]> 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 #10753 from keszybz/pager-no-interrupt
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #if HAVE_BLKID
4 #include <blkid.h>
5 #endif
6 #include <errno.h>
7 #include <getopt.h>
8 #include <grp.h>
9 #include <linux/loop.h>
10 #include <pwd.h>
11 #include <sched.h>
12 #if HAVE_SELINUX
13 #include <selinux/selinux.h>
14 #endif
15 #include <signal.h>
16 #include <stdio.h>
17 #include <stdlib.h>
18 #include <string.h>
19 #include <sys/file.h>
20 #include <sys/mount.h>
21 #include <sys/personality.h>
22 #include <sys/prctl.h>
23 #include <sys/types.h>
24 #include <sys/wait.h>
25 #include <unistd.h>
26
27 #include "sd-bus.h"
28 #include "sd-daemon.h"
29 #include "sd-id128.h"
30
31 #include "alloc-util.h"
32 #include "barrier.h"
33 #include "base-filesystem.h"
34 #include "blkid-util.h"
35 #include "btrfs-util.h"
36 #include "bus-error.h"
37 #include "bus-util.h"
38 #include "cap-list.h"
39 #include "capability-util.h"
40 #include "cgroup-util.h"
41 #include "copy.h"
42 #include "cpu-set-util.h"
43 #include "dev-setup.h"
44 #include "dissect-image.h"
45 #include "env-util.h"
46 #include "fd-util.h"
47 #include "fdset.h"
48 #include "fileio.h"
49 #include "format-util.h"
50 #include "fs-util.h"
51 #include "gpt.h"
52 #include "hexdecoct.h"
53 #include "hostname-util.h"
54 #include "id128-util.h"
55 #include "log.h"
56 #include "loop-util.h"
57 #include "loopback-setup.h"
58 #include "machine-image.h"
59 #include "macro.h"
60 #include "missing.h"
61 #include "mkdir.h"
62 #include "mount-util.h"
63 #include "netlink-util.h"
64 #include "nspawn-cgroup.h"
65 #include "nspawn-def.h"
66 #include "nspawn-expose-ports.h"
67 #include "nspawn-mount.h"
68 #include "nspawn-network.h"
69 #include "nspawn-patch-uid.h"
70 #include "nspawn-register.h"
71 #include "nspawn-seccomp.h"
72 #include "nspawn-settings.h"
73 #include "nspawn-setuid.h"
74 #include "nspawn-stub-pid1.h"
75 #include "os-util.h"
76 #include "pager.h"
77 #include "parse-util.h"
78 #include "path-util.h"
79 #include "process-util.h"
80 #include "ptyfwd.h"
81 #include "random-util.h"
82 #include "raw-clone.h"
83 #include "rlimit-util.h"
84 #include "rm-rf.h"
85 #include "selinux-util.h"
86 #include "signal-util.h"
87 #include "socket-util.h"
88 #include "stat-util.h"
89 #include "stdio-util.h"
90 #include "string-table.h"
91 #include "string-util.h"
92 #include "strv.h"
93 #include "terminal-util.h"
94 #include "umask-util.h"
95 #include "user-util.h"
96 #include "util.h"
97
98 #if HAVE_SPLIT_USR
99 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
100 #else
101 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
102 #endif
103
104 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
105 * nspawn_notify_socket_path is relative to the container
106 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
107 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
108
109 #define EXIT_FORCE_RESTART 133
110
111 typedef enum ContainerStatus {
112 CONTAINER_TERMINATED,
113 CONTAINER_REBOOTED
114 } ContainerStatus;
115
116 static char *arg_directory = NULL;
117 static char *arg_template = NULL;
118 static char *arg_chdir = NULL;
119 static char *arg_pivot_root_new = NULL;
120 static char *arg_pivot_root_old = NULL;
121 static char *arg_user = NULL;
122 static sd_id128_t arg_uuid = {};
123 static char *arg_machine = NULL; /* The name used by the host to refer to this */
124 static char *arg_hostname = NULL; /* The name the payload sees by default */
125 static const char *arg_selinux_context = NULL;
126 static const char *arg_selinux_apifs_context = NULL;
127 static const char *arg_slice = NULL;
128 static bool arg_private_network = false;
129 static bool arg_read_only = false;
130 static StartMode arg_start_mode = START_PID1;
131 static bool arg_ephemeral = false;
132 static LinkJournal arg_link_journal = LINK_AUTO;
133 static bool arg_link_journal_try = false;
134 static uint64_t arg_caps_retain =
135 (1ULL << CAP_AUDIT_CONTROL) |
136 (1ULL << CAP_AUDIT_WRITE) |
137 (1ULL << CAP_CHOWN) |
138 (1ULL << CAP_DAC_OVERRIDE) |
139 (1ULL << CAP_DAC_READ_SEARCH) |
140 (1ULL << CAP_FOWNER) |
141 (1ULL << CAP_FSETID) |
142 (1ULL << CAP_IPC_OWNER) |
143 (1ULL << CAP_KILL) |
144 (1ULL << CAP_LEASE) |
145 (1ULL << CAP_LINUX_IMMUTABLE) |
146 (1ULL << CAP_MKNOD) |
147 (1ULL << CAP_NET_BIND_SERVICE) |
148 (1ULL << CAP_NET_BROADCAST) |
149 (1ULL << CAP_NET_RAW) |
150 (1ULL << CAP_SETFCAP) |
151 (1ULL << CAP_SETGID) |
152 (1ULL << CAP_SETPCAP) |
153 (1ULL << CAP_SETUID) |
154 (1ULL << CAP_SYS_ADMIN) |
155 (1ULL << CAP_SYS_BOOT) |
156 (1ULL << CAP_SYS_CHROOT) |
157 (1ULL << CAP_SYS_NICE) |
158 (1ULL << CAP_SYS_PTRACE) |
159 (1ULL << CAP_SYS_RESOURCE) |
160 (1ULL << CAP_SYS_TTY_CONFIG);
161 static CustomMount *arg_custom_mounts = NULL;
162 static size_t arg_n_custom_mounts = 0;
163 static char **arg_setenv = NULL;
164 static bool arg_quiet = false;
165 static bool arg_register = true;
166 static bool arg_keep_unit = false;
167 static char **arg_network_interfaces = NULL;
168 static char **arg_network_macvlan = NULL;
169 static char **arg_network_ipvlan = NULL;
170 static bool arg_network_veth = false;
171 static char **arg_network_veth_extra = NULL;
172 static char *arg_network_bridge = NULL;
173 static char *arg_network_zone = NULL;
174 static char *arg_network_namespace_path = NULL;
175 static unsigned long arg_personality = PERSONALITY_INVALID;
176 static char *arg_image = NULL;
177 static VolatileMode arg_volatile_mode = VOLATILE_NO;
178 static ExposePort *arg_expose_ports = NULL;
179 static char **arg_property = NULL;
180 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
181 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
182 static bool arg_userns_chown = false;
183 static int arg_kill_signal = 0;
184 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
185 static SettingsMask arg_settings_mask = 0;
186 static int arg_settings_trusted = -1;
187 static char **arg_parameters = NULL;
188 static const char *arg_container_service_name = "systemd-nspawn";
189 static bool arg_notify_ready = false;
190 static bool arg_use_cgns = true;
191 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
192 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
193 static void *arg_root_hash = NULL;
194 static size_t arg_root_hash_size = 0;
195 static char **arg_syscall_whitelist = NULL;
196 static char **arg_syscall_blacklist = NULL;
197 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
198 static bool arg_no_new_privileges = false;
199 static int arg_oom_score_adjust = 0;
200 static bool arg_oom_score_adjust_set = false;
201 static cpu_set_t *arg_cpuset = NULL;
202 static unsigned arg_cpuset_ncpus = 0;
203 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
204 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
205
206 static int help(void) {
207 _cleanup_free_ char *link = NULL;
208 int r;
209
210 (void) pager_open(false);
211
212 r = terminal_urlify_man("systemd-nspawn", "1", &link);
213 if (r < 0)
214 return log_oom();
215
216 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
217 "Spawn a command or OS in a light-weight container.\n\n"
218 " -h --help Show this help\n"
219 " --version Print version string\n"
220 " -q --quiet Do not show status information\n"
221 " -D --directory=PATH Root directory for the container\n"
222 " --template=PATH Initialize root directory from template directory,\n"
223 " if missing\n"
224 " -x --ephemeral Run container with snapshot of root directory, and\n"
225 " remove it after exit\n"
226 " -i --image=PATH File system device or disk image for the container\n"
227 " --root-hash=HASH Specify verity root hash\n"
228 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
229 " -b --boot Boot up full system (i.e. invoke init)\n"
230 " --chdir=PATH Set working directory in the container\n"
231 " --pivot-root=PATH[:PATH]\n"
232 " Pivot root to given directory in the container\n"
233 " -u --user=USER Run the command under specified user or uid\n"
234 " -M --machine=NAME Set the machine name for the container\n"
235 " --hostname=NAME Override the hostname for the container\n"
236 " --uuid=UUID Set a specific machine UUID for the container\n"
237 " -S --slice=SLICE Place the container in the specified slice\n"
238 " --property=NAME=VALUE Set scope unit property\n"
239 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
240 " --private-users[=UIDBASE[:NUIDS]]\n"
241 " Similar, but with user configured UID/GID range\n"
242 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
243 " --private-network Disable network in container\n"
244 " --network-interface=INTERFACE\n"
245 " Assign an existing network interface to the\n"
246 " container\n"
247 " --network-macvlan=INTERFACE\n"
248 " Create a macvlan network interface based on an\n"
249 " existing network interface to the container\n"
250 " --network-ipvlan=INTERFACE\n"
251 " Create a ipvlan network interface based on an\n"
252 " existing network interface to the container\n"
253 " -n --network-veth Add a virtual Ethernet connection between host\n"
254 " and container\n"
255 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
256 " Add an additional virtual Ethernet link between\n"
257 " host and container\n"
258 " --network-bridge=INTERFACE\n"
259 " Add a virtual Ethernet connection to the container\n"
260 " and attach it to an existing bridge on the host\n"
261 " --network-zone=NAME Similar, but attach the new interface to an\n"
262 " an automatically managed bridge interface\n"
263 " --network-namespace-path=PATH\n"
264 " Set network namespace to the one represented by\n"
265 " the specified kernel namespace file node\n"
266 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
267 " Expose a container IP port on the host\n"
268 " -Z --selinux-context=SECLABEL\n"
269 " Set the SELinux security context to be used by\n"
270 " processes in the container\n"
271 " -L --selinux-apifs-context=SECLABEL\n"
272 " Set the SELinux security context to be used by\n"
273 " API/tmpfs file systems in the container\n"
274 " --capability=CAP In addition to the default, retain specified\n"
275 " capability\n"
276 " --drop-capability=CAP Drop the specified capability from the default set\n"
277 " --system-call-filter=LIST|~LIST\n"
278 " Permit/prohibit specific system calls\n"
279 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
280 " --oom-score-adjust=VALUE\n"
281 " Adjust the OOM score value for the payload\n"
282 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
283 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
284 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
285 " host, try-guest, try-host\n"
286 " -j Equivalent to --link-journal=try-guest\n"
287 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
288 " --timezone=MODE Select mode of /etc/localtime initialization\n"
289 " --read-only Mount the root directory read-only\n"
290 " --bind=PATH[:PATH[:OPTIONS]]\n"
291 " Bind mount a file or directory from the host into\n"
292 " the container\n"
293 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
294 " Similar, but creates a read-only bind mount\n"
295 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
296 " --overlay=PATH[:PATH...]:PATH\n"
297 " Create an overlay mount from the host to \n"
298 " the container\n"
299 " --overlay-ro=PATH[:PATH...]:PATH\n"
300 " Similar, but creates a read-only overlay mount\n"
301 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
302 " --register=BOOLEAN Register container as machine\n"
303 " --keep-unit Do not register a scope for the machine, reuse\n"
304 " the service unit nspawn is running in\n"
305 " --volatile[=MODE] Run the system in volatile mode\n"
306 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
307 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
308 "\nSee the %s for details.\n"
309 , program_invocation_short_name
310 , link
311 );
312
313 return 0;
314 }
315
316 static int custom_mount_check_all(void) {
317 size_t i;
318
319 for (i = 0; i < arg_n_custom_mounts; i++) {
320 CustomMount *m = &arg_custom_mounts[i];
321
322 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
323
324 if (arg_userns_chown) {
325 log_error("--private-users-chown may not be combined with custom root mounts.");
326 return -EINVAL;
327 } else if (arg_uid_shift == UID_INVALID) {
328 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
329 return -EINVAL;
330 }
331 }
332 }
333
334 return 0;
335 }
336
337 static int detect_unified_cgroup_hierarchy_from_environment(void) {
338 const char *e;
339 int r;
340
341 /* Allow the user to control whether the unified hierarchy is used */
342 e = getenv("UNIFIED_CGROUP_HIERARCHY");
343 if (e) {
344 r = parse_boolean(e);
345 if (r < 0)
346 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
347 if (r > 0)
348 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
349 else
350 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
351 }
352
353 return 0;
354 }
355
356 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
357 int r;
358
359 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
360 * image actually supports. */
361 r = cg_all_unified();
362 if (r < 0)
363 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
364 if (r > 0) {
365 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
366 * routine only detects 231, so we'll have a false negative here for 230. */
367 r = systemd_installation_has_version(directory, 230);
368 if (r < 0)
369 return log_error_errno(r, "Failed to determine systemd version in container: %m");
370 if (r > 0)
371 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
372 else
373 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
374 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
375 /* Mixed cgroup hierarchy support was added in 233 */
376 r = systemd_installation_has_version(directory, 233);
377 if (r < 0)
378 return log_error_errno(r, "Failed to determine systemd version in container: %m");
379 if (r > 0)
380 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
381 else
382 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
383 } else
384 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
385
386 log_debug("Using %s hierarchy for container.",
387 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
388 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
389
390 return 0;
391 }
392
393 static void parse_share_ns_env(const char *name, unsigned long ns_flag) {
394 int r;
395
396 r = getenv_bool(name);
397 if (r == -ENXIO)
398 return;
399 if (r < 0)
400 log_warning_errno(r, "Failed to parse %s from environment, defaulting to false.", name);
401 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
402 }
403
404 static void parse_mount_settings_env(void) {
405 const char *e;
406 int r;
407
408 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
409 if (r >= 0)
410 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
411 else if (r != -ENXIO)
412 log_warning_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP, ignoring: %m");
413
414 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
415 if (!e)
416 return;
417
418 if (streq(e, "network")) {
419 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
420 return;
421 }
422
423 r = parse_boolean(e);
424 if (r < 0) {
425 log_warning_errno(r, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
426 return;
427 }
428
429 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
430 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
431 }
432
433 static int parse_argv(int argc, char *argv[]) {
434 enum {
435 ARG_VERSION = 0x100,
436 ARG_PRIVATE_NETWORK,
437 ARG_UUID,
438 ARG_READ_ONLY,
439 ARG_CAPABILITY,
440 ARG_DROP_CAPABILITY,
441 ARG_LINK_JOURNAL,
442 ARG_BIND,
443 ARG_BIND_RO,
444 ARG_TMPFS,
445 ARG_OVERLAY,
446 ARG_OVERLAY_RO,
447 ARG_SHARE_SYSTEM,
448 ARG_REGISTER,
449 ARG_KEEP_UNIT,
450 ARG_NETWORK_INTERFACE,
451 ARG_NETWORK_MACVLAN,
452 ARG_NETWORK_IPVLAN,
453 ARG_NETWORK_BRIDGE,
454 ARG_NETWORK_ZONE,
455 ARG_NETWORK_VETH_EXTRA,
456 ARG_NETWORK_NAMESPACE_PATH,
457 ARG_PERSONALITY,
458 ARG_VOLATILE,
459 ARG_TEMPLATE,
460 ARG_PROPERTY,
461 ARG_PRIVATE_USERS,
462 ARG_KILL_SIGNAL,
463 ARG_SETTINGS,
464 ARG_CHDIR,
465 ARG_PIVOT_ROOT,
466 ARG_PRIVATE_USERS_CHOWN,
467 ARG_NOTIFY_READY,
468 ARG_ROOT_HASH,
469 ARG_SYSTEM_CALL_FILTER,
470 ARG_RLIMIT,
471 ARG_HOSTNAME,
472 ARG_NO_NEW_PRIVILEGES,
473 ARG_OOM_SCORE_ADJUST,
474 ARG_CPU_AFFINITY,
475 ARG_RESOLV_CONF,
476 ARG_TIMEZONE,
477 };
478
479 static const struct option options[] = {
480 { "help", no_argument, NULL, 'h' },
481 { "version", no_argument, NULL, ARG_VERSION },
482 { "directory", required_argument, NULL, 'D' },
483 { "template", required_argument, NULL, ARG_TEMPLATE },
484 { "ephemeral", no_argument, NULL, 'x' },
485 { "user", required_argument, NULL, 'u' },
486 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
487 { "as-pid2", no_argument, NULL, 'a' },
488 { "boot", no_argument, NULL, 'b' },
489 { "uuid", required_argument, NULL, ARG_UUID },
490 { "read-only", no_argument, NULL, ARG_READ_ONLY },
491 { "capability", required_argument, NULL, ARG_CAPABILITY },
492 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
493 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
494 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
495 { "bind", required_argument, NULL, ARG_BIND },
496 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
497 { "tmpfs", required_argument, NULL, ARG_TMPFS },
498 { "overlay", required_argument, NULL, ARG_OVERLAY },
499 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
500 { "machine", required_argument, NULL, 'M' },
501 { "hostname", required_argument, NULL, ARG_HOSTNAME },
502 { "slice", required_argument, NULL, 'S' },
503 { "setenv", required_argument, NULL, 'E' },
504 { "selinux-context", required_argument, NULL, 'Z' },
505 { "selinux-apifs-context", required_argument, NULL, 'L' },
506 { "quiet", no_argument, NULL, 'q' },
507 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
508 { "register", required_argument, NULL, ARG_REGISTER },
509 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
510 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
511 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
512 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
513 { "network-veth", no_argument, NULL, 'n' },
514 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
515 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
516 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
517 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
518 { "personality", required_argument, NULL, ARG_PERSONALITY },
519 { "image", required_argument, NULL, 'i' },
520 { "volatile", optional_argument, NULL, ARG_VOLATILE },
521 { "port", required_argument, NULL, 'p' },
522 { "property", required_argument, NULL, ARG_PROPERTY },
523 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
524 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
525 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
526 { "settings", required_argument, NULL, ARG_SETTINGS },
527 { "chdir", required_argument, NULL, ARG_CHDIR },
528 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
529 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
530 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
531 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
532 { "rlimit", required_argument, NULL, ARG_RLIMIT },
533 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
534 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
535 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
536 { "timezone", required_argument, NULL, ARG_TIMEZONE },
537 {}
538 };
539
540 int c, r;
541 const char *p, *e;
542 uint64_t plus = 0, minus = 0;
543 bool mask_all_settings = false, mask_no_settings = false;
544
545 assert(argc >= 0);
546 assert(argv);
547
548 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options, NULL)) >= 0)
549 switch (c) {
550
551 case 'h':
552 return help();
553
554 case ARG_VERSION:
555 return version();
556
557 case 'D':
558 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
559 if (r < 0)
560 return r;
561 break;
562
563 case ARG_TEMPLATE:
564 r = parse_path_argument_and_warn(optarg, false, &arg_template);
565 if (r < 0)
566 return r;
567 break;
568
569 case 'i':
570 r = parse_path_argument_and_warn(optarg, false, &arg_image);
571 if (r < 0)
572 return r;
573 break;
574
575 case 'x':
576 arg_ephemeral = true;
577 arg_settings_mask |= SETTING_EPHEMERAL;
578 break;
579
580 case 'u':
581 r = free_and_strdup(&arg_user, optarg);
582 if (r < 0)
583 return log_oom();
584
585 arg_settings_mask |= SETTING_USER;
586 break;
587
588 case ARG_NETWORK_ZONE: {
589 char *j;
590
591 j = strappend("vz-", optarg);
592 if (!j)
593 return log_oom();
594
595 if (!ifname_valid(j)) {
596 log_error("Network zone name not valid: %s", j);
597 free(j);
598 return -EINVAL;
599 }
600
601 free_and_replace(arg_network_zone, j);
602
603 arg_network_veth = true;
604 arg_private_network = true;
605 arg_settings_mask |= SETTING_NETWORK;
606 break;
607 }
608
609 case ARG_NETWORK_BRIDGE:
610
611 if (!ifname_valid(optarg)) {
612 log_error("Bridge interface name not valid: %s", optarg);
613 return -EINVAL;
614 }
615
616 r = free_and_strdup(&arg_network_bridge, optarg);
617 if (r < 0)
618 return log_oom();
619
620 _fallthrough_;
621 case 'n':
622 arg_network_veth = true;
623 arg_private_network = true;
624 arg_settings_mask |= SETTING_NETWORK;
625 break;
626
627 case ARG_NETWORK_VETH_EXTRA:
628 r = veth_extra_parse(&arg_network_veth_extra, optarg);
629 if (r < 0)
630 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
631
632 arg_private_network = true;
633 arg_settings_mask |= SETTING_NETWORK;
634 break;
635
636 case ARG_NETWORK_INTERFACE:
637 if (!ifname_valid(optarg)) {
638 log_error("Network interface name not valid: %s", optarg);
639 return -EINVAL;
640 }
641
642 if (strv_extend(&arg_network_interfaces, optarg) < 0)
643 return log_oom();
644
645 arg_private_network = true;
646 arg_settings_mask |= SETTING_NETWORK;
647 break;
648
649 case ARG_NETWORK_MACVLAN:
650
651 if (!ifname_valid(optarg)) {
652 log_error("MACVLAN network interface name not valid: %s", optarg);
653 return -EINVAL;
654 }
655
656 if (strv_extend(&arg_network_macvlan, optarg) < 0)
657 return log_oom();
658
659 arg_private_network = true;
660 arg_settings_mask |= SETTING_NETWORK;
661 break;
662
663 case ARG_NETWORK_IPVLAN:
664
665 if (!ifname_valid(optarg)) {
666 log_error("IPVLAN network interface name not valid: %s", optarg);
667 return -EINVAL;
668 }
669
670 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
671 return log_oom();
672
673 _fallthrough_;
674 case ARG_PRIVATE_NETWORK:
675 arg_private_network = true;
676 arg_settings_mask |= SETTING_NETWORK;
677 break;
678
679 case ARG_NETWORK_NAMESPACE_PATH:
680 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
681 if (r < 0)
682 return r;
683
684 break;
685
686 case 'b':
687 if (arg_start_mode == START_PID2) {
688 log_error("--boot and --as-pid2 may not be combined.");
689 return -EINVAL;
690 }
691
692 arg_start_mode = START_BOOT;
693 arg_settings_mask |= SETTING_START_MODE;
694 break;
695
696 case 'a':
697 if (arg_start_mode == START_BOOT) {
698 log_error("--boot and --as-pid2 may not be combined.");
699 return -EINVAL;
700 }
701
702 arg_start_mode = START_PID2;
703 arg_settings_mask |= SETTING_START_MODE;
704 break;
705
706 case ARG_UUID:
707 r = sd_id128_from_string(optarg, &arg_uuid);
708 if (r < 0)
709 return log_error_errno(r, "Invalid UUID: %s", optarg);
710
711 if (sd_id128_is_null(arg_uuid)) {
712 log_error("Machine UUID may not be all zeroes.");
713 return -EINVAL;
714 }
715
716 arg_settings_mask |= SETTING_MACHINE_ID;
717 break;
718
719 case 'S':
720 arg_slice = optarg;
721 break;
722
723 case 'M':
724 if (isempty(optarg))
725 arg_machine = mfree(arg_machine);
726 else {
727 if (!machine_name_is_valid(optarg)) {
728 log_error("Invalid machine name: %s", optarg);
729 return -EINVAL;
730 }
731
732 r = free_and_strdup(&arg_machine, optarg);
733 if (r < 0)
734 return log_oom();
735 }
736 break;
737
738 case ARG_HOSTNAME:
739 if (isempty(optarg))
740 arg_hostname = mfree(arg_hostname);
741 else {
742 if (!hostname_is_valid(optarg, false)) {
743 log_error("Invalid hostname: %s", optarg);
744 return -EINVAL;
745 }
746
747 r = free_and_strdup(&arg_hostname, optarg);
748 if (r < 0)
749 return log_oom();
750 }
751
752 arg_settings_mask |= SETTING_HOSTNAME;
753 break;
754
755 case 'Z':
756 arg_selinux_context = optarg;
757 break;
758
759 case 'L':
760 arg_selinux_apifs_context = optarg;
761 break;
762
763 case ARG_READ_ONLY:
764 arg_read_only = true;
765 arg_settings_mask |= SETTING_READ_ONLY;
766 break;
767
768 case ARG_CAPABILITY:
769 case ARG_DROP_CAPABILITY: {
770 p = optarg;
771 for (;;) {
772 _cleanup_free_ char *t = NULL;
773
774 r = extract_first_word(&p, &t, ",", 0);
775 if (r < 0)
776 return log_error_errno(r, "Failed to parse capability %s.", t);
777
778 if (r == 0)
779 break;
780
781 if (streq(t, "all")) {
782 if (c == ARG_CAPABILITY)
783 plus = (uint64_t) -1;
784 else
785 minus = (uint64_t) -1;
786 } else {
787 int cap;
788
789 cap = capability_from_name(t);
790 if (cap < 0) {
791 log_error("Failed to parse capability %s.", t);
792 return -EINVAL;
793 }
794
795 if (c == ARG_CAPABILITY)
796 plus |= 1ULL << (uint64_t) cap;
797 else
798 minus |= 1ULL << (uint64_t) cap;
799 }
800 }
801
802 arg_settings_mask |= SETTING_CAPABILITY;
803 break;
804 }
805
806 case ARG_NO_NEW_PRIVILEGES:
807 r = parse_boolean(optarg);
808 if (r < 0)
809 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
810
811 arg_no_new_privileges = r;
812 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
813 break;
814
815 case 'j':
816 arg_link_journal = LINK_GUEST;
817 arg_link_journal_try = true;
818 arg_settings_mask |= SETTING_LINK_JOURNAL;
819 break;
820
821 case ARG_LINK_JOURNAL:
822 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
823 if (r < 0) {
824 log_error_errno(r, "Failed to parse link journal mode %s", optarg);
825 return -EINVAL;
826 }
827
828 arg_settings_mask |= SETTING_LINK_JOURNAL;
829 break;
830
831 case ARG_BIND:
832 case ARG_BIND_RO:
833 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
834 if (r < 0)
835 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
836
837 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
838 break;
839
840 case ARG_TMPFS:
841 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
842 if (r < 0)
843 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
844
845 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
846 break;
847
848 case ARG_OVERLAY:
849 case ARG_OVERLAY_RO:
850 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
851 if (r == -EADDRNOTAVAIL)
852 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
853 if (r < 0)
854 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
855
856 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
857 break;
858
859 case 'E': {
860 char **n;
861
862 if (!env_assignment_is_valid(optarg)) {
863 log_error("Environment variable assignment '%s' is not valid.", optarg);
864 return -EINVAL;
865 }
866
867 n = strv_env_set(arg_setenv, optarg);
868 if (!n)
869 return log_oom();
870
871 strv_free_and_replace(arg_setenv, n);
872 arg_settings_mask |= SETTING_ENVIRONMENT;
873 break;
874 }
875
876 case 'q':
877 arg_quiet = true;
878 break;
879
880 case ARG_SHARE_SYSTEM:
881 /* We don't officially support this anymore, except for compat reasons. People should use the
882 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
883 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
884 arg_clone_ns_flags = 0;
885 break;
886
887 case ARG_REGISTER:
888 r = parse_boolean(optarg);
889 if (r < 0) {
890 log_error("Failed to parse --register= argument: %s", optarg);
891 return r;
892 }
893
894 arg_register = r;
895 break;
896
897 case ARG_KEEP_UNIT:
898 arg_keep_unit = true;
899 break;
900
901 case ARG_PERSONALITY:
902
903 arg_personality = personality_from_string(optarg);
904 if (arg_personality == PERSONALITY_INVALID) {
905 log_error("Unknown or unsupported personality '%s'.", optarg);
906 return -EINVAL;
907 }
908
909 arg_settings_mask |= SETTING_PERSONALITY;
910 break;
911
912 case ARG_VOLATILE:
913
914 if (!optarg)
915 arg_volatile_mode = VOLATILE_YES;
916 else if (streq(optarg, "help")) {
917 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
918 return 0;
919 } else {
920 VolatileMode m;
921
922 m = volatile_mode_from_string(optarg);
923 if (m < 0) {
924 log_error("Failed to parse --volatile= argument: %s", optarg);
925 return -EINVAL;
926 } else
927 arg_volatile_mode = m;
928 }
929
930 arg_settings_mask |= SETTING_VOLATILE_MODE;
931 break;
932
933 case 'p':
934 r = expose_port_parse(&arg_expose_ports, optarg);
935 if (r == -EEXIST)
936 return log_error_errno(r, "Duplicate port specification: %s", optarg);
937 if (r < 0)
938 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
939
940 arg_settings_mask |= SETTING_EXPOSE_PORTS;
941 break;
942
943 case ARG_PROPERTY:
944 if (strv_extend(&arg_property, optarg) < 0)
945 return log_oom();
946
947 break;
948
949 case ARG_PRIVATE_USERS: {
950 int boolean = -1;
951
952 if (!optarg)
953 boolean = true;
954 else if (!in_charset(optarg, DIGITS))
955 /* do *not* parse numbers as booleans */
956 boolean = parse_boolean(optarg);
957
958 if (boolean == false) {
959 /* no: User namespacing off */
960 arg_userns_mode = USER_NAMESPACE_NO;
961 arg_uid_shift = UID_INVALID;
962 arg_uid_range = UINT32_C(0x10000);
963 } else if (boolean == true) {
964 /* yes: User namespacing on, UID range is read from root dir */
965 arg_userns_mode = USER_NAMESPACE_FIXED;
966 arg_uid_shift = UID_INVALID;
967 arg_uid_range = UINT32_C(0x10000);
968 } else if (streq(optarg, "pick")) {
969 /* pick: User namespacing on, UID range is picked randomly */
970 arg_userns_mode = USER_NAMESPACE_PICK;
971 arg_uid_shift = UID_INVALID;
972 arg_uid_range = UINT32_C(0x10000);
973 } else {
974 _cleanup_free_ char *buffer = NULL;
975 const char *range, *shift;
976
977 /* anything else: User namespacing on, UID range is explicitly configured */
978
979 range = strchr(optarg, ':');
980 if (range) {
981 buffer = strndup(optarg, range - optarg);
982 if (!buffer)
983 return log_oom();
984 shift = buffer;
985
986 range++;
987 r = safe_atou32(range, &arg_uid_range);
988 if (r < 0)
989 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
990 } else
991 shift = optarg;
992
993 r = parse_uid(shift, &arg_uid_shift);
994 if (r < 0)
995 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
996
997 arg_userns_mode = USER_NAMESPACE_FIXED;
998 }
999
1000 if (arg_uid_range <= 0) {
1001 log_error("UID range cannot be 0.");
1002 return -EINVAL;
1003 }
1004
1005 arg_settings_mask |= SETTING_USERNS;
1006 break;
1007 }
1008
1009 case 'U':
1010 if (userns_supported()) {
1011 arg_userns_mode = USER_NAMESPACE_PICK;
1012 arg_uid_shift = UID_INVALID;
1013 arg_uid_range = UINT32_C(0x10000);
1014
1015 arg_settings_mask |= SETTING_USERNS;
1016 }
1017
1018 break;
1019
1020 case ARG_PRIVATE_USERS_CHOWN:
1021 arg_userns_chown = true;
1022
1023 arg_settings_mask |= SETTING_USERNS;
1024 break;
1025
1026 case ARG_KILL_SIGNAL:
1027 if (streq(optarg, "help")) {
1028 DUMP_STRING_TABLE(signal, int, _NSIG);
1029 return 0;
1030 }
1031
1032 arg_kill_signal = signal_from_string(optarg);
1033 if (arg_kill_signal < 0) {
1034 log_error("Cannot parse signal: %s", optarg);
1035 return -EINVAL;
1036 }
1037
1038 arg_settings_mask |= SETTING_KILL_SIGNAL;
1039 break;
1040
1041 case ARG_SETTINGS:
1042
1043 /* no → do not read files
1044 * yes → read files, do not override cmdline, trust only subset
1045 * override → read files, override cmdline, trust only subset
1046 * trusted → read files, do not override cmdline, trust all
1047 */
1048
1049 r = parse_boolean(optarg);
1050 if (r < 0) {
1051 if (streq(optarg, "trusted")) {
1052 mask_all_settings = false;
1053 mask_no_settings = false;
1054 arg_settings_trusted = true;
1055
1056 } else if (streq(optarg, "override")) {
1057 mask_all_settings = false;
1058 mask_no_settings = true;
1059 arg_settings_trusted = -1;
1060 } else
1061 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1062 } else if (r > 0) {
1063 /* yes */
1064 mask_all_settings = false;
1065 mask_no_settings = false;
1066 arg_settings_trusted = -1;
1067 } else {
1068 /* no */
1069 mask_all_settings = true;
1070 mask_no_settings = false;
1071 arg_settings_trusted = false;
1072 }
1073
1074 break;
1075
1076 case ARG_CHDIR:
1077 if (!path_is_absolute(optarg)) {
1078 log_error("Working directory %s is not an absolute path.", optarg);
1079 return -EINVAL;
1080 }
1081
1082 r = free_and_strdup(&arg_chdir, optarg);
1083 if (r < 0)
1084 return log_oom();
1085
1086 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1087 break;
1088
1089 case ARG_PIVOT_ROOT:
1090 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1091 if (r < 0)
1092 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1093
1094 arg_settings_mask |= SETTING_PIVOT_ROOT;
1095 break;
1096
1097 case ARG_NOTIFY_READY:
1098 r = parse_boolean(optarg);
1099 if (r < 0) {
1100 log_error("%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1101 return -EINVAL;
1102 }
1103 arg_notify_ready = r;
1104 arg_settings_mask |= SETTING_NOTIFY_READY;
1105 break;
1106
1107 case ARG_ROOT_HASH: {
1108 void *k;
1109 size_t l;
1110
1111 r = unhexmem(optarg, strlen(optarg), &k, &l);
1112 if (r < 0)
1113 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1114 if (l < sizeof(sd_id128_t)) {
1115 log_error("Root hash must be at least 128bit long: %s", optarg);
1116 free(k);
1117 return -EINVAL;
1118 }
1119
1120 free(arg_root_hash);
1121 arg_root_hash = k;
1122 arg_root_hash_size = l;
1123 break;
1124 }
1125
1126 case ARG_SYSTEM_CALL_FILTER: {
1127 bool negative;
1128 const char *items;
1129
1130 negative = optarg[0] == '~';
1131 items = negative ? optarg + 1 : optarg;
1132
1133 for (;;) {
1134 _cleanup_free_ char *word = NULL;
1135
1136 r = extract_first_word(&items, &word, NULL, 0);
1137 if (r == 0)
1138 break;
1139 if (r == -ENOMEM)
1140 return log_oom();
1141 if (r < 0)
1142 return log_error_errno(r, "Failed to parse system call filter: %m");
1143
1144 if (negative)
1145 r = strv_extend(&arg_syscall_blacklist, word);
1146 else
1147 r = strv_extend(&arg_syscall_whitelist, word);
1148 if (r < 0)
1149 return log_oom();
1150 }
1151
1152 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1153 break;
1154 }
1155
1156 case ARG_RLIMIT: {
1157 const char *eq;
1158 char *name;
1159 int rl;
1160
1161 if (streq(optarg, "help")) {
1162 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1163 return 0;
1164 }
1165
1166 eq = strchr(optarg, '=');
1167 if (!eq) {
1168 log_error("--rlimit= expects an '=' assignment.");
1169 return -EINVAL;
1170 }
1171
1172 name = strndup(optarg, eq - optarg);
1173 if (!name)
1174 return log_oom();
1175
1176 rl = rlimit_from_string_harder(name);
1177 if (rl < 0) {
1178 log_error("Unknown resource limit: %s", name);
1179 return -EINVAL;
1180 }
1181
1182 if (!arg_rlimit[rl]) {
1183 arg_rlimit[rl] = new0(struct rlimit, 1);
1184 if (!arg_rlimit[rl])
1185 return log_oom();
1186 }
1187
1188 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1189 if (r < 0)
1190 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1191
1192 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1193 break;
1194 }
1195
1196 case ARG_OOM_SCORE_ADJUST:
1197 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1198 if (r < 0)
1199 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1200
1201 arg_oom_score_adjust_set = true;
1202 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1203 break;
1204
1205 case ARG_CPU_AFFINITY: {
1206 _cleanup_cpu_free_ cpu_set_t *cpuset = NULL;
1207
1208 r = parse_cpu_set(optarg, &cpuset);
1209 if (r < 0)
1210 return log_error_errno(r, "Failed to parse CPU affinity mask: %s", optarg);
1211
1212 if (arg_cpuset)
1213 CPU_FREE(arg_cpuset);
1214
1215 arg_cpuset = TAKE_PTR(cpuset);
1216 arg_cpuset_ncpus = r;
1217 arg_settings_mask |= SETTING_CPU_AFFINITY;
1218 break;
1219 }
1220
1221 case ARG_RESOLV_CONF:
1222 if (streq(optarg, "help")) {
1223 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1224 return 0;
1225 }
1226
1227 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1228 if (arg_resolv_conf < 0) {
1229 log_error("Failed to parse /etc/resolv.conf mode: %s", optarg);
1230 return -EINVAL;
1231 }
1232
1233 arg_settings_mask |= SETTING_RESOLV_CONF;
1234 break;
1235
1236 case ARG_TIMEZONE:
1237 if (streq(optarg, "help")) {
1238 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1239 return 0;
1240 }
1241
1242 arg_timezone = timezone_mode_from_string(optarg);
1243 if (arg_timezone < 0) {
1244 log_error("Failed to parse /etc/localtime mode: %s", optarg);
1245 return -EINVAL;
1246 }
1247
1248 arg_settings_mask |= SETTING_TIMEZONE;
1249 break;
1250
1251 case '?':
1252 return -EINVAL;
1253
1254 default:
1255 assert_not_reached("Unhandled option");
1256 }
1257
1258 /* If --network-namespace-path is given with any other network-related option,
1259 * we need to error out, to avoid conflicts between different network options. */
1260 if (arg_network_namespace_path &&
1261 (arg_network_interfaces || arg_network_macvlan ||
1262 arg_network_ipvlan || arg_network_veth_extra ||
1263 arg_network_bridge || arg_network_zone ||
1264 arg_network_veth || arg_private_network)) {
1265 log_error("--network-namespace-path cannot be combined with other network options.");
1266 return -EINVAL;
1267 }
1268
1269 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
1270 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
1271 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
1272 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
1273
1274 if (arg_userns_mode != USER_NAMESPACE_NO)
1275 arg_mount_settings |= MOUNT_USE_USERNS;
1276
1277 if (arg_private_network)
1278 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1279
1280 parse_mount_settings_env();
1281
1282 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1283 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1284 arg_register = false;
1285 if (arg_start_mode != START_PID1) {
1286 log_error("--boot cannot be used without namespacing.");
1287 return -EINVAL;
1288 }
1289 }
1290
1291 if (arg_userns_mode == USER_NAMESPACE_PICK)
1292 arg_userns_chown = true;
1293
1294 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0) {
1295 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1296 * The latter is not technically a user session, but we don't need to labour the point. */
1297 log_error("--keep-unit --register=yes may not be used when invoked from a user session.");
1298 return -EINVAL;
1299 }
1300
1301 if (arg_directory && arg_image) {
1302 log_error("--directory= and --image= may not be combined.");
1303 return -EINVAL;
1304 }
1305
1306 if (arg_template && arg_image) {
1307 log_error("--template= and --image= may not be combined.");
1308 return -EINVAL;
1309 }
1310
1311 if (arg_ephemeral && arg_template && !arg_directory) {
1312 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1313 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1314 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1315 * --directory=". */
1316
1317 arg_directory = TAKE_PTR(arg_template);
1318 }
1319
1320 if (arg_template && !(arg_directory || arg_machine)) {
1321 log_error("--template= needs --directory= or --machine=.");
1322 return -EINVAL;
1323 }
1324
1325 if (arg_ephemeral && arg_template) {
1326 log_error("--ephemeral and --template= may not be combined.");
1327 return -EINVAL;
1328 }
1329
1330 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
1331 log_error("--ephemeral and --link-journal= may not be combined.");
1332 return -EINVAL;
1333 }
1334
1335 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported()) {
1336 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1337 return -EOPNOTSUPP;
1338 }
1339
1340 if (arg_userns_chown && arg_read_only) {
1341 log_error("--read-only and --private-users-chown may not be combined.");
1342 return -EINVAL;
1343 }
1344
1345 if (arg_network_bridge && arg_network_zone) {
1346 log_error("--network-bridge= and --network-zone= may not be combined.");
1347 return -EINVAL;
1348 }
1349
1350 if (argc > optind) {
1351 arg_parameters = strv_copy(argv + optind);
1352 if (!arg_parameters)
1353 return log_oom();
1354
1355 arg_settings_mask |= SETTING_START_MODE;
1356 }
1357
1358 /* Load all settings from .nspawn files */
1359 if (mask_no_settings)
1360 arg_settings_mask = 0;
1361
1362 /* Don't load any settings from .nspawn files */
1363 if (mask_all_settings)
1364 arg_settings_mask = _SETTINGS_MASK_ALL;
1365
1366 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
1367
1368 r = cg_unified_flush();
1369 if (r < 0)
1370 return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
1371
1372 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1373 if (e)
1374 arg_container_service_name = e;
1375
1376 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1377 if (r < 0)
1378 arg_use_cgns = cg_ns_supported();
1379 else
1380 arg_use_cgns = r;
1381
1382 r = custom_mount_check_all();
1383 if (r < 0)
1384 return r;
1385
1386 return 1;
1387 }
1388
1389 static int verify_arguments(void) {
1390 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network) {
1391 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1392 return -EINVAL;
1393 }
1394
1395 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO)) {
1396 log_error("Cannot combine --private-users with read-write mounts.");
1397 return -EINVAL;
1398 }
1399
1400 if (arg_volatile_mode != VOLATILE_NO && arg_read_only) {
1401 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1402 return -EINVAL;
1403 }
1404
1405 if (arg_expose_ports && !arg_private_network) {
1406 log_error("Cannot use --port= without private networking.");
1407 return -EINVAL;
1408 }
1409
1410 #if ! HAVE_LIBIPTC
1411 if (arg_expose_ports) {
1412 log_error("--port= is not supported, compiled without libiptc support.");
1413 return -EOPNOTSUPP;
1414 }
1415 #endif
1416
1417 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1418 arg_kill_signal = SIGRTMIN+3;
1419
1420 return 0;
1421 }
1422
1423 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1424 assert(p);
1425
1426 if (arg_userns_mode == USER_NAMESPACE_NO)
1427 return 0;
1428
1429 if (uid == UID_INVALID && gid == GID_INVALID)
1430 return 0;
1431
1432 if (uid != UID_INVALID) {
1433 uid += arg_uid_shift;
1434
1435 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1436 return -EOVERFLOW;
1437 }
1438
1439 if (gid != GID_INVALID) {
1440 gid += (gid_t) arg_uid_shift;
1441
1442 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1443 return -EOVERFLOW;
1444 }
1445
1446 if (lchown(p, uid, gid) < 0)
1447 return -errno;
1448
1449 return 0;
1450 }
1451
1452 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1453 const char *q;
1454 int r;
1455
1456 q = prefix_roota(root, path);
1457 r = mkdir_errno_wrapper(q, mode);
1458 if (r == -EEXIST)
1459 return 0;
1460 if (r < 0)
1461 return r;
1462
1463 return userns_lchown(q, uid, gid);
1464 }
1465
1466 static const char *timezone_from_path(const char *path) {
1467 const char *z;
1468
1469 z = path_startswith(path, "../usr/share/zoneinfo/");
1470 if (z)
1471 return z;
1472
1473 z = path_startswith(path, "/usr/share/zoneinfo/");
1474 if (z)
1475 return z;
1476
1477 return NULL;
1478 }
1479
1480 static int setup_timezone(const char *dest) {
1481 _cleanup_free_ char *p = NULL, *etc = NULL;
1482 const char *where, *check;
1483 TimezoneMode m;
1484 int r;
1485
1486 assert(dest);
1487
1488 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1489 r = readlink_malloc("/etc/localtime", &p);
1490 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1491 m = arg_read_only && arg_volatile_mode != VOLATILE_YES ? TIMEZONE_OFF : TIMEZONE_DELETE;
1492 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1493 m = arg_read_only && arg_volatile_mode != VOLATILE_YES ? TIMEZONE_BIND : TIMEZONE_COPY;
1494 else if (r < 0) {
1495 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1496 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1497 * file.
1498 *
1499 * Example:
1500 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1501 */
1502 return 0;
1503 } else if (arg_timezone == TIMEZONE_AUTO)
1504 m = arg_read_only && arg_volatile_mode != VOLATILE_YES ? TIMEZONE_BIND : TIMEZONE_SYMLINK;
1505 else
1506 m = arg_timezone;
1507 } else
1508 m = arg_timezone;
1509
1510 if (m == TIMEZONE_OFF)
1511 return 0;
1512
1513 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1514 if (r < 0) {
1515 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1516 return 0;
1517 }
1518
1519 where = strjoina(etc, "/localtime");
1520
1521 switch (m) {
1522
1523 case TIMEZONE_DELETE:
1524 if (unlink(where) < 0)
1525 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1526
1527 return 0;
1528
1529 case TIMEZONE_SYMLINK: {
1530 _cleanup_free_ char *q = NULL;
1531 const char *z, *what;
1532
1533 z = timezone_from_path(p);
1534 if (!z) {
1535 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1536 return 0;
1537 }
1538
1539 r = readlink_malloc(where, &q);
1540 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1541 return 0; /* Already pointing to the right place? Then do nothing .. */
1542
1543 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1544 r = chase_symlinks(check, dest, 0, NULL);
1545 if (r < 0)
1546 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1547 else {
1548 if (unlink(where) < 0 && errno != ENOENT) {
1549 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1550 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1551 return 0;
1552 }
1553
1554 what = strjoina("../usr/share/zoneinfo/", z);
1555 if (symlink(what, where) < 0) {
1556 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1557 errno, "Failed to correct timezone of container, ignoring: %m");
1558 return 0;
1559 }
1560
1561 break;
1562 }
1563
1564 _fallthrough_;
1565 }
1566
1567 case TIMEZONE_BIND: {
1568 _cleanup_free_ char *resolved = NULL;
1569 int found;
1570
1571 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1572 if (found < 0) {
1573 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1574 return 0;
1575 }
1576
1577 if (found == 0) /* missing? */
1578 (void) touch(resolved);
1579
1580 r = mount_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1581 if (r >= 0)
1582 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1583
1584 _fallthrough_;
1585 }
1586
1587 case TIMEZONE_COPY:
1588 /* If mounting failed, try to copy */
1589 r = copy_file_atomic("/etc/localtime", where, 0644, 0, COPY_REFLINK|COPY_REPLACE);
1590 if (r < 0) {
1591 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1592 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
1593 return 0;
1594 }
1595
1596 break;
1597
1598 default:
1599 assert_not_reached("unexpected mode");
1600 }
1601
1602 /* Fix permissions of the symlink or file copy we just created */
1603 r = userns_lchown(where, 0, 0);
1604 if (r < 0)
1605 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
1606
1607 return 0;
1608 }
1609
1610 static int have_resolv_conf(const char *path) {
1611 assert(path);
1612
1613 if (access(path, F_OK) < 0) {
1614 if (errno == ENOENT)
1615 return 0;
1616
1617 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1618 }
1619
1620 return 1;
1621 }
1622
1623 static int resolved_listening(void) {
1624 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1625 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1626 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1627 int r;
1628
1629 /* Check if resolved is listening */
1630
1631 r = sd_bus_open_system(&bus);
1632 if (r < 0)
1633 return log_debug_errno(r, "Failed to open system bus: %m");
1634
1635 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1636 if (r < 0)
1637 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1638 if (r == 0)
1639 return 0;
1640
1641 r = sd_bus_get_property_string(bus,
1642 "org.freedesktop.resolve1",
1643 "/org/freedesktop/resolve1",
1644 "org.freedesktop.resolve1.Manager",
1645 "DNSStubListener",
1646 &error,
1647 &dns_stub_listener_mode);
1648 if (r < 0)
1649 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
1650
1651 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1652 }
1653
1654 static int setup_resolv_conf(const char *dest) {
1655 _cleanup_free_ char *etc = NULL;
1656 const char *where, *what;
1657 ResolvConfMode m;
1658 int r;
1659
1660 assert(dest);
1661
1662 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
1663 if (arg_private_network)
1664 m = RESOLV_CONF_OFF;
1665 else if (have_resolv_conf(STATIC_RESOLV_CONF) > 0 && resolved_listening() > 0)
1666 m = arg_read_only && arg_volatile_mode != VOLATILE_YES ? RESOLV_CONF_BIND_STATIC : RESOLV_CONF_COPY_STATIC;
1667 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1668 m = arg_read_only && arg_volatile_mode != VOLATILE_YES ? RESOLV_CONF_BIND_HOST : RESOLV_CONF_COPY_HOST;
1669 else
1670 m = arg_read_only && arg_volatile_mode != VOLATILE_YES ? RESOLV_CONF_OFF : RESOLV_CONF_DELETE;
1671 } else
1672 m = arg_resolv_conf;
1673
1674 if (m == RESOLV_CONF_OFF)
1675 return 0;
1676
1677 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1678 if (r < 0) {
1679 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1680 return 0;
1681 }
1682
1683 where = strjoina(etc, "/resolv.conf");
1684
1685 if (m == RESOLV_CONF_DELETE) {
1686 if (unlink(where) < 0)
1687 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1688
1689 return 0;
1690 }
1691
1692 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_COPY_STATIC))
1693 what = STATIC_RESOLV_CONF;
1694 else
1695 what = "/etc/resolv.conf";
1696
1697 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC)) {
1698 _cleanup_free_ char *resolved = NULL;
1699 int found;
1700
1701 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1702 if (found < 0) {
1703 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1704 return 0;
1705 }
1706
1707 if (found == 0) /* missing? */
1708 (void) touch(resolved);
1709
1710 r = mount_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
1711 if (r >= 0)
1712 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1713 }
1714
1715 /* If that didn't work, let's copy the file */
1716 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, COPY_REFLINK);
1717 if (r < 0) {
1718 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1719 * resolved or something similar runs inside and the symlink points there.
1720 *
1721 * If the disk image is read-only, there's also no point in complaining.
1722 */
1723 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC) && IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1724 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1725 return 0;
1726 }
1727
1728 r = userns_lchown(where, 0, 0);
1729 if (r < 0)
1730 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1731
1732 return 0;
1733 }
1734
1735 static int setup_boot_id(void) {
1736 _cleanup_(unlink_and_freep) char *from = NULL;
1737 _cleanup_free_ char *path = NULL;
1738 sd_id128_t rnd = SD_ID128_NULL;
1739 const char *to;
1740 int r;
1741
1742 /* Generate a new randomized boot ID, so that each boot-up of
1743 * the container gets a new one */
1744
1745 r = tempfn_random_child(NULL, "proc-sys-kernel-random-boot-id", &path);
1746 if (r < 0)
1747 return log_error_errno(r, "Failed to generate random boot ID path: %m");
1748
1749 r = sd_id128_randomize(&rnd);
1750 if (r < 0)
1751 return log_error_errno(r, "Failed to generate random boot id: %m");
1752
1753 r = id128_write(path, ID128_UUID, rnd, false);
1754 if (r < 0)
1755 return log_error_errno(r, "Failed to write boot id: %m");
1756
1757 from = TAKE_PTR(path);
1758 to = "/proc/sys/kernel/random/boot_id";
1759
1760 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1761 if (r < 0)
1762 return r;
1763
1764 return mount_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1765 }
1766
1767 static int copy_devnodes(const char *dest) {
1768 static const char devnodes[] =
1769 "null\0"
1770 "zero\0"
1771 "full\0"
1772 "random\0"
1773 "urandom\0"
1774 "tty\0"
1775 "net/tun\0";
1776
1777 const char *d;
1778 int r = 0;
1779 _cleanup_umask_ mode_t u;
1780
1781 assert(dest);
1782
1783 u = umask(0000);
1784
1785 /* Create /dev/net, so that we can create /dev/net/tun in it */
1786 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1787 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1788
1789 NULSTR_FOREACH(d, devnodes) {
1790 _cleanup_free_ char *from = NULL, *to = NULL;
1791 struct stat st;
1792
1793 from = strappend("/dev/", d);
1794 if (!from)
1795 return log_oom();
1796
1797 to = prefix_root(dest, from);
1798 if (!to)
1799 return log_oom();
1800
1801 if (stat(from, &st) < 0) {
1802
1803 if (errno != ENOENT)
1804 return log_error_errno(errno, "Failed to stat %s: %m", from);
1805
1806 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1807
1808 log_error("%s is not a char or block device, cannot copy.", from);
1809 return -EIO;
1810
1811 } else {
1812 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
1813
1814 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1815 /* Explicitly warn the user when /dev is already populated. */
1816 if (errno == EEXIST)
1817 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);
1818 if (errno != EPERM)
1819 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1820
1821 /* Some systems abusively restrict mknod but allow bind mounts. */
1822 r = touch(to);
1823 if (r < 0)
1824 return log_error_errno(r, "touch (%s) failed: %m", to);
1825 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1826 if (r < 0)
1827 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1828 }
1829
1830 r = userns_lchown(to, 0, 0);
1831 if (r < 0)
1832 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1833
1834 dn = strjoin("/dev/", S_ISCHR(st.st_mode) ? "char" : "block");
1835 if (!dn)
1836 return log_oom();
1837
1838 r = userns_mkdir(dest, dn, 0755, 0, 0);
1839 if (r < 0)
1840 return log_error_errno(r, "Failed to create '%s': %m", dn);
1841
1842 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
1843 return log_oom();
1844
1845 prefixed = prefix_root(dest, sl);
1846 if (!prefixed)
1847 return log_oom();
1848
1849 t = strjoin("../", d);
1850 if (!t)
1851 return log_oom();
1852
1853 if (symlink(t, prefixed) < 0)
1854 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
1855 }
1856 }
1857
1858 return r;
1859 }
1860
1861 static int setup_pts(const char *dest) {
1862 _cleanup_free_ char *options = NULL;
1863 const char *p;
1864 int r;
1865
1866 #if HAVE_SELINUX
1867 if (arg_selinux_apifs_context)
1868 (void) asprintf(&options,
1869 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1870 arg_uid_shift + TTY_GID,
1871 arg_selinux_apifs_context);
1872 else
1873 #endif
1874 (void) asprintf(&options,
1875 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1876 arg_uid_shift + TTY_GID);
1877
1878 if (!options)
1879 return log_oom();
1880
1881 /* Mount /dev/pts itself */
1882 p = prefix_roota(dest, "/dev/pts");
1883 r = mkdir_errno_wrapper(p, 0755);
1884 if (r < 0)
1885 return log_error_errno(r, "Failed to create /dev/pts: %m");
1886
1887 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
1888 if (r < 0)
1889 return r;
1890 r = userns_lchown(p, 0, 0);
1891 if (r < 0)
1892 return log_error_errno(r, "Failed to chown /dev/pts: %m");
1893
1894 /* Create /dev/ptmx symlink */
1895 p = prefix_roota(dest, "/dev/ptmx");
1896 if (symlink("pts/ptmx", p) < 0)
1897 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1898 r = userns_lchown(p, 0, 0);
1899 if (r < 0)
1900 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
1901
1902 /* And fix /dev/pts/ptmx ownership */
1903 p = prefix_roota(dest, "/dev/pts/ptmx");
1904 r = userns_lchown(p, 0, 0);
1905 if (r < 0)
1906 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
1907
1908 return 0;
1909 }
1910
1911 static int setup_dev_console(const char *dest, const char *console) {
1912 _cleanup_umask_ mode_t u;
1913 const char *to;
1914 int r;
1915
1916 assert(dest);
1917 assert(console);
1918
1919 u = umask(0000);
1920
1921 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
1922 if (r < 0)
1923 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1924
1925 /* We need to bind mount the right tty to /dev/console since
1926 * ptys can only exist on pts file systems. To have something
1927 * to bind mount things on we create a empty regular file. */
1928
1929 to = prefix_roota(dest, "/dev/console");
1930 r = touch(to);
1931 if (r < 0)
1932 return log_error_errno(r, "touch() for /dev/console failed: %m");
1933
1934 return mount_verbose(LOG_ERR, console, to, NULL, MS_BIND, NULL);
1935 }
1936
1937 static int setup_keyring(void) {
1938 key_serial_t keyring;
1939
1940 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
1941 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
1942 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
1943 * these system calls let's make sure we don't leak anything into the container. */
1944
1945 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
1946 if (keyring == -1) {
1947 if (errno == ENOSYS)
1948 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
1949 else if (IN_SET(errno, EACCES, EPERM))
1950 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
1951 else
1952 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
1953 }
1954
1955 return 0;
1956 }
1957
1958 static int setup_kmsg(int kmsg_socket) {
1959 _cleanup_(unlink_and_freep) char *from = NULL;
1960 _cleanup_free_ char *fifo = NULL;
1961 _cleanup_close_ int fd = -1;
1962 _cleanup_umask_ mode_t u;
1963 const char *to;
1964 int r;
1965
1966 assert(kmsg_socket >= 0);
1967
1968 u = umask(0000);
1969
1970 /* We create the kmsg FIFO as as temporary file in /tmp, but immediately delete it after bind mounting it to
1971 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
1972 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
1973 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
1974
1975 r = tempfn_random_child(NULL, "proc-kmsg", &fifo);
1976 if (r < 0)
1977 return log_error_errno(r, "Failed to generate kmsg path: %m");
1978
1979 if (mkfifo(fifo, 0600) < 0)
1980 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
1981
1982 from = TAKE_PTR(fifo);
1983 to = "/proc/kmsg";
1984
1985 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1986 if (r < 0)
1987 return r;
1988
1989 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
1990 if (fd < 0)
1991 return log_error_errno(errno, "Failed to open fifo: %m");
1992
1993 /* Store away the fd in the socket, so that it stays open as long as we run the child */
1994 r = send_one_fd(kmsg_socket, fd, 0);
1995 if (r < 0)
1996 return log_error_errno(r, "Failed to send FIFO fd: %m");
1997
1998 return 0;
1999 }
2000
2001 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2002 union in_addr_union *exposed = userdata;
2003
2004 assert(rtnl);
2005 assert(m);
2006 assert(exposed);
2007
2008 expose_port_execute(rtnl, arg_expose_ports, exposed);
2009 return 0;
2010 }
2011
2012 static int setup_hostname(void) {
2013 int r;
2014
2015 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2016 return 0;
2017
2018 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2019 if (r < 0)
2020 return log_error_errno(r, "Failed to set hostname: %m");
2021
2022 return 0;
2023 }
2024
2025 static int setup_journal(const char *directory) {
2026 sd_id128_t this_id;
2027 _cleanup_free_ char *d = NULL;
2028 const char *p, *q;
2029 bool try;
2030 char id[33], *dirname;
2031 int r;
2032
2033 /* Don't link journals in ephemeral mode */
2034 if (arg_ephemeral)
2035 return 0;
2036
2037 if (arg_link_journal == LINK_NO)
2038 return 0;
2039
2040 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2041
2042 r = sd_id128_get_machine(&this_id);
2043 if (r < 0)
2044 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2045
2046 if (sd_id128_equal(arg_uuid, this_id)) {
2047 log_full(try ? LOG_WARNING : LOG_ERR,
2048 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
2049 if (try)
2050 return 0;
2051 return -EEXIST;
2052 }
2053
2054 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2055 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2056 if (r < 0) {
2057 bool ignore = r == -EROFS && try;
2058 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2059 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2060 return ignore ? 0 : r;
2061 }
2062 }
2063
2064 (void) sd_id128_to_string(arg_uuid, id);
2065
2066 p = strjoina("/var/log/journal/", id);
2067 q = prefix_roota(directory, p);
2068
2069 if (path_is_mount_point(p, NULL, 0) > 0) {
2070 if (try)
2071 return 0;
2072
2073 log_error("%s: already a mount point, refusing to use for journal", p);
2074 return -EEXIST;
2075 }
2076
2077 if (path_is_mount_point(q, NULL, 0) > 0) {
2078 if (try)
2079 return 0;
2080
2081 log_error("%s: already a mount point, refusing to use for journal", q);
2082 return -EEXIST;
2083 }
2084
2085 r = readlink_and_make_absolute(p, &d);
2086 if (r >= 0) {
2087 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2088 path_equal(d, q)) {
2089
2090 r = userns_mkdir(directory, p, 0755, 0, 0);
2091 if (r < 0)
2092 log_warning_errno(r, "Failed to create directory %s: %m", q);
2093 return 0;
2094 }
2095
2096 if (unlink(p) < 0)
2097 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2098 } else if (r == -EINVAL) {
2099
2100 if (arg_link_journal == LINK_GUEST &&
2101 rmdir(p) < 0) {
2102
2103 if (errno == ENOTDIR) {
2104 log_error("%s already exists and is neither a symlink nor a directory", p);
2105 return r;
2106 } else
2107 return log_error_errno(errno, "Failed to remove %s: %m", p);
2108 }
2109 } else if (r != -ENOENT)
2110 return log_error_errno(r, "readlink(%s) failed: %m", p);
2111
2112 if (arg_link_journal == LINK_GUEST) {
2113
2114 if (symlink(q, p) < 0) {
2115 if (try) {
2116 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2117 return 0;
2118 } else
2119 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2120 }
2121
2122 r = userns_mkdir(directory, p, 0755, 0, 0);
2123 if (r < 0)
2124 log_warning_errno(r, "Failed to create directory %s: %m", q);
2125 return 0;
2126 }
2127
2128 if (arg_link_journal == LINK_HOST) {
2129 /* don't create parents here — if the host doesn't have
2130 * permanent journal set up, don't force it here */
2131
2132 r = mkdir_errno_wrapper(p, 0755);
2133 if (r < 0 && r != -EEXIST) {
2134 if (try) {
2135 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2136 return 0;
2137 } else
2138 return log_error_errno(r, "Failed to create %s: %m", p);
2139 }
2140
2141 } else if (access(p, F_OK) < 0)
2142 return 0;
2143
2144 if (dir_is_empty(q) == 0)
2145 log_warning("%s is not empty, proceeding anyway.", q);
2146
2147 r = userns_mkdir(directory, p, 0755, 0, 0);
2148 if (r < 0)
2149 return log_error_errno(r, "Failed to create %s: %m", q);
2150
2151 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2152 if (r < 0)
2153 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2154
2155 return 0;
2156 }
2157
2158 static int drop_capabilities(void) {
2159 return capability_bounding_set_drop(arg_caps_retain, false);
2160 }
2161
2162 static int reset_audit_loginuid(void) {
2163 _cleanup_free_ char *p = NULL;
2164 int r;
2165
2166 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2167 return 0;
2168
2169 r = read_one_line_file("/proc/self/loginuid", &p);
2170 if (r == -ENOENT)
2171 return 0;
2172 if (r < 0)
2173 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2174
2175 /* Already reset? */
2176 if (streq(p, "4294967295"))
2177 return 0;
2178
2179 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2180 if (r < 0) {
2181 log_error_errno(r,
2182 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2183 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2184 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2185 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2186 "using systemd-nspawn. Sleeping for 5s... (%m)");
2187
2188 sleep(5);
2189 }
2190
2191 return 0;
2192 }
2193
2194 static int setup_propagate(const char *root) {
2195 const char *p, *q;
2196 int r;
2197
2198 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2199 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2200 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2201 (void) mkdir_p(p, 0600);
2202
2203 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
2204 if (r < 0)
2205 return log_error_errno(r, "Failed to create /run/systemd: %m");
2206
2207 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
2208 if (r < 0)
2209 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
2210
2211 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2212 if (r < 0)
2213 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
2214
2215 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
2216 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2217 if (r < 0)
2218 return r;
2219
2220 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2221 if (r < 0)
2222 return r;
2223
2224 /* machined will MS_MOVE into that directory, and that's only
2225 * supported for non-shared mounts. */
2226 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2227 }
2228
2229 static int setup_machine_id(const char *directory) {
2230 const char *etc_machine_id;
2231 sd_id128_t id;
2232 int r;
2233
2234 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2235 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2236 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2237 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2238 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2239 * container behaves nicely). */
2240
2241 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2242
2243 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
2244 if (r < 0) {
2245 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2246 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2247
2248 if (sd_id128_is_null(arg_uuid)) {
2249 r = sd_id128_randomize(&arg_uuid);
2250 if (r < 0)
2251 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2252 }
2253 } else {
2254 if (sd_id128_is_null(id)) {
2255 log_error("Machine ID in container image is zero, refusing.");
2256 return -EINVAL;
2257 }
2258
2259 arg_uuid = id;
2260 }
2261
2262 return 0;
2263 }
2264
2265 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2266 int r;
2267
2268 assert(directory);
2269
2270 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
2271 return 0;
2272
2273 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2274 if (r == -EOPNOTSUPP)
2275 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.");
2276 if (r == -EBADE)
2277 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2278 if (r < 0)
2279 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2280 if (r == 0)
2281 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2282 else
2283 log_debug("Patched directory tree to match UID/GID range.");
2284
2285 return r;
2286 }
2287
2288 /*
2289 * Return values:
2290 * < 0 : wait_for_terminate() failed to get the state of the
2291 * container, the container was terminated by a signal, or
2292 * failed for an unknown reason. No change is made to the
2293 * container argument.
2294 * > 0 : The program executed in the container terminated with an
2295 * error. The exit code of the program executed in the
2296 * container is returned. The container argument has been set
2297 * to CONTAINER_TERMINATED.
2298 * 0 : The container is being rebooted, has been shut down or exited
2299 * successfully. The container argument has been set to either
2300 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2301 *
2302 * That is, success is indicated by a return value of zero, and an
2303 * error is indicated by a non-zero value.
2304 */
2305 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2306 siginfo_t status;
2307 int r;
2308
2309 r = wait_for_terminate(pid, &status);
2310 if (r < 0)
2311 return log_warning_errno(r, "Failed to wait for container: %m");
2312
2313 switch (status.si_code) {
2314
2315 case CLD_EXITED:
2316 if (status.si_status == 0)
2317 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2318 else
2319 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2320
2321 *container = CONTAINER_TERMINATED;
2322 return status.si_status;
2323
2324 case CLD_KILLED:
2325 if (status.si_status == SIGINT) {
2326 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2327 *container = CONTAINER_TERMINATED;
2328 return 0;
2329
2330 } else if (status.si_status == SIGHUP) {
2331 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2332 *container = CONTAINER_REBOOTED;
2333 return 0;
2334 }
2335
2336 _fallthrough_;
2337 case CLD_DUMPED:
2338 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2339 return -EIO;
2340
2341 default:
2342 log_error("Container %s failed due to unknown reason.", arg_machine);
2343 return -EIO;
2344 }
2345 }
2346
2347 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2348 pid_t pid;
2349
2350 pid = PTR_TO_PID(userdata);
2351 if (pid > 0) {
2352 if (kill(pid, arg_kill_signal) >= 0) {
2353 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2354 sd_event_source_set_userdata(s, NULL);
2355 return 0;
2356 }
2357 }
2358
2359 sd_event_exit(sd_event_source_get_event(s), 0);
2360 return 0;
2361 }
2362
2363 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2364 pid_t pid;
2365
2366 assert(s);
2367 assert(ssi);
2368
2369 pid = PTR_TO_PID(userdata);
2370
2371 for (;;) {
2372 siginfo_t si = {};
2373
2374 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2375 return log_error_errno(errno, "Failed to waitid(): %m");
2376 if (si.si_pid == 0) /* No pending children. */
2377 break;
2378 if (si.si_pid == pid) {
2379 /* The main process we care for has exited. Return from
2380 * signal handler but leave the zombie. */
2381 sd_event_exit(sd_event_source_get_event(s), 0);
2382 break;
2383 }
2384
2385 /* Reap all other children. */
2386 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2387 }
2388
2389 return 0;
2390 }
2391
2392 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2393 pid_t pid;
2394
2395 assert(m);
2396
2397 pid = PTR_TO_PID(userdata);
2398
2399 if (arg_kill_signal > 0) {
2400 log_info("Container termination requested. Attempting to halt container.");
2401 (void) kill(pid, arg_kill_signal);
2402 } else {
2403 log_info("Container termination requested. Exiting.");
2404 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2405 }
2406
2407 return 0;
2408 }
2409
2410 static int determine_names(void) {
2411 int r;
2412
2413 if (arg_template && !arg_directory && arg_machine) {
2414
2415 /* If --template= was specified then we should not
2416 * search for a machine, but instead create a new one
2417 * in /var/lib/machine. */
2418
2419 arg_directory = strjoin("/var/lib/machines/", arg_machine);
2420 if (!arg_directory)
2421 return log_oom();
2422 }
2423
2424 if (!arg_image && !arg_directory) {
2425 if (arg_machine) {
2426 _cleanup_(image_unrefp) Image *i = NULL;
2427
2428 r = image_find(IMAGE_MACHINE, arg_machine, &i);
2429 if (r == -ENOENT)
2430 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
2431 if (r < 0)
2432 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2433
2434 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2435 r = free_and_strdup(&arg_image, i->path);
2436 else
2437 r = free_and_strdup(&arg_directory, i->path);
2438 if (r < 0)
2439 return log_oom();
2440
2441 if (!arg_ephemeral)
2442 arg_read_only = arg_read_only || i->read_only;
2443 } else {
2444 r = safe_getcwd(&arg_directory);
2445 if (r < 0)
2446 return log_error_errno(r, "Failed to determine current directory: %m");
2447 }
2448
2449 if (!arg_directory && !arg_image) {
2450 log_error("Failed to determine path, please use -D or -i.");
2451 return -EINVAL;
2452 }
2453 }
2454
2455 if (!arg_machine) {
2456 if (arg_directory && path_equal(arg_directory, "/"))
2457 arg_machine = gethostname_malloc();
2458 else {
2459 if (arg_image) {
2460 char *e;
2461
2462 arg_machine = strdup(basename(arg_image));
2463
2464 /* Truncate suffix if there is one */
2465 e = endswith(arg_machine, ".raw");
2466 if (e)
2467 *e = 0;
2468 } else
2469 arg_machine = strdup(basename(arg_directory));
2470 }
2471 if (!arg_machine)
2472 return log_oom();
2473
2474 hostname_cleanup(arg_machine);
2475 if (!machine_name_is_valid(arg_machine)) {
2476 log_error("Failed to determine machine name automatically, please use -M.");
2477 return -EINVAL;
2478 }
2479
2480 if (arg_ephemeral) {
2481 char *b;
2482
2483 /* Add a random suffix when this is an
2484 * ephemeral machine, so that we can run many
2485 * instances at once without manually having
2486 * to specify -M each time. */
2487
2488 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2489 return log_oom();
2490
2491 free(arg_machine);
2492 arg_machine = b;
2493 }
2494 }
2495
2496 return 0;
2497 }
2498
2499 static int chase_symlinks_and_update(char **p, unsigned flags) {
2500 char *chased;
2501 int r;
2502
2503 assert(p);
2504
2505 if (!*p)
2506 return 0;
2507
2508 r = chase_symlinks(*p, NULL, flags, &chased);
2509 if (r < 0)
2510 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2511
2512 free_and_replace(*p, chased);
2513 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2514 }
2515
2516 static int determine_uid_shift(const char *directory) {
2517 int r;
2518
2519 if (arg_userns_mode == USER_NAMESPACE_NO) {
2520 arg_uid_shift = 0;
2521 return 0;
2522 }
2523
2524 if (arg_uid_shift == UID_INVALID) {
2525 struct stat st;
2526
2527 r = stat(directory, &st);
2528 if (r < 0)
2529 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2530
2531 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2532
2533 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
2534 log_error("UID and GID base of %s don't match.", directory);
2535 return -EINVAL;
2536 }
2537
2538 arg_uid_range = UINT32_C(0x10000);
2539 }
2540
2541 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
2542 log_error("UID base too high for UID range.");
2543 return -EINVAL;
2544 }
2545
2546 return 0;
2547 }
2548
2549 static int inner_child(
2550 Barrier *barrier,
2551 const char *directory,
2552 bool secondary,
2553 int kmsg_socket,
2554 int rtnl_socket,
2555 FDSet *fds) {
2556
2557 _cleanup_free_ char *home = NULL;
2558 char as_uuid[37];
2559 size_t n_env = 1;
2560 const char *envp[] = {
2561 "PATH=" DEFAULT_PATH_COMPAT,
2562 NULL, /* container */
2563 NULL, /* TERM */
2564 NULL, /* HOME */
2565 NULL, /* USER */
2566 NULL, /* LOGNAME */
2567 NULL, /* container_uuid */
2568 NULL, /* LISTEN_FDS */
2569 NULL, /* LISTEN_PID */
2570 NULL, /* NOTIFY_SOCKET */
2571 NULL
2572 };
2573 const char *exec_target;
2574 _cleanup_strv_free_ char **env_use = NULL;
2575 int r;
2576
2577 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2578 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2579 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2580 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2581 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2582 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2583 * namespace.
2584 *
2585 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2586 * unshare(). See below. */
2587
2588 assert(barrier);
2589 assert(directory);
2590 assert(kmsg_socket >= 0);
2591
2592 if (arg_userns_mode != USER_NAMESPACE_NO) {
2593 /* Tell the parent, that it now can write the UID map. */
2594 (void) barrier_place(barrier); /* #1 */
2595
2596 /* Wait until the parent wrote the UID map */
2597 if (!barrier_place_and_sync(barrier)) { /* #2 */
2598 log_error("Parent died too early");
2599 return -ESRCH;
2600 }
2601 }
2602
2603 r = reset_uid_gid();
2604 if (r < 0)
2605 return log_error_errno(r, "Couldn't become new root: %m");
2606
2607 r = mount_all(NULL,
2608 arg_mount_settings | MOUNT_IN_USERNS,
2609 arg_uid_shift,
2610 arg_selinux_apifs_context);
2611 if (r < 0)
2612 return r;
2613
2614 if (!arg_network_namespace_path && arg_private_network) {
2615 r = unshare(CLONE_NEWNET);
2616 if (r < 0)
2617 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2618
2619 /* Tell the parent that it can setup network interfaces. */
2620 (void) barrier_place(barrier); /* #3 */
2621 }
2622
2623 r = mount_sysfs(NULL, arg_mount_settings);
2624 if (r < 0)
2625 return r;
2626
2627 /* Wait until we are cgroup-ified, so that we
2628 * can mount the right cgroup path writable */
2629 if (!barrier_place_and_sync(barrier)) { /* #4 */
2630 log_error("Parent died too early");
2631 return -ESRCH;
2632 }
2633
2634 if (arg_use_cgns && cg_ns_supported()) {
2635 r = unshare(CLONE_NEWCGROUP);
2636 if (r < 0)
2637 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2638 r = mount_cgroups(
2639 "",
2640 arg_unified_cgroup_hierarchy,
2641 arg_userns_mode != USER_NAMESPACE_NO,
2642 arg_uid_shift,
2643 arg_uid_range,
2644 arg_selinux_apifs_context,
2645 true);
2646 if (r < 0)
2647 return r;
2648 } else {
2649 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2650 if (r < 0)
2651 return r;
2652 }
2653
2654 r = setup_boot_id();
2655 if (r < 0)
2656 return r;
2657
2658 r = setup_kmsg(kmsg_socket);
2659 if (r < 0)
2660 return r;
2661 kmsg_socket = safe_close(kmsg_socket);
2662
2663 if (setsid() < 0)
2664 return log_error_errno(errno, "setsid() failed: %m");
2665
2666 if (arg_private_network)
2667 loopback_setup();
2668
2669 if (arg_expose_ports) {
2670 r = expose_port_send_rtnl(rtnl_socket);
2671 if (r < 0)
2672 return r;
2673 rtnl_socket = safe_close(rtnl_socket);
2674 }
2675
2676 if (arg_oom_score_adjust_set) {
2677 r = set_oom_score_adjust(arg_oom_score_adjust);
2678 if (r < 0)
2679 return log_error_errno(r, "Failed to adjust OOM score: %m");
2680 }
2681
2682 if (arg_cpuset)
2683 if (sched_setaffinity(0, CPU_ALLOC_SIZE(arg_cpuset_ncpus), arg_cpuset) < 0)
2684 return log_error_errno(errno, "Failed to set CPU affinity: %m");
2685
2686 r = drop_capabilities();
2687 if (r < 0)
2688 return log_error_errno(r, "drop_capabilities() failed: %m");
2689
2690 (void) setup_hostname();
2691
2692 if (arg_personality != PERSONALITY_INVALID) {
2693 r = safe_personality(arg_personality);
2694 if (r < 0)
2695 return log_error_errno(r, "personality() failed: %m");
2696 } else if (secondary) {
2697 r = safe_personality(PER_LINUX32);
2698 if (r < 0)
2699 return log_error_errno(r, "personality() failed: %m");
2700 }
2701
2702 #if HAVE_SELINUX
2703 if (arg_selinux_context)
2704 if (setexeccon(arg_selinux_context) < 0)
2705 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
2706 #endif
2707
2708 r = change_uid_gid(arg_user, &home);
2709 if (r < 0)
2710 return r;
2711
2712 if (arg_no_new_privileges)
2713 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
2714 return log_error_errno(errno, "Failed to disable new privileges: %m");
2715
2716 /* LXC sets container=lxc, so follow the scheme here */
2717 envp[n_env++] = strjoina("container=", arg_container_service_name);
2718
2719 envp[n_env] = strv_find_prefix(environ, "TERM=");
2720 if (envp[n_env])
2721 n_env++;
2722
2723 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
2724 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
2725 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
2726 return log_oom();
2727
2728 assert(!sd_id128_is_null(arg_uuid));
2729
2730 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
2731 return log_oom();
2732
2733 if (fdset_size(fds) > 0) {
2734 r = fdset_cloexec(fds, false);
2735 if (r < 0)
2736 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
2737
2738 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
2739 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
2740 return log_oom();
2741 }
2742 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
2743 return log_oom();
2744
2745 env_use = strv_env_merge(2, envp, arg_setenv);
2746 if (!env_use)
2747 return log_oom();
2748
2749 /* Let the parent know that we are ready and
2750 * wait until the parent is ready with the
2751 * setup, too... */
2752 if (!barrier_place_and_sync(barrier)) { /* #5 */
2753 log_error("Parent died too early");
2754 return -ESRCH;
2755 }
2756
2757 if (arg_chdir)
2758 if (chdir(arg_chdir) < 0)
2759 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
2760
2761 if (arg_start_mode == START_PID2) {
2762 r = stub_pid1(arg_uuid);
2763 if (r < 0)
2764 return r;
2765 }
2766
2767 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
2768 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
2769 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
2770 log_close();
2771 log_set_open_when_needed(true);
2772
2773 (void) fdset_close_others(fds);
2774
2775 if (arg_start_mode == START_BOOT) {
2776 char **a;
2777 size_t m;
2778
2779 /* Automatically search for the init system */
2780
2781 m = strv_length(arg_parameters);
2782 a = newa(char*, m + 2);
2783 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
2784 a[1 + m] = NULL;
2785
2786 a[0] = (char*) "/usr/lib/systemd/systemd";
2787 execve(a[0], a, env_use);
2788
2789 a[0] = (char*) "/lib/systemd/systemd";
2790 execve(a[0], a, env_use);
2791
2792 a[0] = (char*) "/sbin/init";
2793 execve(a[0], a, env_use);
2794
2795 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2796 } else if (!strv_isempty(arg_parameters)) {
2797 const char *dollar_path;
2798
2799 exec_target = arg_parameters[0];
2800
2801 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
2802 * binary. */
2803 dollar_path = strv_env_get(env_use, "PATH");
2804 if (dollar_path) {
2805 if (putenv((char*) dollar_path) != 0)
2806 return log_error_errno(errno, "Failed to update $PATH: %m");
2807 }
2808
2809 execvpe(arg_parameters[0], arg_parameters, env_use);
2810 } else {
2811 if (!arg_chdir)
2812 /* If we cannot change the directory, we'll end up in /, that is expected. */
2813 (void) chdir(home ?: "/root");
2814
2815 execle("/bin/bash", "-bash", NULL, env_use);
2816 execle("/bin/sh", "-sh", NULL, env_use);
2817
2818 exec_target = "/bin/bash, /bin/sh";
2819 }
2820
2821 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
2822 }
2823
2824 static int setup_sd_notify_child(void) {
2825 _cleanup_close_ int fd = -1;
2826 union sockaddr_union sa = {
2827 .un.sun_family = AF_UNIX,
2828 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
2829 };
2830 int r;
2831
2832 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2833 if (fd < 0)
2834 return log_error_errno(errno, "Failed to allocate notification socket: %m");
2835
2836 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
2837 (void) sockaddr_un_unlink(&sa.un);
2838
2839 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
2840 if (r < 0)
2841 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
2842
2843 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
2844 if (r < 0)
2845 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
2846
2847 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
2848 if (r < 0)
2849 return log_error_errno(r, "SO_PASSCRED failed: %m");
2850
2851 return TAKE_FD(fd);
2852 }
2853
2854 static int outer_child(
2855 Barrier *barrier,
2856 const char *directory,
2857 const char *console,
2858 DissectedImage *dissected_image,
2859 bool interactive,
2860 bool secondary,
2861 int pid_socket,
2862 int uuid_socket,
2863 int notify_socket,
2864 int kmsg_socket,
2865 int rtnl_socket,
2866 int uid_shift_socket,
2867 int unified_cgroup_hierarchy_socket,
2868 FDSet *fds,
2869 int netns_fd) {
2870
2871 _cleanup_close_ int fd = -1;
2872 int r, which_failed;
2873 pid_t pid;
2874 ssize_t l;
2875
2876 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
2877 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
2878 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
2879 * initializations a second child (the "inner" one) is forked off it, and it exits. */
2880
2881 assert(barrier);
2882 assert(directory);
2883 assert(console);
2884 assert(pid_socket >= 0);
2885 assert(uuid_socket >= 0);
2886 assert(notify_socket >= 0);
2887 assert(kmsg_socket >= 0);
2888
2889 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
2890 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
2891
2892 if (interactive) {
2893 int terminal;
2894
2895 terminal = open_terminal(console, O_RDWR);
2896 if (terminal < 0)
2897 return log_error_errno(terminal, "Failed to open console: %m");
2898
2899 /* Make sure we can continue logging to the original stderr, even if stderr points elsewhere now */
2900 r = log_dup_console();
2901 if (r < 0)
2902 return log_error_errno(r, "Failed to duplicate stderr: %m");
2903
2904 r = rearrange_stdio(terminal, terminal, terminal); /* invalidates 'terminal' on success and failure */
2905 if (r < 0)
2906 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2907 }
2908
2909 r = reset_audit_loginuid();
2910 if (r < 0)
2911 return r;
2912
2913 /* Mark everything as slave, so that we still
2914 * receive mounts from the real root, but don't
2915 * propagate mounts to the real root. */
2916 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
2917 if (r < 0)
2918 return r;
2919
2920 if (dissected_image) {
2921 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
2922 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
2923 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
2924 * makes sure ESP partitions and userns are compatible. */
2925
2926 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2927 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
2928 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0)|
2929 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
2930 if (r < 0)
2931 return r;
2932 }
2933
2934 r = determine_uid_shift(directory);
2935 if (r < 0)
2936 return r;
2937
2938 if (arg_userns_mode != USER_NAMESPACE_NO) {
2939 /* Let the parent know which UID shift we read from the image */
2940 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
2941 if (l < 0)
2942 return log_error_errno(errno, "Failed to send UID shift: %m");
2943 if (l != sizeof(arg_uid_shift)) {
2944 log_error("Short write while sending UID shift.");
2945 return -EIO;
2946 }
2947
2948 if (arg_userns_mode == USER_NAMESPACE_PICK) {
2949 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2950 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2951 * not it will pick a different one, and send it back to us. */
2952
2953 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
2954 if (l < 0)
2955 return log_error_errno(errno, "Failed to recv UID shift: %m");
2956 if (l != sizeof(arg_uid_shift)) {
2957 log_error("Short read while receiving UID shift.");
2958 return -EIO;
2959 }
2960 }
2961
2962 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
2963 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
2964 }
2965
2966 if (dissected_image) {
2967 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
2968 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2969 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2970 if (r < 0)
2971 return r;
2972 }
2973
2974 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
2975 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
2976
2977 r = detect_unified_cgroup_hierarchy_from_image(directory);
2978 if (r < 0)
2979 return r;
2980
2981 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
2982 if (l < 0)
2983 return log_error_errno(errno, "Failed to send cgroup mode: %m");
2984 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
2985 log_error("Short write while sending cgroup mode.");
2986 return -EIO;
2987 }
2988
2989 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
2990 }
2991
2992 /* Turn directory into bind mount */
2993 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
2994 if (r < 0)
2995 return r;
2996
2997 r = setup_pivot_root(
2998 directory,
2999 arg_pivot_root_new,
3000 arg_pivot_root_old);
3001 if (r < 0)
3002 return r;
3003
3004 r = setup_volatile(
3005 directory,
3006 arg_volatile_mode,
3007 arg_userns_mode != USER_NAMESPACE_NO,
3008 arg_uid_shift,
3009 arg_uid_range,
3010 arg_selinux_context);
3011 if (r < 0)
3012 return r;
3013
3014 r = setup_volatile_state(
3015 directory,
3016 arg_volatile_mode,
3017 arg_userns_mode != USER_NAMESPACE_NO,
3018 arg_uid_shift,
3019 arg_uid_range,
3020 arg_selinux_context);
3021 if (r < 0)
3022 return r;
3023
3024 /* Mark everything as shared so our mounts get propagated down. This is
3025 * required to make new bind mounts available in systemd services
3026 * inside the containter that create a new mount namespace.
3027 * See https://github.com/systemd/systemd/issues/3860
3028 * Further submounts (such as /dev) done after this will inherit the
3029 * shared propagation mode. */
3030 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3031 if (r < 0)
3032 return r;
3033
3034 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3035 if (r < 0)
3036 return r;
3037
3038 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3039 if (r < 0)
3040 return r;
3041
3042 if (arg_read_only) {
3043 r = bind_remount_recursive(directory, true, NULL);
3044 if (r < 0)
3045 return log_error_errno(r, "Failed to make tree read-only: %m");
3046 }
3047
3048 r = mount_all(directory,
3049 arg_mount_settings,
3050 arg_uid_shift,
3051 arg_selinux_apifs_context);
3052 if (r < 0)
3053 return r;
3054
3055 r = copy_devnodes(directory);
3056 if (r < 0)
3057 return r;
3058
3059 dev_setup(directory, arg_uid_shift, arg_uid_shift);
3060
3061 r = setup_pts(directory);
3062 if (r < 0)
3063 return r;
3064
3065 r = setup_propagate(directory);
3066 if (r < 0)
3067 return r;
3068
3069 r = setup_dev_console(directory, console);
3070 if (r < 0)
3071 return r;
3072
3073 r = setup_keyring();
3074 if (r < 0)
3075 return r;
3076
3077 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
3078 if (r < 0)
3079 return r;
3080
3081 r = setup_timezone(directory);
3082 if (r < 0)
3083 return r;
3084
3085 r = setup_resolv_conf(directory);
3086 if (r < 0)
3087 return r;
3088
3089 r = setup_machine_id(directory);
3090 if (r < 0)
3091 return r;
3092
3093 r = setup_journal(directory);
3094 if (r < 0)
3095 return r;
3096
3097 r = mount_custom(
3098 directory,
3099 arg_custom_mounts,
3100 arg_n_custom_mounts,
3101 arg_userns_mode != USER_NAMESPACE_NO,
3102 arg_uid_shift,
3103 arg_uid_range,
3104 arg_selinux_apifs_context);
3105 if (r < 0)
3106 return r;
3107
3108 if (!arg_use_cgns || !cg_ns_supported()) {
3109 r = mount_cgroups(
3110 directory,
3111 arg_unified_cgroup_hierarchy,
3112 arg_userns_mode != USER_NAMESPACE_NO,
3113 arg_uid_shift,
3114 arg_uid_range,
3115 arg_selinux_apifs_context,
3116 false);
3117 if (r < 0)
3118 return r;
3119 }
3120
3121 r = mount_move_root(directory);
3122 if (r < 0)
3123 return log_error_errno(r, "Failed to move root directory: %m");
3124
3125 fd = setup_sd_notify_child();
3126 if (fd < 0)
3127 return fd;
3128
3129 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3130 if (r < 0)
3131 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3132
3133 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3134 arg_clone_ns_flags |
3135 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3136 if (pid < 0)
3137 return log_error_errno(errno, "Failed to fork inner child: %m");
3138 if (pid == 0) {
3139 pid_socket = safe_close(pid_socket);
3140 uuid_socket = safe_close(uuid_socket);
3141 notify_socket = safe_close(notify_socket);
3142 uid_shift_socket = safe_close(uid_shift_socket);
3143
3144 /* The inner child has all namespaces that are
3145 * requested, so that we all are owned by the user if
3146 * user namespaces are turned on. */
3147
3148 if (arg_network_namespace_path) {
3149 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3150 if (r < 0)
3151 return log_error_errno(r, "Failed to join network namespace: %m");
3152 }
3153
3154 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
3155 if (r < 0)
3156 _exit(EXIT_FAILURE);
3157
3158 _exit(EXIT_SUCCESS);
3159 }
3160
3161 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3162 if (l < 0)
3163 return log_error_errno(errno, "Failed to send PID: %m");
3164 if (l != sizeof(pid)) {
3165 log_error("Short write while sending PID.");
3166 return -EIO;
3167 }
3168
3169 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3170 if (l < 0)
3171 return log_error_errno(errno, "Failed to send machine ID: %m");
3172 if (l != sizeof(arg_uuid)) {
3173 log_error("Short write while sending machine ID.");
3174 return -EIO;
3175 }
3176
3177 l = send_one_fd(notify_socket, fd, 0);
3178 if (l < 0)
3179 return log_error_errno(errno, "Failed to send notify fd: %m");
3180
3181 pid_socket = safe_close(pid_socket);
3182 uuid_socket = safe_close(uuid_socket);
3183 notify_socket = safe_close(notify_socket);
3184 kmsg_socket = safe_close(kmsg_socket);
3185 rtnl_socket = safe_close(rtnl_socket);
3186 netns_fd = safe_close(netns_fd);
3187
3188 return 0;
3189 }
3190
3191 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
3192 bool tried_hashed = false;
3193 unsigned n_tries = 100;
3194 uid_t candidate;
3195 int r;
3196
3197 assert(shift);
3198 assert(ret_lock_file);
3199 assert(arg_userns_mode == USER_NAMESPACE_PICK);
3200 assert(arg_uid_range == 0x10000U);
3201
3202 candidate = *shift;
3203
3204 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3205
3206 for (;;) {
3207 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
3208 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
3209
3210 if (--n_tries <= 0)
3211 return -EBUSY;
3212
3213 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
3214 goto next;
3215 if ((candidate & UINT32_C(0xFFFF)) != 0)
3216 goto next;
3217
3218 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3219 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3220 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3221 goto next;
3222 if (r < 0)
3223 return r;
3224
3225 /* Make some superficial checks whether the range is currently known in the user database */
3226 if (getpwuid(candidate))
3227 goto next;
3228 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3229 goto next;
3230 if (getgrgid(candidate))
3231 goto next;
3232 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3233 goto next;
3234
3235 *ret_lock_file = lf;
3236 lf = (struct LockFile) LOCK_FILE_INIT;
3237 *shift = candidate;
3238 return 0;
3239
3240 next:
3241 if (arg_machine && !tried_hashed) {
3242 /* Try to hash the base from the container name */
3243
3244 static const uint8_t hash_key[] = {
3245 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3246 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3247 };
3248
3249 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3250
3251 tried_hashed = true;
3252 } else
3253 random_bytes(&candidate, sizeof(candidate));
3254
3255 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
3256 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3257 }
3258 }
3259
3260 static int setup_uid_map(pid_t pid) {
3261 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3262 int r;
3263
3264 assert(pid > 1);
3265
3266 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3267 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3268 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3269 if (r < 0)
3270 return log_error_errno(r, "Failed to write UID map: %m");
3271
3272 /* We always assign the same UID and GID ranges */
3273 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3274 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3275 if (r < 0)
3276 return log_error_errno(r, "Failed to write GID map: %m");
3277
3278 return 0;
3279 }
3280
3281 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3282 char buf[NOTIFY_BUFFER_MAX+1];
3283 char *p = NULL;
3284 struct iovec iovec = {
3285 .iov_base = buf,
3286 .iov_len = sizeof(buf)-1,
3287 };
3288 union {
3289 struct cmsghdr cmsghdr;
3290 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
3291 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
3292 } control = {};
3293 struct msghdr msghdr = {
3294 .msg_iov = &iovec,
3295 .msg_iovlen = 1,
3296 .msg_control = &control,
3297 .msg_controllen = sizeof(control),
3298 };
3299 struct cmsghdr *cmsg;
3300 struct ucred *ucred = NULL;
3301 ssize_t n;
3302 pid_t inner_child_pid;
3303 _cleanup_strv_free_ char **tags = NULL;
3304
3305 assert(userdata);
3306
3307 inner_child_pid = PTR_TO_PID(userdata);
3308
3309 if (revents != EPOLLIN) {
3310 log_warning("Got unexpected poll event for notify fd.");
3311 return 0;
3312 }
3313
3314 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3315 if (n < 0) {
3316 if (IN_SET(errno, EAGAIN, EINTR))
3317 return 0;
3318
3319 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3320 }
3321 cmsg_close_all(&msghdr);
3322
3323 CMSG_FOREACH(cmsg, &msghdr) {
3324 if (cmsg->cmsg_level == SOL_SOCKET &&
3325 cmsg->cmsg_type == SCM_CREDENTIALS &&
3326 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3327
3328 ucred = (struct ucred*) CMSG_DATA(cmsg);
3329 }
3330 }
3331
3332 if (!ucred || ucred->pid != inner_child_pid) {
3333 log_debug("Received notify message without valid credentials. Ignoring.");
3334 return 0;
3335 }
3336
3337 if ((size_t) n >= sizeof(buf)) {
3338 log_warning("Received notify message exceeded maximum size. Ignoring.");
3339 return 0;
3340 }
3341
3342 buf[n] = 0;
3343 tags = strv_split(buf, "\n\r");
3344 if (!tags)
3345 return log_oom();
3346
3347 if (strv_find(tags, "READY=1"))
3348 sd_notifyf(false, "READY=1\n");
3349
3350 p = strv_find_startswith(tags, "STATUS=");
3351 if (p)
3352 sd_notifyf(false, "STATUS=Container running: %s", p);
3353
3354 return 0;
3355 }
3356
3357 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3358 int r;
3359
3360 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3361 if (r < 0)
3362 return log_error_errno(r, "Failed to allocate notify event source: %m");
3363
3364 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3365
3366 return 0;
3367 }
3368
3369 static int merge_settings(Settings *settings, const char *path) {
3370 int rl;
3371
3372 assert(settings);
3373 assert(path);
3374
3375 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3376 * that this steals the fields of the Settings* structure, and hence modifies it. */
3377
3378 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3379 settings->start_mode >= 0) {
3380 arg_start_mode = settings->start_mode;
3381 strv_free_and_replace(arg_parameters, settings->parameters);
3382 }
3383
3384 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
3385 arg_ephemeral = settings->ephemeral;
3386
3387 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3388 settings->pivot_root_new) {
3389 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3390 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3391 }
3392
3393 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3394 settings->working_directory)
3395 free_and_replace(arg_chdir, settings->working_directory);
3396
3397 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3398 settings->environment)
3399 strv_free_and_replace(arg_setenv, settings->environment);
3400
3401 if ((arg_settings_mask & SETTING_USER) == 0 &&
3402 settings->user)
3403 free_and_replace(arg_user, settings->user);
3404
3405 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3406 uint64_t plus;
3407
3408 plus = settings->capability;
3409 if (settings_private_network(settings))
3410 plus |= (1ULL << CAP_NET_ADMIN);
3411
3412 if (!arg_settings_trusted && plus != 0) {
3413 if (settings->capability != 0)
3414 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
3415 } else
3416 arg_caps_retain |= plus;
3417
3418 arg_caps_retain &= ~settings->drop_capability;
3419 }
3420
3421 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3422 settings->kill_signal > 0)
3423 arg_kill_signal = settings->kill_signal;
3424
3425 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3426 settings->personality != PERSONALITY_INVALID)
3427 arg_personality = settings->personality;
3428
3429 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3430 !sd_id128_is_null(settings->machine_id)) {
3431
3432 if (!arg_settings_trusted)
3433 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
3434 else
3435 arg_uuid = settings->machine_id;
3436 }
3437
3438 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3439 settings->read_only >= 0)
3440 arg_read_only = settings->read_only;
3441
3442 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3443 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3444 arg_volatile_mode = settings->volatile_mode;
3445
3446 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3447 settings->n_custom_mounts > 0) {
3448
3449 if (!arg_settings_trusted)
3450 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
3451 else {
3452 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3453 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3454 arg_n_custom_mounts = settings->n_custom_mounts;
3455 settings->n_custom_mounts = 0;
3456 }
3457 }
3458
3459 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3460 (settings->private_network >= 0 ||
3461 settings->network_veth >= 0 ||
3462 settings->network_bridge ||
3463 settings->network_zone ||
3464 settings->network_interfaces ||
3465 settings->network_macvlan ||
3466 settings->network_ipvlan ||
3467 settings->network_veth_extra)) {
3468
3469 if (!arg_settings_trusted)
3470 log_warning("Ignoring network settings, file %s is not trusted.", path);
3471 else {
3472 arg_network_veth = settings_network_veth(settings);
3473 arg_private_network = settings_private_network(settings);
3474
3475 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
3476 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
3477 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
3478 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
3479
3480 free_and_replace(arg_network_bridge, settings->network_bridge);
3481 free_and_replace(arg_network_zone, settings->network_zone);
3482 }
3483 }
3484
3485 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3486 settings->expose_ports) {
3487
3488 if (!arg_settings_trusted)
3489 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
3490 else {
3491 expose_port_free_all(arg_expose_ports);
3492 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3493 }
3494 }
3495
3496 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3497 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3498
3499 if (!arg_settings_trusted)
3500 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
3501 else {
3502 arg_userns_mode = settings->userns_mode;
3503 arg_uid_shift = settings->uid_shift;
3504 arg_uid_range = settings->uid_range;
3505 arg_userns_chown = settings->userns_chown;
3506 }
3507 }
3508
3509 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3510 arg_notify_ready = settings->notify_ready;
3511
3512 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3513
3514 if (!arg_settings_trusted && !strv_isempty(arg_syscall_whitelist))
3515 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
3516 else {
3517 strv_free_and_replace(arg_syscall_whitelist, settings->syscall_whitelist);
3518 strv_free_and_replace(arg_syscall_blacklist, settings->syscall_blacklist);
3519 }
3520 }
3521
3522 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
3523 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
3524 continue;
3525
3526 if (!settings->rlimit[rl])
3527 continue;
3528
3529 if (!arg_settings_trusted) {
3530 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
3531 continue;
3532 }
3533
3534 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
3535 }
3536
3537 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
3538 settings->hostname)
3539 free_and_replace(arg_hostname, settings->hostname);
3540
3541 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
3542 settings->no_new_privileges >= 0)
3543 arg_no_new_privileges = settings->no_new_privileges;
3544
3545 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
3546 settings->oom_score_adjust_set) {
3547
3548 if (!arg_settings_trusted)
3549 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
3550 else {
3551 arg_oom_score_adjust = settings->oom_score_adjust;
3552 arg_oom_score_adjust_set = true;
3553 }
3554 }
3555
3556 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
3557 settings->cpuset) {
3558
3559 if (!arg_settings_trusted)
3560 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
3561 else {
3562 if (arg_cpuset)
3563 CPU_FREE(arg_cpuset);
3564 arg_cpuset = TAKE_PTR(settings->cpuset);
3565 arg_cpuset_ncpus = settings->cpuset_ncpus;
3566 }
3567 }
3568
3569 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
3570 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
3571 arg_resolv_conf = settings->resolv_conf;
3572
3573 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
3574 settings->link_journal != _LINK_JOURNAL_INVALID) {
3575
3576 if (!arg_settings_trusted)
3577 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
3578 else {
3579 arg_link_journal = settings->link_journal;
3580 arg_link_journal_try = settings->link_journal_try;
3581 }
3582 }
3583
3584 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
3585 settings->timezone != _TIMEZONE_MODE_INVALID)
3586 arg_timezone = settings->timezone;
3587
3588 return 0;
3589 }
3590
3591 static int load_settings(void) {
3592 _cleanup_(settings_freep) Settings *settings = NULL;
3593 _cleanup_fclose_ FILE *f = NULL;
3594 _cleanup_free_ char *p = NULL;
3595 const char *fn, *i;
3596 int r;
3597
3598 /* If all settings are masked, there's no point in looking for
3599 * the settings file */
3600 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3601 return 0;
3602
3603 fn = strjoina(arg_machine, ".nspawn");
3604
3605 /* We first look in the admin's directories in /etc and /run */
3606 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3607 _cleanup_free_ char *j = NULL;
3608
3609 j = strjoin(i, "/", fn);
3610 if (!j)
3611 return log_oom();
3612
3613 f = fopen(j, "re");
3614 if (f) {
3615 p = TAKE_PTR(j);
3616
3617 /* By default, we trust configuration from /etc and /run */
3618 if (arg_settings_trusted < 0)
3619 arg_settings_trusted = true;
3620
3621 break;
3622 }
3623
3624 if (errno != ENOENT)
3625 return log_error_errno(errno, "Failed to open %s: %m", j);
3626 }
3627
3628 if (!f) {
3629 /* After that, let's look for a file next to the
3630 * actual image we shall boot. */
3631
3632 if (arg_image) {
3633 p = file_in_same_dir(arg_image, fn);
3634 if (!p)
3635 return log_oom();
3636 } else if (arg_directory) {
3637 p = file_in_same_dir(arg_directory, fn);
3638 if (!p)
3639 return log_oom();
3640 }
3641
3642 if (p) {
3643 f = fopen(p, "re");
3644 if (!f && errno != ENOENT)
3645 return log_error_errno(errno, "Failed to open %s: %m", p);
3646
3647 /* By default, we do not trust configuration from /var/lib/machines */
3648 if (arg_settings_trusted < 0)
3649 arg_settings_trusted = false;
3650 }
3651 }
3652
3653 if (!f)
3654 return 0;
3655
3656 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
3657
3658 r = settings_load(f, p, &settings);
3659 if (r < 0)
3660 return r;
3661
3662 return merge_settings(settings, p);
3663 }
3664
3665 static int run(int master,
3666 const char* console,
3667 DissectedImage *dissected_image,
3668 bool interactive,
3669 bool secondary,
3670 FDSet *fds,
3671 char veth_name[IFNAMSIZ], bool *veth_created,
3672 union in_addr_union *exposed,
3673 pid_t *pid, int *ret) {
3674
3675 static const struct sigaction sa = {
3676 .sa_handler = nop_signal_handler,
3677 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
3678 };
3679
3680 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
3681 _cleanup_close_ int etc_passwd_lock = -1;
3682 _cleanup_close_pair_ int
3683 kmsg_socket_pair[2] = { -1, -1 },
3684 rtnl_socket_pair[2] = { -1, -1 },
3685 pid_socket_pair[2] = { -1, -1 },
3686 uuid_socket_pair[2] = { -1, -1 },
3687 notify_socket_pair[2] = { -1, -1 },
3688 uid_shift_socket_pair[2] = { -1, -1 },
3689 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
3690
3691 _cleanup_close_ int notify_socket= -1;
3692 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
3693 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
3694 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
3695 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
3696 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
3697 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
3698 ContainerStatus container_status = 0;
3699 char last_char = 0;
3700 int ifi = 0, r;
3701 ssize_t l;
3702 sigset_t mask_chld;
3703 _cleanup_close_ int netns_fd = -1;
3704
3705 assert_se(sigemptyset(&mask_chld) == 0);
3706 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
3707
3708 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3709 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3710 * check with getpwuid() if the specific user already exists. Note that /etc might be
3711 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3712 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3713 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3714 * really ours. */
3715
3716 etc_passwd_lock = take_etc_passwd_lock(NULL);
3717 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
3718 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
3719 }
3720
3721 r = barrier_create(&barrier);
3722 if (r < 0)
3723 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
3724
3725 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
3726 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
3727
3728 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
3729 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
3730
3731 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
3732 return log_error_errno(errno, "Failed to create pid socket pair: %m");
3733
3734 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
3735 return log_error_errno(errno, "Failed to create id socket pair: %m");
3736
3737 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
3738 return log_error_errno(errno, "Failed to create notify socket pair: %m");
3739
3740 if (arg_userns_mode != USER_NAMESPACE_NO)
3741 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
3742 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
3743
3744 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
3745 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
3746 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
3747
3748 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3749 * parent's blocking calls and give it a chance to call wait() and terminate. */
3750 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
3751 if (r < 0)
3752 return log_error_errno(errno, "Failed to change the signal mask: %m");
3753
3754 r = sigaction(SIGCHLD, &sa, NULL);
3755 if (r < 0)
3756 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
3757
3758 if (arg_network_namespace_path) {
3759 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
3760 if (netns_fd < 0)
3761 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
3762
3763 r = fd_is_network_ns(netns_fd);
3764 if (r == -EUCLEAN)
3765 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);
3766 else if (r < 0)
3767 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
3768 else if (r == 0) {
3769 log_error("Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
3770 return -EINVAL;
3771 }
3772 }
3773
3774 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
3775 if (*pid < 0)
3776 return log_error_errno(errno, "clone() failed%s: %m",
3777 errno == EINVAL ?
3778 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3779
3780 if (*pid == 0) {
3781 /* The outer child only has a file system namespace. */
3782 barrier_set_role(&barrier, BARRIER_CHILD);
3783
3784 master = safe_close(master);
3785
3786 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
3787 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3788 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
3789 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
3790 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
3791 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
3792 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
3793
3794 (void) reset_all_signal_handlers();
3795 (void) reset_signal_mask();
3796
3797 r = outer_child(&barrier,
3798 arg_directory,
3799 console,
3800 dissected_image,
3801 interactive,
3802 secondary,
3803 pid_socket_pair[1],
3804 uuid_socket_pair[1],
3805 notify_socket_pair[1],
3806 kmsg_socket_pair[1],
3807 rtnl_socket_pair[1],
3808 uid_shift_socket_pair[1],
3809 unified_cgroup_hierarchy_socket_pair[1],
3810 fds,
3811 netns_fd);
3812 if (r < 0)
3813 _exit(EXIT_FAILURE);
3814
3815 _exit(EXIT_SUCCESS);
3816 }
3817
3818 barrier_set_role(&barrier, BARRIER_PARENT);
3819
3820 fds = fdset_free(fds);
3821
3822 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
3823 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
3824 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
3825 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
3826 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
3827 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
3828 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
3829
3830 if (arg_userns_mode != USER_NAMESPACE_NO) {
3831 /* The child just let us know the UID shift it might have read from the image. */
3832 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
3833 if (l < 0)
3834 return log_error_errno(errno, "Failed to read UID shift: %m");
3835 if (l != sizeof arg_uid_shift) {
3836 log_error("Short read while reading UID shift.");
3837 return -EIO;
3838 }
3839
3840 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3841 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3842 * image, but if that's already in use, pick a new one, and report back to the child,
3843 * which one we now picked. */
3844
3845 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
3846 if (r < 0)
3847 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
3848
3849 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
3850 if (l < 0)
3851 return log_error_errno(errno, "Failed to send UID shift: %m");
3852 if (l != sizeof arg_uid_shift) {
3853 log_error("Short write while writing UID shift.");
3854 return -EIO;
3855 }
3856 }
3857 }
3858
3859 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3860 /* The child let us know the support cgroup mode it might have read from the image. */
3861 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
3862 if (l < 0)
3863 return log_error_errno(errno, "Failed to read cgroup mode: %m");
3864 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
3865 log_error("Short read while reading cgroup mode (%zu bytes).%s",
3866 l, l == 0 ? " The child is most likely dead." : "");
3867 return -EIO;
3868 }
3869 }
3870
3871 /* Wait for the outer child. */
3872 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
3873 if (r < 0)
3874 return r;
3875 if (r != EXIT_SUCCESS)
3876 return -EIO;
3877
3878 /* And now retrieve the PID of the inner child. */
3879 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
3880 if (l < 0)
3881 return log_error_errno(errno, "Failed to read inner child PID: %m");
3882 if (l != sizeof *pid) {
3883 log_error("Short read while reading inner child PID.");
3884 return -EIO;
3885 }
3886
3887 /* We also retrieve container UUID in case it was generated by outer child */
3888 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
3889 if (l < 0)
3890 return log_error_errno(errno, "Failed to read container machine ID: %m");
3891 if (l != sizeof(arg_uuid)) {
3892 log_error("Short read while reading container machined ID.");
3893 return -EIO;
3894 }
3895
3896 /* We also retrieve the socket used for notifications generated by outer child */
3897 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
3898 if (notify_socket < 0)
3899 return log_error_errno(notify_socket,
3900 "Failed to receive notification socket from the outer child: %m");
3901
3902 log_debug("Init process invoked as PID "PID_FMT, *pid);
3903
3904 if (arg_userns_mode != USER_NAMESPACE_NO) {
3905 if (!barrier_place_and_sync(&barrier)) { /* #1 */
3906 log_error("Child died too early.");
3907 return -ESRCH;
3908 }
3909
3910 r = setup_uid_map(*pid);
3911 if (r < 0)
3912 return r;
3913
3914 (void) barrier_place(&barrier); /* #2 */
3915 }
3916
3917 if (arg_private_network) {
3918 if (!arg_network_namespace_path) {
3919 /* Wait until the child has unshared its network namespace. */
3920 if (!barrier_place_and_sync(&barrier)) { /* #3 */
3921 log_error("Child died too early");
3922 return -ESRCH;
3923 }
3924 }
3925
3926 r = move_network_interfaces(*pid, arg_network_interfaces);
3927 if (r < 0)
3928 return r;
3929
3930 if (arg_network_veth) {
3931 r = setup_veth(arg_machine, *pid, veth_name,
3932 arg_network_bridge || arg_network_zone);
3933 if (r < 0)
3934 return r;
3935 else if (r > 0)
3936 ifi = r;
3937
3938 if (arg_network_bridge) {
3939 /* Add the interface to a bridge */
3940 r = setup_bridge(veth_name, arg_network_bridge, false);
3941 if (r < 0)
3942 return r;
3943 if (r > 0)
3944 ifi = r;
3945 } else if (arg_network_zone) {
3946 /* Add the interface to a bridge, possibly creating it */
3947 r = setup_bridge(veth_name, arg_network_zone, true);
3948 if (r < 0)
3949 return r;
3950 if (r > 0)
3951 ifi = r;
3952 }
3953 }
3954
3955 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
3956 if (r < 0)
3957 return r;
3958
3959 /* We created the primary and extra veth links now; let's remember this, so that we know to
3960 remove them later on. Note that we don't bother with removing veth links that were created
3961 here when their setup failed half-way, because in that case the kernel should be able to
3962 remove them on its own, since they cannot be referenced by anything yet. */
3963 *veth_created = true;
3964
3965 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
3966 if (r < 0)
3967 return r;
3968
3969 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
3970 if (r < 0)
3971 return r;
3972 }
3973
3974 if (arg_register || !arg_keep_unit) {
3975 r = sd_bus_default_system(&bus);
3976 if (r < 0)
3977 return log_error_errno(r, "Failed to open system bus: %m");
3978
3979 r = sd_bus_set_close_on_exit(bus, false);
3980 if (r < 0)
3981 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
3982 }
3983
3984 if (!arg_keep_unit) {
3985 /* When a new scope is created for this container, then we'll be registered as its controller, in which
3986 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
3987 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
3988
3989 r = sd_bus_match_signal_async(
3990 bus,
3991 NULL,
3992 "org.freedesktop.systemd1",
3993 NULL,
3994 "org.freedesktop.systemd1.Scope",
3995 "RequestStop",
3996 on_request_stop, NULL, PID_TO_PTR(*pid));
3997 if (r < 0)
3998 return log_error_errno(r, "Failed to request RequestStop match: %m");
3999 }
4000
4001 if (arg_register) {
4002 r = register_machine(
4003 bus,
4004 arg_machine,
4005 *pid,
4006 arg_directory,
4007 arg_uuid,
4008 ifi,
4009 arg_slice,
4010 arg_custom_mounts, arg_n_custom_mounts,
4011 arg_kill_signal,
4012 arg_property,
4013 arg_keep_unit,
4014 arg_container_service_name);
4015 if (r < 0)
4016 return r;
4017
4018 } else if (!arg_keep_unit) {
4019 r = allocate_scope(
4020 bus,
4021 arg_machine,
4022 *pid,
4023 arg_slice,
4024 arg_custom_mounts, arg_n_custom_mounts,
4025 arg_kill_signal,
4026 arg_property);
4027 if (r < 0)
4028 return r;
4029
4030 } else if (arg_slice || arg_property)
4031 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4032
4033 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4034 if (r < 0)
4035 return r;
4036
4037 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
4038 if (r < 0)
4039 return r;
4040
4041 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4042 if (r < 0)
4043 return r;
4044
4045 /* Notify the child that the parent is ready with all
4046 * its setup (including cgroup-ification), and that
4047 * the child can now hand over control to the code to
4048 * run inside the container. */
4049 (void) barrier_place(&barrier); /* #4 */
4050
4051 /* Block SIGCHLD here, before notifying child.
4052 * process_pty() will handle it with the other signals. */
4053 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4054
4055 /* Reset signal to default */
4056 r = default_signals(SIGCHLD, -1);
4057 if (r < 0)
4058 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4059
4060 r = sd_event_new(&event);
4061 if (r < 0)
4062 return log_error_errno(r, "Failed to get default event source: %m");
4063
4064 (void) sd_event_set_watchdog(event, true);
4065
4066 if (bus) {
4067 r = sd_bus_attach_event(bus, event, 0);
4068 if (r < 0)
4069 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4070 }
4071
4072 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
4073 if (r < 0)
4074 return r;
4075
4076 /* Let the child know that we are ready and wait that the child is completely ready now. */
4077 if (!barrier_place_and_sync(&barrier)) { /* #5 */
4078 log_error("Child died too early.");
4079 return -ESRCH;
4080 }
4081
4082 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4083 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4084 etc_passwd_lock = safe_close(etc_passwd_lock);
4085
4086 sd_notifyf(false,
4087 "STATUS=Container running.\n"
4088 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
4089 if (!arg_notify_ready)
4090 (void) sd_notify(false, "READY=1\n");
4091
4092 if (arg_kill_signal > 0) {
4093 /* Try to kill the init system on SIGINT or SIGTERM */
4094 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4095 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
4096 } else {
4097 /* Immediately exit */
4098 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4099 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4100 }
4101
4102 /* Exit when the child exits */
4103 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
4104
4105 if (arg_expose_ports) {
4106 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
4107 if (r < 0)
4108 return r;
4109
4110 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
4111 }
4112
4113 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4114
4115 r = pty_forward_new(event, master,
4116 PTY_FORWARD_IGNORE_VHANGUP | (interactive ? 0 : PTY_FORWARD_READ_ONLY),
4117 &forward);
4118 if (r < 0)
4119 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4120
4121 r = sd_event_loop(event);
4122 if (r < 0)
4123 return log_error_errno(r, "Failed to run event loop: %m");
4124
4125 pty_forward_get_last_char(forward, &last_char);
4126
4127 forward = pty_forward_free(forward);
4128
4129 if (!arg_quiet && last_char != '\n')
4130 putc('\n', stdout);
4131
4132 /* Kill if it is not dead yet anyway */
4133 if (bus) {
4134 if (arg_register)
4135 terminate_machine(bus, arg_machine);
4136 else if (!arg_keep_unit)
4137 terminate_scope(bus, arg_machine);
4138 }
4139
4140 /* Normally redundant, but better safe than sorry */
4141 (void) kill(*pid, SIGKILL);
4142
4143 r = wait_for_container(*pid, &container_status);
4144 *pid = 0;
4145
4146 if (r < 0)
4147 /* We failed to wait for the container, or the container exited abnormally. */
4148 return r;
4149 if (r > 0 || container_status == CONTAINER_TERMINATED) {
4150 /* r > 0 → The container exited with a non-zero status.
4151 * As a special case, we need to replace 133 with a different value,
4152 * because 133 is special-cased in the service file to reboot the container.
4153 * otherwise → The container exited with zero status and a reboot was not requested.
4154 */
4155 if (r == EXIT_FORCE_RESTART)
4156 r = EXIT_FAILURE; /* replace 133 with the general failure code */
4157 *ret = r;
4158 return 0; /* finito */
4159 }
4160
4161 /* CONTAINER_REBOOTED, loop again */
4162
4163 if (arg_keep_unit) {
4164 /* Special handling if we are running as a service: instead of simply
4165 * restarting the machine we want to restart the entire service, so let's
4166 * inform systemd about this with the special exit code 133. The service
4167 * file uses RestartForceExitStatus=133 so that this results in a full
4168 * nspawn restart. This is necessary since we might have cgroup parameters
4169 * set we want to have flushed out. */
4170 *ret = EXIT_FORCE_RESTART;
4171 return 0; /* finito */
4172 }
4173
4174 expose_port_flush(arg_expose_ports, exposed);
4175
4176 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4177 *veth_created = false;
4178 return 1; /* loop again */
4179 }
4180
4181 static int initialize_rlimits(void) {
4182 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4183 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4184 * container execution environments. */
4185
4186 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
4187 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
4188 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
4189 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
4190 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
4191 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
4192 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
4193 [RLIMIT_MEMLOCK] = { 65536, 65536 },
4194 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
4195 [RLIMIT_NICE] = { 0, 0 },
4196 [RLIMIT_NOFILE] = { 1024, 4096 },
4197 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
4198 [RLIMIT_RTPRIO] = { 0, 0 },
4199 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
4200 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
4201
4202 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4203 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4204 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4205 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4206 * that PID 1 changes a number of other resource limits during early initialization which is why we
4207 * don't read the other limits from PID 1 but prefer the static table above. */
4208 };
4209
4210 int rl;
4211
4212 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
4213 /* Let's only fill in what the user hasn't explicitly configured anyway */
4214 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
4215 const struct rlimit *v;
4216 struct rlimit buffer;
4217
4218 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
4219 /* For these two let's read the limits off PID 1. See above for an explanation. */
4220
4221 if (prlimit(1, rl, NULL, &buffer) < 0)
4222 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
4223
4224 v = &buffer;
4225 } else
4226 v = kernel_defaults + rl;
4227
4228 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
4229 if (!arg_rlimit[rl])
4230 return log_oom();
4231 }
4232
4233 if (DEBUG_LOGGING) {
4234 _cleanup_free_ char *k = NULL;
4235
4236 (void) rlimit_format(arg_rlimit[rl], &k);
4237 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
4238 }
4239 }
4240
4241 return 0;
4242 }
4243
4244 int main(int argc, char *argv[]) {
4245 _cleanup_free_ char *console = NULL;
4246 _cleanup_close_ int master = -1;
4247 _cleanup_fdset_free_ FDSet *fds = NULL;
4248 int r, n_fd_passed, ret = EXIT_SUCCESS;
4249 char veth_name[IFNAMSIZ] = "";
4250 bool secondary = false, remove_directory = false, remove_image = false;
4251 pid_t pid = 0;
4252 union in_addr_union exposed = {};
4253 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4254 bool interactive, veth_created = false, remove_tmprootdir = false;
4255 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
4256 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
4257 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4258 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
4259
4260 log_parse_environment();
4261 log_open();
4262
4263 /* Make sure rename_process() in the stub init process can work */
4264 saved_argv = argv;
4265 saved_argc = argc;
4266
4267 r = parse_argv(argc, argv);
4268 if (r <= 0)
4269 goto finish;
4270
4271 r = must_be_root();
4272 if (r < 0)
4273 goto finish;
4274
4275 r = initialize_rlimits();
4276 if (r < 0)
4277 goto finish;
4278
4279 r = determine_names();
4280 if (r < 0)
4281 goto finish;
4282
4283 r = load_settings();
4284 if (r < 0)
4285 goto finish;
4286
4287 r = verify_arguments();
4288 if (r < 0)
4289 goto finish;
4290
4291 r = detect_unified_cgroup_hierarchy_from_environment();
4292 if (r < 0)
4293 goto finish;
4294
4295 n_fd_passed = sd_listen_fds(false);
4296 if (n_fd_passed > 0) {
4297 r = fdset_new_listen_fds(&fds, false);
4298 if (r < 0) {
4299 log_error_errno(r, "Failed to collect file descriptors: %m");
4300 goto finish;
4301 }
4302 }
4303
4304 /* The "default" umask. This is appropriate for most file and directory
4305 * operations performed by nspawn, and is the umask that will be used for
4306 * the child. Functions like copy_devnodes() change the umask temporarily. */
4307 umask(0022);
4308
4309 if (arg_directory) {
4310 assert(!arg_image);
4311
4312 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
4313 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4314 r = -EINVAL;
4315 goto finish;
4316 }
4317
4318 if (arg_ephemeral) {
4319 _cleanup_free_ char *np = NULL;
4320
4321 r = chase_symlinks_and_update(&arg_directory, 0);
4322 if (r < 0)
4323 goto finish;
4324
4325 /* If the specified path is a mount point we
4326 * generate the new snapshot immediately
4327 * inside it under a random name. However if
4328 * the specified is not a mount point we
4329 * create the new snapshot in the parent
4330 * directory, just next to it. */
4331 r = path_is_mount_point(arg_directory, NULL, 0);
4332 if (r < 0) {
4333 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4334 goto finish;
4335 }
4336 if (r > 0)
4337 r = tempfn_random_child(arg_directory, "machine.", &np);
4338 else
4339 r = tempfn_random(arg_directory, "machine.", &np);
4340 if (r < 0) {
4341 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
4342 goto finish;
4343 }
4344
4345 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4346 if (r < 0) {
4347 log_error_errno(r, "Failed to lock %s: %m", np);
4348 goto finish;
4349 }
4350
4351 r = btrfs_subvol_snapshot(arg_directory, np,
4352 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4353 BTRFS_SNAPSHOT_FALLBACK_COPY |
4354 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4355 BTRFS_SNAPSHOT_RECURSIVE |
4356 BTRFS_SNAPSHOT_QUOTA);
4357 if (r < 0) {
4358 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4359 goto finish;
4360 }
4361
4362 free_and_replace(arg_directory, np);
4363
4364 remove_directory = true;
4365
4366 } else {
4367 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
4368 if (r < 0)
4369 goto finish;
4370
4371 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4372 if (r == -EBUSY) {
4373 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4374 goto finish;
4375 }
4376 if (r < 0) {
4377 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4378 goto finish;
4379 }
4380
4381 if (arg_template) {
4382 r = chase_symlinks_and_update(&arg_template, 0);
4383 if (r < 0)
4384 goto finish;
4385
4386 r = btrfs_subvol_snapshot(arg_template, arg_directory,
4387 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4388 BTRFS_SNAPSHOT_FALLBACK_COPY |
4389 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4390 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
4391 BTRFS_SNAPSHOT_RECURSIVE |
4392 BTRFS_SNAPSHOT_QUOTA);
4393 if (r == -EEXIST)
4394 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4395 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4396 else if (r < 0) {
4397 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4398 goto finish;
4399 } else
4400 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4401 "Populated %s from template %s.", arg_directory, arg_template);
4402 }
4403 }
4404
4405 if (arg_start_mode == START_BOOT) {
4406 const char *p;
4407
4408 if (arg_pivot_root_new)
4409 p = prefix_roota(arg_directory, arg_pivot_root_new);
4410 else
4411 p = arg_directory;
4412
4413 if (path_is_os_tree(p) <= 0) {
4414 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
4415 r = -EINVAL;
4416 goto finish;
4417 }
4418 } else {
4419 const char *p, *q;
4420
4421 if (arg_pivot_root_new)
4422 p = prefix_roota(arg_directory, arg_pivot_root_new);
4423 else
4424 p = arg_directory;
4425
4426 q = strjoina(p, "/usr/");
4427
4428 if (laccess(q, F_OK) < 0) {
4429 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
4430 r = -EINVAL;
4431 goto finish;
4432 }
4433 }
4434
4435 } else {
4436 assert(arg_image);
4437 assert(!arg_template);
4438
4439 r = chase_symlinks_and_update(&arg_image, 0);
4440 if (r < 0)
4441 goto finish;
4442
4443 if (arg_ephemeral) {
4444 _cleanup_free_ char *np = NULL;
4445
4446 r = tempfn_random(arg_image, "machine.", &np);
4447 if (r < 0) {
4448 log_error_errno(r, "Failed to generate name for image snapshot: %m");
4449 goto finish;
4450 }
4451
4452 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4453 if (r < 0) {
4454 r = log_error_errno(r, "Failed to create image lock: %m");
4455 goto finish;
4456 }
4457
4458 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, COPY_REFLINK);
4459 if (r < 0) {
4460 r = log_error_errno(r, "Failed to copy image file: %m");
4461 goto finish;
4462 }
4463
4464 free_and_replace(arg_image, np);
4465
4466 remove_image = true;
4467 } else {
4468 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4469 if (r == -EBUSY) {
4470 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4471 goto finish;
4472 }
4473 if (r < 0) {
4474 r = log_error_errno(r, "Failed to create image lock: %m");
4475 goto finish;
4476 }
4477
4478 if (!arg_root_hash) {
4479 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4480 if (r < 0) {
4481 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4482 goto finish;
4483 }
4484 }
4485 }
4486
4487 if (!mkdtemp(tmprootdir)) {
4488 r = log_error_errno(errno, "Failed to create temporary directory: %m");
4489 goto finish;
4490 }
4491
4492 remove_tmprootdir = true;
4493
4494 arg_directory = strdup(tmprootdir);
4495 if (!arg_directory) {
4496 r = log_oom();
4497 goto finish;
4498 }
4499
4500 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4501 if (r < 0) {
4502 log_error_errno(r, "Failed to set up loopback block device: %m");
4503 goto finish;
4504 }
4505
4506 r = dissect_image_and_warn(
4507 loop->fd,
4508 arg_image,
4509 arg_root_hash, arg_root_hash_size,
4510 DISSECT_IMAGE_REQUIRE_ROOT,
4511 &dissected_image);
4512 if (r == -ENOPKG) {
4513 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4514 log_notice("Note that the disk image needs to\n"
4515 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4516 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4517 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4518 " d) or contain a file system without a partition table\n"
4519 "in order to be bootable with systemd-nspawn.");
4520 goto finish;
4521 }
4522 if (r < 0)
4523 goto finish;
4524
4525 if (!arg_root_hash && dissected_image->can_verity)
4526 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4527
4528 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
4529 if (r < 0)
4530 goto finish;
4531
4532 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4533 if (remove_image && unlink(arg_image) >= 0)
4534 remove_image = false;
4535 }
4536
4537 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
4538 if (r < 0)
4539 goto finish;
4540
4541 interactive =
4542 isatty(STDIN_FILENO) > 0 &&
4543 isatty(STDOUT_FILENO) > 0;
4544
4545 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NONBLOCK);
4546 if (master < 0) {
4547 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4548 goto finish;
4549 }
4550
4551 r = ptsname_malloc(master, &console);
4552 if (r < 0) {
4553 r = log_error_errno(r, "Failed to determine tty name: %m");
4554 goto finish;
4555 }
4556
4557 if (arg_selinux_apifs_context) {
4558 r = mac_selinux_apply(console, arg_selinux_apifs_context);
4559 if (r < 0)
4560 goto finish;
4561 }
4562
4563 if (unlockpt(master) < 0) {
4564 r = log_error_errno(errno, "Failed to unlock tty: %m");
4565 goto finish;
4566 }
4567
4568 if (!arg_quiet)
4569 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4570 arg_machine, arg_image ?: arg_directory);
4571
4572 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
4573
4574 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
4575 r = log_error_errno(errno, "Failed to become subreaper: %m");
4576 goto finish;
4577 }
4578
4579 for (;;) {
4580 r = run(master,
4581 console,
4582 dissected_image,
4583 interactive, secondary,
4584 fds,
4585 veth_name, &veth_created,
4586 &exposed,
4587 &pid, &ret);
4588 if (r <= 0)
4589 break;
4590 }
4591
4592 finish:
4593 sd_notify(false,
4594 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
4595 "STOPPING=1\nSTATUS=Terminating...");
4596
4597 if (pid > 0)
4598 (void) kill(pid, SIGKILL);
4599
4600 /* Try to flush whatever is still queued in the pty */
4601 if (master >= 0) {
4602 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
4603 master = safe_close(master);
4604 }
4605
4606 if (pid > 0)
4607 (void) wait_for_terminate(pid, NULL);
4608
4609 pager_close();
4610
4611 if (remove_directory && arg_directory) {
4612 int k;
4613
4614 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
4615 if (k < 0)
4616 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
4617 }
4618
4619 if (remove_image && arg_image) {
4620 if (unlink(arg_image) < 0)
4621 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
4622 }
4623
4624 if (remove_tmprootdir) {
4625 if (rmdir(tmprootdir) < 0)
4626 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
4627 }
4628
4629 if (arg_machine) {
4630 const char *p;
4631
4632 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
4633 (void) rm_rf(p, REMOVE_ROOT);
4634 }
4635
4636 expose_port_flush(arg_expose_ports, &exposed);
4637
4638 if (veth_created)
4639 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4640 (void) remove_bridge(arg_network_zone);
4641
4642 free(arg_directory);
4643 free(arg_template);
4644 free(arg_image);
4645 free(arg_machine);
4646 free(arg_hostname);
4647 free(arg_user);
4648 free(arg_pivot_root_new);
4649 free(arg_pivot_root_old);
4650 free(arg_chdir);
4651 strv_free(arg_setenv);
4652 free(arg_network_bridge);
4653 strv_free(arg_network_interfaces);
4654 strv_free(arg_network_macvlan);
4655 strv_free(arg_network_ipvlan);
4656 strv_free(arg_network_veth_extra);
4657 strv_free(arg_parameters);
4658 free(arg_network_zone);
4659 free(arg_network_namespace_path);
4660 strv_free(arg_property);
4661 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4662 expose_port_free_all(arg_expose_ports);
4663 free(arg_root_hash);
4664 rlimit_free_all(arg_rlimit);
4665 strv_free(arg_syscall_whitelist);
4666 strv_free(arg_syscall_blacklist);
4667 arg_cpuset = cpu_set_mfree(arg_cpuset);
4668
4669 return r < 0 ? EXIT_FAILURE : ret;
4670 }
Unnamed repository; edit this file 'description' to name the repository.