 case ARG_NETWORK_ZONE: {
839 char *j;
840
841 j = strjoin("vz-", optarg);
842 if (!j)
843 return log_oom();
844
845 if (!ifname_valid(j)) {
846 log_error("Network zone name not valid: %s", j);
847 free(j);
848 return -EINVAL;
849 }
850
851 free_and_replace(arg_network_zone, j);
852
853 arg_network_veth = true;
854 arg_private_network = true;
855 arg_settings_mask |= SETTING_NETWORK;
856 break;
857 }
858
859 case ARG_NETWORK_BRIDGE:
860
861 if (!ifname_valid(optarg))
862 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
863 "Bridge interface name not valid: %s", optarg);
864
865 r = free_and_strdup(&arg_network_bridge, optarg);
866 if (r < 0)
867 return log_oom();
868
869 _fallthrough_;
870 case 'n':
871 arg_network_veth = true;
872 arg_private_network = true;
873 arg_settings_mask |= SETTING_NETWORK;
874 break;
875
876 case ARG_NETWORK_VETH_EXTRA:
877 r = veth_extra_parse(&arg_network_veth_extra, optarg);
878 if (r < 0)
879 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
880
881 arg_private_network = true;
882 arg_settings_mask |= SETTING_NETWORK;
883 break;
884
885 case ARG_NETWORK_INTERFACE:
886 if (!ifname_valid(optarg))
887 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
888 "Network interface name not valid: %s", optarg);
889
890 r = test_network_interface_initialized(optarg);
891 if (r < 0)
892 return r;
893
894 if (strv_extend(&arg_network_interfaces, optarg) < 0)
895 return log_oom();
896
897 arg_private_network = true;
898 arg_settings_mask |= SETTING_NETWORK;
899 break;
900
901 case ARG_NETWORK_MACVLAN:
902
903 if (!ifname_valid(optarg))
904 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
905 "MACVLAN network interface name not valid: %s", optarg);
906
907 r = test_network_interface_initialized(optarg);
908 if (r < 0)
909 return r;
910
911 if (strv_extend(&arg_network_macvlan, optarg) < 0)
912 return log_oom();
913
914 arg_private_network = true;
915 arg_settings_mask |= SETTING_NETWORK;
916 break;
917
918 case ARG_NETWORK_IPVLAN:
919
920 if (!ifname_valid(optarg))
921 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
922 "IPVLAN network interface name not valid: %s", optarg);
923
924 r = test_network_interface_initialized(optarg);
925 if (r < 0)
926 return r;
927
928 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
929 return log_oom();
930
931 _fallthrough_;
932 case ARG_PRIVATE_NETWORK:
933 arg_private_network = true;
934 arg_settings_mask |= SETTING_NETWORK;
935 break;
936
937 case ARG_NETWORK_NAMESPACE_PATH:
938 r = parse_path_argument(optarg, false, &arg_network_namespace_path);
939 if (r < 0)
940 return r;
941
942 arg_settings_mask |= SETTING_NETWORK;
943 break;
944
945 case 'b':
946 if (arg_start_mode == START_PID2)
947 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
948 "--boot and --as-pid2 may not be combined.");
949
950 arg_start_mode = START_BOOT;
951 arg_settings_mask |= SETTING_START_MODE;
952 break;
953
954 case 'a':
955 if (arg_start_mode == START_BOOT)
956 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
957 "--boot and --as-pid2 may not be combined.");
958
959 arg_start_mode = START_PID2;
960 arg_settings_mask |= SETTING_START_MODE;
961 break;
962
963 case ARG_UUID:
964 r = sd_id128_from_string(optarg, &arg_uuid);
965 if (r < 0)
966 return log_error_errno(r, "Invalid UUID: %s", optarg);
967
968 if (sd_id128_is_null(arg_uuid))
969 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
970 "Machine UUID may not be all zeroes.");
971
972 arg_settings_mask |= SETTING_MACHINE_ID;
973 break;
974
975 case 'S': {
976 _cleanup_free_ char *mangled = NULL;
977
978 r = unit_name_mangle_with_suffix(optarg, NULL, UNIT_NAME_MANGLE_WARN, ".slice", &mangled);
979 if (r < 0)
980 return log_oom();
981
982 free_and_replace(arg_slice, mangled);
983 arg_settings_mask |= SETTING_SLICE;
984 break;
985 }
986
987 case 'M':
988 if (isempty(optarg))
989 arg_machine = mfree(arg_machine);
990 else {
991 if (!hostname_is_valid(optarg, 0))
992 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
993 "Invalid machine name: %s", optarg);
994
995 r = free_and_strdup(&arg_machine, optarg);
996 if (r < 0)
997 return log_oom();
998 }
999 break;
1000
1001 case ARG_HOSTNAME:
1002 if (isempty(optarg))
1003 arg_hostname = mfree(arg_hostname);
1004 else {
1005 if (!hostname_is_valid(optarg, 0))
1006 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1007 "Invalid hostname: %s", optarg);
1008
1009 r = free_and_strdup(&arg_hostname, optarg);
1010 if (r < 0)
1011 return log_oom();
1012 }
1013
1014 arg_settings_mask |= SETTING_HOSTNAME;
1015 break;
1016
1017 case 'Z':
1018 arg_selinux_context = optarg;
1019 break;
1020
1021 case 'L':
1022 arg_selinux_apifs_context = optarg;
1023 break;
1024
1025 case ARG_READ_ONLY:
1026 arg_read_only = true;
1027 arg_settings_mask |= SETTING_READ_ONLY;
1028 break;
1029
1030 case ARG_AMBIENT_CAPABILITY: {
1031 uint64_t m;
1032 r = parse_capability_spec(optarg, &m);
1033 if (r <= 0)
1034 return r;
1035 arg_caps_ambient |= m;
1036 arg_settings_mask |= SETTING_CAPABILITY;
1037 break;
1038 }
1039 case ARG_CAPABILITY:
1040 case ARG_DROP_CAPABILITY: {
1041 uint64_t m;
1042 r = parse_capability_spec(optarg, &m);
1043 if (r <= 0)
1044 return r;
1045
1046 if (c == ARG_CAPABILITY)
1047 plus |= m;
1048 else
1049 minus |= m;
1050 arg_settings_mask |= SETTING_CAPABILITY;
1051 break;
1052 }
1053 case ARG_NO_NEW_PRIVILEGES:
1054 r = parse_boolean(optarg);
1055 if (r < 0)
1056 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
1057
1058 arg_no_new_privileges = r;
1059 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
1060 break;
1061
1062 case 'j':
1063 arg_link_journal = LINK_GUEST;
1064 arg_link_journal_try = true;
1065 arg_settings_mask |= SETTING_LINK_JOURNAL;
1066 break;
1067
1068 case ARG_LINK_JOURNAL:
1069 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
1070 if (r < 0)
1071 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
1072
1073 arg_settings_mask |= SETTING_LINK_JOURNAL;
1074 break;
1075
1076 case ARG_BIND:
1077 case ARG_BIND_RO:
1078 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
1079 if (r < 0)
1080 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
1081
1082 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1083 break;
1084
1085 case ARG_TMPFS:
1086 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1087 if (r < 0)
1088 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
1089
1090 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1091 break;
1092
1093 case ARG_OVERLAY:
1094 case ARG_OVERLAY_RO:
1095 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
1096 if (r == -EADDRNOTAVAIL)
1097 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1098 if (r < 0)
1099 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
1100
1101 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1102 break;
1103
1104 case ARG_INACCESSIBLE:
1105 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1106 if (r < 0)
1107 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
1108
1109 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1110 break;
1111
1112 case 'E': {
1113 if (!env_assignment_is_valid(optarg))
1114 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1115 "Environment variable assignment '%s' is not valid.", optarg);
1116 r = strv_env_replace_strdup(&arg_setenv, optarg);
1117 if (r < 0)
1118 return r;
1119
1120 arg_settings_mask |= SETTING_ENVIRONMENT;
1121 break;
1122 }
1123
1124 case 'q':
1125 arg_quiet = true;
1126 break;
1127
1128 case ARG_SHARE_SYSTEM:
1129 /* We don't officially support this anymore, except for compat reasons. People should use the
1130 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1131 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1132 arg_clone_ns_flags = 0;
1133 break;
1134
1135 case ARG_REGISTER:
1136 r = parse_boolean(optarg);
1137 if (r < 0) {
1138 log_error("Failed to parse --register= argument: %s", optarg);
1139 return r;
1140 }
1141
1142 arg_register = r;
1143 break;
1144
1145 case ARG_KEEP_UNIT:
1146 arg_keep_unit = true;
1147 break;
1148
1149 case ARG_PERSONALITY:
1150
1151 arg_personality = personality_from_string(optarg);
1152 if (arg_personality == PERSONALITY_INVALID)
1153 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1154 "Unknown or unsupported personality '%s'.", optarg);
1155
1156 arg_settings_mask |= SETTING_PERSONALITY;
1157 break;
1158
1159 case ARG_VOLATILE:
1160
1161 if (!optarg)
1162 arg_volatile_mode = VOLATILE_YES;
1163 else if (streq(optarg, "help")) {
1164 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1165 return 0;
1166 } else {
1167 VolatileMode m;
1168
1169 m = volatile_mode_from_string(optarg);
1170 if (m < 0)
1171 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1172 "Failed to parse --volatile= argument: %s", optarg);
1173 else
1174 arg_volatile_mode = m;
1175 }
1176
1177 arg_settings_mask |= SETTING_VOLATILE_MODE;
1178 break;
1179
1180 case 'p':
1181 r = expose_port_parse(&arg_expose_ports, optarg);
1182 if (r == -EEXIST)
1183 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1184 if (r < 0)
1185 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1186
1187 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1188 break;
1189
1190 case ARG_PROPERTY:
1191 if (strv_extend(&arg_property, optarg) < 0)
1192 return log_oom();
1193
1194 break;
1195
1196 case ARG_PRIVATE_USERS: {
1197 int boolean = -1;
1198
1199 if (!optarg)
1200 boolean = true;
1201 else if (!in_charset(optarg, DIGITS))
1202 /* do *not* parse numbers as booleans */
1203 boolean = parse_boolean(optarg);
1204
1205 if (boolean == false) {
1206 /* no: User namespacing off */
1207 arg_userns_mode = USER_NAMESPACE_NO;
1208 arg_uid_shift = UID_INVALID;
1209 arg_uid_range = UINT32_C(0x10000);
1210 } else if (boolean == true) {
1211 /* yes: User namespacing on, UID range is read from root dir */
1212 arg_userns_mode = USER_NAMESPACE_FIXED;
1213 arg_uid_shift = UID_INVALID;
1214 arg_uid_range = UINT32_C(0x10000);
1215 } else if (streq(optarg, "pick")) {
1216 /* pick: User namespacing on, UID range is picked randomly */
1217 arg_userns_mode = USER_NAMESPACE_PICK;
1218 arg_uid_shift = UID_INVALID;
1219 arg_uid_range = UINT32_C(0x10000);
1220 } else {
1221 _cleanup_free_ char *buffer = NULL;
1222 const char *range, *shift;
1223
1224 /* anything else: User namespacing on, UID range is explicitly configured */
1225
1226 range = strchr(optarg, ':');
1227 if (range) {
1228 buffer = strndup(optarg, range - optarg);
1229 if (!buffer)
1230 return log_oom();
1231 shift = buffer;
1232
1233 range++;
1234 r = safe_atou32(range, &arg_uid_range);
1235 if (r < 0)
1236 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
1237 } else
1238 shift = optarg;
1239
1240 r = parse_uid(shift, &arg_uid_shift);
1241 if (r < 0)
1242 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
1243
1244 arg_userns_mode = USER_NAMESPACE_FIXED;
1245 }
1246
1247 if (arg_uid_range <= 0)
1248 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1249 "UID range cannot be 0.");
1250
1251 arg_settings_mask |= SETTING_USERNS;
1252 break;
1253 }
1254
1255 case 'U':
1256 if (userns_supported()) {
1257 arg_userns_mode = USER_NAMESPACE_PICK;
1258 arg_uid_shift = UID_INVALID;
1259 arg_uid_range = UINT32_C(0x10000);
1260
1261 arg_settings_mask |= SETTING_USERNS;
1262 }
1263
1264 break;
1265
1266 case ARG_PRIVATE_USERS_CHOWN:
1267 arg_userns_chown = true;
1268
1269 arg_settings_mask |= SETTING_USERNS;
1270 break;
1271
1272 case ARG_KILL_SIGNAL:
1273 if (streq(optarg, "help")) {
1274 DUMP_STRING_TABLE(signal, int, _NSIG);
1275 return 0;
1276 }
1277
1278 arg_kill_signal = signal_from_string(optarg);
1279 if (arg_kill_signal < 0)
1280 return log_error_errno(arg_kill_signal, "Cannot parse signal: %s", optarg);
1281
1282 arg_settings_mask |= SETTING_KILL_SIGNAL;
1283 break;
1284
1285 case ARG_SETTINGS:
1286
1287 /* no → do not read files
1288 * yes → read files, do not override cmdline, trust only subset
1289 * override → read files, override cmdline, trust only subset
1290 * trusted → read files, do not override cmdline, trust all
1291 */
1292
1293 r = parse_boolean(optarg);
1294 if (r < 0) {
1295 if (streq(optarg, "trusted")) {
1296 mask_all_settings = false;
1297 mask_no_settings = false;
1298 arg_settings_trusted = true;
1299
1300 } else if (streq(optarg, "override")) {
1301 mask_all_settings = false;
1302 mask_no_settings = true;
1303 arg_settings_trusted = -1;
1304 } else
1305 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1306 } else if (r > 0) {
1307 /* yes */
1308 mask_all_settings = false;
1309 mask_no_settings = false;
1310 arg_settings_trusted = -1;
1311 } else {
1312 /* no */
1313 mask_all_settings = true;
1314 mask_no_settings = false;
1315 arg_settings_trusted = false;
1316 }
1317
1318 break;
1319
1320 case ARG_CHDIR:
1321 if (!path_is_absolute(optarg))
1322 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1323 "Working directory %s is not an absolute path.", optarg);
1324
1325 r = free_and_strdup(&arg_chdir, optarg);
1326 if (r < 0)
1327 return log_oom();
1328
1329 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1330 break;
1331
1332 case ARG_PIVOT_ROOT:
1333 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1334 if (r < 0)
1335 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1336
1337 arg_settings_mask |= SETTING_PIVOT_ROOT;
1338 break;
1339
1340 case ARG_NOTIFY_READY:
1341 r = parse_boolean(optarg);
1342 if (r < 0)
1343 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1344 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1345 arg_notify_ready = r;
1346 arg_settings_mask |= SETTING_NOTIFY_READY;
1347 break;
1348
1349 case ARG_ROOT_HASH: {
1350 _cleanup_free_ void *k = NULL;
1351 size_t l;
1352
1353 r = unhexmem(optarg, strlen(optarg), &k, &l);
1354 if (r < 0)
1355 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1356 if (l < sizeof(sd_id128_t))
1357 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128bit long: %s", optarg);
1358
1359 free_and_replace(arg_verity_settings.root_hash, k);
1360 arg_verity_settings.root_hash_size = l;
1361 break;
1362 }
1363
1364 case ARG_ROOT_HASH_SIG: {
1365 char *value;
1366 size_t l;
1367 void *p;
1368
1369 if ((value = startswith(optarg, "base64:"))) {
1370 r = unbase64mem(value, strlen(value), &p, &l);
1371 if (r < 0)
1372 return log_error_errno(r, "Failed to parse root hash signature '%s': %m", optarg);
1373
1374 } else {
1375 r = read_full_file(optarg, (char**) &p, &l);
1376 if (r < 0)
1377 return log_error_errno(r, "Failed parse root hash signature file '%s': %m", optarg);
1378 }
1379
1380 free_and_replace(arg_verity_settings.root_hash_sig, p);
1381 arg_verity_settings.root_hash_sig_size = l;
1382 break;
1383 }
1384
1385 case ARG_VERITY_DATA:
1386 r = parse_path_argument(optarg, false, &arg_verity_settings.data_path);
1387 if (r < 0)
1388 return r;
1389 break;
1390
1391 case ARG_SYSTEM_CALL_FILTER: {
1392 bool negative;
1393 const char *items;
1394
1395 negative = optarg[0] == '~';
1396 items = negative ? optarg + 1 : optarg;
1397
1398 for (;;) {
1399 _cleanup_free_ char *word = NULL;
1400
1401 r = extract_first_word(&items, &word, NULL, 0);
1402 if (r == 0)
1403 break;
1404 if (r == -ENOMEM)
1405 return log_oom();
1406 if (r < 0)
1407 return log_error_errno(r, "Failed to parse system call filter: %m");
1408
1409 if (negative)
1410 r = strv_extend(&arg_syscall_deny_list, word);
1411 else
1412 r = strv_extend(&arg_syscall_allow_list, word);
1413 if (r < 0)
1414 return log_oom();
1415 }
1416
1417 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1418 break;
1419 }
1420
1421 case ARG_RLIMIT: {
1422 const char *eq;
1423 _cleanup_free_ char *name = NULL;
1424 int rl;
1425
1426 if (streq(optarg, "help")) {
1427 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1428 return 0;
1429 }
1430
1431 eq = strchr(optarg, '=');
1432 if (!eq)
1433 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1434 "--rlimit= expects an '=' assignment.");
1435
1436 name = strndup(optarg, eq - optarg);
1437 if (!name)
1438 return log_oom();
1439
1440 rl = rlimit_from_string_harder(name);
1441 if (rl < 0)
1442 return log_error_errno(rl, "Unknown resource limit: %s", name);
1443
1444 if (!arg_rlimit[rl]) {
1445 arg_rlimit[rl] = new0(struct rlimit, 1);
1446 if (!arg_rlimit[rl])
1447 return log_oom();
1448 }
1449
1450 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1451 if (r < 0)
1452 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1453
1454 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1455 break;
1456 }
1457
1458 case ARG_OOM_SCORE_ADJUST:
1459 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1460 if (r < 0)
1461 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1462
1463 arg_oom_score_adjust_set = true;
1464 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1465 break;
1466
1467 case ARG_CPU_AFFINITY: {
1468 CPUSet cpuset;
1469
1470 r = parse_cpu_set(optarg, &cpuset);
1471 if (r < 0)
1472 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
1473
1474 cpu_set_reset(&arg_cpu_set);
1475 arg_cpu_set = cpuset;
1476 arg_settings_mask |= SETTING_CPU_AFFINITY;
1477 break;
1478 }
1479
1480 case ARG_RESOLV_CONF:
1481 if (streq(optarg, "help")) {
1482 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1483 return 0;
1484 }
1485
1486 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1487 if (arg_resolv_conf < 0)
1488 return log_error_errno(arg_resolv_conf,
1489 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1490
1491 arg_settings_mask |= SETTING_RESOLV_CONF;
1492 break;
1493
1494 case ARG_TIMEZONE:
1495 if (streq(optarg, "help")) {
1496 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1497 return 0;
1498 }
1499
1500 arg_timezone = timezone_mode_from_string(optarg);
1501 if (arg_timezone < 0)
1502 return log_error_errno(arg_timezone,
1503 "Failed to parse /etc/localtime mode: %s", optarg);
1504
1505 arg_settings_mask |= SETTING_TIMEZONE;
1506 break;
1507
1508 case ARG_CONSOLE:
1509 r = handle_arg_console(optarg);
1510 if (r <= 0)
1511 return r;
1512 break;
1513
1514 case 'P':
1515 case ARG_PIPE:
1516 r = handle_arg_console("pipe");
1517 if (r <= 0)
1518 return r;
1519 break;
1520
1521 case ARG_NO_PAGER:
1522 arg_pager_flags |= PAGER_DISABLE;
1523 break;
1524
1525 case ARG_SET_CREDENTIAL: {
1526 _cleanup_free_ char *word = NULL, *data = NULL;
1527 const char *p = optarg;
1528 Credential *a;
1529 size_t i;
1530 int l;
1531
1532 r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1533 if (r == -ENOMEM)
1534 return log_oom();
1535 if (r < 0)
1536 return log_error_errno(r, "Failed to parse --set-credential= parameter: %m");
1537 if (r == 0 || !p)
1538 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --set-credential=: %s", optarg);
1539
1540 if (!credential_name_valid(word))
1541 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
1542
1543 for (i = 0; i < arg_n_credentials; i++)
1544 if (streq(arg_credentials[i].id, word))
1545 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
1546
1547 l = cunescape(p, UNESCAPE_ACCEPT_NUL, &data);
1548 if (l < 0)
1549 return log_error_errno(l, "Failed to unescape credential data: %s", p);
1550
1551 a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
1552 if (!a)
1553 return log_oom();
1554
1555 a[arg_n_credentials++] = (Credential) {
1556 .id = TAKE_PTR(word),
1557 .data = TAKE_PTR(data),
1558 .size = l,
1559 };
1560
1561 arg_credentials = a;
1562
1563 arg_settings_mask |= SETTING_CREDENTIALS;
1564 break;
1565 }
1566
1567 case ARG_LOAD_CREDENTIAL: {
1568 ReadFullFileFlags flags = READ_FULL_FILE_SECURE;
1569 _cleanup_(erase_and_freep) char *data = NULL;
1570 _cleanup_free_ char *word = NULL, *j = NULL;
1571 const char *p = optarg;
1572 Credential *a;
1573 size_t size, i;
1574
1575 r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1576 if (r == -ENOMEM)
1577 return log_oom();
1578 if (r < 0)
1579 return log_error_errno(r, "Failed to parse --set-credential= parameter: %m");
1580 if (r == 0 || !p)
1581 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --set-credential=: %s", optarg);
1582
1583 if (!credential_name_valid(word))
1584 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
1585
1586 for (i = 0; i < arg_n_credentials; i++)
1587 if (streq(arg_credentials[i].id, word))
1588 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
1589
1590 if (path_is_absolute(p))
1591 flags |= READ_FULL_FILE_CONNECT_SOCKET;
1592 else {
1593 const char *e;
1594
1595 e = getenv("CREDENTIALS_DIRECTORY");
1596 if (!e)
1597 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential not available (no credentials passed at all): %s", word);
1598
1599 j = path_join(e, p);
1600 if (!j)
1601 return log_oom();
1602 }
1603
1604 r = read_full_file_full(AT_FDCWD, j ?: p, UINT64_MAX, SIZE_MAX,
1605 flags,
1606 NULL,
1607 &data, &size);
1608 if (r < 0)
1609 return log_error_errno(r, "Failed to read credential '%s': %m", j ?: p);
1610
1611 a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
1612 if (!a)
1613 return log_oom();
1614
1615 a[arg_n_credentials++] = (Credential) {
1616 .id = TAKE_PTR(word),
1617 .data = TAKE_PTR(data),
1618 .size = size,
1619 };
1620
1621 arg_credentials = a;
1622
1623 arg_settings_mask |= SETTING_CREDENTIALS;
1624 break;
1625 }
1626
1627 case '?':
1628 return -EINVAL;
1629
1630 default:
1631 assert_not_reached("Unhandled option");
1632 }
1633
1634 if (argc > optind) {
1635 strv_free(arg_parameters);
1636 arg_parameters = strv_copy(argv + optind);
1637 if (!arg_parameters)
1638 return log_oom();
1639
1640 arg_settings_mask |= SETTING_START_MODE;
1641 }
1642
1643 if (arg_ephemeral && arg_template && !arg_directory)
1644 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1645 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1646 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1647 * --directory=". */
1648 arg_directory = TAKE_PTR(arg_template);
1649
1650 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0)) & ~minus;
1651
1652 /* Make sure to parse environment before we reset the settings mask below */
1653 r = parse_environment();
1654 if (r < 0)
1655 return r;
1656
1657 /* Load all settings from .nspawn files */
1658 if (mask_no_settings)
1659 arg_settings_mask = 0;
1660
1661 /* Don't load any settings from .nspawn files */
1662 if (mask_all_settings)
1663 arg_settings_mask = _SETTINGS_MASK_ALL;
1664
1665 return 1;
1666 }
1667
1668 static int verify_arguments(void) {
1669 int r;
1670
1671 if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
1672 /* If we are running the stub init in the container, we don't need to look at what the init
1673 * in the container supports, because we are not using it. Let's immediately pick the right
1674 * setting based on the host system configuration.
1675 *
1676 * We only do this, if the user didn't use an environment variable to override the detection.
1677 */
1678
1679 r = cg_all_unified();
1680 if (r < 0)
1681 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
1682 if (r > 0)
1683 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
1684 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
1685 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
1686 else
1687 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
1688 }
1689
1690 if (arg_userns_mode != USER_NAMESPACE_NO)
1691 arg_mount_settings |= MOUNT_USE_USERNS;
1692
1693 if (arg_private_network)
1694 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1695
1696 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1697 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1698 arg_register = false;
1699 if (arg_start_mode != START_PID1)
1700 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1701 }
1702
1703 if (arg_userns_mode == USER_NAMESPACE_PICK)
1704 arg_userns_chown = true;
1705
1706 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1707 arg_kill_signal = SIGRTMIN+3;
1708
1709 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 */
1710 arg_read_only = true;
1711
1712 if (has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts))
1713 arg_read_only = true;
1714
1715 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1716 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1717 * The latter is not technically a user session, but we don't need to labour the point. */
1718 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1719
1720 if (arg_directory && arg_image)
1721 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1722
1723 if (arg_template && arg_image)
1724 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1725
1726 if (arg_template && !(arg_directory || arg_machine))
1727 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1728
1729 if (arg_ephemeral && arg_template)
1730 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1731
1732 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
1733 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
1734
1735 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1736 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1737
1738 if (arg_userns_chown && arg_read_only)
1739 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1740 "--read-only and --private-users-chown may not be combined.");
1741
1742 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1743 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1744 * copy-up (in case of overlay) making the entire exercise pointless. */
1745 if (arg_userns_chown && arg_volatile_mode != VOLATILE_NO)
1746 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-chown may not be combined.");
1747
1748 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1749 * we need to error out, to avoid conflicts between different network options. */
1750 if (arg_network_namespace_path &&
1751 (arg_network_interfaces || arg_network_macvlan ||
1752 arg_network_ipvlan || arg_network_veth_extra ||
1753 arg_network_bridge || arg_network_zone ||
1754 arg_network_veth))
1755 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1756
1757 if (arg_network_bridge && arg_network_zone)
1758 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1759 "--network-bridge= and --network-zone= may not be combined.");
1760
1761 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1762 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1763
1764 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1765 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1766
1767 if (arg_expose_ports && !arg_private_network)
1768 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1769
1770 if (arg_caps_ambient) {
1771 if (arg_caps_ambient == (uint64_t)-1)
1772 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= does not support the value all.");
1773
1774 if ((arg_caps_ambient & arg_caps_retain) != arg_caps_ambient)
1775 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not fully covered by Capability= setting.");
1776
1777 if (arg_start_mode == START_BOOT)
1778 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not useful for boot mode.");
1779 }
1780
1781 r = custom_mount_check_all();
1782 if (r < 0)
1783 return r;
1784
1785 return 0;
1786 }
1787
1788 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1789 assert(p);
1790
1791 if (arg_userns_mode == USER_NAMESPACE_NO)
1792 return 0;
1793
1794 if (uid == UID_INVALID && gid == GID_INVALID)
1795 return 0;
1796
1797 if (uid != UID_INVALID) {
1798 uid += arg_uid_shift;
1799
1800 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1801 return -EOVERFLOW;
1802 }
1803
1804 if (gid != GID_INVALID) {
1805 gid += (gid_t) arg_uid_shift;
1806
1807 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1808 return -EOVERFLOW;
1809 }
1810
1811 if (lchown(p, uid, gid) < 0)
1812 return -errno;
1813
1814 return 0;
1815 }
1816
1817 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1818 const char *q;
1819 int r;
1820
1821 q = prefix_roota(root, path);
1822 r = mkdir_errno_wrapper(q, mode);
1823 if (r == -EEXIST)
1824 return 0;
1825 if (r < 0)
1826 return r;
1827
1828 return userns_lchown(q, uid, gid);
1829 }
1830
1831 static const char *timezone_from_path(const char *path) {
1832 return PATH_STARTSWITH_SET(
1833 path,
1834 "../usr/share/zoneinfo/",
1835 "/usr/share/zoneinfo/");
1836 }
1837
1838 static bool etc_writable(void) {
1839 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1840 }
1841
1842 static int setup_timezone(const char *dest) {
1843 _cleanup_free_ char *p = NULL, *etc = NULL;
1844 const char *where, *check;
1845 TimezoneMode m;
1846 int r;
1847
1848 assert(dest);
1849
1850 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1851 r = readlink_malloc("/etc/localtime", &p);
1852 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1853 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1854 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1855 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1856 else if (r < 0) {
1857 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1858 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1859 * file.
1860 *
1861 * Example:
1862 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1863 */
1864 return 0;
1865 } else if (arg_timezone == TIMEZONE_AUTO)
1866 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1867 else
1868 m = arg_timezone;
1869 } else
1870 m = arg_timezone;
1871
1872 if (m == TIMEZONE_OFF)
1873 return 0;
1874
1875 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1876 if (r < 0) {
1877 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1878 return 0;
1879 }
1880
1881 where = strjoina(etc, "/localtime");
1882
1883 switch (m) {
1884
1885 case TIMEZONE_DELETE:
1886 if (unlink(where) < 0)
1887 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1888
1889 return 0;
1890
1891 case TIMEZONE_SYMLINK: {
1892 _cleanup_free_ char *q = NULL;
1893 const char *z, *what;
1894
1895 z = timezone_from_path(p);
1896 if (!z) {
1897 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1898 return 0;
1899 }
1900
1901 r = readlink_malloc(where, &q);
1902 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1903 return 0; /* Already pointing to the right place? Then do nothing .. */
1904
1905 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1906 r = chase_symlinks(check, dest, 0, NULL, NULL);
1907 if (r < 0)
1908 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1909 else {
1910 if (unlink(where) < 0 && errno != ENOENT) {
1911 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1912 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1913 return 0;
1914 }
1915
1916 what = strjoina("../usr/share/zoneinfo/", z);
1917 if (symlink(what, where) < 0) {
1918 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1919 errno, "Failed to correct timezone of container, ignoring: %m");
1920 return 0;
1921 }
1922
1923 break;
1924 }
1925
1926 _fallthrough_;
1927 }
1928
1929 case TIMEZONE_BIND: {
1930 _cleanup_free_ char *resolved = NULL;
1931 int found;
1932
1933 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
1934 if (found < 0) {
1935 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1936 return 0;
1937 }
1938
1939 if (found == 0) /* missing? */
1940 (void) touch(resolved);
1941
1942 r = mount_nofollow_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1943 if (r >= 0)
1944 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1945
1946 _fallthrough_;
1947 }
1948
1949 case TIMEZONE_COPY:
1950 /* If mounting failed, try to copy */
1951 r = copy_file_atomic("/etc/localtime", where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
1952 if (r < 0) {
1953 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1954 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
1955 return 0;
1956 }
1957
1958 break;
1959
1960 default:
1961 assert_not_reached("unexpected mode");
1962 }
1963
1964 /* Fix permissions of the symlink or file copy we just created */
1965 r = userns_lchown(where, 0, 0);
1966 if (r < 0)
1967 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
1968
1969 return 0;
1970 }
1971
1972 static int have_resolv_conf(const char *path) {
1973 assert(path);
1974
1975 if (access(path, F_OK) < 0) {
1976 if (errno == ENOENT)
1977 return 0;
1978
1979 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1980 }
1981
1982 return 1;
1983 }
1984
1985 static int resolved_listening(void) {
1986 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1987 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1988 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1989 int r;
1990
1991 /* Check if resolved is listening */
1992
1993 r = sd_bus_open_system(&bus);
1994 if (r < 0)
1995 return log_debug_errno(r, "Failed to open system bus: %m");
1996
1997 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1998 if (r < 0)
1999 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2000 if (r == 0)
2001 return 0;
2002
2003 r = sd_bus_get_property_string(bus,
2004 "org.freedesktop.resolve1",
2005 "/org/freedesktop/resolve1",
2006 "org.freedesktop.resolve1.Manager",
2007 "DNSStubListener",
2008 &error,
2009 &dns_stub_listener_mode);
2010 if (r < 0)
2011 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
2012
2013 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
2014 }
2015
2016 static int setup_resolv_conf(const char *dest) {
2017 _cleanup_free_ char *etc = NULL;
2018 const char *where, *what;
2019 ResolvConfMode m;
2020 int r;
2021
2022 assert(dest);
2023
2024 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
2025 if (arg_private_network)
2026 m = RESOLV_CONF_OFF;
2027 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF) > 0 && resolved_listening() > 0)
2028 m = etc_writable() ? RESOLV_CONF_COPY_STUB : RESOLV_CONF_BIND_STUB;
2029 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2030 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
2031 else
2032 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
2033
2034 } else
2035 m = arg_resolv_conf;
2036
2037 if (m == RESOLV_CONF_OFF)
2038 return 0;
2039
2040 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
2041 if (r < 0) {
2042 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
2043 return 0;
2044 }
2045
2046 where = strjoina(etc, "/resolv.conf");
2047
2048 if (m == RESOLV_CONF_DELETE) {
2049 if (unlink(where) < 0)
2050 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
2051
2052 return 0;
2053 }
2054
2055 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_COPY_STATIC))
2056 what = PRIVATE_STATIC_RESOLV_CONF;
2057 else if (IN_SET(m, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_COPY_UPLINK))
2058 what = PRIVATE_UPLINK_RESOLV_CONF;
2059 else if (IN_SET(m, RESOLV_CONF_BIND_STUB, RESOLV_CONF_REPLACE_STUB, RESOLV_CONF_COPY_STUB))
2060 what = PRIVATE_STUB_RESOLV_CONF;
2061 else
2062 what = "/etc/resolv.conf";
2063
2064 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_BIND_STUB)) {
2065 _cleanup_free_ char *resolved = NULL;
2066 int found;
2067
2068 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
2069 if (found < 0) {
2070 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2071 return 0;
2072 }
2073
2074 if (found == 0) /* missing? */
2075 (void) touch(resolved);
2076
2077 r = mount_nofollow_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
2078 if (r >= 0)
2079 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
2080
2081 /* If that didn't work, let's copy the file */
2082 }
2083
2084 if (IN_SET(m, RESOLV_CONF_REPLACE_HOST, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_REPLACE_STUB))
2085 r = copy_file_atomic(what, where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
2086 else
2087 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, 0, COPY_REFLINK);
2088 if (r < 0) {
2089 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2090 * resolved or something similar runs inside and the symlink points there.
2091 *
2092 * If the disk image is read-only, there's also no point in complaining.
2093 */
2094 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC, RESOLV_CONF_COPY_UPLINK, RESOLV_CONF_COPY_STUB) &&
2095 IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
2096 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
2097 return 0;
2098 }
2099
2100 r = userns_lchown(where, 0, 0);
2101 if (r < 0)
2102 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
2103
2104 return 0;
2105 }
2106
2107 static int setup_boot_id(void) {
2108 _cleanup_(unlink_and_freep) char *from = NULL;
2109 _cleanup_free_ char *path = NULL;
2110 sd_id128_t rnd = SD_ID128_NULL;
2111 const char *to;
2112 int r;
2113
2114 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2115
2116 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
2117 if (r < 0)
2118 return log_error_errno(r, "Failed to generate random boot ID path: %m");
2119
2120 r = sd_id128_randomize(&rnd);
2121 if (r < 0)
2122 return log_error_errno(r, "Failed to generate random boot id: %m");
2123
2124 r = id128_write(path, ID128_UUID, rnd, false);
2125 if (r < 0)
2126 return log_error_errno(r, "Failed to write boot id: %m");
2127
2128 from = TAKE_PTR(path);
2129 to = "/proc/sys/kernel/random/boot_id";
2130
2131 r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
2132 if (r < 0)
2133 return r;
2134
2135 return mount_nofollow_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2136 }
2137
2138 static int copy_devnodes(const char *dest) {
2139 static const char devnodes[] =
2140 "null\0"
2141 "zero\0"
2142 "full\0"
2143 "random\0"
2144 "urandom\0"
2145 "tty\0"
2146 "net/tun\0";
2147
2148 _cleanup_umask_ mode_t u;
2149 const char *d;
2150 int r = 0;
2151
2152 assert(dest);
2153
2154 u = umask(0000);
2155
2156 /* Create /dev/net, so that we can create /dev/net/tun in it */
2157 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
2158 return log_error_errno(r, "Failed to create /dev/net directory: %m");
2159
2160 NULSTR_FOREACH(d, devnodes) {
2161 _cleanup_free_ char *from = NULL, *to = NULL;
2162 struct stat st;
2163
2164 from = path_join("/dev/", d);
2165 if (!from)
2166 return log_oom();
2167
2168 to = path_join(dest, from);
2169 if (!to)
2170 return log_oom();
2171
2172 if (stat(from, &st) < 0) {
2173
2174 if (errno != ENOENT)
2175 return log_error_errno(errno, "Failed to stat %s: %m", from);
2176
2177 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
2178 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2179 "%s is not a char or block device, cannot copy.", from);
2180 else {
2181 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
2182
2183 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
2184 /* Explicitly warn the user when /dev is already populated. */
2185 if (errno == EEXIST)
2186 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);
2187 if (errno != EPERM)
2188 return log_error_errno(errno, "mknod(%s) failed: %m", to);
2189
2190 /* Some systems abusively restrict mknod but allow bind mounts. */
2191 r = touch(to);
2192 if (r < 0)
2193 return log_error_errno(r, "touch (%s) failed: %m", to);
2194 r = mount_nofollow_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
2195 if (r < 0)
2196 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
2197 }
2198
2199 r = userns_lchown(to, 0, 0);
2200 if (r < 0)
2201 return log_error_errno(r, "chown() of device node %s failed: %m", to);
2202
2203 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
2204 if (!dn)
2205 return log_oom();
2206
2207 r = userns_mkdir(dest, dn, 0755, 0, 0);
2208 if (r < 0)
2209 return log_error_errno(r, "Failed to create '%s': %m", dn);
2210
2211 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
2212 return log_oom();
2213
2214 prefixed = path_join(dest, sl);
2215 if (!prefixed)
2216 return log_oom();
2217
2218 t = path_join("..", d);
2219 if (!t)
2220 return log_oom();
2221
2222 if (symlink(t, prefixed) < 0)
2223 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
2224 }
2225 }
2226
2227 return r;
2228 }
2229
2230 static int make_extra_nodes(const char *dest) {
2231 _cleanup_umask_ mode_t u;
2232 size_t i;
2233 int r;
2234
2235 u = 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 = mkdir_errno_wrapper(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 = -1;
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 static int setup_credentials(const char *root) {
2369 const char *q;
2370 int r;
2371
2372 if (arg_n_credentials <= 0)
2373 return 0;
2374
2375 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2376 if (r < 0)
2377 return log_error_errno(r, "Failed to create /run/host: %m");
2378
2379 r = userns_mkdir(root, "/run/host/credentials", 0700, 0, 0);
2380 if (r < 0)
2381 return log_error_errno(r, "Failed to create /run/host/credentials: %m");
2382
2383 q = prefix_roota(root, "/run/host/credentials");
2384 r = mount_nofollow_verbose(LOG_ERR, NULL, q, "ramfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0700");
2385 if (r < 0)
2386 return r;
2387
2388 for (size_t i = 0; i < arg_n_credentials; i++) {
2389 _cleanup_free_ char *j = NULL;
2390 _cleanup_close_ int fd = -1;
2391
2392 j = path_join(q, arg_credentials[i].id);
2393 if (!j)
2394 return log_oom();
2395
2396 fd = open(j, O_CREAT|O_EXCL|O_WRONLY|O_CLOEXEC|O_NOFOLLOW, 0600);
2397 if (fd < 0)
2398 return log_error_errno(errno, "Failed to create credential file %s: %m", j);
2399
2400 r = loop_write(fd, arg_credentials[i].data, arg_credentials[i].size, /* do_poll= */ false);
2401 if (r < 0)
2402 return log_error_errno(r, "Failed to write credential to file %s: %m", j);
2403
2404 if (fchmod(fd, 0400) < 0)
2405 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", j);
2406
2407 if (arg_userns_mode != USER_NAMESPACE_NO) {
2408 if (fchown(fd, arg_uid_shift, arg_uid_shift) < 0)
2409 return log_error_errno(errno, "Failed to adjust ownership of %s: %m", j);
2410 }
2411 }
2412
2413 if (chmod(q, 0500) < 0)
2414 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", q);
2415
2416 r = userns_lchown(q, 0, 0);
2417 if (r < 0)
2418 return r;
2419
2420 /* Make both mount and superblock read-only now */
2421 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2422 if (r < 0)
2423 return r;
2424
2425 return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0500");
2426 }
2427
2428 static int setup_kmsg(int kmsg_socket) {
2429 _cleanup_(unlink_and_freep) char *from = NULL;
2430 _cleanup_free_ char *fifo = NULL;
2431 _cleanup_close_ int fd = -1;
2432 _cleanup_umask_ mode_t u;
2433 int r;
2434
2435 assert(kmsg_socket >= 0);
2436
2437 u = umask(0000);
2438
2439 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2440 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2441 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2442 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2443
2444 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2445 if (r < 0)
2446 return log_error_errno(r, "Failed to generate kmsg path: %m");
2447
2448 if (mkfifo(fifo, 0600) < 0)
2449 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2450
2451 from = TAKE_PTR(fifo);
2452
2453 r = mount_nofollow_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2454 if (r < 0)
2455 return r;
2456
2457 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2458 if (fd < 0)
2459 return log_error_errno(errno, "Failed to open fifo: %m");
2460
2461 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2462 r = send_one_fd(kmsg_socket, fd, 0);
2463 if (r < 0)
2464 return log_error_errno(r, "Failed to send FIFO fd: %m");
2465
2466 return 0;
2467 }
2468
2469 struct ExposeArgs {
2470 union in_addr_union address;
2471 struct FirewallContext *fw_ctx;
2472 };
2473
2474 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2475 struct ExposeArgs *args = userdata;
2476
2477 assert(rtnl);
2478 assert(m);
2479 assert(args);
2480
2481 expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, &args->address);
2482 return 0;
2483 }
2484
2485 static int setup_hostname(void) {
2486 int r;
2487
2488 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2489 return 0;
2490
2491 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2492 if (r < 0)
2493 return log_error_errno(r, "Failed to set hostname: %m");
2494
2495 return 0;
2496 }
2497
2498 static int setup_journal(const char *directory) {
2499 _cleanup_free_ char *d = NULL;
2500 char id[SD_ID128_STRING_MAX];
2501 const char *dirname, *p, *q;
2502 sd_id128_t this_id;
2503 bool try;
2504 int r;
2505
2506 /* Don't link journals in ephemeral mode */
2507 if (arg_ephemeral)
2508 return 0;
2509
2510 if (arg_link_journal == LINK_NO)
2511 return 0;
2512
2513 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2514
2515 r = sd_id128_get_machine(&this_id);
2516 if (r < 0)
2517 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2518
2519 if (sd_id128_equal(arg_uuid, this_id)) {
2520 log_full(try ? LOG_WARNING : LOG_ERR,
2521 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
2522 if (try)
2523 return 0;
2524 return -EEXIST;
2525 }
2526
2527 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2528 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2529 if (r < 0) {
2530 bool ignore = r == -EROFS && try;
2531 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2532 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2533 return ignore ? 0 : r;
2534 }
2535 }
2536
2537 (void) sd_id128_to_string(arg_uuid, id);
2538
2539 p = strjoina("/var/log/journal/", id);
2540 q = prefix_roota(directory, p);
2541
2542 if (path_is_mount_point(p, NULL, 0) > 0) {
2543 if (try)
2544 return 0;
2545
2546 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2547 "%s: already a mount point, refusing to use for journal", p);
2548 }
2549
2550 if (path_is_mount_point(q, NULL, 0) > 0) {
2551 if (try)
2552 return 0;
2553
2554 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2555 "%s: already a mount point, refusing to use for journal", q);
2556 }
2557
2558 r = readlink_and_make_absolute(p, &d);
2559 if (r >= 0) {
2560 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2561 path_equal(d, q)) {
2562
2563 r = userns_mkdir(directory, p, 0755, 0, 0);
2564 if (r < 0)
2565 log_warning_errno(r, "Failed to create directory %s: %m", q);
2566 return 0;
2567 }
2568
2569 if (unlink(p) < 0)
2570 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2571 } else if (r == -EINVAL) {
2572
2573 if (arg_link_journal == LINK_GUEST &&
2574 rmdir(p) < 0) {
2575
2576 if (errno == ENOTDIR) {
2577 log_error("%s already exists and is neither a symlink nor a directory", p);
2578 return r;
2579 } else
2580 return log_error_errno(errno, "Failed to remove %s: %m", p);
2581 }
2582 } else if (r != -ENOENT)
2583 return log_error_errno(r, "readlink(%s) failed: %m", p);
2584
2585 if (arg_link_journal == LINK_GUEST) {
2586
2587 if (symlink(q, p) < 0) {
2588 if (try) {
2589 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2590 return 0;
2591 } else
2592 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2593 }
2594
2595 r = userns_mkdir(directory, p, 0755, 0, 0);
2596 if (r < 0)
2597 log_warning_errno(r, "Failed to create directory %s: %m", q);
2598 return 0;
2599 }
2600
2601 if (arg_link_journal == LINK_HOST) {
2602 /* don't create parents here — if the host doesn't have
2603 * permanent journal set up, don't force it here */
2604
2605 r = mkdir_errno_wrapper(p, 0755);
2606 if (r < 0 && r != -EEXIST) {
2607 if (try) {
2608 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2609 return 0;
2610 } else
2611 return log_error_errno(r, "Failed to create %s: %m", p);
2612 }
2613
2614 } else if (access(p, F_OK) < 0)
2615 return 0;
2616
2617 if (dir_is_empty(q) == 0)
2618 log_warning("%s is not empty, proceeding anyway.", q);
2619
2620 r = userns_mkdir(directory, p, 0755, 0, 0);
2621 if (r < 0)
2622 return log_error_errno(r, "Failed to create %s: %m", q);
2623
2624 r = mount_nofollow_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2625 if (r < 0)
2626 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2627
2628 return 0;
2629 }
2630
2631 static int drop_capabilities(uid_t uid) {
2632 CapabilityQuintet q;
2633
2634 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2635 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2636 * arg_caps_retain. */
2637
2638 if (capability_quintet_is_set(&arg_full_capabilities)) {
2639 q = arg_full_capabilities;
2640
2641 if (q.bounding == (uint64_t) -1)
2642 q.bounding = uid == 0 ? arg_caps_retain : 0;
2643
2644 if (q.effective == (uint64_t) -1)
2645 q.effective = uid == 0 ? q.bounding : 0;
2646
2647 if (q.inheritable == (uint64_t) -1)
2648 q.inheritable = uid == 0 ? q.bounding : arg_caps_ambient;
2649
2650 if (q.permitted == (uint64_t) -1)
2651 q.permitted = uid == 0 ? q.bounding : arg_caps_ambient;
2652
2653 if (q.ambient == (uint64_t) -1 && ambient_capabilities_supported())
2654 q.ambient = arg_caps_ambient;
2655
2656 if (capability_quintet_mangle(&q))
2657 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2658
2659 } else {
2660 q = (CapabilityQuintet) {
2661 .bounding = arg_caps_retain,
2662 .effective = uid == 0 ? arg_caps_retain : 0,
2663 .inheritable = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2664 .permitted = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2665 .ambient = ambient_capabilities_supported() ? arg_caps_ambient : (uint64_t) -1,
2666 };
2667
2668 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2669 * in order to maintain the same behavior as systemd < 242. */
2670 if (capability_quintet_mangle(&q))
2671 log_full(arg_quiet ? LOG_DEBUG : LOG_WARNING,
2672 "Some capabilities will not be set because they are not in the current bounding set.");
2673
2674 }
2675
2676 return capability_quintet_enforce(&q);
2677 }
2678
2679 static int reset_audit_loginuid(void) {
2680 _cleanup_free_ char *p = NULL;
2681 int r;
2682
2683 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2684 return 0;
2685
2686 r = read_one_line_file("/proc/self/loginuid", &p);
2687 if (r == -ENOENT)
2688 return 0;
2689 if (r < 0)
2690 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2691
2692 /* Already reset? */
2693 if (streq(p, "4294967295"))
2694 return 0;
2695
2696 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2697 if (r < 0) {
2698 log_error_errno(r,
2699 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2700 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2701 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2702 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2703 "using systemd-nspawn. Sleeping for 5s... (%m)");
2704
2705 sleep(5);
2706 }
2707
2708 return 0;
2709 }
2710
2711 static int setup_propagate(const char *root) {
2712 const char *p, *q;
2713 int r;
2714
2715 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2716 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2717 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2718 (void) mkdir_p(p, 0600);
2719
2720 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2721 if (r < 0)
2722 return log_error_errno(r, "Failed to create /run/host: %m");
2723
2724 r = userns_mkdir(root, "/run/host/incoming", 0600, 0, 0);
2725 if (r < 0)
2726 return log_error_errno(r, "Failed to create /run/host/incoming: %m");
2727
2728 q = prefix_roota(root, "/run/host/incoming");
2729 r = mount_nofollow_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2730 if (r < 0)
2731 return r;
2732
2733 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2734 if (r < 0)
2735 return r;
2736
2737 /* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
2738 return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2739 }
2740
2741 static int setup_machine_id(const char *directory) {
2742 const char *etc_machine_id;
2743 sd_id128_t id;
2744 int r;
2745
2746 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2747 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2748 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2749 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2750 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2751 * container behaves nicely). */
2752
2753 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2754
2755 r = id128_read(etc_machine_id, ID128_PLAIN_OR_UNINIT, &id);
2756 if (r < 0) {
2757 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2758 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2759
2760 if (sd_id128_is_null(arg_uuid)) {
2761 r = sd_id128_randomize(&arg_uuid);
2762 if (r < 0)
2763 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2764 }
2765 } else {
2766 if (sd_id128_is_null(id))
2767 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2768 "Machine ID in container image is zero, refusing.");
2769
2770 arg_uuid = id;
2771 }
2772
2773 return 0;
2774 }
2775
2776 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2777 int r;
2778
2779 assert(directory);
2780
2781 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
2782 return 0;
2783
2784 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2785 if (r == -EOPNOTSUPP)
2786 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.");
2787 if (r == -EBADE)
2788 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2789 if (r < 0)
2790 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2791 if (r == 0)
2792 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2793 else
2794 log_debug("Patched directory tree to match UID/GID range.");
2795
2796 return r;
2797 }
2798
2799 /*
2800 * Return values:
2801 * < 0 : wait_for_terminate() failed to get the state of the
2802 * container, the container was terminated by a signal, or
2803 * failed for an unknown reason. No change is made to the
2804 * container argument.
2805 * > 0 : The program executed in the container terminated with an
2806 * error. The exit code of the program executed in the
2807 * container is returned. The container argument has been set
2808 * to CONTAINER_TERMINATED.
2809 * 0 : The container is being rebooted, has been shut down or exited
2810 * successfully. The container argument has been set to either
2811 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2812 *
2813 * That is, success is indicated by a return value of zero, and an
2814 * error is indicated by a non-zero value.
2815 */
2816 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2817 siginfo_t status;
2818 int r;
2819
2820 r = wait_for_terminate(pid, &status);
2821 if (r < 0)
2822 return log_warning_errno(r, "Failed to wait for container: %m");
2823
2824 switch (status.si_code) {
2825
2826 case CLD_EXITED:
2827 if (status.si_status == 0)
2828 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2829 else
2830 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2831
2832 *container = CONTAINER_TERMINATED;
2833 return status.si_status;
2834
2835 case CLD_KILLED:
2836 if (status.si_status == SIGINT) {
2837 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2838 *container = CONTAINER_TERMINATED;
2839 return 0;
2840
2841 } else if (status.si_status == SIGHUP) {
2842 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2843 *container = CONTAINER_REBOOTED;
2844 return 0;
2845 }
2846
2847 _fallthrough_;
2848 case CLD_DUMPED:
2849 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2850 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2851
2852 default:
2853 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2854 "Container %s failed due to unknown reason.", arg_machine);
2855 }
2856 }
2857
2858 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2859 pid_t pid;
2860
2861 pid = PTR_TO_PID(userdata);
2862 if (pid > 0) {
2863 if (kill(pid, arg_kill_signal) >= 0) {
2864 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2865 sd_event_source_set_userdata(s, NULL);
2866 return 0;
2867 }
2868 }
2869
2870 sd_event_exit(sd_event_source_get_event(s), 0);
2871 return 0;
2872 }
2873
2874 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2875 pid_t pid;
2876
2877 assert(s);
2878 assert(ssi);
2879
2880 pid = PTR_TO_PID(userdata);
2881
2882 for (;;) {
2883 siginfo_t si = {};
2884
2885 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2886 return log_error_errno(errno, "Failed to waitid(): %m");
2887 if (si.si_pid == 0) /* No pending children. */
2888 break;
2889 if (si.si_pid == pid) {
2890 /* The main process we care for has exited. Return from
2891 * signal handler but leave the zombie. */
2892 sd_event_exit(sd_event_source_get_event(s), 0);
2893 break;
2894 }
2895
2896 /* Reap all other children. */
2897 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2898 }
2899
2900 return 0;
2901 }
2902
2903 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2904 pid_t pid;
2905
2906 assert(m);
2907
2908 pid = PTR_TO_PID(userdata);
2909
2910 if (arg_kill_signal > 0) {
2911 log_info("Container termination requested. Attempting to halt container.");
2912 (void) kill(pid, arg_kill_signal);
2913 } else {
2914 log_info("Container termination requested. Exiting.");
2915 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2916 }
2917
2918 return 0;
2919 }
2920
2921 static int determine_names(void) {
2922 int r;
2923
2924 if (arg_template && !arg_directory && arg_machine) {
2925
2926 /* If --template= was specified then we should not
2927 * search for a machine, but instead create a new one
2928 * in /var/lib/machine. */
2929
2930 arg_directory = path_join("/var/lib/machines", arg_machine);
2931 if (!arg_directory)
2932 return log_oom();
2933 }
2934
2935 if (!arg_image && !arg_directory) {
2936 if (arg_machine) {
2937 _cleanup_(image_unrefp) Image *i = NULL;
2938
2939 r = image_find(IMAGE_MACHINE, arg_machine, NULL, &i);
2940 if (r == -ENOENT)
2941 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
2942 if (r < 0)
2943 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2944
2945 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2946 r = free_and_strdup(&arg_image, i->path);
2947 else
2948 r = free_and_strdup(&arg_directory, i->path);
2949 if (r < 0)
2950 return log_oom();
2951
2952 if (!arg_ephemeral)
2953 arg_read_only = arg_read_only || i->read_only;
2954 } else {
2955 r = safe_getcwd(&arg_directory);
2956 if (r < 0)
2957 return log_error_errno(r, "Failed to determine current directory: %m");
2958 }
2959
2960 if (!arg_directory && !arg_image)
2961 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
2962 }
2963
2964 if (!arg_machine) {
2965 if (arg_directory && path_equal(arg_directory, "/"))
2966 arg_machine = gethostname_malloc();
2967 else {
2968 if (arg_image) {
2969 char *e;
2970
2971 arg_machine = strdup(basename(arg_image));
2972
2973 /* Truncate suffix if there is one */
2974 e = endswith(arg_machine, ".raw");
2975 if (e)
2976 *e = 0;
2977 } else
2978 arg_machine = strdup(basename(arg_directory));
2979 }
2980 if (!arg_machine)
2981 return log_oom();
2982
2983 hostname_cleanup(arg_machine);
2984 if (!hostname_is_valid(arg_machine, 0))
2985 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
2986
2987 if (arg_ephemeral) {
2988 char *b;
2989
2990 /* Add a random suffix when this is an
2991 * ephemeral machine, so that we can run many
2992 * instances at once without manually having
2993 * to specify -M each time. */
2994
2995 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2996 return log_oom();
2997
2998 free(arg_machine);
2999 arg_machine = b;
3000 }
3001 }
3002
3003 return 0;
3004 }
3005
3006 static int chase_symlinks_and_update(char **p, unsigned flags) {
3007 char *chased;
3008 int r;
3009
3010 assert(p);
3011
3012 if (!*p)
3013 return 0;
3014
3015 r = chase_symlinks(*p, NULL, flags, &chased, NULL);
3016 if (r < 0)
3017 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
3018
3019 return free_and_replace(*p, chased);
3020 }
3021
3022 static int determine_uid_shift(const char *directory) {
3023 int r;
3024
3025 if (arg_userns_mode == USER_NAMESPACE_NO) {
3026 arg_uid_shift = 0;
3027 return 0;
3028 }
3029
3030 if (arg_uid_shift == UID_INVALID) {
3031 struct stat st;
3032
3033 r = stat(directory, &st);
3034 if (r < 0)
3035 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
3036
3037 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
3038
3039 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
3040 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3041 "UID and GID base of %s don't match.", directory);
3042
3043 arg_uid_range = UINT32_C(0x10000);
3044 }
3045
3046 if (arg_uid_shift > (uid_t) -1 - arg_uid_range)
3047 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3048 "UID base too high for UID range.");
3049
3050 return 0;
3051 }
3052
3053 static unsigned long effective_clone_ns_flags(void) {
3054 unsigned long flags = arg_clone_ns_flags;
3055
3056 if (arg_private_network)
3057 flags |= CLONE_NEWNET;
3058 if (arg_use_cgns)
3059 flags |= CLONE_NEWCGROUP;
3060 if (arg_userns_mode != USER_NAMESPACE_NO)
3061 flags |= CLONE_NEWUSER;
3062
3063 return flags;
3064 }
3065
3066 static int patch_sysctl(void) {
3067
3068 /* This table is inspired by runc's sysctl() function */
3069 static const struct {
3070 const char *key;
3071 bool prefix;
3072 unsigned long clone_flags;
3073 } safe_sysctl[] = {
3074 { "kernel.hostname", false, CLONE_NEWUTS },
3075 { "kernel.domainname", false, CLONE_NEWUTS },
3076 { "kernel.msgmax", false, CLONE_NEWIPC },
3077 { "kernel.msgmnb", false, CLONE_NEWIPC },
3078 { "kernel.msgmni", false, CLONE_NEWIPC },
3079 { "kernel.sem", false, CLONE_NEWIPC },
3080 { "kernel.shmall", false, CLONE_NEWIPC },
3081 { "kernel.shmmax", false, CLONE_NEWIPC },
3082 { "kernel.shmmni", false, CLONE_NEWIPC },
3083 { "fs.mqueue.", true, CLONE_NEWIPC },
3084 { "net.", true, CLONE_NEWNET },
3085 };
3086
3087 unsigned long flags;
3088 char **k, **v;
3089 int r;
3090
3091 flags = effective_clone_ns_flags();
3092
3093 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
3094 bool good = false;
3095 size_t i;
3096
3097 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
3098
3099 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
3100 continue;
3101
3102 if (safe_sysctl[i].prefix)
3103 good = startswith(*k, safe_sysctl[i].key);
3104 else
3105 good = streq(*k, safe_sysctl[i].key);
3106
3107 if (good)
3108 break;
3109 }
3110
3111 if (!good)
3112 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
3113
3114 r = sysctl_write(*k, *v);
3115 if (r < 0)
3116 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
3117 }
3118
3119 return 0;
3120 }
3121
3122 static int inner_child(
3123 Barrier *barrier,
3124 const char *directory,
3125 bool secondary,
3126 int kmsg_socket,
3127 int rtnl_socket,
3128 int master_pty_socket,
3129 FDSet *fds,
3130 char **os_release_pairs) {
3131
3132 _cleanup_free_ char *home = NULL;
3133 char as_uuid[ID128_UUID_STRING_MAX];
3134 size_t n_env = 1;
3135 const char *envp[] = {
3136 "PATH=" DEFAULT_PATH_COMPAT,
3137 NULL, /* container */
3138 NULL, /* TERM */
3139 NULL, /* HOME */
3140 NULL, /* USER */
3141 NULL, /* LOGNAME */
3142 NULL, /* container_uuid */
3143 NULL, /* LISTEN_FDS */
3144 NULL, /* LISTEN_PID */
3145 NULL, /* NOTIFY_SOCKET */
3146 NULL, /* CREDENTIALS_DIRECTORY */
3147 NULL
3148 };
3149 const char *exec_target;
3150 _cleanup_strv_free_ char **env_use = NULL;
3151 int r, which_failed;
3152
3153 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3154 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3155 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3156 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3157 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3158 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3159 * namespace.
3160 *
3161 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3162 * unshare(). See below. */
3163
3164 assert(barrier);
3165 assert(directory);
3166 assert(kmsg_socket >= 0);
3167
3168 log_debug("Inner child is initializing.");
3169
3170 if (arg_userns_mode != USER_NAMESPACE_NO) {
3171 /* Tell the parent, that it now can write the UID map. */
3172 (void) barrier_place(barrier); /* #1 */
3173
3174 /* Wait until the parent wrote the UID map */
3175 if (!barrier_place_and_sync(barrier)) /* #2 */
3176 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3177
3178 /* Become the new root user inside our namespace */
3179 r = reset_uid_gid();
3180 if (r < 0)
3181 return log_error_errno(r, "Couldn't become new root: %m");
3182
3183 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3184 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3185 * propagation, but simply create new peer groups for all our mounts). */
3186 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SHARED|MS_REC, NULL);
3187 if (r < 0)
3188 return r;
3189 }
3190
3191 r = mount_all(NULL,
3192 arg_mount_settings | MOUNT_IN_USERNS,
3193 arg_uid_shift,
3194 arg_selinux_apifs_context);
3195 if (r < 0)
3196 return r;
3197
3198 if (!arg_network_namespace_path && arg_private_network) {
3199 r = unshare(CLONE_NEWNET);
3200 if (r < 0)
3201 return log_error_errno(errno, "Failed to unshare network namespace: %m");
3202
3203 /* Tell the parent that it can setup network interfaces. */
3204 (void) barrier_place(barrier); /* #3 */
3205 }
3206
3207 r = mount_sysfs(NULL, arg_mount_settings);
3208 if (r < 0)
3209 return r;
3210
3211 /* Wait until we are cgroup-ified, so that we
3212 * can mount the right cgroup path writable */
3213 if (!barrier_place_and_sync(barrier)) /* #4 */
3214 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3215 "Parent died too early");
3216
3217 if (arg_use_cgns) {
3218 r = unshare(CLONE_NEWCGROUP);
3219 if (r < 0)
3220 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
3221 r = mount_cgroups(
3222 "",
3223 arg_unified_cgroup_hierarchy,
3224 arg_userns_mode != USER_NAMESPACE_NO,
3225 arg_uid_shift,
3226 arg_uid_range,
3227 arg_selinux_apifs_context,
3228 true);
3229 } else
3230 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
3231 if (r < 0)
3232 return r;
3233
3234 r = setup_boot_id();
3235 if (r < 0)
3236 return r;
3237
3238 r = setup_kmsg(kmsg_socket);
3239 if (r < 0)
3240 return r;
3241 kmsg_socket = safe_close(kmsg_socket);
3242
3243 r = mount_custom(
3244 "/",
3245 arg_custom_mounts,
3246 arg_n_custom_mounts,
3247 0,
3248 arg_selinux_apifs_context,
3249 MOUNT_NON_ROOT_ONLY | MOUNT_IN_USERNS);
3250 if (r < 0)
3251 return r;
3252
3253 if (setsid() < 0)
3254 return log_error_errno(errno, "setsid() failed: %m");
3255
3256 if (arg_private_network)
3257 (void) loopback_setup();
3258
3259 if (arg_expose_ports) {
3260 r = expose_port_send_rtnl(rtnl_socket);
3261 if (r < 0)
3262 return r;
3263 rtnl_socket = safe_close(rtnl_socket);
3264 }
3265
3266 if (arg_console_mode != CONSOLE_PIPE) {
3267 _cleanup_close_ int master = -1;
3268 _cleanup_free_ char *console = NULL;
3269
3270 /* Allocate a pty and make it available as /dev/console. */
3271 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
3272 if (master < 0)
3273 return log_error_errno(master, "Failed to allocate a pty: %m");
3274
3275 r = setup_dev_console(console);
3276 if (r < 0)
3277 return log_error_errno(r, "Failed to set up /dev/console: %m");
3278
3279 r = send_one_fd(master_pty_socket, master, 0);
3280 if (r < 0)
3281 return log_error_errno(r, "Failed to send master fd: %m");
3282 master_pty_socket = safe_close(master_pty_socket);
3283
3284 r = setup_stdio_as_dev_console();
3285 if (r < 0)
3286 return r;
3287 }
3288
3289 r = patch_sysctl();
3290 if (r < 0)
3291 return r;
3292
3293 if (arg_oom_score_adjust_set) {
3294 r = set_oom_score_adjust(arg_oom_score_adjust);
3295 if (r < 0)
3296 return log_error_errno(r, "Failed to adjust OOM score: %m");
3297 }
3298
3299 if (arg_cpu_set.set)
3300 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
3301 return log_error_errno(errno, "Failed to set CPU affinity: %m");
3302
3303 (void) setup_hostname();
3304
3305 if (arg_personality != PERSONALITY_INVALID) {
3306 r = safe_personality(arg_personality);
3307 if (r < 0)
3308 return log_error_errno(r, "personality() failed: %m");
3309 } else if (secondary) {
3310 r = safe_personality(PER_LINUX32);
3311 if (r < 0)
3312 return log_error_errno(r, "personality() failed: %m");
3313 }
3314
3315 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3316 if (r < 0)
3317 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3318
3319 #if HAVE_SECCOMP
3320 if (arg_seccomp) {
3321
3322 if (is_seccomp_available()) {
3323
3324 r = seccomp_load(arg_seccomp);
3325 if (ERRNO_IS_SECCOMP_FATAL(r))
3326 return log_error_errno(r, "Failed to install seccomp filter: %m");
3327 if (r < 0)
3328 log_debug_errno(r, "Failed to install seccomp filter: %m");
3329 }
3330 } else
3331 #endif
3332 {
3333 r = setup_seccomp(arg_caps_retain, arg_syscall_allow_list, arg_syscall_deny_list);
3334 if (r < 0)
3335 return r;
3336 }
3337
3338 #if HAVE_SELINUX
3339 if (arg_selinux_context)
3340 if (setexeccon(arg_selinux_context) < 0)
3341 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3342 #endif
3343
3344 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3345 * if we need to later on. */
3346 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3347 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3348
3349 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3350 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids, arg_console_mode != CONSOLE_PIPE);
3351 else
3352 r = change_uid_gid(arg_user, arg_console_mode != CONSOLE_PIPE, &home);
3353 if (r < 0)
3354 return r;
3355
3356 r = drop_capabilities(getuid());
3357 if (r < 0)
3358 return log_error_errno(r, "Dropping capabilities failed: %m");
3359
3360 if (arg_no_new_privileges)
3361 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3362 return log_error_errno(errno, "Failed to disable new privileges: %m");
3363
3364 /* LXC sets container=lxc, so follow the scheme here */
3365 envp[n_env++] = strjoina("container=", arg_container_service_name);
3366
3367 envp[n_env] = strv_find_prefix(environ, "TERM=");
3368 if (envp[n_env])
3369 n_env++;
3370
3371 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3372 if (asprintf((char**)(envp + n_env++), "HOME=%s", home ?: "/root") < 0)
3373 return log_oom();
3374
3375 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3376 if (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ?: "root") < 0 ||
3377 asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3378 return log_oom();
3379
3380 assert(!sd_id128_is_null(arg_uuid));
3381
3382 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
3383 return log_oom();
3384
3385 if (fdset_size(fds) > 0) {
3386 r = fdset_cloexec(fds, false);
3387 if (r < 0)
3388 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3389
3390 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3391 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3392 return log_oom();
3393 }
3394 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3395 return log_oom();
3396
3397 if (arg_n_credentials > 0) {
3398 envp[n_env] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3399 if (!envp[n_env])
3400 return log_oom();
3401 n_env++;
3402 }
3403
3404 env_use = strv_env_merge(3, envp, os_release_pairs, arg_setenv);
3405 if (!env_use)
3406 return log_oom();
3407
3408 /* Let the parent know that we are ready and
3409 * wait until the parent is ready with the
3410 * setup, too... */
3411 if (!barrier_place_and_sync(barrier)) /* #5 */
3412 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3413
3414 if (arg_chdir)
3415 if (chdir(arg_chdir) < 0)
3416 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3417
3418 if (arg_start_mode == START_PID2) {
3419 r = stub_pid1(arg_uuid);
3420 if (r < 0)
3421 return r;
3422 }
3423
3424 if (arg_console_mode != CONSOLE_PIPE) {
3425 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3426 * are configured for that. Acquire it as controlling tty. */
3427 if (ioctl(STDIN_FILENO, TIOCSCTTY) < 0)
3428 return log_error_errno(errno, "Failed to acquire controlling TTY: %m");
3429 }
3430
3431 log_debug("Inner child completed, invoking payload.");
3432
3433 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3434 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3435 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3436 log_close();
3437 log_set_open_when_needed(true);
3438
3439 (void) fdset_close_others(fds);
3440
3441 if (arg_start_mode == START_BOOT) {
3442 char **a;
3443 size_t m;
3444
3445 /* Automatically search for the init system */
3446
3447 m = strv_length(arg_parameters);
3448 a = newa(char*, m + 2);
3449 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3450 a[1 + m] = NULL;
3451
3452 a[0] = (char*) "/usr/lib/systemd/systemd";
3453 execve(a[0], a, env_use);
3454
3455 a[0] = (char*) "/lib/systemd/systemd";
3456 execve(a[0], a, env_use);
3457
3458 a[0] = (char*) "/sbin/init";
3459 execve(a[0], a, env_use);
3460
3461 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3462 } else if (!strv_isempty(arg_parameters)) {
3463 const char *dollar_path;
3464
3465 exec_target = arg_parameters[0];
3466
3467 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3468 * binary. */
3469 dollar_path = strv_env_get(env_use, "PATH");
3470 if (dollar_path) {
3471 if (setenv("PATH", dollar_path, 1) < 0)
3472 return log_error_errno(errno, "Failed to update $PATH: %m");
3473 }
3474
3475 execvpe(arg_parameters[0], arg_parameters, env_use);
3476 } else {
3477 if (!arg_chdir)
3478 /* If we cannot change the directory, we'll end up in /, that is expected. */
3479 (void) chdir(home ?: "/root");
3480
3481 execle("/bin/bash", "-bash", NULL, env_use);
3482 execle("/bin/sh", "-sh", NULL, env_use);
3483
3484 exec_target = "/bin/bash, /bin/sh";
3485 }
3486
3487 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3488 }
3489
3490 static int setup_notify_child(void) {
3491 _cleanup_close_ int fd = -1;
3492 union sockaddr_union sa = {
3493 .un.sun_family = AF_UNIX,
3494 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3495 };
3496 int r;
3497
3498 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3499 if (fd < 0)
3500 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3501
3502 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3503 (void) sockaddr_un_unlink(&sa.un);
3504
3505 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3506 if (r < 0)
3507 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3508
3509 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3510 if (r < 0)
3511 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3512
3513 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3514 if (r < 0)
3515 return log_error_errno(r, "SO_PASSCRED failed: %m");
3516
3517 return TAKE_FD(fd);
3518 }
3519
3520 static int outer_child(
3521 Barrier *barrier,
3522 const char *directory,
3523 DissectedImage *dissected_image,
3524 bool secondary,
3525 int pid_socket,
3526 int uuid_socket,
3527 int notify_socket,
3528 int kmsg_socket,
3529 int rtnl_socket,
3530 int uid_shift_socket,
3531 int master_pty_socket,
3532 int unified_cgroup_hierarchy_socket,
3533 FDSet *fds,
3534 int netns_fd) {
3535
3536 _cleanup_strv_free_ char **os_release_pairs = NULL;
3537 _cleanup_close_ int fd = -1;
3538 const char *p;
3539 pid_t pid;
3540 ssize_t l;
3541 int r;
3542
3543 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3544 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3545 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3546 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3547
3548 assert(barrier);
3549 assert(directory);
3550 assert(pid_socket >= 0);
3551 assert(uuid_socket >= 0);
3552 assert(notify_socket >= 0);
3553 assert(master_pty_socket >= 0);
3554 assert(kmsg_socket >= 0);
3555
3556 log_debug("Outer child is initializing.");
3557
3558 r = load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs);
3559 if (r < 0)
3560 log_debug_errno(r, "Failed to read os-release from host for container, ignoring: %m");
3561
3562 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3563 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3564
3565 r = reset_audit_loginuid();
3566 if (r < 0)
3567 return r;
3568
3569 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3570 * mounts to the real root. */
3571 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3572 if (r < 0)
3573 return r;
3574
3575 if (dissected_image) {
3576 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3577 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3578 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3579 * makes sure ESP partitions and userns are compatible. */
3580
3581 r = dissected_image_mount_and_warn(
3582 dissected_image, directory, arg_uid_shift,
3583 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
3584 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK)|
3585 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3586 if (r < 0)
3587 return r;
3588 }
3589
3590 r = determine_uid_shift(directory);
3591 if (r < 0)
3592 return r;
3593
3594 if (arg_userns_mode != USER_NAMESPACE_NO) {
3595 /* Let the parent know which UID shift we read from the image */
3596 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3597 if (l < 0)
3598 return log_error_errno(errno, "Failed to send UID shift: %m");
3599 if (l != sizeof(arg_uid_shift))
3600 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3601 "Short write while sending UID shift.");
3602
3603 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3604 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3605 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3606 * not it will pick a different one, and send it back to us. */
3607
3608 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3609 if (l < 0)
3610 return log_error_errno(errno, "Failed to recv UID shift: %m");
3611 if (l != sizeof(arg_uid_shift))
3612 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3613 "Short read while receiving UID shift.");
3614 }
3615
3616 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3617 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3618 }
3619
3620 if (path_equal(directory, "/")) {
3621 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3622 * place, so that we can make changes to its mount structure (for example, to implement
3623 * --volatile=) without this interfering with our ability to access files such as
3624 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3625 * (instead of a temporary directory, since we are living in our own mount namspace here
3626 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3627 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3628
3629 r = mount_nofollow_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
3630 if (r < 0)
3631 return r;
3632
3633 directory = "/run/systemd/nspawn-root";
3634 }
3635
3636 r = setup_pivot_root(
3637 directory,
3638 arg_pivot_root_new,
3639 arg_pivot_root_old);
3640 if (r < 0)
3641 return r;
3642
3643 r = setup_volatile_mode(
3644 directory,
3645 arg_volatile_mode,
3646 arg_uid_shift,
3647 arg_selinux_apifs_context);
3648 if (r < 0)
3649 return r;
3650
3651 r = mount_custom(
3652 directory,
3653 arg_custom_mounts,
3654 arg_n_custom_mounts,
3655 arg_uid_shift,
3656 arg_selinux_apifs_context,
3657 MOUNT_ROOT_ONLY);
3658 if (r < 0)
3659 return r;
3660
3661 /* Make sure we always have a mount that we can move to root later on. */
3662 if (!path_is_mount_point(directory, NULL, 0)) {
3663 r = mount_nofollow_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
3664 if (r < 0)
3665 return r;
3666 }
3667
3668 if (dissected_image) {
3669 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3670 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3671 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK));
3672 if (r == -EUCLEAN)
3673 return log_error_errno(r, "File system check for image failed: %m");
3674 if (r < 0)
3675 return log_error_errno(r, "Failed to mount image file system: %m");
3676 }
3677
3678 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3679 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3680
3681 r = detect_unified_cgroup_hierarchy_from_image(directory);
3682 if (r < 0)
3683 return r;
3684
3685 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3686 if (l < 0)
3687 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3688 if (l != sizeof(arg_unified_cgroup_hierarchy))
3689 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3690 "Short write while sending cgroup mode.");
3691
3692 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3693 }
3694
3695 /* Mark everything as shared so our mounts get propagated down. This is
3696 * required to make new bind mounts available in systemd services
3697 * inside the container that create a new mount namespace.
3698 * See https://github.com/systemd/systemd/issues/3860
3699 * Further submounts (such as /dev) done after this will inherit the
3700 * shared propagation mode.
3701 *
3702 * IMPORTANT: Do not overmount the root directory anymore from now on to
3703 * enable moving the root directory mount to root later on.
3704 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3705 */
3706 r = mount_nofollow_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3707 if (r < 0)
3708 return r;
3709
3710 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3711 if (r < 0)
3712 return r;
3713
3714 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3715 if (r < 0)
3716 return r;
3717
3718 if (arg_read_only && arg_volatile_mode == VOLATILE_NO &&
3719 !has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts)) {
3720 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3721 if (r < 0)
3722 return log_error_errno(r, "Failed to make tree read-only: %m");
3723 }
3724
3725 r = mount_all(directory,
3726 arg_mount_settings,
3727 arg_uid_shift,
3728 arg_selinux_apifs_context);
3729 if (r < 0)
3730 return r;
3731
3732 r = copy_devnodes(directory);
3733 if (r < 0)
3734 return r;
3735
3736 r = make_extra_nodes(directory);
3737 if (r < 0)
3738 return r;
3739
3740 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3741
3742 p = prefix_roota(directory, "/run/host");
3743 (void) make_inaccessible_nodes(p, arg_uid_shift, arg_uid_shift);
3744
3745 r = setup_pts(directory);
3746 if (r < 0)
3747 return r;
3748
3749 r = setup_propagate(directory);
3750 if (r < 0)
3751 return r;
3752
3753 r = setup_keyring();
3754 if (r < 0)
3755 return r;
3756
3757 r = setup_credentials(directory);
3758 if (r < 0)
3759 return r;
3760
3761 r = mount_custom(
3762 directory,
3763 arg_custom_mounts,
3764 arg_n_custom_mounts,
3765 arg_uid_shift,
3766 arg_selinux_apifs_context,
3767 MOUNT_NON_ROOT_ONLY);
3768 if (r < 0)
3769 return r;
3770
3771 r = setup_timezone(directory);
3772 if (r < 0)
3773 return r;
3774
3775 r = setup_resolv_conf(directory);
3776 if (r < 0)
3777 return r;
3778
3779 r = setup_machine_id(directory);
3780 if (r < 0)
3781 return r;
3782
3783 r = setup_journal(directory);
3784 if (r < 0)
3785 return r;
3786
3787 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3788 p = prefix_roota(directory, "/run/host/container-manager");
3789 (void) write_string_file(p, arg_container_service_name, WRITE_STRING_FILE_CREATE);
3790
3791 /* The same stuff as the $container_uuid env var */
3792 p = prefix_roota(directory, "/run/host/container-uuid");
3793 (void) write_string_filef(p, WRITE_STRING_FILE_CREATE, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid));
3794
3795 if (!arg_use_cgns) {
3796 r = mount_cgroups(
3797 directory,
3798 arg_unified_cgroup_hierarchy,
3799 arg_userns_mode != USER_NAMESPACE_NO,
3800 arg_uid_shift,
3801 arg_uid_range,
3802 arg_selinux_apifs_context,
3803 false);
3804 if (r < 0)
3805 return r;
3806 }
3807
3808 r = mount_move_root(directory);
3809 if (r < 0)
3810 return log_error_errno(r, "Failed to move root directory: %m");
3811
3812 fd = setup_notify_child();
3813 if (fd < 0)
3814 return fd;
3815
3816 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3817 arg_clone_ns_flags |
3818 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3819 if (pid < 0)
3820 return log_error_errno(errno, "Failed to fork inner child: %m");
3821 if (pid == 0) {
3822 pid_socket = safe_close(pid_socket);
3823 uuid_socket = safe_close(uuid_socket);
3824 notify_socket = safe_close(notify_socket);
3825 uid_shift_socket = safe_close(uid_shift_socket);
3826
3827 /* The inner child has all namespaces that are requested, so that we all are owned by the
3828 * user if user namespaces are turned on. */
3829
3830 if (arg_network_namespace_path) {
3831 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3832 if (r < 0)
3833 return log_error_errno(r, "Failed to join network namespace: %m");
3834 }
3835
3836 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds, os_release_pairs);
3837 if (r < 0)
3838 _exit(EXIT_FAILURE);
3839
3840 _exit(EXIT_SUCCESS);
3841 }
3842
3843 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3844 if (l < 0)
3845 return log_error_errno(errno, "Failed to send PID: %m");
3846 if (l != sizeof(pid))
3847 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3848 "Short write while sending PID.");
3849
3850 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3851 if (l < 0)
3852 return log_error_errno(errno, "Failed to send machine ID: %m");
3853 if (l != sizeof(arg_uuid))
3854 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3855 "Short write while sending machine ID.");
3856
3857 l = send_one_fd(notify_socket, fd, 0);
3858 if (l < 0)
3859 return log_error_errno(l, "Failed to send notify fd: %m");
3860
3861 pid_socket = safe_close(pid_socket);
3862 uuid_socket = safe_close(uuid_socket);
3863 notify_socket = safe_close(notify_socket);
3864 master_pty_socket = safe_close(master_pty_socket);
3865 kmsg_socket = safe_close(kmsg_socket);
3866 rtnl_socket = safe_close(rtnl_socket);
3867 netns_fd = safe_close(netns_fd);
3868
3869 return 0;
3870 }
3871
3872 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
3873 bool tried_hashed = false;
3874 unsigned n_tries = 100;
3875 uid_t candidate;
3876 int r;
3877
3878 assert(shift);
3879 assert(ret_lock_file);
3880 assert(arg_userns_mode == USER_NAMESPACE_PICK);
3881 assert(arg_uid_range == 0x10000U);
3882
3883 candidate = *shift;
3884
3885 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3886
3887 for (;;) {
3888 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
3889 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
3890
3891 if (--n_tries <= 0)
3892 return -EBUSY;
3893
3894 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
3895 goto next;
3896 if ((candidate & UINT32_C(0xFFFF)) != 0)
3897 goto next;
3898
3899 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3900 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3901 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3902 goto next;
3903 if (r < 0)
3904 return r;
3905
3906 /* Make some superficial checks whether the range is currently known in the user database */
3907 if (getpwuid(candidate))
3908 goto next;
3909 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3910 goto next;
3911 if (getgrgid(candidate))
3912 goto next;
3913 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3914 goto next;
3915
3916 *ret_lock_file = lf;
3917 lf = (struct LockFile) LOCK_FILE_INIT;
3918 *shift = candidate;
3919 return 0;
3920
3921 next:
3922 if (arg_machine && !tried_hashed) {
3923 /* Try to hash the base from the container name */
3924
3925 static const uint8_t hash_key[] = {
3926 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3927 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3928 };
3929
3930 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3931
3932 tried_hashed = true;
3933 } else
3934 random_bytes(&candidate, sizeof(candidate));
3935
3936 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
3937 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3938 }
3939 }
3940
3941 static int setup_uid_map(pid_t pid) {
3942 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3943 int r;
3944
3945 assert(pid > 1);
3946
3947 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3948 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3949 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3950 if (r < 0)
3951 return log_error_errno(r, "Failed to write UID map: %m");
3952
3953 /* We always assign the same UID and GID ranges */
3954 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3955 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3956 if (r < 0)
3957 return log_error_errno(r, "Failed to write GID map: %m");
3958
3959 return 0;
3960 }
3961
3962 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3963 char buf[NOTIFY_BUFFER_MAX+1];
3964 char *p = NULL;
3965 struct iovec iovec = {
3966 .iov_base = buf,
3967 .iov_len = sizeof(buf)-1,
3968 };
3969 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
3970 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
3971 struct msghdr msghdr = {
3972 .msg_iov = &iovec,
3973 .msg_iovlen = 1,
3974 .msg_control = &control,
3975 .msg_controllen = sizeof(control),
3976 };
3977 struct ucred *ucred;
3978 ssize_t n;
3979 pid_t inner_child_pid;
3980 _cleanup_strv_free_ char **tags = NULL;
3981
3982 assert(userdata);
3983
3984 inner_child_pid = PTR_TO_PID(userdata);
3985
3986 if (revents != EPOLLIN) {
3987 log_warning("Got unexpected poll event for notify fd.");
3988 return 0;
3989 }
3990
3991 n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3992 if (IN_SET(n, -EAGAIN, -EINTR))
3993 return 0;
3994 if (n == -EXFULL) {
3995 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
3996 return 0;
3997 }
3998 if (n < 0)
3999 return log_warning_errno(n, "Couldn't read notification socket: %m");
4000
4001 cmsg_close_all(&msghdr);
4002
4003 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
4004 if (!ucred || ucred->pid != inner_child_pid) {
4005 log_debug("Received notify message without valid credentials. Ignoring.");
4006 return 0;
4007 }
4008
4009 if ((size_t) n >= sizeof(buf)) {
4010 log_warning("Received notify message exceeded maximum size. Ignoring.");
4011 return 0;
4012 }
4013
4014 buf[n] = 0;
4015 tags = strv_split(buf, "\n\r");
4016 if (!tags)
4017 return log_oom();
4018
4019 if (strv_find(tags, "READY=1"))
4020 (void) sd_notifyf(false, "READY=1\n");
4021
4022 p = strv_find_startswith(tags, "STATUS=");
4023 if (p)
4024 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
4025
4026 return 0;
4027 }
4028
4029 static int setup_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
4030 int r;
4031
4032 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
4033 if (r < 0)
4034 return log_error_errno(r, "Failed to allocate notify event source: %m");
4035
4036 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
4037
4038 return 0;
4039 }
4040
4041 static int merge_settings(Settings *settings, const char *path) {
4042 int rl;
4043
4044 assert(settings);
4045 assert(path);
4046
4047 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4048 * that this steals the fields of the Settings* structure, and hence modifies it. */
4049
4050 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
4051 settings->start_mode >= 0) {
4052 arg_start_mode = settings->start_mode;
4053 strv_free_and_replace(arg_parameters, settings->parameters);
4054 }
4055
4056 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
4057 arg_ephemeral = settings->ephemeral;
4058
4059 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
4060 settings->root) {
4061
4062 if (!arg_settings_trusted)
4063 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
4064 else
4065 free_and_replace(arg_directory, settings->root);
4066 }
4067
4068 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
4069 settings->pivot_root_new) {
4070 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
4071 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
4072 }
4073
4074 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
4075 settings->working_directory)
4076 free_and_replace(arg_chdir, settings->working_directory);
4077
4078 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
4079 settings->environment)
4080 strv_free_and_replace(arg_setenv, settings->environment);
4081
4082 if ((arg_settings_mask & SETTING_USER) == 0) {
4083
4084 if (settings->user)
4085 free_and_replace(arg_user, settings->user);
4086
4087 if (uid_is_valid(settings->uid))
4088 arg_uid = settings->uid;
4089 if (gid_is_valid(settings->gid))
4090 arg_gid = settings->gid;
4091 if (settings->n_supplementary_gids > 0) {
4092 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
4093 arg_n_supplementary_gids = settings->n_supplementary_gids;
4094 }
4095 }
4096
4097 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
4098 uint64_t plus, minus;
4099 uint64_t network_minus = 0;
4100 uint64_t ambient;
4101
4102 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4103 * Settings structure */
4104
4105 plus = settings->capability;
4106 minus = settings->drop_capability;
4107
4108 if ((arg_settings_mask & SETTING_NETWORK) == 0) {
4109 if (settings_private_network(settings))
4110 plus |= UINT64_C(1) << CAP_NET_ADMIN;
4111 else
4112 network_minus |= UINT64_C(1) << CAP_NET_ADMIN;
4113 }
4114
4115 if (!arg_settings_trusted && plus != 0) {
4116 if (settings->capability != 0)
4117 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
4118 } else {
4119 arg_caps_retain &= ~network_minus;
4120 arg_caps_retain |= plus;
4121 }
4122
4123 arg_caps_retain &= ~minus;
4124
4125 /* Copy the full capabilities over too */
4126 if (capability_quintet_is_set(&settings->full_capabilities)) {
4127 if (!arg_settings_trusted)
4128 log_warning("Ignoring capability settings, file %s is not trusted.", path);
4129 else
4130 arg_full_capabilities = settings->full_capabilities;
4131 }
4132
4133 ambient = settings->ambient_capability;
4134 if (!arg_settings_trusted && ambient != 0)
4135 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path);
4136 else
4137 arg_caps_ambient |= ambient;
4138 }
4139
4140 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
4141 settings->kill_signal > 0)
4142 arg_kill_signal = settings->kill_signal;
4143
4144 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
4145 settings->personality != PERSONALITY_INVALID)
4146 arg_personality = settings->personality;
4147
4148 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
4149 !sd_id128_is_null(settings->machine_id)) {
4150
4151 if (!arg_settings_trusted)
4152 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
4153 else
4154 arg_uuid = settings->machine_id;
4155 }
4156
4157 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
4158 settings->read_only >= 0)
4159 arg_read_only = settings->read_only;
4160
4161 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
4162 settings->volatile_mode != _VOLATILE_MODE_INVALID)
4163 arg_volatile_mode = settings->volatile_mode;
4164
4165 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
4166 settings->n_custom_mounts > 0) {
4167
4168 if (!arg_settings_trusted)
4169 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
4170 else {
4171 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4172 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
4173 arg_n_custom_mounts = settings->n_custom_mounts;
4174 settings->n_custom_mounts = 0;
4175 }
4176 }
4177
4178 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4179 (settings->private_network >= 0 ||
4180 settings->network_veth >= 0 ||
4181 settings->network_bridge ||
4182 settings->network_zone ||
4183 settings->network_interfaces ||
4184 settings->network_macvlan ||
4185 settings->network_ipvlan ||
4186 settings->network_veth_extra ||
4187 settings->network_namespace_path)) {
4188
4189 if (!arg_settings_trusted)
4190 log_warning("Ignoring network settings, file %s is not trusted.", path);
4191 else {
4192 arg_network_veth = settings_network_veth(settings);
4193 arg_private_network = settings_private_network(settings);
4194
4195 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
4196 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
4197 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
4198 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
4199
4200 free_and_replace(arg_network_bridge, settings->network_bridge);
4201 free_and_replace(arg_network_zone, settings->network_zone);
4202
4203 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
4204 }
4205 }
4206
4207 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
4208 settings->expose_ports) {
4209
4210 if (!arg_settings_trusted)
4211 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
4212 else {
4213 expose_port_free_all(arg_expose_ports);
4214 arg_expose_ports = TAKE_PTR(settings->expose_ports);
4215 }
4216 }
4217
4218 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
4219 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
4220
4221 if (!arg_settings_trusted)
4222 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
4223 else {
4224 arg_userns_mode = settings->userns_mode;
4225 arg_uid_shift = settings->uid_shift;
4226 arg_uid_range = settings->uid_range;
4227 arg_userns_chown = settings->userns_chown;
4228 }
4229 }
4230
4231 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
4232 arg_notify_ready = settings->notify_ready;
4233
4234 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
4235
4236 if (!arg_settings_trusted && !strv_isempty(settings->syscall_allow_list))
4237 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
4238 else {
4239 strv_free_and_replace(arg_syscall_allow_list, settings->syscall_allow_list);
4240 strv_free_and_replace(arg_syscall_deny_list, settings->syscall_deny_list);
4241 }
4242
4243 #if HAVE_SECCOMP
4244 if (!arg_settings_trusted && settings->seccomp)
4245 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
4246 else {
4247 seccomp_release(arg_seccomp);
4248 arg_seccomp = TAKE_PTR(settings->seccomp);
4249 }
4250 #endif
4251 }
4252
4253 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
4254 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
4255 continue;
4256
4257 if (!settings->rlimit[rl])
4258 continue;
4259
4260 if (!arg_settings_trusted) {
4261 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
4262 continue;
4263 }
4264
4265 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
4266 }
4267
4268 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
4269 settings->hostname)
4270 free_and_replace(arg_hostname, settings->hostname);
4271
4272 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
4273 settings->no_new_privileges >= 0)
4274 arg_no_new_privileges = settings->no_new_privileges;
4275
4276 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
4277 settings->oom_score_adjust_set) {
4278
4279 if (!arg_settings_trusted)
4280 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
4281 else {
4282 arg_oom_score_adjust = settings->oom_score_adjust;
4283 arg_oom_score_adjust_set = true;
4284 }
4285 }
4286
4287 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
4288 settings->cpu_set.set) {
4289
4290 if (!arg_settings_trusted)
4291 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
4292 else {
4293 cpu_set_reset(&arg_cpu_set);
4294 arg_cpu_set = settings->cpu_set;
4295 settings->cpu_set = (CPUSet) {};
4296 }
4297 }
4298
4299 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
4300 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
4301 arg_resolv_conf = settings->resolv_conf;
4302
4303 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
4304 settings->link_journal != _LINK_JOURNAL_INVALID) {
4305
4306 if (!arg_settings_trusted)
4307 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
4308 else {
4309 arg_link_journal = settings->link_journal;
4310 arg_link_journal_try = settings->link_journal_try;
4311 }
4312 }
4313
4314 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
4315 settings->timezone != _TIMEZONE_MODE_INVALID)
4316 arg_timezone = settings->timezone;
4317
4318 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
4319 settings->slice) {
4320
4321 if (!arg_settings_trusted)
4322 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
4323 else
4324 free_and_replace(arg_slice, settings->slice);
4325 }
4326
4327 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
4328 settings->use_cgns >= 0) {
4329
4330 if (!arg_settings_trusted)
4331 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
4332 else
4333 arg_use_cgns = settings->use_cgns;
4334 }
4335
4336 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
4337 settings->clone_ns_flags != (unsigned long) -1) {
4338
4339 if (!arg_settings_trusted)
4340 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
4341 else
4342 arg_clone_ns_flags = settings->clone_ns_flags;
4343 }
4344
4345 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
4346 settings->console_mode >= 0) {
4347
4348 if (!arg_settings_trusted)
4349 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
4350 else
4351 arg_console_mode = settings->console_mode;
4352 }
4353
4354 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4355 * don't consult arg_settings_mask for them. */
4356
4357 sd_bus_message_unref(arg_property_message);
4358 arg_property_message = TAKE_PTR(settings->properties);
4359
4360 arg_console_width = settings->console_width;
4361 arg_console_height = settings->console_height;
4362
4363 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
4364 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
4365 arg_n_extra_nodes = settings->n_extra_nodes;
4366
4367 return 0;
4368 }
4369
4370 static int load_settings(void) {
4371 _cleanup_(settings_freep) Settings *settings = NULL;
4372 _cleanup_fclose_ FILE *f = NULL;
4373 _cleanup_free_ char *p = NULL;
4374 const char *fn, *i;
4375 int r;
4376
4377 if (arg_oci_bundle)
4378 return 0;
4379
4380 /* If all settings are masked, there's no point in looking for
4381 * the settings file */
4382 if (FLAGS_SET(arg_settings_mask, _SETTINGS_MASK_ALL))
4383 return 0;
4384
4385 fn = strjoina(arg_machine, ".nspawn");
4386
4387 /* We first look in the admin's directories in /etc and /run */
4388 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4389 _cleanup_free_ char *j = NULL;
4390
4391 j = path_join(i, fn);
4392 if (!j)
4393 return log_oom();
4394
4395 f = fopen(j, "re");
4396 if (f) {
4397 p = TAKE_PTR(j);
4398
4399 /* By default, we trust configuration from /etc and /run */
4400 if (arg_settings_trusted < 0)
4401 arg_settings_trusted = true;
4402
4403 break;
4404 }
4405
4406 if (errno != ENOENT)
4407 return log_error_errno(errno, "Failed to open %s: %m", j);
4408 }
4409
4410 if (!f) {
4411 /* After that, let's look for a file next to the
4412 * actual image we shall boot. */
4413
4414 if (arg_image) {
4415 p = file_in_same_dir(arg_image, fn);
4416 if (!p)
4417 return log_oom();
4418 } else if (arg_directory && !path_equal(arg_directory, "/")) {
4419 p = file_in_same_dir(arg_directory, fn);
4420 if (!p)
4421 return log_oom();
4422 }
4423
4424 if (p) {
4425 f = fopen(p, "re");
4426 if (!f && errno != ENOENT)
4427 return log_error_errno(errno, "Failed to open %s: %m", p);
4428
4429 /* By default, we do not trust configuration from /var/lib/machines */
4430 if (arg_settings_trusted < 0)
4431 arg_settings_trusted = false;
4432 }
4433 }
4434
4435 if (!f)
4436 return 0;
4437
4438 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4439
4440 r = settings_load(f, p, &settings);
4441 if (r < 0)
4442 return r;
4443
4444 return merge_settings(settings, p);
4445 }
4446
4447 static int load_oci_bundle(void) {
4448 _cleanup_(settings_freep) Settings *settings = NULL;
4449 int r;
4450
4451 if (!arg_oci_bundle)
4452 return 0;
4453
4454 /* By default let's trust OCI bundles */
4455 if (arg_settings_trusted < 0)
4456 arg_settings_trusted = true;
4457
4458 r = oci_load(NULL, arg_oci_bundle, &settings);
4459 if (r < 0)
4460 return r;
4461
4462 return merge_settings(settings, arg_oci_bundle);
4463 }
4464
4465 static int run_container(
4466 DissectedImage *dissected_image,
4467 bool secondary,
4468 FDSet *fds,
4469 char veth_name[IFNAMSIZ], bool *veth_created,
4470 struct ExposeArgs *expose_args,
4471 int *master, pid_t *pid, int *ret) {
4472
4473 static const struct sigaction sa = {
4474 .sa_handler = nop_signal_handler,
4475 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4476 };
4477
4478 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4479 _cleanup_close_ int etc_passwd_lock = -1;
4480 _cleanup_close_pair_ int
4481 kmsg_socket_pair[2] = { -1, -1 },
4482 rtnl_socket_pair[2] = { -1, -1 },
4483 pid_socket_pair[2] = { -1, -1 },
4484 uuid_socket_pair[2] = { -1, -1 },
4485 notify_socket_pair[2] = { -1, -1 },
4486 uid_shift_socket_pair[2] = { -1, -1 },
4487 master_pty_socket_pair[2] = { -1, -1 },
4488 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4489
4490 _cleanup_close_ int notify_socket = -1;
4491 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4492 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4493 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4494 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4495 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4496 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4497 ContainerStatus container_status = 0;
4498 int ifi = 0, r;
4499 ssize_t l;
4500 sigset_t mask_chld;
4501 _cleanup_close_ int child_netns_fd = -1;
4502
4503 assert_se(sigemptyset(&mask_chld) == 0);
4504 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4505
4506 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4507 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4508 * check with getpwuid() if the specific user already exists. Note that /etc might be
4509 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4510 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4511 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4512 * really ours. */
4513
4514 etc_passwd_lock = take_etc_passwd_lock(NULL);
4515 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4516 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4517 }
4518
4519 r = barrier_create(&barrier);
4520 if (r < 0)
4521 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4522
4523 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4524 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4525
4526 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4527 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4528
4529 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4530 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4531
4532 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4533 return log_error_errno(errno, "Failed to create id socket pair: %m");
4534
4535 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4536 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4537
4538 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
4539 return log_error_errno(errno, "Failed to create console socket pair: %m");
4540
4541 if (arg_userns_mode != USER_NAMESPACE_NO)
4542 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4543 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4544
4545 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4546 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4547 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4548
4549 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4550 * parent's blocking calls and give it a chance to call wait() and terminate. */
4551 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4552 if (r < 0)
4553 return log_error_errno(errno, "Failed to change the signal mask: %m");
4554
4555 r = sigaction(SIGCHLD, &sa, NULL);
4556 if (r < 0)
4557 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4558
4559 if (arg_network_namespace_path) {
4560 child_netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4561 if (child_netns_fd < 0)
4562 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4563
4564 r = fd_is_network_ns(child_netns_fd);
4565 if (r == -EUCLEAN)
4566 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);
4567 else if (r < 0)
4568 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4569 else if (r == 0)
4570 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4571 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4572 }
4573
4574 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4575 if (*pid < 0)
4576 return log_error_errno(errno, "clone() failed%s: %m",
4577 errno == EINVAL ?
4578 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4579
4580 if (*pid == 0) {
4581 /* The outer child only has a file system namespace. */
4582 barrier_set_role(&barrier, BARRIER_CHILD);
4583
4584 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4585 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4586 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4587 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4588 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
4589 master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
4590 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4591 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
4592
4593 (void) reset_all_signal_handlers();
4594 (void) reset_signal_mask();
4595
4596 r = outer_child(&barrier,
4597 arg_directory,
4598 dissected_image,
4599 secondary,
4600 pid_socket_pair[1],
4601 uuid_socket_pair[1],
4602 notify_socket_pair[1],
4603 kmsg_socket_pair[1],
4604 rtnl_socket_pair[1],
4605 uid_shift_socket_pair[1],
4606 master_pty_socket_pair[1],
4607 unified_cgroup_hierarchy_socket_pair[1],
4608 fds,
4609 child_netns_fd);
4610 if (r < 0)
4611 _exit(EXIT_FAILURE);
4612
4613 _exit(EXIT_SUCCESS);
4614 }
4615
4616 barrier_set_role(&barrier, BARRIER_PARENT);
4617
4618 fdset_close(fds);
4619
4620 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4621 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4622 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4623 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4624 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
4625 master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
4626 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
4627 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
4628
4629 if (arg_userns_mode != USER_NAMESPACE_NO) {
4630 /* The child just let us know the UID shift it might have read from the image. */
4631 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4632 if (l < 0)
4633 return log_error_errno(errno, "Failed to read UID shift: %m");
4634 if (l != sizeof arg_uid_shift)
4635 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
4636
4637 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4638 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4639 * image, but if that's already in use, pick a new one, and report back to the child,
4640 * which one we now picked. */
4641
4642 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4643 if (r < 0)
4644 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4645
4646 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4647 if (l < 0)
4648 return log_error_errno(errno, "Failed to send UID shift: %m");
4649 if (l != sizeof arg_uid_shift)
4650 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
4651 }
4652 }
4653
4654 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4655 /* The child let us know the support cgroup mode it might have read from the image. */
4656 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4657 if (l < 0)
4658 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4659 if (l != sizeof(arg_unified_cgroup_hierarchy))
4660 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zu bytes).%s",
4661 l, l == 0 ? " The child is most likely dead." : "");
4662 }
4663
4664 /* Wait for the outer child. */
4665 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4666 if (r < 0)
4667 return r;
4668 if (r != EXIT_SUCCESS)
4669 return -EIO;
4670
4671 /* And now retrieve the PID of the inner child. */
4672 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4673 if (l < 0)
4674 return log_error_errno(errno, "Failed to read inner child PID: %m");
4675 if (l != sizeof *pid)
4676 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
4677
4678 /* We also retrieve container UUID in case it was generated by outer child */
4679 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4680 if (l < 0)
4681 return log_error_errno(errno, "Failed to read container machine ID: %m");
4682 if (l != sizeof(arg_uuid))
4683 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
4684
4685 /* We also retrieve the socket used for notifications generated by outer child */
4686 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4687 if (notify_socket < 0)
4688 return log_error_errno(notify_socket,
4689 "Failed to receive notification socket from the outer child: %m");
4690
4691 log_debug("Init process invoked as PID "PID_FMT, *pid);
4692
4693 if (arg_userns_mode != USER_NAMESPACE_NO) {
4694 if (!barrier_place_and_sync(&barrier)) /* #1 */
4695 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4696
4697 r = setup_uid_map(*pid);
4698 if (r < 0)
4699 return r;
4700
4701 (void) barrier_place(&barrier); /* #2 */
4702 }
4703
4704 if (arg_private_network) {
4705 if (!arg_network_namespace_path) {
4706 /* Wait until the child has unshared its network namespace. */
4707 if (!barrier_place_and_sync(&barrier)) /* #3 */
4708 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
4709 }
4710
4711 if (child_netns_fd < 0) {
4712 /* Make sure we have an open file descriptor to the child's network
4713 * namespace so it stays alive even if the child exits. */
4714 r = namespace_open(*pid, NULL, NULL, &child_netns_fd, NULL, NULL);
4715 if (r < 0)
4716 return log_error_errno(r, "Failed to open child network namespace: %m");
4717 }
4718
4719 r = move_network_interfaces(child_netns_fd, arg_network_interfaces);
4720 if (r < 0)
4721 return r;
4722
4723 if (arg_network_veth) {
4724 r = setup_veth(arg_machine, *pid, veth_name,
4725 arg_network_bridge || arg_network_zone);
4726 if (r < 0)
4727 return r;
4728 else if (r > 0)
4729 ifi = r;
4730
4731 if (arg_network_bridge) {
4732 /* Add the interface to a bridge */
4733 r = setup_bridge(veth_name, arg_network_bridge, false);
4734 if (r < 0)
4735 return r;
4736 if (r > 0)
4737 ifi = r;
4738 } else if (arg_network_zone) {
4739 /* Add the interface to a bridge, possibly creating it */
4740 r = setup_bridge(veth_name, arg_network_zone, true);
4741 if (r < 0)
4742 return r;
4743 if (r > 0)
4744 ifi = r;
4745 }
4746 }
4747
4748 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
4749 if (r < 0)
4750 return r;
4751
4752 /* We created the primary and extra veth links now; let's remember this, so that we know to
4753 remove them later on. Note that we don't bother with removing veth links that were created
4754 here when their setup failed half-way, because in that case the kernel should be able to
4755 remove them on its own, since they cannot be referenced by anything yet. */
4756 *veth_created = true;
4757
4758 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
4759 if (r < 0)
4760 return r;
4761
4762 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
4763 if (r < 0)
4764 return r;
4765 }
4766
4767 if (arg_register || !arg_keep_unit) {
4768 r = sd_bus_default_system(&bus);
4769 if (r < 0)
4770 return log_error_errno(r, "Failed to open system bus: %m");
4771
4772 r = sd_bus_set_close_on_exit(bus, false);
4773 if (r < 0)
4774 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
4775 }
4776
4777 if (!arg_keep_unit) {
4778 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4779 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4780 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4781
4782 r = sd_bus_match_signal_async(
4783 bus,
4784 NULL,
4785 "org.freedesktop.systemd1",
4786 NULL,
4787 "org.freedesktop.systemd1.Scope",
4788 "RequestStop",
4789 on_request_stop, NULL, PID_TO_PTR(*pid));
4790 if (r < 0)
4791 return log_error_errno(r, "Failed to request RequestStop match: %m");
4792 }
4793
4794 if (arg_register) {
4795 r = register_machine(
4796 bus,
4797 arg_machine,
4798 *pid,
4799 arg_directory,
4800 arg_uuid,
4801 ifi,
4802 arg_slice,
4803 arg_custom_mounts, arg_n_custom_mounts,
4804 arg_kill_signal,
4805 arg_property,
4806 arg_property_message,
4807 arg_keep_unit,
4808 arg_container_service_name);
4809 if (r < 0)
4810 return r;
4811
4812 } else if (!arg_keep_unit) {
4813 r = allocate_scope(
4814 bus,
4815 arg_machine,
4816 *pid,
4817 arg_slice,
4818 arg_custom_mounts, arg_n_custom_mounts,
4819 arg_kill_signal,
4820 arg_property,
4821 arg_property_message);
4822 if (r < 0)
4823 return r;
4824
4825 } else if (arg_slice || arg_property)
4826 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4827
4828 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
4829 if (r < 0)
4830 return r;
4831
4832 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4833 if (r < 0)
4834 return r;
4835
4836 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4837 if (r < 0)
4838 return r;
4839
4840 /* Notify the child that the parent is ready with all
4841 * its setup (including cgroup-ification), and that
4842 * the child can now hand over control to the code to
4843 * run inside the container. */
4844 (void) barrier_place(&barrier); /* #4 */
4845
4846 /* Block SIGCHLD here, before notifying child.
4847 * process_pty() will handle it with the other signals. */
4848 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4849
4850 /* Reset signal to default */
4851 r = default_signals(SIGCHLD, -1);
4852 if (r < 0)
4853 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4854
4855 r = sd_event_new(&event);
4856 if (r < 0)
4857 return log_error_errno(r, "Failed to get default event source: %m");
4858
4859 (void) sd_event_set_watchdog(event, true);
4860
4861 if (bus) {
4862 r = sd_bus_attach_event(bus, event, 0);
4863 if (r < 0)
4864 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4865 }
4866
4867 r = setup_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
4868 if (r < 0)
4869 return r;
4870
4871 /* Let the child know that we are ready and wait that the child is completely ready now. */
4872 if (!barrier_place_and_sync(&barrier)) /* #5 */
4873 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4874
4875 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
4876 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4877 etc_passwd_lock = safe_close(etc_passwd_lock);
4878
4879 (void) sd_notifyf(false,
4880 "STATUS=Container running.\n"
4881 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
4882 if (!arg_notify_ready)
4883 (void) sd_notify(false, "READY=1\n");
4884
4885 if (arg_kill_signal > 0) {
4886 /* Try to kill the init system on SIGINT or SIGTERM */
4887 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4888 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
4889 } else {
4890 /* Immediately exit */
4891 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4892 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4893 }
4894
4895 /* Exit when the child exits */
4896 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
4897
4898 if (arg_expose_ports) {
4899 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, expose_args, &rtnl);
4900 if (r < 0)
4901 return r;
4902
4903 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, &expose_args->address);
4904 }
4905
4906 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4907
4908 if (arg_console_mode != CONSOLE_PIPE) {
4909 _cleanup_close_ int fd = -1;
4910 PTYForwardFlags flags = 0;
4911
4912 /* Retrieve the master pty allocated by inner child */
4913 fd = receive_one_fd(master_pty_socket_pair[0], 0);
4914 if (fd < 0)
4915 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
4916
4917 switch (arg_console_mode) {
4918
4919 case CONSOLE_READ_ONLY:
4920 flags |= PTY_FORWARD_READ_ONLY;
4921
4922 _fallthrough_;
4923
4924 case CONSOLE_INTERACTIVE:
4925 flags |= PTY_FORWARD_IGNORE_VHANGUP;
4926
4927 r = pty_forward_new(event, fd, flags, &forward);
4928 if (r < 0)
4929 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4930
4931 if (arg_console_width != (unsigned) -1 || arg_console_height != (unsigned) -1)
4932 (void) pty_forward_set_width_height(forward,
4933 arg_console_width,
4934 arg_console_height);
4935 break;
4936
4937 default:
4938 assert(arg_console_mode == CONSOLE_PASSIVE);
4939 }
4940
4941 *master = TAKE_FD(fd);
4942 }
4943
4944 r = sd_event_loop(event);
4945 if (r < 0)
4946 return log_error_errno(r, "Failed to run event loop: %m");
4947
4948 if (forward) {
4949 char last_char = 0;
4950
4951 (void) pty_forward_get_last_char(forward, &last_char);
4952 forward = pty_forward_free(forward);
4953
4954 if (!arg_quiet && last_char != '\n')
4955 putc('\n', stdout);
4956 }
4957
4958 /* Kill if it is not dead yet anyway */
4959 if (!arg_register && !arg_keep_unit && bus)
4960 terminate_scope(bus, arg_machine);
4961
4962 /* Normally redundant, but better safe than sorry */
4963 (void) kill(*pid, SIGKILL);
4964
4965 if (arg_private_network) {
4966 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
4967 * to avoid having to move the parent to the child network namespace. */
4968 r = safe_fork(NULL, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_WAIT|FORK_LOG, NULL);
4969 if (r < 0)
4970 return r;
4971
4972 if (r == 0) {
4973 _cleanup_close_ int parent_netns_fd = -1;
4974
4975 r = namespace_open(getpid(), NULL, NULL, &parent_netns_fd, NULL, NULL);
4976 if (r < 0) {
4977 log_error_errno(r, "Failed to open parent network namespace: %m");
4978 _exit(EXIT_FAILURE);
4979 }
4980
4981 r = namespace_enter(-1, -1, child_netns_fd, -1, -1);
4982 if (r < 0) {
4983 log_error_errno(r, "Failed to enter child network namespace: %m");
4984 _exit(EXIT_FAILURE);
4985 }
4986
4987 r = move_network_interfaces(parent_netns_fd, arg_network_interfaces);
4988 if (r < 0)
4989 log_error_errno(r, "Failed to move network interfaces back to parent network namespace: %m");
4990
4991 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
4992 }
4993 }
4994
4995 r = wait_for_container(*pid, &container_status);
4996 *pid = 0;
4997
4998 /* Tell machined that we are gone. */
4999 if (bus)
5000 (void) unregister_machine(bus, arg_machine);
5001
5002 if (r < 0)
5003 /* We failed to wait for the container, or the container exited abnormally. */
5004 return r;
5005 if (r > 0 || container_status == CONTAINER_TERMINATED) {
5006 /* r > 0 → The container exited with a non-zero status.
5007 * As a special case, we need to replace 133 with a different value,
5008 * because 133 is special-cased in the service file to reboot the container.
5009 * otherwise → The container exited with zero status and a reboot was not requested.
5010 */
5011 if (r == EXIT_FORCE_RESTART)
5012 r = EXIT_FAILURE; /* replace 133 with the general failure code */
5013 *ret = r;
5014 return 0; /* finito */
5015 }
5016
5017 /* CONTAINER_REBOOTED, loop again */
5018
5019 if (arg_keep_unit) {
5020 /* Special handling if we are running as a service: instead of simply
5021 * restarting the machine we want to restart the entire service, so let's
5022 * inform systemd about this with the special exit code 133. The service
5023 * file uses RestartForceExitStatus=133 so that this results in a full
5024 * nspawn restart. This is necessary since we might have cgroup parameters
5025 * set we want to have flushed out. */
5026 *ret = EXIT_FORCE_RESTART;
5027 return 0; /* finito */
5028 }
5029
5030 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, &expose_args->address);
5031
5032 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5033 *veth_created = false;
5034 return 1; /* loop again */
5035 }
5036
5037 static int initialize_rlimits(void) {
5038 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
5039 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5040 * container execution environments. */
5041
5042 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
5043 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
5044 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
5045 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
5046 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
5047 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
5048 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
5049 [RLIMIT_MEMLOCK] = { 65536, 65536 },
5050 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
5051 [RLIMIT_NICE] = { 0, 0 },
5052 [RLIMIT_NOFILE] = { 1024, 4096 },
5053 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
5054 [RLIMIT_RTPRIO] = { 0, 0 },
5055 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
5056 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
5057
5058 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5059 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5060 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5061 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5062 * that PID 1 changes a number of other resource limits during early initialization which is why we
5063 * don't read the other limits from PID 1 but prefer the static table above. */
5064 };
5065
5066 int rl;
5067
5068 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
5069 /* Let's only fill in what the user hasn't explicitly configured anyway */
5070 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
5071 const struct rlimit *v;
5072 struct rlimit buffer;
5073
5074 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
5075 /* For these two let's read the limits off PID 1. See above for an explanation. */
5076
5077 if (prlimit(1, rl, NULL, &buffer) < 0)
5078 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
5079
5080 v = &buffer;
5081 } else
5082 v = kernel_defaults + rl;
5083
5084 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
5085 if (!arg_rlimit[rl])
5086 return log_oom();
5087 }
5088
5089 if (DEBUG_LOGGING) {
5090 _cleanup_free_ char *k = NULL;
5091
5092 (void) rlimit_format(arg_rlimit[rl], &k);
5093 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
5094 }
5095 }
5096
5097 return 0;
5098 }
5099
5100 static int cant_be_in_netns(void) {
5101 union sockaddr_union sa = {
5102 .un = {
5103 .sun_family = AF_UNIX,
5104 .sun_path = "/run/udev/control",
5105 },
5106 };
5107 char udev_path[STRLEN("/proc//ns/net") + DECIMAL_STR_MAX(pid_t)];
5108 _cleanup_free_ char *udev_ns = NULL, *our_ns = NULL;
5109 _cleanup_close_ int fd = -1;
5110 struct ucred ucred;
5111 int r;
5112
5113 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5114 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5115 * nice message. */
5116
5117 if (!arg_image) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5118 return 0;
5119
5120 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_NONBLOCK|SOCK_CLOEXEC, 0);
5121 if (fd < 0)
5122 return log_error_errno(errno, "Failed to allocate udev control socket: %m");
5123
5124 if (connect(fd, &sa.un, SOCKADDR_UN_LEN(sa.un)) < 0) {
5125
5126 if (errno == ENOENT || ERRNO_IS_DISCONNECT(errno))
5127 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5128 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5129
5130 return log_error_errno(errno, "Failed to connect socket to udev control socket: %m");
5131 }
5132
5133 r = getpeercred(fd, &ucred);
5134 if (r < 0)
5135 return log_error_errno(r, "Failed to determine peer of udev control socket: %m");
5136
5137 xsprintf(udev_path, "/proc/" PID_FMT "/ns/net", ucred.pid);
5138 r = readlink_malloc(udev_path, &udev_ns);
5139 if (r < 0)
5140 return log_error_errno(r, "Failed to read network namespace of udev: %m");
5141
5142 r = readlink_malloc("/proc/self/ns/net", &our_ns);
5143 if (r < 0)
5144 return log_error_errno(r, "Failed to read our own network namespace: %m");
5145
5146 if (!streq(our_ns, udev_ns))
5147 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5148 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5149 return 0;
5150 }
5151
5152 static int run(int argc, char *argv[]) {
5153 bool secondary = false, remove_directory = false, remove_image = false,
5154 veth_created = false, remove_tmprootdir = false;
5155 _cleanup_close_ int master = -1;
5156 _cleanup_fdset_free_ FDSet *fds = NULL;
5157 int r, n_fd_passed, ret = EXIT_SUCCESS;
5158 char veth_name[IFNAMSIZ] = "";
5159 struct ExposeArgs expose_args = {};
5160 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
5161 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
5162 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
5163 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
5164 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
5165 _cleanup_(fw_ctx_freep) FirewallContext *fw_ctx = NULL;
5166 pid_t pid = 0;
5167
5168 log_parse_environment();
5169 log_open();
5170
5171 r = parse_argv(argc, argv);
5172 if (r <= 0)
5173 goto finish;
5174
5175 if (geteuid() != 0) {
5176 r = log_warning_errno(SYNTHETIC_ERRNO(EPERM),
5177 argc >= 2 ? "Need to be root." :
5178 "Need to be root (and some arguments are usually required).\nHint: try --help");
5179 goto finish;
5180 }
5181
5182 r = cant_be_in_netns();
5183 if (r < 0)
5184 goto finish;
5185
5186 r = initialize_rlimits();
5187 if (r < 0)
5188 goto finish;
5189
5190 r = load_oci_bundle();
5191 if (r < 0)
5192 goto finish;
5193
5194 r = determine_names();
5195 if (r < 0)
5196 goto finish;
5197
5198 r = load_settings();
5199 if (r < 0)
5200 goto finish;
5201
5202 r = cg_unified();
5203 if (r < 0) {
5204 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5205 goto finish;
5206 }
5207
5208 r = verify_arguments();
5209 if (r < 0)
5210 goto finish;
5211
5212 /* Reapply environment settings. */
5213 (void) detect_unified_cgroup_hierarchy_from_environment();
5214
5215 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5216 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5217 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5218 (void) ignore_signals(SIGPIPE, -1);
5219
5220 n_fd_passed = sd_listen_fds(false);
5221 if (n_fd_passed > 0) {
5222 r = fdset_new_listen_fds(&fds, false);
5223 if (r < 0) {
5224 log_error_errno(r, "Failed to collect file descriptors: %m");
5225 goto finish;
5226 }
5227 }
5228
5229 /* The "default" umask. This is appropriate for most file and directory
5230 * operations performed by nspawn, and is the umask that will be used for
5231 * the child. Functions like copy_devnodes() change the umask temporarily. */
5232 umask(0022);
5233
5234 if (arg_directory) {
5235 assert(!arg_image);
5236
5237 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5238 * /var from the host will propagate into container dynamically (because bad things happen if
5239 * two systems write to the same /var). Let's allow it for the special cases where /var is
5240 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5241 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
5242 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5243 r = -EINVAL;
5244 goto finish;
5245 }
5246
5247 if (arg_ephemeral) {
5248 _cleanup_free_ char *np = NULL;
5249
5250 r = chase_symlinks_and_update(&arg_directory, 0);
5251 if (r < 0)
5252 goto finish;
5253
5254 /* If the specified path is a mount point we generate the new snapshot immediately
5255 * inside it under a random name. However if the specified is not a mount point we
5256 * create the new snapshot in the parent directory, just next to it. */
5257 r = path_is_mount_point(arg_directory, NULL, 0);
5258 if (r < 0) {
5259 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
5260 goto finish;
5261 }
5262 if (r > 0)
5263 r = tempfn_random_child(arg_directory, "machine.", &np);
5264 else
5265 r = tempfn_random(arg_directory, "machine.", &np);
5266 if (r < 0) {
5267 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
5268 goto finish;
5269 }
5270
5271 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5272 * only owned by us and no one else. */
5273 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5274 if (r < 0) {
5275 log_error_errno(r, "Failed to lock %s: %m", np);
5276 goto finish;
5277 }
5278
5279 {
5280 BLOCK_SIGNALS(SIGINT);
5281 r = btrfs_subvol_snapshot(arg_directory, np,
5282 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5283 BTRFS_SNAPSHOT_FALLBACK_COPY |
5284 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5285 BTRFS_SNAPSHOT_RECURSIVE |
5286 BTRFS_SNAPSHOT_QUOTA |
5287 BTRFS_SNAPSHOT_SIGINT);
5288 }
5289 if (r == -EINTR) {
5290 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
5291 goto finish;
5292 }
5293 if (r < 0) {
5294 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
5295 goto finish;
5296 }
5297
5298 free_and_replace(arg_directory, np);
5299 remove_directory = true;
5300 } else {
5301 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
5302 if (r < 0)
5303 goto finish;
5304
5305 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5306 if (r == -EBUSY) {
5307 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
5308 goto finish;
5309 }
5310 if (r < 0) {
5311 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
5312 goto finish;
5313 }
5314
5315 if (arg_template) {
5316 r = chase_symlinks_and_update(&arg_template, 0);
5317 if (r < 0)
5318 goto finish;
5319
5320 {
5321 BLOCK_SIGNALS(SIGINT);
5322 r = btrfs_subvol_snapshot(arg_template, arg_directory,
5323 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5324 BTRFS_SNAPSHOT_FALLBACK_COPY |
5325 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5326 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
5327 BTRFS_SNAPSHOT_RECURSIVE |
5328 BTRFS_SNAPSHOT_QUOTA |
5329 BTRFS_SNAPSHOT_SIGINT);
5330 }
5331 if (r == -EEXIST)
5332 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5333 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
5334 else if (r == -EINTR) {
5335 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
5336 goto finish;
5337 } else if (r < 0) {
5338 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
5339 goto finish;
5340 } else
5341 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5342 "Populated %s from template %s.", arg_directory, arg_template);
5343 }
5344 }
5345
5346 if (arg_start_mode == START_BOOT) {
5347 const char *p;
5348
5349 if (arg_pivot_root_new)
5350 p = prefix_roota(arg_directory, arg_pivot_root_new);
5351 else
5352 p = arg_directory;
5353
5354 if (path_is_os_tree(p) <= 0) {
5355 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
5356 r = -EINVAL;
5357 goto finish;
5358 }
5359 } else {
5360 const char *p, *q;
5361
5362 if (arg_pivot_root_new)
5363 p = prefix_roota(arg_directory, arg_pivot_root_new);
5364 else
5365 p = arg_directory;
5366
5367 q = strjoina(p, "/usr/");
5368
5369 if (laccess(q, F_OK) < 0) {
5370 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
5371 r = -EINVAL;
5372 goto finish;
5373 }
5374 }
5375
5376 } else {
5377 DissectImageFlags dissect_image_flags = DISSECT_IMAGE_REQUIRE_ROOT | DISSECT_IMAGE_RELAX_VAR_CHECK;
5378 assert(arg_image);
5379 assert(!arg_template);
5380
5381 r = chase_symlinks_and_update(&arg_image, 0);
5382 if (r < 0)
5383 goto finish;
5384
5385 if (arg_ephemeral) {
5386 _cleanup_free_ char *np = NULL;
5387
5388 r = tempfn_random(arg_image, "machine.", &np);
5389 if (r < 0) {
5390 log_error_errno(r, "Failed to generate name for image snapshot: %m");
5391 goto finish;
5392 }
5393
5394 /* Always take an exclusive lock on our own ephemeral copy. */
5395 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5396 if (r < 0) {
5397 r = log_error_errno(r, "Failed to create image lock: %m");
5398 goto finish;
5399 }
5400
5401 {
5402 BLOCK_SIGNALS(SIGINT);
5403 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME|COPY_SIGINT);
5404 }
5405 if (r == -EINTR) {
5406 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
5407 goto finish;
5408 }
5409 if (r < 0) {
5410 r = log_error_errno(r, "Failed to copy image file: %m");
5411 goto finish;
5412 }
5413
5414 free_and_replace(arg_image, np);
5415 remove_image = true;
5416 } else {
5417 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5418 if (r == -EBUSY) {
5419 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
5420 goto finish;
5421 }
5422 if (r < 0) {
5423 r = log_error_errno(r, "Failed to create image lock: %m");
5424 goto finish;
5425 }
5426
5427 r = verity_settings_load(
5428 &arg_verity_settings,
5429 arg_image, NULL, NULL);
5430 if (r < 0) {
5431 log_error_errno(r, "Failed to read verity artefacts for %s: %m", arg_image);
5432 goto finish;
5433 }
5434
5435 if (arg_verity_settings.data_path)
5436 dissect_image_flags |= DISSECT_IMAGE_NO_PARTITION_TABLE;
5437 }
5438
5439 if (!mkdtemp(tmprootdir)) {
5440 r = log_error_errno(errno, "Failed to create temporary directory: %m");
5441 goto finish;
5442 }
5443
5444 remove_tmprootdir = true;
5445
5446 arg_directory = strdup(tmprootdir);
5447 if (!arg_directory) {
5448 r = log_oom();
5449 goto finish;
5450 }
5451
5452 r = loop_device_make_by_path(
5453 arg_image,
5454 arg_read_only ? O_RDONLY : O_RDWR,
5455 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
5456 &loop);
5457 if (r < 0) {
5458 log_error_errno(r, "Failed to set up loopback block device: %m");
5459 goto finish;
5460 }
5461
5462 r = dissect_image_and_warn(
5463 loop->fd,
5464 arg_image,
5465 &arg_verity_settings,
5466 NULL,
5467 dissect_image_flags,
5468 &dissected_image);
5469 if (r == -ENOPKG) {
5470 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5471 log_notice("Note that the disk image needs to\n"
5472 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5473 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5474 " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
5475 " d) or contain a file system without a partition table\n"
5476 "in order to be bootable with systemd-nspawn.");
5477 goto finish;
5478 }
5479 if (r < 0)
5480 goto finish;
5481
5482 if (!arg_verity_settings.root_hash && dissected_image->can_verity)
5483 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
5484
5485 r = dissected_image_decrypt_interactively(
5486 dissected_image,
5487 NULL,
5488 &arg_verity_settings,
5489 0,
5490 &decrypted_image);
5491 if (r < 0)
5492 goto finish;
5493
5494 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5495 if (remove_image && unlink(arg_image) >= 0)
5496 remove_image = false;
5497 }
5498
5499 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5500 if (r < 0)
5501 goto finish;
5502
5503 if (arg_console_mode < 0)
5504 arg_console_mode =
5505 isatty(STDIN_FILENO) > 0 &&
5506 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
5507
5508 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5509 arg_quiet = true;
5510
5511 if (!arg_quiet)
5512 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5513 arg_machine, arg_image ?: arg_directory);
5514
5515 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
5516
5517 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
5518 r = log_error_errno(errno, "Failed to become subreaper: %m");
5519 goto finish;
5520 }
5521
5522 if (arg_expose_ports) {
5523 r = fw_ctx_new(&fw_ctx);
5524 if (r < 0) {
5525 log_error_errno(r, "Cannot expose configured ports, firewall initialization failed: %m");
5526 goto finish;
5527 }
5528 expose_args.fw_ctx = fw_ctx;
5529 }
5530 for (;;) {
5531 r = run_container(dissected_image,
5532 secondary,
5533 fds,
5534 veth_name, &veth_created,
5535 &expose_args, &master,
5536 &pid, &ret);
5537 if (r <= 0)
5538 break;
5539 }
5540
5541 finish:
5542 (void) sd_notify(false,
5543 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5544 "STOPPING=1\nSTATUS=Terminating...");
5545
5546 if (pid > 0)
5547 (void) kill(pid, SIGKILL);
5548
5549 /* Try to flush whatever is still queued in the pty */
5550 if (master >= 0) {
5551 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
5552 master = safe_close(master);
5553 }
5554
5555 if (pid > 0)
5556 (void) wait_for_terminate(pid, NULL);
5557
5558 pager_close();
5559
5560 if (remove_directory && arg_directory) {
5561 int k;
5562
5563 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5564 if (k < 0)
5565 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5566 }
5567
5568 if (remove_image && arg_image) {
5569 if (unlink(arg_image) < 0)
5570 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5571 }
5572
5573 if (remove_tmprootdir) {
5574 if (rmdir(tmprootdir) < 0)
5575 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5576 }
5577
5578 if (arg_machine) {
5579 const char *p;
5580
5581 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5582 (void) rm_rf(p, REMOVE_ROOT);
5583 }
5584
5585 expose_port_flush(&fw_ctx, arg_expose_ports, &expose_args.address);
5586
5587 if (veth_created)
5588 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5589 (void) remove_bridge(arg_network_zone);
5590
5591 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5592 expose_port_free_all(arg_expose_ports);
5593 rlimit_free_all(arg_rlimit);
5594 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
5595 credential_free_all(arg_credentials, arg_n_credentials);
5596
5597 if (r < 0)
5598 return r;
5599
5600 return ret;
5601 }
5602
5603 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.