]> 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
util: introduce save_argc_argv() helper
[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/fs.h>
10 #include <linux/loop.h>
11 #include <pwd.h>
12 #include <sched.h>
13 #if HAVE_SELINUX
14 #include <selinux/selinux.h>
15 #endif
16 #include <signal.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 #include <sys/file.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 "main-func.h"
61 #include "missing.h"
62 #include "mkdir.h"
63 #include "mount-util.h"
64 #include "mountpoint-util.h"
65 #include "namespace-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-def.h"
69 #include "nspawn-expose-ports.h"
70 #include "nspawn-mount.h"
71 #include "nspawn-network.h"
72 #include "nspawn-oci.h"
73 #include "nspawn-patch-uid.h"
74 #include "nspawn-register.h"
75 #include "nspawn-seccomp.h"
76 #include "nspawn-settings.h"
77 #include "nspawn-setuid.h"
78 #include "nspawn-stub-pid1.h"
79 #include "nulstr-util.h"
80 #include "os-util.h"
81 #include "pager.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "pretty-print.h"
85 #include "process-util.h"
86 #include "ptyfwd.h"
87 #include "random-util.h"
88 #include "raw-clone.h"
89 #include "rlimit-util.h"
90 #include "rm-rf.h"
91 #if HAVE_SECCOMP
92 #include "seccomp-util.h"
93 #endif
94 #include "selinux-util.h"
95 #include "signal-util.h"
96 #include "socket-util.h"
97 #include "stat-util.h"
98 #include "stdio-util.h"
99 #include "string-table.h"
100 #include "string-util.h"
101 #include "strv.h"
102 #include "sysctl-util.h"
103 #include "terminal-util.h"
104 #include "tmpfile-util.h"
105 #include "umask-util.h"
106 #include "user-util.h"
107 #include "util.h"
108
109 #if HAVE_SPLIT_USR
110 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
111 #else
112 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
113 #endif
114
115 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
116 * nspawn_notify_socket_path is relative to the container
117 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
119
120 #define EXIT_FORCE_RESTART 133
121
122 typedef enum ContainerStatus {
123 CONTAINER_TERMINATED,
124 CONTAINER_REBOOTED,
125 } ContainerStatus;
126
127 static char *arg_directory = NULL;
128 static char *arg_template = NULL;
129 static char *arg_chdir = NULL;
130 static char *arg_pivot_root_new = NULL;
131 static char *arg_pivot_root_old = NULL;
132 static char *arg_user = NULL;
133 static uid_t arg_uid = UID_INVALID;
134 static gid_t arg_gid = GID_INVALID;
135 static gid_t* arg_supplementary_gids = NULL;
136 static size_t arg_n_supplementary_gids = 0;
137 static sd_id128_t arg_uuid = {};
138 static char *arg_machine = NULL; /* The name used by the host to refer to this */
139 static char *arg_hostname = NULL; /* The name the payload sees by default */
140 static const char *arg_selinux_context = NULL;
141 static const char *arg_selinux_apifs_context = NULL;
142 static char *arg_slice = NULL;
143 static bool arg_private_network = false;
144 static bool arg_read_only = false;
145 static StartMode arg_start_mode = START_PID1;
146 static bool arg_ephemeral = false;
147 static LinkJournal arg_link_journal = LINK_AUTO;
148 static bool arg_link_journal_try = false;
149 static uint64_t arg_caps_retain =
150 (1ULL << CAP_AUDIT_CONTROL) |
151 (1ULL << CAP_AUDIT_WRITE) |
152 (1ULL << CAP_CHOWN) |
153 (1ULL << CAP_DAC_OVERRIDE) |
154 (1ULL << CAP_DAC_READ_SEARCH) |
155 (1ULL << CAP_FOWNER) |
156 (1ULL << CAP_FSETID) |
157 (1ULL << CAP_IPC_OWNER) |
158 (1ULL << CAP_KILL) |
159 (1ULL << CAP_LEASE) |
160 (1ULL << CAP_LINUX_IMMUTABLE) |
161 (1ULL << CAP_MKNOD) |
162 (1ULL << CAP_NET_BIND_SERVICE) |
163 (1ULL << CAP_NET_BROADCAST) |
164 (1ULL << CAP_NET_RAW) |
165 (1ULL << CAP_SETFCAP) |
166 (1ULL << CAP_SETGID) |
167 (1ULL << CAP_SETPCAP) |
168 (1ULL << CAP_SETUID) |
169 (1ULL << CAP_SYS_ADMIN) |
170 (1ULL << CAP_SYS_BOOT) |
171 (1ULL << CAP_SYS_CHROOT) |
172 (1ULL << CAP_SYS_NICE) |
173 (1ULL << CAP_SYS_PTRACE) |
174 (1ULL << CAP_SYS_RESOURCE) |
175 (1ULL << CAP_SYS_TTY_CONFIG);
176 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
177 static CustomMount *arg_custom_mounts = NULL;
178 static size_t arg_n_custom_mounts = 0;
179 static char **arg_setenv = NULL;
180 static bool arg_quiet = false;
181 static bool arg_register = true;
182 static bool arg_keep_unit = false;
183 static char **arg_network_interfaces = NULL;
184 static char **arg_network_macvlan = NULL;
185 static char **arg_network_ipvlan = NULL;
186 static bool arg_network_veth = false;
187 static char **arg_network_veth_extra = NULL;
188 static char *arg_network_bridge = NULL;
189 static char *arg_network_zone = NULL;
190 static char *arg_network_namespace_path = NULL;
191 static PagerFlags arg_pager_flags = 0;
192 static unsigned long arg_personality = PERSONALITY_INVALID;
193 static char *arg_image = NULL;
194 static char *arg_oci_bundle = NULL;
195 static VolatileMode arg_volatile_mode = VOLATILE_NO;
196 static ExposePort *arg_expose_ports = NULL;
197 static char **arg_property = NULL;
198 static sd_bus_message *arg_property_message = NULL;
199 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
200 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
201 static bool arg_userns_chown = false;
202 static int arg_kill_signal = 0;
203 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
204 static SettingsMask arg_settings_mask = 0;
205 static int arg_settings_trusted = -1;
206 static char **arg_parameters = NULL;
207 static const char *arg_container_service_name = "systemd-nspawn";
208 static bool arg_notify_ready = false;
209 static bool arg_use_cgns = true;
210 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
211 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
212 static void *arg_root_hash = NULL;
213 static size_t arg_root_hash_size = 0;
214 static char **arg_syscall_whitelist = NULL;
215 static char **arg_syscall_blacklist = NULL;
216 #if HAVE_SECCOMP
217 static scmp_filter_ctx arg_seccomp = NULL;
218 #endif
219 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
220 static bool arg_no_new_privileges = false;
221 static int arg_oom_score_adjust = 0;
222 static bool arg_oom_score_adjust_set = false;
223 static cpu_set_t *arg_cpuset = NULL;
224 static unsigned arg_cpuset_ncpus = 0;
225 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
226 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
227 static unsigned arg_console_width = (unsigned) -1, arg_console_height = (unsigned) -1;
228 static DeviceNode* arg_extra_nodes = NULL;
229 static size_t arg_n_extra_nodes = 0;
230 static char **arg_sysctl = NULL;
231 static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
232
233 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
234 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
235 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
236 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
237 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
238 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
239 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
240 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
241 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
250 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
252 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
253 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
254 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
255 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
256 STATIC_DESTRUCTOR_REGISTER(arg_root_hash, freep);
257 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist, strv_freep);
258 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist, strv_freep);
259 #if HAVE_SECCOMP
260 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
261 #endif
262 STATIC_DESTRUCTOR_REGISTER(arg_cpuset, CPU_FREEp);
263 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
264
265 static int help(void) {
266 _cleanup_free_ char *link = NULL;
267 int r;
268
269 (void) pager_open(arg_pager_flags);
270
271 r = terminal_urlify_man("systemd-nspawn", "1", &link);
272 if (r < 0)
273 return log_oom();
274
275 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
276 "Spawn a command or OS in a light-weight container.\n\n"
277 " -h --help Show this help\n"
278 " --version Print version string\n"
279 " -q --quiet Do not show status information\n"
280 " --no-pager Do not pipe output into a pager\n"
281 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
282 "%3$sImage:%4$s\n"
283 " -D --directory=PATH Root directory for the container\n"
284 " --template=PATH Initialize root directory from template directory,\n"
285 " if missing\n"
286 " -x --ephemeral Run container with snapshot of root directory, and\n"
287 " remove it after exit\n"
288 " -i --image=PATH File system device or disk image for the container\n"
289 " --oci-bundle=PATH OCI bundle directory\n"
290 " --read-only Mount the root directory read-only\n"
291 " --volatile[=MODE] Run the system in volatile mode\n"
292 " --root-hash=HASH Specify verity root hash\n"
293 " --pivot-root=PATH[:PATH]\n"
294 " Pivot root to given directory in the container\n\n"
295 "%3$sExecution:%4$s\n"
296 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
297 " -b --boot Boot up full system (i.e. invoke init)\n"
298 " --chdir=PATH Set working directory in the container\n"
299 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
300 " -u --user=USER Run the command under specified user or UID\n"
301 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
302 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
303 "%3$sSystem Identity:%4$s\n"
304 " -M --machine=NAME Set the machine name for the container\n"
305 " --hostname=NAME Override the hostname for the container\n"
306 " --uuid=UUID Set a specific machine UUID for the container\n\n"
307 "%3$sProperties:%4$s\n"
308 " -S --slice=SLICE Place the container in the specified slice\n"
309 " --property=NAME=VALUE Set scope unit property\n"
310 " --register=BOOLEAN Register container as machine\n"
311 " --keep-unit Do not register a scope for the machine, reuse\n"
312 " the service unit nspawn is running in\n\n"
313 "%3$sUser Namespacing:%4$s\n"
314 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
315 " --private-users[=UIDBASE[:NUIDS]]\n"
316 " Similar, but with user configured UID/GID range\n"
317 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
318 "%3$sNetworking:%4$s\n"
319 " --private-network Disable network in container\n"
320 " --network-interface=INTERFACE\n"
321 " Assign an existing network interface to the\n"
322 " container\n"
323 " --network-macvlan=INTERFACE\n"
324 " Create a macvlan network interface based on an\n"
325 " existing network interface to the container\n"
326 " --network-ipvlan=INTERFACE\n"
327 " Create a ipvlan network interface based on an\n"
328 " existing network interface to the container\n"
329 " -n --network-veth Add a virtual Ethernet connection between host\n"
330 " and container\n"
331 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
332 " Add an additional virtual Ethernet link between\n"
333 " host and container\n"
334 " --network-bridge=INTERFACE\n"
335 " Add a virtual Ethernet connection to the container\n"
336 " and attach it to an existing bridge on the host\n"
337 " --network-zone=NAME Similar, but attach the new interface to an\n"
338 " an automatically managed bridge interface\n"
339 " --network-namespace-path=PATH\n"
340 " Set network namespace to the one represented by\n"
341 " the specified kernel namespace file node\n"
342 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
343 " Expose a container IP port on the host\n\n"
344 "%3$sSecurity:%4$s\n"
345 " --capability=CAP In addition to the default, retain specified\n"
346 " capability\n"
347 " --drop-capability=CAP Drop the specified capability from the default set\n"
348 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
349 " --system-call-filter=LIST|~LIST\n"
350 " Permit/prohibit specific system calls\n"
351 " -Z --selinux-context=SECLABEL\n"
352 " Set the SELinux security context to be used by\n"
353 " processes in the container\n"
354 " -L --selinux-apifs-context=SECLABEL\n"
355 " Set the SELinux security context to be used by\n"
356 " API/tmpfs file systems in the container\n\n"
357 "%3$sResources:%4$s\n"
358 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
359 " --oom-score-adjust=VALUE\n"
360 " Adjust the OOM score value for the payload\n"
361 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
362 " --personality=ARCH Pick personality for this container\n\n"
363 "%3$sIntegration:%4$s\n"
364 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
365 " --timezone=MODE Select mode of /etc/localtime initialization\n"
366 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
367 " host, try-guest, try-host\n"
368 " -j Equivalent to --link-journal=try-guest\n\n"
369 "%3$sMounts:%4$s\n"
370 " --bind=PATH[:PATH[:OPTIONS]]\n"
371 " Bind mount a file or directory from the host into\n"
372 " the container\n"
373 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
374 " Similar, but creates a read-only bind mount\n"
375 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
376 " it\n"
377 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
378 " --overlay=PATH[:PATH...]:PATH\n"
379 " Create an overlay mount from the host to \n"
380 " the container\n"
381 " --overlay-ro=PATH[:PATH...]:PATH\n"
382 " Similar, but creates a read-only overlay mount\n\n"
383 "%3$sInput/Output:%4$s\n"
384 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
385 " set up for the container.\n"
386 " -P --pipe Equivalent to --console=pipe\n"
387 "\nSee the %2$s for details.\n"
388 , program_invocation_short_name
389 , link
390 , ansi_underline(), ansi_normal());
391
392 return 0;
393 }
394
395 static int custom_mount_check_all(void) {
396 size_t i;
397
398 for (i = 0; i < arg_n_custom_mounts; i++) {
399 CustomMount *m = &arg_custom_mounts[i];
400
401 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
402 if (arg_userns_chown)
403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
404 "--private-users-chown may not be combined with custom root mounts.");
405 else if (arg_uid_shift == UID_INVALID)
406 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
407 "--private-users with automatic UID shift may not be combined with custom root mounts.");
408 }
409 }
410
411 return 0;
412 }
413
414 static int detect_unified_cgroup_hierarchy_from_environment(void) {
415 const char *e;
416 int r;
417
418 /* Allow the user to control whether the unified hierarchy is used */
419 e = getenv("UNIFIED_CGROUP_HIERARCHY");
420 if (e) {
421 r = parse_boolean(e);
422 if (r < 0)
423 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
424 if (r > 0)
425 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
426 else
427 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
428 }
429
430 return 0;
431 }
432
433 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
434 int r;
435
436 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
437 * image actually supports. */
438 r = cg_all_unified();
439 if (r < 0)
440 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
441 if (r > 0) {
442 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
443 * routine only detects 231, so we'll have a false negative here for 230. */
444 r = systemd_installation_has_version(directory, 230);
445 if (r < 0)
446 return log_error_errno(r, "Failed to determine systemd version in container: %m");
447 if (r > 0)
448 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
449 else
450 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
451 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
452 /* Mixed cgroup hierarchy support was added in 233 */
453 r = systemd_installation_has_version(directory, 233);
454 if (r < 0)
455 return log_error_errno(r, "Failed to determine systemd version in container: %m");
456 if (r > 0)
457 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
458 else
459 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
460 } else
461 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
462
463 log_debug("Using %s hierarchy for container.",
464 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
465 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
466
467 return 0;
468 }
469
470 static void parse_share_ns_env(const char *name, unsigned long ns_flag) {
471 int r;
472
473 r = getenv_bool(name);
474 if (r == -ENXIO)
475 return;
476 if (r < 0)
477 log_warning_errno(r, "Failed to parse %s from environment, defaulting to false.", name);
478
479 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
480 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
481 }
482
483 static void parse_mount_settings_env(void) {
484 const char *e;
485 int r;
486
487 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
488 if (r >= 0)
489 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
490 else if (r != -ENXIO)
491 log_warning_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP, ignoring: %m");
492
493 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
494 if (!e)
495 return;
496
497 if (streq(e, "network")) {
498 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
499 return;
500 }
501
502 r = parse_boolean(e);
503 if (r < 0) {
504 log_warning_errno(r, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
505 return;
506 }
507
508 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
509 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
510 }
511
512 static void parse_environment(void) {
513 const char *e;
514 int r;
515
516 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
517 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
518 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
519 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
520
521 parse_mount_settings_env();
522
523 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
524 * even if it is supported. If not supported, it has no effect. */
525 if (!cg_ns_supported())
526 arg_use_cgns = false;
527 else {
528 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
529 if (r < 0) {
530 if (r != -ENXIO)
531 log_warning_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS, ignoring: %m");
532
533 arg_use_cgns = true;
534 } else {
535 arg_use_cgns = r > 0;
536 arg_settings_mask |= SETTING_USE_CGNS;
537 }
538 }
539
540 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
541 if (e)
542 arg_container_service_name = e;
543
544 detect_unified_cgroup_hierarchy_from_environment();
545 }
546
547 static int parse_argv(int argc, char *argv[]) {
548 enum {
549 ARG_VERSION = 0x100,
550 ARG_PRIVATE_NETWORK,
551 ARG_UUID,
552 ARG_READ_ONLY,
553 ARG_CAPABILITY,
554 ARG_DROP_CAPABILITY,
555 ARG_LINK_JOURNAL,
556 ARG_BIND,
557 ARG_BIND_RO,
558 ARG_TMPFS,
559 ARG_OVERLAY,
560 ARG_OVERLAY_RO,
561 ARG_INACCESSIBLE,
562 ARG_SHARE_SYSTEM,
563 ARG_REGISTER,
564 ARG_KEEP_UNIT,
565 ARG_NETWORK_INTERFACE,
566 ARG_NETWORK_MACVLAN,
567 ARG_NETWORK_IPVLAN,
568 ARG_NETWORK_BRIDGE,
569 ARG_NETWORK_ZONE,
570 ARG_NETWORK_VETH_EXTRA,
571 ARG_NETWORK_NAMESPACE_PATH,
572 ARG_PERSONALITY,
573 ARG_VOLATILE,
574 ARG_TEMPLATE,
575 ARG_PROPERTY,
576 ARG_PRIVATE_USERS,
577 ARG_KILL_SIGNAL,
578 ARG_SETTINGS,
579 ARG_CHDIR,
580 ARG_PIVOT_ROOT,
581 ARG_PRIVATE_USERS_CHOWN,
582 ARG_NOTIFY_READY,
583 ARG_ROOT_HASH,
584 ARG_SYSTEM_CALL_FILTER,
585 ARG_RLIMIT,
586 ARG_HOSTNAME,
587 ARG_NO_NEW_PRIVILEGES,
588 ARG_OOM_SCORE_ADJUST,
589 ARG_CPU_AFFINITY,
590 ARG_RESOLV_CONF,
591 ARG_TIMEZONE,
592 ARG_CONSOLE,
593 ARG_PIPE,
594 ARG_OCI_BUNDLE,
595 ARG_NO_PAGER,
596 };
597
598 static const struct option options[] = {
599 { "help", no_argument, NULL, 'h' },
600 { "version", no_argument, NULL, ARG_VERSION },
601 { "directory", required_argument, NULL, 'D' },
602 { "template", required_argument, NULL, ARG_TEMPLATE },
603 { "ephemeral", no_argument, NULL, 'x' },
604 { "user", required_argument, NULL, 'u' },
605 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
606 { "as-pid2", no_argument, NULL, 'a' },
607 { "boot", no_argument, NULL, 'b' },
608 { "uuid", required_argument, NULL, ARG_UUID },
609 { "read-only", no_argument, NULL, ARG_READ_ONLY },
610 { "capability", required_argument, NULL, ARG_CAPABILITY },
611 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
612 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
613 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
614 { "bind", required_argument, NULL, ARG_BIND },
615 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
616 { "tmpfs", required_argument, NULL, ARG_TMPFS },
617 { "overlay", required_argument, NULL, ARG_OVERLAY },
618 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
619 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
620 { "machine", required_argument, NULL, 'M' },
621 { "hostname", required_argument, NULL, ARG_HOSTNAME },
622 { "slice", required_argument, NULL, 'S' },
623 { "setenv", required_argument, NULL, 'E' },
624 { "selinux-context", required_argument, NULL, 'Z' },
625 { "selinux-apifs-context", required_argument, NULL, 'L' },
626 { "quiet", no_argument, NULL, 'q' },
627 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
628 { "register", required_argument, NULL, ARG_REGISTER },
629 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
630 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
631 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
632 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
633 { "network-veth", no_argument, NULL, 'n' },
634 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
635 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
636 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
637 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
638 { "personality", required_argument, NULL, ARG_PERSONALITY },
639 { "image", required_argument, NULL, 'i' },
640 { "volatile", optional_argument, NULL, ARG_VOLATILE },
641 { "port", required_argument, NULL, 'p' },
642 { "property", required_argument, NULL, ARG_PROPERTY },
643 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
644 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
645 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
646 { "settings", required_argument, NULL, ARG_SETTINGS },
647 { "chdir", required_argument, NULL, ARG_CHDIR },
648 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
649 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
650 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
651 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
652 { "rlimit", required_argument, NULL, ARG_RLIMIT },
653 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
654 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
655 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
656 { "timezone", required_argument, NULL, ARG_TIMEZONE },
657 { "console", required_argument, NULL, ARG_CONSOLE },
658 { "pipe", no_argument, NULL, ARG_PIPE },
659 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
660 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
661 {}
662 };
663
664 int c, r;
665 const char *p;
666 uint64_t plus = 0, minus = 0;
667 bool mask_all_settings = false, mask_no_settings = false;
668
669 assert(argc >= 0);
670 assert(argv);
671
672 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
673 switch (c) {
674
675 case 'h':
676 return help();
677
678 case ARG_VERSION:
679 return version();
680
681 case 'D':
682 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
683 if (r < 0)
684 return r;
685
686 arg_settings_mask |= SETTING_DIRECTORY;
687 break;
688
689 case ARG_TEMPLATE:
690 r = parse_path_argument_and_warn(optarg, false, &arg_template);
691 if (r < 0)
692 return r;
693
694 arg_settings_mask |= SETTING_DIRECTORY;
695 break;
696
697 case 'i':
698 r = parse_path_argument_and_warn(optarg, false, &arg_image);
699 if (r < 0)
700 return r;
701
702 arg_settings_mask |= SETTING_DIRECTORY;
703 break;
704
705 case ARG_OCI_BUNDLE:
706 r = parse_path_argument_and_warn(optarg, false, &arg_oci_bundle);
707 if (r < 0)
708 return r;
709
710 break;
711
712 case 'x':
713 arg_ephemeral = true;
714 arg_settings_mask |= SETTING_EPHEMERAL;
715 break;
716
717 case 'u':
718 r = free_and_strdup(&arg_user, optarg);
719 if (r < 0)
720 return log_oom();
721
722 arg_settings_mask |= SETTING_USER;
723 break;
724
725 case ARG_NETWORK_ZONE: {
726 char *j;
727
728 j = strappend("vz-", optarg);
729 if (!j)
730 return log_oom();
731
732 if (!ifname_valid(j)) {
733 log_error("Network zone name not valid: %s", j);
734 free(j);
735 return -EINVAL;
736 }
737
738 free_and_replace(arg_network_zone, j);
739
740 arg_network_veth = true;
741 arg_private_network = true;
742 arg_settings_mask |= SETTING_NETWORK;
743 break;
744 }
745
746 case ARG_NETWORK_BRIDGE:
747
748 if (!ifname_valid(optarg))
749 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
750 "Bridge interface name not valid: %s", optarg);
751
752 r = free_and_strdup(&arg_network_bridge, optarg);
753 if (r < 0)
754 return log_oom();
755
756 _fallthrough_;
757 case 'n':
758 arg_network_veth = true;
759 arg_private_network = true;
760 arg_settings_mask |= SETTING_NETWORK;
761 break;
762
763 case ARG_NETWORK_VETH_EXTRA:
764 r = veth_extra_parse(&arg_network_veth_extra, optarg);
765 if (r < 0)
766 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
767
768 arg_private_network = true;
769 arg_settings_mask |= SETTING_NETWORK;
770 break;
771
772 case ARG_NETWORK_INTERFACE:
773 if (!ifname_valid(optarg))
774 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
775 "Network interface name not valid: %s", optarg);
776
777 if (strv_extend(&arg_network_interfaces, optarg) < 0)
778 return log_oom();
779
780 arg_private_network = true;
781 arg_settings_mask |= SETTING_NETWORK;
782 break;
783
784 case ARG_NETWORK_MACVLAN:
785
786 if (!ifname_valid(optarg))
787 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
788 "MACVLAN network interface name not valid: %s", optarg);
789
790 if (strv_extend(&arg_network_macvlan, optarg) < 0)
791 return log_oom();
792
793 arg_private_network = true;
794 arg_settings_mask |= SETTING_NETWORK;
795 break;
796
797 case ARG_NETWORK_IPVLAN:
798
799 if (!ifname_valid(optarg))
800 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
801 "IPVLAN network interface name not valid: %s", optarg);
802
803 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
804 return log_oom();
805
806 _fallthrough_;
807 case ARG_PRIVATE_NETWORK:
808 arg_private_network = true;
809 arg_settings_mask |= SETTING_NETWORK;
810 break;
811
812 case ARG_NETWORK_NAMESPACE_PATH:
813 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
814 if (r < 0)
815 return r;
816
817 arg_settings_mask |= SETTING_NETWORK;
818 break;
819
820 case 'b':
821 if (arg_start_mode == START_PID2)
822 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
823 "--boot and --as-pid2 may not be combined.");
824
825 arg_start_mode = START_BOOT;
826 arg_settings_mask |= SETTING_START_MODE;
827 break;
828
829 case 'a':
830 if (arg_start_mode == START_BOOT)
831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
832 "--boot and --as-pid2 may not be combined.");
833
834 arg_start_mode = START_PID2;
835 arg_settings_mask |= SETTING_START_MODE;
836 break;
837
838 case ARG_UUID:
839 r = sd_id128_from_string(optarg, &arg_uuid);
840 if (r < 0)
841 return log_error_errno(r, "Invalid UUID: %s", optarg);
842
843 if (sd_id128_is_null(arg_uuid))
844 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
845 "Machine UUID may not be all zeroes.");
846
847 arg_settings_mask |= SETTING_MACHINE_ID;
848 break;
849
850 case 'S':
851 r = free_and_strdup(&arg_slice, optarg);
852 if (r < 0)
853 return log_oom();
854
855 arg_settings_mask |= SETTING_SLICE;
856 break;
857
858 case 'M':
859 if (isempty(optarg))
860 arg_machine = mfree(arg_machine);
861 else {
862 if (!machine_name_is_valid(optarg))
863 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
864 "Invalid machine name: %s", optarg);
865
866 r = free_and_strdup(&arg_machine, optarg);
867 if (r < 0)
868 return log_oom();
869 }
870 break;
871
872 case ARG_HOSTNAME:
873 if (isempty(optarg))
874 arg_hostname = mfree(arg_hostname);
875 else {
876 if (!hostname_is_valid(optarg, false))
877 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
878 "Invalid hostname: %s", optarg);
879
880 r = free_and_strdup(&arg_hostname, optarg);
881 if (r < 0)
882 return log_oom();
883 }
884
885 arg_settings_mask |= SETTING_HOSTNAME;
886 break;
887
888 case 'Z':
889 arg_selinux_context = optarg;
890 break;
891
892 case 'L':
893 arg_selinux_apifs_context = optarg;
894 break;
895
896 case ARG_READ_ONLY:
897 arg_read_only = true;
898 arg_settings_mask |= SETTING_READ_ONLY;
899 break;
900
901 case ARG_CAPABILITY:
902 case ARG_DROP_CAPABILITY: {
903 p = optarg;
904 for (;;) {
905 _cleanup_free_ char *t = NULL;
906
907 r = extract_first_word(&p, &t, ",", 0);
908 if (r < 0)
909 return log_error_errno(r, "Failed to parse capability %s.", t);
910 if (r == 0)
911 break;
912
913 if (streq(t, "all")) {
914 if (c == ARG_CAPABILITY)
915 plus = (uint64_t) -1;
916 else
917 minus = (uint64_t) -1;
918 } else {
919 r = capability_from_name(t);
920 if (r < 0)
921 return log_error_errno(r, "Failed to parse capability %s.", t);
922
923 if (c == ARG_CAPABILITY)
924 plus |= 1ULL << r;
925 else
926 minus |= 1ULL << r;
927 }
928 }
929
930 arg_settings_mask |= SETTING_CAPABILITY;
931 break;
932 }
933
934 case ARG_NO_NEW_PRIVILEGES:
935 r = parse_boolean(optarg);
936 if (r < 0)
937 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
938
939 arg_no_new_privileges = r;
940 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
941 break;
942
943 case 'j':
944 arg_link_journal = LINK_GUEST;
945 arg_link_journal_try = true;
946 arg_settings_mask |= SETTING_LINK_JOURNAL;
947 break;
948
949 case ARG_LINK_JOURNAL:
950 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
951 if (r < 0) {
952 log_error_errno(r, "Failed to parse link journal mode %s", optarg);
953 return -EINVAL;
954 }
955
956 arg_settings_mask |= SETTING_LINK_JOURNAL;
957 break;
958
959 case ARG_BIND:
960 case ARG_BIND_RO:
961 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
962 if (r < 0)
963 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
964
965 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
966 break;
967
968 case ARG_TMPFS:
969 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
970 if (r < 0)
971 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
972
973 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
974 break;
975
976 case ARG_OVERLAY:
977 case ARG_OVERLAY_RO:
978 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
979 if (r == -EADDRNOTAVAIL)
980 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
981 if (r < 0)
982 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
983
984 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
985 break;
986
987 case ARG_INACCESSIBLE:
988 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
989 if (r < 0)
990 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
991
992 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
993 break;
994
995 case 'E': {
996 char **n;
997
998 if (!env_assignment_is_valid(optarg))
999 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1000 "Environment variable assignment '%s' is not valid.", optarg);
1001
1002 n = strv_env_set(arg_setenv, optarg);
1003 if (!n)
1004 return log_oom();
1005
1006 strv_free_and_replace(arg_setenv, n);
1007 arg_settings_mask |= SETTING_ENVIRONMENT;
1008 break;
1009 }
1010
1011 case 'q':
1012 arg_quiet = true;
1013 break;
1014
1015 case ARG_SHARE_SYSTEM:
1016 /* We don't officially support this anymore, except for compat reasons. People should use the
1017 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1018 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1019 arg_clone_ns_flags = 0;
1020 break;
1021
1022 case ARG_REGISTER:
1023 r = parse_boolean(optarg);
1024 if (r < 0) {
1025 log_error("Failed to parse --register= argument: %s", optarg);
1026 return r;
1027 }
1028
1029 arg_register = r;
1030 break;
1031
1032 case ARG_KEEP_UNIT:
1033 arg_keep_unit = true;
1034 break;
1035
1036 case ARG_PERSONALITY:
1037
1038 arg_personality = personality_from_string(optarg);
1039 if (arg_personality == PERSONALITY_INVALID)
1040 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1041 "Unknown or unsupported personality '%s'.", optarg);
1042
1043 arg_settings_mask |= SETTING_PERSONALITY;
1044 break;
1045
1046 case ARG_VOLATILE:
1047
1048 if (!optarg)
1049 arg_volatile_mode = VOLATILE_YES;
1050 else if (streq(optarg, "help")) {
1051 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1052 return 0;
1053 } else {
1054 VolatileMode m;
1055
1056 m = volatile_mode_from_string(optarg);
1057 if (m < 0)
1058 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1059 "Failed to parse --volatile= argument: %s", optarg);
1060 else
1061 arg_volatile_mode = m;
1062 }
1063
1064 arg_settings_mask |= SETTING_VOLATILE_MODE;
1065 break;
1066
1067 case 'p':
1068 r = expose_port_parse(&arg_expose_ports, optarg);
1069 if (r == -EEXIST)
1070 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1071 if (r < 0)
1072 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1073
1074 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1075 break;
1076
1077 case ARG_PROPERTY:
1078 if (strv_extend(&arg_property, optarg) < 0)
1079 return log_oom();
1080
1081 break;
1082
1083 case ARG_PRIVATE_USERS: {
1084 int boolean = -1;
1085
1086 if (!optarg)
1087 boolean = true;
1088 else if (!in_charset(optarg, DIGITS))
1089 /* do *not* parse numbers as booleans */
1090 boolean = parse_boolean(optarg);
1091
1092 if (boolean == false) {
1093 /* no: User namespacing off */
1094 arg_userns_mode = USER_NAMESPACE_NO;
1095 arg_uid_shift = UID_INVALID;
1096 arg_uid_range = UINT32_C(0x10000);
1097 } else if (boolean == true) {
1098 /* yes: User namespacing on, UID range is read from root dir */
1099 arg_userns_mode = USER_NAMESPACE_FIXED;
1100 arg_uid_shift = UID_INVALID;
1101 arg_uid_range = UINT32_C(0x10000);
1102 } else if (streq(optarg, "pick")) {
1103 /* pick: User namespacing on, UID range is picked randomly */
1104 arg_userns_mode = USER_NAMESPACE_PICK;
1105 arg_uid_shift = UID_INVALID;
1106 arg_uid_range = UINT32_C(0x10000);
1107 } else {
1108 _cleanup_free_ char *buffer = NULL;
1109 const char *range, *shift;
1110
1111 /* anything else: User namespacing on, UID range is explicitly configured */
1112
1113 range = strchr(optarg, ':');
1114 if (range) {
1115 buffer = strndup(optarg, range - optarg);
1116 if (!buffer)
1117 return log_oom();
1118 shift = buffer;
1119
1120 range++;
1121 r = safe_atou32(range, &arg_uid_range);
1122 if (r < 0)
1123 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
1124 } else
1125 shift = optarg;
1126
1127 r = parse_uid(shift, &arg_uid_shift);
1128 if (r < 0)
1129 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
1130
1131 arg_userns_mode = USER_NAMESPACE_FIXED;
1132 }
1133
1134 if (arg_uid_range <= 0)
1135 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1136 "UID range cannot be 0.");
1137
1138 arg_settings_mask |= SETTING_USERNS;
1139 break;
1140 }
1141
1142 case 'U':
1143 if (userns_supported()) {
1144 arg_userns_mode = USER_NAMESPACE_PICK;
1145 arg_uid_shift = UID_INVALID;
1146 arg_uid_range = UINT32_C(0x10000);
1147
1148 arg_settings_mask |= SETTING_USERNS;
1149 }
1150
1151 break;
1152
1153 case ARG_PRIVATE_USERS_CHOWN:
1154 arg_userns_chown = true;
1155
1156 arg_settings_mask |= SETTING_USERNS;
1157 break;
1158
1159 case ARG_KILL_SIGNAL:
1160 if (streq(optarg, "help")) {
1161 DUMP_STRING_TABLE(signal, int, _NSIG);
1162 return 0;
1163 }
1164
1165 arg_kill_signal = signal_from_string(optarg);
1166 if (arg_kill_signal < 0)
1167 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1168 "Cannot parse signal: %s", optarg);
1169
1170 arg_settings_mask |= SETTING_KILL_SIGNAL;
1171 break;
1172
1173 case ARG_SETTINGS:
1174
1175 /* no → do not read files
1176 * yes → read files, do not override cmdline, trust only subset
1177 * override → read files, override cmdline, trust only subset
1178 * trusted → read files, do not override cmdline, trust all
1179 */
1180
1181 r = parse_boolean(optarg);
1182 if (r < 0) {
1183 if (streq(optarg, "trusted")) {
1184 mask_all_settings = false;
1185 mask_no_settings = false;
1186 arg_settings_trusted = true;
1187
1188 } else if (streq(optarg, "override")) {
1189 mask_all_settings = false;
1190 mask_no_settings = true;
1191 arg_settings_trusted = -1;
1192 } else
1193 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1194 } else if (r > 0) {
1195 /* yes */
1196 mask_all_settings = false;
1197 mask_no_settings = false;
1198 arg_settings_trusted = -1;
1199 } else {
1200 /* no */
1201 mask_all_settings = true;
1202 mask_no_settings = false;
1203 arg_settings_trusted = false;
1204 }
1205
1206 break;
1207
1208 case ARG_CHDIR:
1209 if (!path_is_absolute(optarg))
1210 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1211 "Working directory %s is not an absolute path.", optarg);
1212
1213 r = free_and_strdup(&arg_chdir, optarg);
1214 if (r < 0)
1215 return log_oom();
1216
1217 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1218 break;
1219
1220 case ARG_PIVOT_ROOT:
1221 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1222 if (r < 0)
1223 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1224
1225 arg_settings_mask |= SETTING_PIVOT_ROOT;
1226 break;
1227
1228 case ARG_NOTIFY_READY:
1229 r = parse_boolean(optarg);
1230 if (r < 0)
1231 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1232 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1233 arg_notify_ready = r;
1234 arg_settings_mask |= SETTING_NOTIFY_READY;
1235 break;
1236
1237 case ARG_ROOT_HASH: {
1238 void *k;
1239 size_t l;
1240
1241 r = unhexmem(optarg, strlen(optarg), &k, &l);
1242 if (r < 0)
1243 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1244 if (l < sizeof(sd_id128_t)) {
1245 log_error("Root hash must be at least 128bit long: %s", optarg);
1246 free(k);
1247 return -EINVAL;
1248 }
1249
1250 free(arg_root_hash);
1251 arg_root_hash = k;
1252 arg_root_hash_size = l;
1253 break;
1254 }
1255
1256 case ARG_SYSTEM_CALL_FILTER: {
1257 bool negative;
1258 const char *items;
1259
1260 negative = optarg[0] == '~';
1261 items = negative ? optarg + 1 : optarg;
1262
1263 for (;;) {
1264 _cleanup_free_ char *word = NULL;
1265
1266 r = extract_first_word(&items, &word, NULL, 0);
1267 if (r == 0)
1268 break;
1269 if (r == -ENOMEM)
1270 return log_oom();
1271 if (r < 0)
1272 return log_error_errno(r, "Failed to parse system call filter: %m");
1273
1274 if (negative)
1275 r = strv_extend(&arg_syscall_blacklist, word);
1276 else
1277 r = strv_extend(&arg_syscall_whitelist, word);
1278 if (r < 0)
1279 return log_oom();
1280 }
1281
1282 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1283 break;
1284 }
1285
1286 case ARG_RLIMIT: {
1287 const char *eq;
1288 char *name;
1289 int rl;
1290
1291 if (streq(optarg, "help")) {
1292 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1293 return 0;
1294 }
1295
1296 eq = strchr(optarg, '=');
1297 if (!eq)
1298 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1299 "--rlimit= expects an '=' assignment.");
1300
1301 name = strndup(optarg, eq - optarg);
1302 if (!name)
1303 return log_oom();
1304
1305 rl = rlimit_from_string_harder(name);
1306 if (rl < 0)
1307 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1308 "Unknown resource limit: %s", name);
1309
1310 if (!arg_rlimit[rl]) {
1311 arg_rlimit[rl] = new0(struct rlimit, 1);
1312 if (!arg_rlimit[rl])
1313 return log_oom();
1314 }
1315
1316 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1317 if (r < 0)
1318 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1319
1320 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1321 break;
1322 }
1323
1324 case ARG_OOM_SCORE_ADJUST:
1325 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1326 if (r < 0)
1327 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1328
1329 arg_oom_score_adjust_set = true;
1330 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1331 break;
1332
1333 case ARG_CPU_AFFINITY: {
1334 _cleanup_cpu_free_ cpu_set_t *cpuset = NULL;
1335
1336 r = parse_cpu_set(optarg, &cpuset);
1337 if (r < 0)
1338 return log_error_errno(r, "Failed to parse CPU affinity mask: %s", optarg);
1339
1340 if (arg_cpuset)
1341 CPU_FREE(arg_cpuset);
1342
1343 arg_cpuset = TAKE_PTR(cpuset);
1344 arg_cpuset_ncpus = r;
1345 arg_settings_mask |= SETTING_CPU_AFFINITY;
1346 break;
1347 }
1348
1349 case ARG_RESOLV_CONF:
1350 if (streq(optarg, "help")) {
1351 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1352 return 0;
1353 }
1354
1355 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1356 if (arg_resolv_conf < 0)
1357 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1358 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1359
1360 arg_settings_mask |= SETTING_RESOLV_CONF;
1361 break;
1362
1363 case ARG_TIMEZONE:
1364 if (streq(optarg, "help")) {
1365 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1366 return 0;
1367 }
1368
1369 arg_timezone = timezone_mode_from_string(optarg);
1370 if (arg_timezone < 0)
1371 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1372 "Failed to parse /etc/localtime mode: %s", optarg);
1373
1374 arg_settings_mask |= SETTING_TIMEZONE;
1375 break;
1376
1377 case ARG_CONSOLE:
1378 if (streq(optarg, "interactive"))
1379 arg_console_mode = CONSOLE_INTERACTIVE;
1380 else if (streq(optarg, "read-only"))
1381 arg_console_mode = CONSOLE_READ_ONLY;
1382 else if (streq(optarg, "passive"))
1383 arg_console_mode = CONSOLE_PASSIVE;
1384 else if (streq(optarg, "pipe"))
1385 arg_console_mode = CONSOLE_PIPE;
1386 else if (streq(optarg, "help"))
1387 puts("interactive\n"
1388 "read-only\n"
1389 "passive\n"
1390 "pipe");
1391 else {
1392 log_error("Unknown console mode: %s", optarg);
1393 return -EINVAL;
1394 }
1395
1396 arg_settings_mask |= SETTING_CONSOLE_MODE;
1397 break;
1398
1399 case 'P':
1400 case ARG_PIPE:
1401 arg_console_mode = CONSOLE_PIPE;
1402 arg_settings_mask |= SETTING_CONSOLE_MODE;
1403 break;
1404
1405 case ARG_NO_PAGER:
1406 arg_pager_flags |= PAGER_DISABLE;
1407 break;
1408
1409 case '?':
1410 return -EINVAL;
1411
1412 default:
1413 assert_not_reached("Unhandled option");
1414 }
1415
1416 if (argc > optind) {
1417 strv_free(arg_parameters);
1418 arg_parameters = strv_copy(argv + optind);
1419 if (!arg_parameters)
1420 return log_oom();
1421
1422 arg_settings_mask |= SETTING_START_MODE;
1423 }
1424
1425 if (arg_ephemeral && arg_template && !arg_directory)
1426 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1427 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1428 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1429 * --directory=". */
1430 arg_directory = TAKE_PTR(arg_template);
1431
1432 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0)) & ~minus;
1433
1434 /* Make sure to parse environment before we reset the settings mask below */
1435 parse_environment();
1436
1437 /* Load all settings from .nspawn files */
1438 if (mask_no_settings)
1439 arg_settings_mask = 0;
1440
1441 /* Don't load any settings from .nspawn files */
1442 if (mask_all_settings)
1443 arg_settings_mask = _SETTINGS_MASK_ALL;
1444
1445 return 1;
1446 }
1447
1448 static int verify_arguments(void) {
1449 int r;
1450
1451 if (arg_userns_mode != USER_NAMESPACE_NO)
1452 arg_mount_settings |= MOUNT_USE_USERNS;
1453
1454 if (arg_private_network)
1455 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1456
1457 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1458 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1459 arg_register = false;
1460 if (arg_start_mode != START_PID1)
1461 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1462 }
1463
1464 if (arg_userns_mode == USER_NAMESPACE_PICK)
1465 arg_userns_chown = true;
1466
1467 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1468 arg_kill_signal = SIGRTMIN+3;
1469
1470 if (arg_volatile_mode != VOLATILE_NO) /* Make sure all file systems contained in the image are mounted read-only if we are in volatile mode */
1471 arg_read_only = true;
1472
1473 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1474 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1475 * The latter is not technically a user session, but we don't need to labour the point. */
1476 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1477
1478 if (arg_directory && arg_image)
1479 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1480
1481 if (arg_template && arg_image)
1482 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1483
1484 if (arg_template && !(arg_directory || arg_machine))
1485 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1486
1487 if (arg_ephemeral && arg_template)
1488 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1489
1490 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
1491 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
1492
1493 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1494 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1495
1496 if (arg_userns_chown && arg_read_only)
1497 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1498 "--read-only and --private-users-chown may not be combined.");
1499
1500 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1501 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1502 * copy-up (in case of overlay) making the entire excercise pointless. */
1503 if (arg_userns_chown && arg_volatile_mode != VOLATILE_NO)
1504 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-chown may not be combined.");
1505
1506 /* If --network-namespace-path is given with any other network-related option, we need to error out,
1507 * to avoid conflicts between different network options. */
1508 if (arg_network_namespace_path &&
1509 (arg_network_interfaces || arg_network_macvlan ||
1510 arg_network_ipvlan || arg_network_veth_extra ||
1511 arg_network_bridge || arg_network_zone ||
1512 arg_network_veth || arg_private_network))
1513 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1514
1515 if (arg_network_bridge && arg_network_zone)
1516 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1517 "--network-bridge= and --network-zone= may not be combined.");
1518
1519 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1520 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1521
1522 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1523 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1524
1525 if (arg_expose_ports && !arg_private_network)
1526 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1527
1528 #if ! HAVE_LIBIPTC
1529 if (arg_expose_ports)
1530 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--port= is not supported, compiled without libiptc support.");
1531 #endif
1532
1533 r = custom_mount_check_all();
1534 if (r < 0)
1535 return r;
1536
1537 return 0;
1538 }
1539
1540 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1541 assert(p);
1542
1543 if (arg_userns_mode == USER_NAMESPACE_NO)
1544 return 0;
1545
1546 if (uid == UID_INVALID && gid == GID_INVALID)
1547 return 0;
1548
1549 if (uid != UID_INVALID) {
1550 uid += arg_uid_shift;
1551
1552 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1553 return -EOVERFLOW;
1554 }
1555
1556 if (gid != GID_INVALID) {
1557 gid += (gid_t) arg_uid_shift;
1558
1559 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1560 return -EOVERFLOW;
1561 }
1562
1563 if (lchown(p, uid, gid) < 0)
1564 return -errno;
1565
1566 return 0;
1567 }
1568
1569 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1570 const char *q;
1571 int r;
1572
1573 q = prefix_roota(root, path);
1574 r = mkdir_errno_wrapper(q, mode);
1575 if (r == -EEXIST)
1576 return 0;
1577 if (r < 0)
1578 return r;
1579
1580 return userns_lchown(q, uid, gid);
1581 }
1582
1583 static const char *timezone_from_path(const char *path) {
1584 return PATH_STARTSWITH_SET(
1585 path,
1586 "../usr/share/zoneinfo/",
1587 "/usr/share/zoneinfo/");
1588 }
1589
1590 static bool etc_writable(void) {
1591 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1592 }
1593
1594 static int setup_timezone(const char *dest) {
1595 _cleanup_free_ char *p = NULL, *etc = NULL;
1596 const char *where, *check;
1597 TimezoneMode m;
1598 int r;
1599
1600 assert(dest);
1601
1602 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1603 r = readlink_malloc("/etc/localtime", &p);
1604 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1605 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1606 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1607 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1608 else if (r < 0) {
1609 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1610 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1611 * file.
1612 *
1613 * Example:
1614 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1615 */
1616 return 0;
1617 } else if (arg_timezone == TIMEZONE_AUTO)
1618 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1619 else
1620 m = arg_timezone;
1621 } else
1622 m = arg_timezone;
1623
1624 if (m == TIMEZONE_OFF)
1625 return 0;
1626
1627 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1628 if (r < 0) {
1629 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1630 return 0;
1631 }
1632
1633 where = strjoina(etc, "/localtime");
1634
1635 switch (m) {
1636
1637 case TIMEZONE_DELETE:
1638 if (unlink(where) < 0)
1639 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1640
1641 return 0;
1642
1643 case TIMEZONE_SYMLINK: {
1644 _cleanup_free_ char *q = NULL;
1645 const char *z, *what;
1646
1647 z = timezone_from_path(p);
1648 if (!z) {
1649 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1650 return 0;
1651 }
1652
1653 r = readlink_malloc(where, &q);
1654 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1655 return 0; /* Already pointing to the right place? Then do nothing .. */
1656
1657 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1658 r = chase_symlinks(check, dest, 0, NULL);
1659 if (r < 0)
1660 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1661 else {
1662 if (unlink(where) < 0 && errno != ENOENT) {
1663 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1664 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1665 return 0;
1666 }
1667
1668 what = strjoina("../usr/share/zoneinfo/", z);
1669 if (symlink(what, where) < 0) {
1670 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1671 errno, "Failed to correct timezone of container, ignoring: %m");
1672 return 0;
1673 }
1674
1675 break;
1676 }
1677
1678 _fallthrough_;
1679 }
1680
1681 case TIMEZONE_BIND: {
1682 _cleanup_free_ char *resolved = NULL;
1683 int found;
1684
1685 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1686 if (found < 0) {
1687 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1688 return 0;
1689 }
1690
1691 if (found == 0) /* missing? */
1692 (void) touch(resolved);
1693
1694 r = mount_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1695 if (r >= 0)
1696 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1697
1698 _fallthrough_;
1699 }
1700
1701 case TIMEZONE_COPY:
1702 /* If mounting failed, try to copy */
1703 r = copy_file_atomic("/etc/localtime", where, 0644, 0, COPY_REFLINK|COPY_REPLACE);
1704 if (r < 0) {
1705 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1706 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
1707 return 0;
1708 }
1709
1710 break;
1711
1712 default:
1713 assert_not_reached("unexpected mode");
1714 }
1715
1716 /* Fix permissions of the symlink or file copy we just created */
1717 r = userns_lchown(where, 0, 0);
1718 if (r < 0)
1719 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
1720
1721 return 0;
1722 }
1723
1724 static int have_resolv_conf(const char *path) {
1725 assert(path);
1726
1727 if (access(path, F_OK) < 0) {
1728 if (errno == ENOENT)
1729 return 0;
1730
1731 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1732 }
1733
1734 return 1;
1735 }
1736
1737 static int resolved_listening(void) {
1738 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1739 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1740 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1741 int r;
1742
1743 /* Check if resolved is listening */
1744
1745 r = sd_bus_open_system(&bus);
1746 if (r < 0)
1747 return log_debug_errno(r, "Failed to open system bus: %m");
1748
1749 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1750 if (r < 0)
1751 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1752 if (r == 0)
1753 return 0;
1754
1755 r = sd_bus_get_property_string(bus,
1756 "org.freedesktop.resolve1",
1757 "/org/freedesktop/resolve1",
1758 "org.freedesktop.resolve1.Manager",
1759 "DNSStubListener",
1760 &error,
1761 &dns_stub_listener_mode);
1762 if (r < 0)
1763 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
1764
1765 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1766 }
1767
1768 static int setup_resolv_conf(const char *dest) {
1769 _cleanup_free_ char *etc = NULL;
1770 const char *where, *what;
1771 ResolvConfMode m;
1772 int r;
1773
1774 assert(dest);
1775
1776 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
1777 if (arg_private_network)
1778 m = RESOLV_CONF_OFF;
1779 else if (have_resolv_conf(STATIC_RESOLV_CONF) > 0 && resolved_listening() > 0)
1780 m = etc_writable() ? RESOLV_CONF_COPY_STATIC : RESOLV_CONF_BIND_STATIC;
1781 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1782 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
1783 else
1784 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
1785 } else
1786 m = arg_resolv_conf;
1787
1788 if (m == RESOLV_CONF_OFF)
1789 return 0;
1790
1791 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1792 if (r < 0) {
1793 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1794 return 0;
1795 }
1796
1797 where = strjoina(etc, "/resolv.conf");
1798
1799 if (m == RESOLV_CONF_DELETE) {
1800 if (unlink(where) < 0)
1801 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1802
1803 return 0;
1804 }
1805
1806 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_COPY_STATIC))
1807 what = STATIC_RESOLV_CONF;
1808 else
1809 what = "/etc/resolv.conf";
1810
1811 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC)) {
1812 _cleanup_free_ char *resolved = NULL;
1813 int found;
1814
1815 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1816 if (found < 0) {
1817 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1818 return 0;
1819 }
1820
1821 if (found == 0) /* missing? */
1822 (void) touch(resolved);
1823
1824 r = mount_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
1825 if (r >= 0)
1826 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1827 }
1828
1829 /* If that didn't work, let's copy the file */
1830 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, COPY_REFLINK);
1831 if (r < 0) {
1832 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1833 * resolved or something similar runs inside and the symlink points there.
1834 *
1835 * If the disk image is read-only, there's also no point in complaining.
1836 */
1837 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,
1838 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1839 return 0;
1840 }
1841
1842 r = userns_lchown(where, 0, 0);
1843 if (r < 0)
1844 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1845
1846 return 0;
1847 }
1848
1849 static int setup_boot_id(void) {
1850 _cleanup_(unlink_and_freep) char *from = NULL;
1851 _cleanup_free_ char *path = NULL;
1852 sd_id128_t rnd = SD_ID128_NULL;
1853 const char *to;
1854 int r;
1855
1856 /* Generate a new randomized boot ID, so that each boot-up of
1857 * the container gets a new one */
1858
1859 r = tempfn_random_child(NULL, "proc-sys-kernel-random-boot-id", &path);
1860 if (r < 0)
1861 return log_error_errno(r, "Failed to generate random boot ID path: %m");
1862
1863 r = sd_id128_randomize(&rnd);
1864 if (r < 0)
1865 return log_error_errno(r, "Failed to generate random boot id: %m");
1866
1867 r = id128_write(path, ID128_UUID, rnd, false);
1868 if (r < 0)
1869 return log_error_errno(r, "Failed to write boot id: %m");
1870
1871 from = TAKE_PTR(path);
1872 to = "/proc/sys/kernel/random/boot_id";
1873
1874 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1875 if (r < 0)
1876 return r;
1877
1878 return mount_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1879 }
1880
1881 static int copy_devnodes(const char *dest) {
1882 static const char devnodes[] =
1883 "null\0"
1884 "zero\0"
1885 "full\0"
1886 "random\0"
1887 "urandom\0"
1888 "tty\0"
1889 "net/tun\0";
1890
1891 _cleanup_umask_ mode_t u;
1892 const char *d;
1893 int r = 0;
1894
1895 assert(dest);
1896
1897 u = umask(0000);
1898
1899 /* Create /dev/net, so that we can create /dev/net/tun in it */
1900 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1901 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1902
1903 NULSTR_FOREACH(d, devnodes) {
1904 _cleanup_free_ char *from = NULL, *to = NULL;
1905 struct stat st;
1906
1907 from = strappend("/dev/", d);
1908 if (!from)
1909 return log_oom();
1910
1911 to = prefix_root(dest, from);
1912 if (!to)
1913 return log_oom();
1914
1915 if (stat(from, &st) < 0) {
1916
1917 if (errno != ENOENT)
1918 return log_error_errno(errno, "Failed to stat %s: %m", from);
1919
1920 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
1921 return log_error_errno(SYNTHETIC_ERRNO(EIO),
1922 "%s is not a char or block device, cannot copy.", from);
1923 else {
1924 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
1925
1926 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1927 /* Explicitly warn the user when /dev is already populated. */
1928 if (errno == EEXIST)
1929 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);
1930 if (errno != EPERM)
1931 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1932
1933 /* Some systems abusively restrict mknod but allow bind mounts. */
1934 r = touch(to);
1935 if (r < 0)
1936 return log_error_errno(r, "touch (%s) failed: %m", to);
1937 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1938 if (r < 0)
1939 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1940 }
1941
1942 r = userns_lchown(to, 0, 0);
1943 if (r < 0)
1944 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1945
1946 dn = strjoin("/dev/", S_ISCHR(st.st_mode) ? "char" : "block");
1947 if (!dn)
1948 return log_oom();
1949
1950 r = userns_mkdir(dest, dn, 0755, 0, 0);
1951 if (r < 0)
1952 return log_error_errno(r, "Failed to create '%s': %m", dn);
1953
1954 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
1955 return log_oom();
1956
1957 prefixed = prefix_root(dest, sl);
1958 if (!prefixed)
1959 return log_oom();
1960
1961 t = strjoin("../", d);
1962 if (!t)
1963 return log_oom();
1964
1965 if (symlink(t, prefixed) < 0)
1966 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
1967 }
1968 }
1969
1970 return r;
1971 }
1972
1973 static int make_extra_nodes(const char *dest) {
1974 _cleanup_umask_ mode_t u;
1975 size_t i;
1976 int r;
1977
1978 u = umask(0000);
1979
1980 for (i = 0; i < arg_n_extra_nodes; i++) {
1981 _cleanup_free_ char *path = NULL;
1982 DeviceNode *n = arg_extra_nodes + i;
1983
1984 path = prefix_root(dest, n->path);
1985 if (!path)
1986 return log_oom();
1987
1988 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
1989 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
1990
1991 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
1992 if (r < 0)
1993 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
1994 }
1995
1996 return 0;
1997 }
1998
1999 static int setup_pts(const char *dest) {
2000 _cleanup_free_ char *options = NULL;
2001 const char *p;
2002 int r;
2003
2004 #if HAVE_SELINUX
2005 if (arg_selinux_apifs_context)
2006 (void) asprintf(&options,
2007 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
2008 arg_uid_shift + TTY_GID,
2009 arg_selinux_apifs_context);
2010 else
2011 #endif
2012 (void) asprintf(&options,
2013 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
2014 arg_uid_shift + TTY_GID);
2015
2016 if (!options)
2017 return log_oom();
2018
2019 /* Mount /dev/pts itself */
2020 p = prefix_roota(dest, "/dev/pts");
2021 r = mkdir_errno_wrapper(p, 0755);
2022 if (r < 0)
2023 return log_error_errno(r, "Failed to create /dev/pts: %m");
2024
2025 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2026 if (r < 0)
2027 return r;
2028 r = userns_lchown(p, 0, 0);
2029 if (r < 0)
2030 return log_error_errno(r, "Failed to chown /dev/pts: %m");
2031
2032 /* Create /dev/ptmx symlink */
2033 p = prefix_roota(dest, "/dev/ptmx");
2034 if (symlink("pts/ptmx", p) < 0)
2035 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
2036 r = userns_lchown(p, 0, 0);
2037 if (r < 0)
2038 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
2039
2040 /* And fix /dev/pts/ptmx ownership */
2041 p = prefix_roota(dest, "/dev/pts/ptmx");
2042 r = userns_lchown(p, 0, 0);
2043 if (r < 0)
2044 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
2045
2046 return 0;
2047 }
2048
2049 static int setup_dev_console(const char *dest, const char *console) {
2050 _cleanup_umask_ mode_t u;
2051 const char *to;
2052 int r;
2053
2054 assert(dest);
2055
2056 u = umask(0000);
2057
2058 if (!console)
2059 return 0;
2060
2061 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
2062 if (r < 0)
2063 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
2064
2065 /* We need to bind mount the right tty to /dev/console since
2066 * ptys can only exist on pts file systems. To have something
2067 * to bind mount things on we create a empty regular file. */
2068
2069 to = prefix_roota(dest, "/dev/console");
2070 r = touch(to);
2071 if (r < 0)
2072 return log_error_errno(r, "touch() for /dev/console failed: %m");
2073
2074 return mount_verbose(LOG_ERR, console, to, NULL, MS_BIND, NULL);
2075 }
2076
2077 static int setup_keyring(void) {
2078 key_serial_t keyring;
2079
2080 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2081 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2082 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2083 * these system calls let's make sure we don't leak anything into the container. */
2084
2085 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2086 if (keyring == -1) {
2087 if (errno == ENOSYS)
2088 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2089 else if (IN_SET(errno, EACCES, EPERM))
2090 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2091 else
2092 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2093 }
2094
2095 return 0;
2096 }
2097
2098 static int setup_kmsg(int kmsg_socket) {
2099 _cleanup_(unlink_and_freep) char *from = NULL;
2100 _cleanup_free_ char *fifo = NULL;
2101 _cleanup_close_ int fd = -1;
2102 _cleanup_umask_ mode_t u;
2103 const char *to;
2104 int r;
2105
2106 assert(kmsg_socket >= 0);
2107
2108 u = umask(0000);
2109
2110 /* We create the kmsg FIFO as as temporary file in /tmp, but immediately delete it after bind mounting it to
2111 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2112 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2113 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2114
2115 r = tempfn_random_child(NULL, "proc-kmsg", &fifo);
2116 if (r < 0)
2117 return log_error_errno(r, "Failed to generate kmsg path: %m");
2118
2119 if (mkfifo(fifo, 0600) < 0)
2120 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2121
2122 from = TAKE_PTR(fifo);
2123 to = "/proc/kmsg";
2124
2125 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
2126 if (r < 0)
2127 return r;
2128
2129 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2130 if (fd < 0)
2131 return log_error_errno(errno, "Failed to open fifo: %m");
2132
2133 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2134 r = send_one_fd(kmsg_socket, fd, 0);
2135 if (r < 0)
2136 return log_error_errno(r, "Failed to send FIFO fd: %m");
2137
2138 return 0;
2139 }
2140
2141 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2142 union in_addr_union *exposed = userdata;
2143
2144 assert(rtnl);
2145 assert(m);
2146 assert(exposed);
2147
2148 expose_port_execute(rtnl, arg_expose_ports, exposed);
2149 return 0;
2150 }
2151
2152 static int setup_hostname(void) {
2153 int r;
2154
2155 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2156 return 0;
2157
2158 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2159 if (r < 0)
2160 return log_error_errno(r, "Failed to set hostname: %m");
2161
2162 return 0;
2163 }
2164
2165 static int setup_journal(const char *directory) {
2166 _cleanup_free_ char *d = NULL;
2167 const char *dirname, *p, *q;
2168 sd_id128_t this_id;
2169 char id[33];
2170 bool try;
2171 int r;
2172
2173 /* Don't link journals in ephemeral mode */
2174 if (arg_ephemeral)
2175 return 0;
2176
2177 if (arg_link_journal == LINK_NO)
2178 return 0;
2179
2180 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2181
2182 r = sd_id128_get_machine(&this_id);
2183 if (r < 0)
2184 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2185
2186 if (sd_id128_equal(arg_uuid, this_id)) {
2187 log_full(try ? LOG_WARNING : LOG_ERR,
2188 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
2189 if (try)
2190 return 0;
2191 return -EEXIST;
2192 }
2193
2194 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2195 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2196 if (r < 0) {
2197 bool ignore = r == -EROFS && try;
2198 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2199 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2200 return ignore ? 0 : r;
2201 }
2202 }
2203
2204 (void) sd_id128_to_string(arg_uuid, id);
2205
2206 p = strjoina("/var/log/journal/", id);
2207 q = prefix_roota(directory, p);
2208
2209 if (path_is_mount_point(p, NULL, 0) > 0) {
2210 if (try)
2211 return 0;
2212
2213 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2214 "%s: already a mount point, refusing to use for journal", p);
2215 }
2216
2217 if (path_is_mount_point(q, NULL, 0) > 0) {
2218 if (try)
2219 return 0;
2220
2221 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2222 "%s: already a mount point, refusing to use for journal", q);
2223 }
2224
2225 r = readlink_and_make_absolute(p, &d);
2226 if (r >= 0) {
2227 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2228 path_equal(d, q)) {
2229
2230 r = userns_mkdir(directory, p, 0755, 0, 0);
2231 if (r < 0)
2232 log_warning_errno(r, "Failed to create directory %s: %m", q);
2233 return 0;
2234 }
2235
2236 if (unlink(p) < 0)
2237 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2238 } else if (r == -EINVAL) {
2239
2240 if (arg_link_journal == LINK_GUEST &&
2241 rmdir(p) < 0) {
2242
2243 if (errno == ENOTDIR) {
2244 log_error("%s already exists and is neither a symlink nor a directory", p);
2245 return r;
2246 } else
2247 return log_error_errno(errno, "Failed to remove %s: %m", p);
2248 }
2249 } else if (r != -ENOENT)
2250 return log_error_errno(r, "readlink(%s) failed: %m", p);
2251
2252 if (arg_link_journal == LINK_GUEST) {
2253
2254 if (symlink(q, p) < 0) {
2255 if (try) {
2256 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2257 return 0;
2258 } else
2259 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2260 }
2261
2262 r = userns_mkdir(directory, p, 0755, 0, 0);
2263 if (r < 0)
2264 log_warning_errno(r, "Failed to create directory %s: %m", q);
2265 return 0;
2266 }
2267
2268 if (arg_link_journal == LINK_HOST) {
2269 /* don't create parents here — if the host doesn't have
2270 * permanent journal set up, don't force it here */
2271
2272 r = mkdir_errno_wrapper(p, 0755);
2273 if (r < 0 && r != -EEXIST) {
2274 if (try) {
2275 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2276 return 0;
2277 } else
2278 return log_error_errno(r, "Failed to create %s: %m", p);
2279 }
2280
2281 } else if (access(p, F_OK) < 0)
2282 return 0;
2283
2284 if (dir_is_empty(q) == 0)
2285 log_warning("%s is not empty, proceeding anyway.", q);
2286
2287 r = userns_mkdir(directory, p, 0755, 0, 0);
2288 if (r < 0)
2289 return log_error_errno(r, "Failed to create %s: %m", q);
2290
2291 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2292 if (r < 0)
2293 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2294
2295 return 0;
2296 }
2297
2298 static int drop_capabilities(uid_t uid) {
2299 CapabilityQuintet q;
2300
2301 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2302 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2303 * arg_caps_retain. */
2304
2305 if (capability_quintet_is_set(&arg_full_capabilities)) {
2306 q = arg_full_capabilities;
2307
2308 if (q.bounding == (uint64_t) -1)
2309 q.bounding = uid == 0 ? arg_caps_retain : 0;
2310
2311 if (q.effective == (uint64_t) -1)
2312 q.effective = uid == 0 ? q.bounding : 0;
2313
2314 if (q.inheritable == (uint64_t) -1)
2315 q.inheritable = uid == 0 ? q.bounding : 0;
2316
2317 if (q.permitted == (uint64_t) -1)
2318 q.permitted = uid == 0 ? q.bounding : 0;
2319
2320 if (q.ambient == (uint64_t) -1 && ambient_capabilities_supported())
2321 q.ambient = 0;
2322 } else
2323 q = (CapabilityQuintet) {
2324 .bounding = arg_caps_retain,
2325 .effective = uid == 0 ? arg_caps_retain : 0,
2326 .inheritable = uid == 0 ? arg_caps_retain : 0,
2327 .permitted = uid == 0 ? arg_caps_retain : 0,
2328 .ambient = ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2329 };
2330
2331 return capability_quintet_enforce(&q);
2332 }
2333
2334 static int reset_audit_loginuid(void) {
2335 _cleanup_free_ char *p = NULL;
2336 int r;
2337
2338 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2339 return 0;
2340
2341 r = read_one_line_file("/proc/self/loginuid", &p);
2342 if (r == -ENOENT)
2343 return 0;
2344 if (r < 0)
2345 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2346
2347 /* Already reset? */
2348 if (streq(p, "4294967295"))
2349 return 0;
2350
2351 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2352 if (r < 0) {
2353 log_error_errno(r,
2354 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2355 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2356 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2357 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2358 "using systemd-nspawn. Sleeping for 5s... (%m)");
2359
2360 sleep(5);
2361 }
2362
2363 return 0;
2364 }
2365
2366 static int setup_propagate(const char *root) {
2367 const char *p, *q;
2368 int r;
2369
2370 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2371 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2372 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2373 (void) mkdir_p(p, 0600);
2374
2375 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
2376 if (r < 0)
2377 return log_error_errno(r, "Failed to create /run/systemd: %m");
2378
2379 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
2380 if (r < 0)
2381 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
2382
2383 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2384 if (r < 0)
2385 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
2386
2387 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
2388 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2389 if (r < 0)
2390 return r;
2391
2392 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2393 if (r < 0)
2394 return r;
2395
2396 /* machined will MS_MOVE into that directory, and that's only
2397 * supported for non-shared mounts. */
2398 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2399 }
2400
2401 static int setup_machine_id(const char *directory) {
2402 const char *etc_machine_id;
2403 sd_id128_t id;
2404 int r;
2405
2406 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2407 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2408 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2409 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2410 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2411 * container behaves nicely). */
2412
2413 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2414
2415 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
2416 if (r < 0) {
2417 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2418 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2419
2420 if (sd_id128_is_null(arg_uuid)) {
2421 r = sd_id128_randomize(&arg_uuid);
2422 if (r < 0)
2423 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2424 }
2425 } else {
2426 if (sd_id128_is_null(id))
2427 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2428 "Machine ID in container image is zero, refusing.");
2429
2430 arg_uuid = id;
2431 }
2432
2433 return 0;
2434 }
2435
2436 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2437 int r;
2438
2439 assert(directory);
2440
2441 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
2442 return 0;
2443
2444 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2445 if (r == -EOPNOTSUPP)
2446 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.");
2447 if (r == -EBADE)
2448 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2449 if (r < 0)
2450 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2451 if (r == 0)
2452 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2453 else
2454 log_debug("Patched directory tree to match UID/GID range.");
2455
2456 return r;
2457 }
2458
2459 /*
2460 * Return values:
2461 * < 0 : wait_for_terminate() failed to get the state of the
2462 * container, the container was terminated by a signal, or
2463 * failed for an unknown reason. No change is made to the
2464 * container argument.
2465 * > 0 : The program executed in the container terminated with an
2466 * error. The exit code of the program executed in the
2467 * container is returned. The container argument has been set
2468 * to CONTAINER_TERMINATED.
2469 * 0 : The container is being rebooted, has been shut down or exited
2470 * successfully. The container argument has been set to either
2471 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2472 *
2473 * That is, success is indicated by a return value of zero, and an
2474 * error is indicated by a non-zero value.
2475 */
2476 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2477 siginfo_t status;
2478 int r;
2479
2480 r = wait_for_terminate(pid, &status);
2481 if (r < 0)
2482 return log_warning_errno(r, "Failed to wait for container: %m");
2483
2484 switch (status.si_code) {
2485
2486 case CLD_EXITED:
2487 if (status.si_status == 0)
2488 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2489 else
2490 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2491
2492 *container = CONTAINER_TERMINATED;
2493 return status.si_status;
2494
2495 case CLD_KILLED:
2496 if (status.si_status == SIGINT) {
2497 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2498 *container = CONTAINER_TERMINATED;
2499 return 0;
2500
2501 } else if (status.si_status == SIGHUP) {
2502 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2503 *container = CONTAINER_REBOOTED;
2504 return 0;
2505 }
2506
2507 _fallthrough_;
2508 case CLD_DUMPED:
2509 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2510 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2511
2512 default:
2513 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2514 "Container %s failed due to unknown reason.", arg_machine);
2515 }
2516 }
2517
2518 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2519 pid_t pid;
2520
2521 pid = PTR_TO_PID(userdata);
2522 if (pid > 0) {
2523 if (kill(pid, arg_kill_signal) >= 0) {
2524 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2525 sd_event_source_set_userdata(s, NULL);
2526 return 0;
2527 }
2528 }
2529
2530 sd_event_exit(sd_event_source_get_event(s), 0);
2531 return 0;
2532 }
2533
2534 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2535 pid_t pid;
2536
2537 assert(s);
2538 assert(ssi);
2539
2540 pid = PTR_TO_PID(userdata);
2541
2542 for (;;) {
2543 siginfo_t si = {};
2544
2545 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2546 return log_error_errno(errno, "Failed to waitid(): %m");
2547 if (si.si_pid == 0) /* No pending children. */
2548 break;
2549 if (si.si_pid == pid) {
2550 /* The main process we care for has exited. Return from
2551 * signal handler but leave the zombie. */
2552 sd_event_exit(sd_event_source_get_event(s), 0);
2553 break;
2554 }
2555
2556 /* Reap all other children. */
2557 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2558 }
2559
2560 return 0;
2561 }
2562
2563 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2564 pid_t pid;
2565
2566 assert(m);
2567
2568 pid = PTR_TO_PID(userdata);
2569
2570 if (arg_kill_signal > 0) {
2571 log_info("Container termination requested. Attempting to halt container.");
2572 (void) kill(pid, arg_kill_signal);
2573 } else {
2574 log_info("Container termination requested. Exiting.");
2575 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2576 }
2577
2578 return 0;
2579 }
2580
2581 static int determine_names(void) {
2582 int r;
2583
2584 if (arg_template && !arg_directory && arg_machine) {
2585
2586 /* If --template= was specified then we should not
2587 * search for a machine, but instead create a new one
2588 * in /var/lib/machine. */
2589
2590 arg_directory = strjoin("/var/lib/machines/", arg_machine);
2591 if (!arg_directory)
2592 return log_oom();
2593 }
2594
2595 if (!arg_image && !arg_directory) {
2596 if (arg_machine) {
2597 _cleanup_(image_unrefp) Image *i = NULL;
2598
2599 r = image_find(IMAGE_MACHINE, arg_machine, &i);
2600 if (r == -ENOENT)
2601 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
2602 if (r < 0)
2603 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2604
2605 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2606 r = free_and_strdup(&arg_image, i->path);
2607 else
2608 r = free_and_strdup(&arg_directory, i->path);
2609 if (r < 0)
2610 return log_oom();
2611
2612 if (!arg_ephemeral)
2613 arg_read_only = arg_read_only || i->read_only;
2614 } else {
2615 r = safe_getcwd(&arg_directory);
2616 if (r < 0)
2617 return log_error_errno(r, "Failed to determine current directory: %m");
2618 }
2619
2620 if (!arg_directory && !arg_image) {
2621 log_error("Failed to determine path, please use -D or -i.");
2622 return -EINVAL;
2623 }
2624 }
2625
2626 if (!arg_machine) {
2627 if (arg_directory && path_equal(arg_directory, "/"))
2628 arg_machine = gethostname_malloc();
2629 else {
2630 if (arg_image) {
2631 char *e;
2632
2633 arg_machine = strdup(basename(arg_image));
2634
2635 /* Truncate suffix if there is one */
2636 e = endswith(arg_machine, ".raw");
2637 if (e)
2638 *e = 0;
2639 } else
2640 arg_machine = strdup(basename(arg_directory));
2641 }
2642 if (!arg_machine)
2643 return log_oom();
2644
2645 hostname_cleanup(arg_machine);
2646 if (!machine_name_is_valid(arg_machine)) {
2647 log_error("Failed to determine machine name automatically, please use -M.");
2648 return -EINVAL;
2649 }
2650
2651 if (arg_ephemeral) {
2652 char *b;
2653
2654 /* Add a random suffix when this is an
2655 * ephemeral machine, so that we can run many
2656 * instances at once without manually having
2657 * to specify -M each time. */
2658
2659 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2660 return log_oom();
2661
2662 free(arg_machine);
2663 arg_machine = b;
2664 }
2665 }
2666
2667 return 0;
2668 }
2669
2670 static int chase_symlinks_and_update(char **p, unsigned flags) {
2671 char *chased;
2672 int r;
2673
2674 assert(p);
2675
2676 if (!*p)
2677 return 0;
2678
2679 r = chase_symlinks(*p, NULL, flags, &chased);
2680 if (r < 0)
2681 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2682
2683 free_and_replace(*p, chased);
2684 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2685 }
2686
2687 static int determine_uid_shift(const char *directory) {
2688 int r;
2689
2690 if (arg_userns_mode == USER_NAMESPACE_NO) {
2691 arg_uid_shift = 0;
2692 return 0;
2693 }
2694
2695 if (arg_uid_shift == UID_INVALID) {
2696 struct stat st;
2697
2698 r = stat(directory, &st);
2699 if (r < 0)
2700 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2701
2702 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2703
2704 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
2705 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2706 "UID and GID base of %s don't match.", directory);
2707
2708 arg_uid_range = UINT32_C(0x10000);
2709 }
2710
2711 if (arg_uid_shift > (uid_t) -1 - arg_uid_range)
2712 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2713 "UID base too high for UID range.");
2714
2715 return 0;
2716 }
2717
2718 static unsigned long effective_clone_ns_flags(void) {
2719 unsigned long flags = arg_clone_ns_flags;
2720
2721 if (arg_private_network)
2722 flags |= CLONE_NEWNET;
2723 if (arg_use_cgns)
2724 flags |= CLONE_NEWCGROUP;
2725 if (arg_userns_mode != USER_NAMESPACE_NO)
2726 flags |= CLONE_NEWUSER;
2727
2728 return flags;
2729 }
2730
2731 static int patch_sysctl(void) {
2732
2733 /* This table is inspired by runc's sysctl() function */
2734 static const struct {
2735 const char *key;
2736 bool prefix;
2737 unsigned long clone_flags;
2738 } safe_sysctl[] = {
2739 { "kernel.hostname", false, CLONE_NEWUTS },
2740 { "kernel.domainname", false, CLONE_NEWUTS },
2741 { "kernel.msgmax", false, CLONE_NEWIPC },
2742 { "kernel.msgmnb", false, CLONE_NEWIPC },
2743 { "kernel.msgmni", false, CLONE_NEWIPC },
2744 { "kernel.sem", false, CLONE_NEWIPC },
2745 { "kernel.shmall", false, CLONE_NEWIPC },
2746 { "kernel.shmmax", false, CLONE_NEWIPC },
2747 { "kernel.shmmni", false, CLONE_NEWIPC },
2748 { "fs.mqueue.", true, CLONE_NEWIPC },
2749 { "net.", true, CLONE_NEWNET },
2750 };
2751
2752 unsigned long flags;
2753 char **k, **v;
2754 int r;
2755
2756 flags = effective_clone_ns_flags();
2757
2758 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
2759 bool good = false;
2760 size_t i;
2761
2762 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
2763
2764 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
2765 continue;
2766
2767 if (safe_sysctl[i].prefix)
2768 good = startswith(*k, safe_sysctl[i].key);
2769 else
2770 good = streq(*k, safe_sysctl[i].key);
2771
2772 if (good)
2773 break;
2774 }
2775
2776 if (!good) {
2777 log_error("Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
2778 return -EPERM;
2779 }
2780
2781 r = sysctl_write(*k, *v);
2782 if (r < 0)
2783 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
2784 }
2785
2786 return 0;
2787 }
2788
2789 static int inner_child(
2790 Barrier *barrier,
2791 const char *directory,
2792 bool secondary,
2793 int kmsg_socket,
2794 int rtnl_socket,
2795 FDSet *fds) {
2796
2797 _cleanup_free_ char *home = NULL;
2798 char as_uuid[37];
2799 size_t n_env = 1;
2800 const char *envp[] = {
2801 "PATH=" DEFAULT_PATH_COMPAT,
2802 NULL, /* container */
2803 NULL, /* TERM */
2804 NULL, /* HOME */
2805 NULL, /* USER */
2806 NULL, /* LOGNAME */
2807 NULL, /* container_uuid */
2808 NULL, /* LISTEN_FDS */
2809 NULL, /* LISTEN_PID */
2810 NULL, /* NOTIFY_SOCKET */
2811 NULL
2812 };
2813 const char *exec_target;
2814 _cleanup_strv_free_ char **env_use = NULL;
2815 int r, which_failed;
2816
2817 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2818 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2819 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2820 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2821 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2822 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2823 * namespace.
2824 *
2825 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2826 * unshare(). See below. */
2827
2828 assert(barrier);
2829 assert(directory);
2830 assert(kmsg_socket >= 0);
2831
2832 log_debug("Inner child is initializing.");
2833
2834 if (arg_userns_mode != USER_NAMESPACE_NO) {
2835 /* Tell the parent, that it now can write the UID map. */
2836 (void) barrier_place(barrier); /* #1 */
2837
2838 /* Wait until the parent wrote the UID map */
2839 if (!barrier_place_and_sync(barrier)) /* #2 */
2840 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2841 "Parent died too early");
2842 }
2843
2844 r = reset_uid_gid();
2845 if (r < 0)
2846 return log_error_errno(r, "Couldn't become new root: %m");
2847
2848 r = mount_all(NULL,
2849 arg_mount_settings | MOUNT_IN_USERNS,
2850 arg_uid_shift,
2851 arg_selinux_apifs_context);
2852 if (r < 0)
2853 return r;
2854
2855 if (!arg_network_namespace_path && arg_private_network) {
2856 r = unshare(CLONE_NEWNET);
2857 if (r < 0)
2858 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2859
2860 /* Tell the parent that it can setup network interfaces. */
2861 (void) barrier_place(barrier); /* #3 */
2862 }
2863
2864 r = mount_sysfs(NULL, arg_mount_settings);
2865 if (r < 0)
2866 return r;
2867
2868 /* Wait until we are cgroup-ified, so that we
2869 * can mount the right cgroup path writable */
2870 if (!barrier_place_and_sync(barrier)) /* #4 */
2871 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2872 "Parent died too early");
2873
2874 if (arg_use_cgns) {
2875 r = unshare(CLONE_NEWCGROUP);
2876 if (r < 0)
2877 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2878 r = mount_cgroups(
2879 "",
2880 arg_unified_cgroup_hierarchy,
2881 arg_userns_mode != USER_NAMESPACE_NO,
2882 arg_uid_shift,
2883 arg_uid_range,
2884 arg_selinux_apifs_context,
2885 true);
2886 if (r < 0)
2887 return r;
2888 } else {
2889 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2890 if (r < 0)
2891 return r;
2892 }
2893
2894 r = setup_boot_id();
2895 if (r < 0)
2896 return r;
2897
2898 r = setup_kmsg(kmsg_socket);
2899 if (r < 0)
2900 return r;
2901 kmsg_socket = safe_close(kmsg_socket);
2902
2903 r = mount_custom(
2904 "/",
2905 arg_custom_mounts,
2906 arg_n_custom_mounts,
2907 false,
2908 0,
2909 0,
2910 arg_selinux_apifs_context,
2911 true);
2912 if (r < 0)
2913 return r;
2914
2915 if (setsid() < 0)
2916 return log_error_errno(errno, "setsid() failed: %m");
2917
2918 if (arg_private_network)
2919 loopback_setup();
2920
2921 if (arg_expose_ports) {
2922 r = expose_port_send_rtnl(rtnl_socket);
2923 if (r < 0)
2924 return r;
2925 rtnl_socket = safe_close(rtnl_socket);
2926 }
2927
2928 r = patch_sysctl();
2929 if (r < 0)
2930 return r;
2931
2932 if (arg_oom_score_adjust_set) {
2933 r = set_oom_score_adjust(arg_oom_score_adjust);
2934 if (r < 0)
2935 return log_error_errno(r, "Failed to adjust OOM score: %m");
2936 }
2937
2938 if (arg_cpuset)
2939 if (sched_setaffinity(0, CPU_ALLOC_SIZE(arg_cpuset_ncpus), arg_cpuset) < 0)
2940 return log_error_errno(errno, "Failed to set CPU affinity: %m");
2941
2942 (void) setup_hostname();
2943
2944 if (arg_personality != PERSONALITY_INVALID) {
2945 r = safe_personality(arg_personality);
2946 if (r < 0)
2947 return log_error_errno(r, "personality() failed: %m");
2948 } else if (secondary) {
2949 r = safe_personality(PER_LINUX32);
2950 if (r < 0)
2951 return log_error_errno(r, "personality() failed: %m");
2952 }
2953
2954 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
2955 if (r < 0)
2956 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
2957
2958 #if HAVE_SECCOMP
2959 if (arg_seccomp) {
2960
2961 if (is_seccomp_available()) {
2962
2963 r = seccomp_load(arg_seccomp);
2964 if (IN_SET(r, -EPERM, -EACCES))
2965 return log_error_errno(r, "Failed to install seccomp filter: %m");
2966 if (r < 0)
2967 log_debug_errno(r, "Failed to install seccomp filter: %m");
2968 }
2969 } else
2970 #endif
2971 {
2972 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
2973 if (r < 0)
2974 return r;
2975 }
2976
2977 #if HAVE_SELINUX
2978 if (arg_selinux_context)
2979 if (setexeccon(arg_selinux_context) < 0)
2980 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
2981 #endif
2982
2983 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
2984 * if we need to later on. */
2985 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
2986 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
2987
2988 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
2989 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids);
2990 else
2991 r = change_uid_gid(arg_user, &home);
2992 if (r < 0)
2993 return r;
2994
2995 r = drop_capabilities(getuid());
2996 if (r < 0)
2997 return log_error_errno(r, "Dropping capabilities failed: %m");
2998
2999 if (arg_no_new_privileges)
3000 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3001 return log_error_errno(errno, "Failed to disable new privileges: %m");
3002
3003 /* LXC sets container=lxc, so follow the scheme here */
3004 envp[n_env++] = strjoina("container=", arg_container_service_name);
3005
3006 envp[n_env] = strv_find_prefix(environ, "TERM=");
3007 if (envp[n_env])
3008 n_env++;
3009
3010 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3011 if (asprintf((char**)(envp + n_env++), "HOME=%s", home ?: "/root") < 0)
3012 return log_oom();
3013
3014 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3015 if (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ?: "root") < 0 ||
3016 asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3017 return log_oom();
3018
3019 assert(!sd_id128_is_null(arg_uuid));
3020
3021 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
3022 return log_oom();
3023
3024 if (fdset_size(fds) > 0) {
3025 r = fdset_cloexec(fds, false);
3026 if (r < 0)
3027 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3028
3029 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3030 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3031 return log_oom();
3032 }
3033 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3034 return log_oom();
3035
3036 env_use = strv_env_merge(2, envp, arg_setenv);
3037 if (!env_use)
3038 return log_oom();
3039
3040 /* Let the parent know that we are ready and
3041 * wait until the parent is ready with the
3042 * setup, too... */
3043 if (!barrier_place_and_sync(barrier)) /* #5 */
3044 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3045 "Parent died too early");
3046
3047 if (arg_chdir)
3048 if (chdir(arg_chdir) < 0)
3049 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3050
3051 if (arg_start_mode == START_PID2) {
3052 r = stub_pid1(arg_uuid);
3053 if (r < 0)
3054 return r;
3055 }
3056
3057 log_debug("Inner child completed, invoking payload.");
3058
3059 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3060 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3061 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3062 log_close();
3063 log_set_open_when_needed(true);
3064
3065 (void) fdset_close_others(fds);
3066
3067 if (arg_start_mode == START_BOOT) {
3068 char **a;
3069 size_t m;
3070
3071 /* Automatically search for the init system */
3072
3073 m = strv_length(arg_parameters);
3074 a = newa(char*, m + 2);
3075 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3076 a[1 + m] = NULL;
3077
3078 a[0] = (char*) "/usr/lib/systemd/systemd";
3079 execve(a[0], a, env_use);
3080
3081 a[0] = (char*) "/lib/systemd/systemd";
3082 execve(a[0], a, env_use);
3083
3084 a[0] = (char*) "/sbin/init";
3085 execve(a[0], a, env_use);
3086
3087 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3088 } else if (!strv_isempty(arg_parameters)) {
3089 const char *dollar_path;
3090
3091 exec_target = arg_parameters[0];
3092
3093 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3094 * binary. */
3095 dollar_path = strv_env_get(env_use, "PATH");
3096 if (dollar_path) {
3097 if (putenv((char*) dollar_path) != 0)
3098 return log_error_errno(errno, "Failed to update $PATH: %m");
3099 }
3100
3101 execvpe(arg_parameters[0], arg_parameters, env_use);
3102 } else {
3103 if (!arg_chdir)
3104 /* If we cannot change the directory, we'll end up in /, that is expected. */
3105 (void) chdir(home ?: "/root");
3106
3107 execle("/bin/bash", "-bash", NULL, env_use);
3108 execle("/bin/sh", "-sh", NULL, env_use);
3109
3110 exec_target = "/bin/bash, /bin/sh";
3111 }
3112
3113 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3114 }
3115
3116 static int setup_sd_notify_child(void) {
3117 _cleanup_close_ int fd = -1;
3118 union sockaddr_union sa = {
3119 .un.sun_family = AF_UNIX,
3120 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3121 };
3122 int r;
3123
3124 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3125 if (fd < 0)
3126 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3127
3128 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3129 (void) sockaddr_un_unlink(&sa.un);
3130
3131 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3132 if (r < 0)
3133 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3134
3135 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3136 if (r < 0)
3137 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3138
3139 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3140 if (r < 0)
3141 return log_error_errno(r, "SO_PASSCRED failed: %m");
3142
3143 return TAKE_FD(fd);
3144 }
3145
3146 static int outer_child(
3147 Barrier *barrier,
3148 const char *directory,
3149 const char *console,
3150 DissectedImage *dissected_image,
3151 bool secondary,
3152 int pid_socket,
3153 int uuid_socket,
3154 int notify_socket,
3155 int kmsg_socket,
3156 int rtnl_socket,
3157 int uid_shift_socket,
3158 int unified_cgroup_hierarchy_socket,
3159 FDSet *fds,
3160 int netns_fd) {
3161
3162 _cleanup_close_ int fd = -1;
3163 pid_t pid;
3164 ssize_t l;
3165 int r;
3166
3167 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3168 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3169 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3170 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3171
3172 assert(barrier);
3173 assert(directory);
3174 assert(pid_socket >= 0);
3175 assert(uuid_socket >= 0);
3176 assert(notify_socket >= 0);
3177 assert(kmsg_socket >= 0);
3178
3179 log_debug("Outer child is initializing.");
3180
3181 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3182 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3183
3184 if (arg_console_mode != CONSOLE_PIPE) {
3185 int terminal;
3186
3187 assert(console);
3188
3189 terminal = open_terminal(console, O_RDWR);
3190 if (terminal < 0)
3191 return log_error_errno(terminal, "Failed to open console: %m");
3192
3193 /* Make sure we can continue logging to the original stderr, even if stderr points elsewhere now */
3194 r = log_dup_console();
3195 if (r < 0)
3196 return log_error_errno(r, "Failed to duplicate stderr: %m");
3197
3198 r = rearrange_stdio(terminal, terminal, terminal); /* invalidates 'terminal' on success and failure */
3199 if (r < 0)
3200 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
3201 }
3202
3203 r = reset_audit_loginuid();
3204 if (r < 0)
3205 return r;
3206
3207 /* Mark everything as slave, so that we still
3208 * receive mounts from the real root, but don't
3209 * propagate mounts to the real root. */
3210 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3211 if (r < 0)
3212 return r;
3213
3214 if (dissected_image) {
3215 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3216 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3217 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3218 * makes sure ESP partitions and userns are compatible. */
3219
3220 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3221 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
3222 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0)|
3223 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3224 if (r < 0)
3225 return r;
3226 }
3227
3228 r = determine_uid_shift(directory);
3229 if (r < 0)
3230 return r;
3231
3232 if (arg_userns_mode != USER_NAMESPACE_NO) {
3233 /* Let the parent know which UID shift we read from the image */
3234 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3235 if (l < 0)
3236 return log_error_errno(errno, "Failed to send UID shift: %m");
3237 if (l != sizeof(arg_uid_shift))
3238 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3239 "Short write while sending UID shift.");
3240
3241 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3242 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3243 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3244 * not it will pick a different one, and send it back to us. */
3245
3246 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3247 if (l < 0)
3248 return log_error_errno(errno, "Failed to recv UID shift: %m");
3249 if (l != sizeof(arg_uid_shift))
3250 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3251 "Short read while receiving UID shift.");
3252 }
3253
3254 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3255 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3256 }
3257
3258 if (!dissected_image) {
3259 /* Turn directory into bind mount */
3260 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
3261 if (r < 0)
3262 return r;
3263 }
3264
3265 r = setup_pivot_root(
3266 directory,
3267 arg_pivot_root_new,
3268 arg_pivot_root_old);
3269 if (r < 0)
3270 return r;
3271
3272 r = setup_volatile_mode(
3273 directory,
3274 arg_volatile_mode,
3275 arg_userns_mode != USER_NAMESPACE_NO,
3276 arg_uid_shift,
3277 arg_uid_range,
3278 arg_selinux_context);
3279 if (r < 0)
3280 return r;
3281
3282 if (dissected_image) {
3283 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3284 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3285 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
3286 if (r < 0)
3287 return r;
3288 }
3289
3290 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3291 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3292
3293 r = detect_unified_cgroup_hierarchy_from_image(directory);
3294 if (r < 0)
3295 return r;
3296
3297 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3298 if (l < 0)
3299 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3300 if (l != sizeof(arg_unified_cgroup_hierarchy))
3301 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3302 "Short write while sending cgroup mode.");
3303
3304 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3305 }
3306
3307 /* Mark everything as shared so our mounts get propagated down. This is
3308 * required to make new bind mounts available in systemd services
3309 * inside the containter that create a new mount namespace.
3310 * See https://github.com/systemd/systemd/issues/3860
3311 * Further submounts (such as /dev) done after this will inherit the
3312 * shared propagation mode. */
3313 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3314 if (r < 0)
3315 return r;
3316
3317 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3318 if (r < 0)
3319 return r;
3320
3321 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3322 if (r < 0)
3323 return r;
3324
3325 if (arg_read_only && arg_volatile_mode == VOLATILE_NO) {
3326 r = bind_remount_recursive(directory, true, NULL);
3327 if (r < 0)
3328 return log_error_errno(r, "Failed to make tree read-only: %m");
3329 }
3330
3331 r = mount_all(directory,
3332 arg_mount_settings,
3333 arg_uid_shift,
3334 arg_selinux_apifs_context);
3335 if (r < 0)
3336 return r;
3337
3338 r = copy_devnodes(directory);
3339 if (r < 0)
3340 return r;
3341
3342 r = make_extra_nodes(directory);
3343 if (r < 0)
3344 return r;
3345
3346 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3347 (void) make_inaccessible_nodes(directory, arg_uid_shift, arg_uid_shift);
3348
3349 r = setup_pts(directory);
3350 if (r < 0)
3351 return r;
3352
3353 r = setup_propagate(directory);
3354 if (r < 0)
3355 return r;
3356
3357 r = setup_dev_console(directory, console);
3358 if (r < 0)
3359 return r;
3360
3361 r = setup_keyring();
3362 if (r < 0)
3363 return r;
3364
3365 r = setup_timezone(directory);
3366 if (r < 0)
3367 return r;
3368
3369 r = setup_resolv_conf(directory);
3370 if (r < 0)
3371 return r;
3372
3373 r = setup_machine_id(directory);
3374 if (r < 0)
3375 return r;
3376
3377 r = setup_journal(directory);
3378 if (r < 0)
3379 return r;
3380
3381 r = mount_custom(
3382 directory,
3383 arg_custom_mounts,
3384 arg_n_custom_mounts,
3385 arg_userns_mode != USER_NAMESPACE_NO,
3386 arg_uid_shift,
3387 arg_uid_range,
3388 arg_selinux_apifs_context,
3389 false);
3390 if (r < 0)
3391 return r;
3392
3393 if (!arg_use_cgns) {
3394 r = mount_cgroups(
3395 directory,
3396 arg_unified_cgroup_hierarchy,
3397 arg_userns_mode != USER_NAMESPACE_NO,
3398 arg_uid_shift,
3399 arg_uid_range,
3400 arg_selinux_apifs_context,
3401 false);
3402 if (r < 0)
3403 return r;
3404 }
3405
3406 r = mount_move_root(directory);
3407 if (r < 0)
3408 return log_error_errno(r, "Failed to move root directory: %m");
3409
3410 fd = setup_sd_notify_child();
3411 if (fd < 0)
3412 return fd;
3413
3414 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3415 arg_clone_ns_flags |
3416 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3417 if (pid < 0)
3418 return log_error_errno(errno, "Failed to fork inner child: %m");
3419 if (pid == 0) {
3420 pid_socket = safe_close(pid_socket);
3421 uuid_socket = safe_close(uuid_socket);
3422 notify_socket = safe_close(notify_socket);
3423 uid_shift_socket = safe_close(uid_shift_socket);
3424
3425 /* The inner child has all namespaces that are
3426 * requested, so that we all are owned by the user if
3427 * user namespaces are turned on. */
3428
3429 if (arg_network_namespace_path) {
3430 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3431 if (r < 0)
3432 return log_error_errno(r, "Failed to join network namespace: %m");
3433 }
3434
3435 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
3436 if (r < 0)
3437 _exit(EXIT_FAILURE);
3438
3439 _exit(EXIT_SUCCESS);
3440 }
3441
3442 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3443 if (l < 0)
3444 return log_error_errno(errno, "Failed to send PID: %m");
3445 if (l != sizeof(pid))
3446 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3447 "Short write while sending PID.");
3448
3449 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3450 if (l < 0)
3451 return log_error_errno(errno, "Failed to send machine ID: %m");
3452 if (l != sizeof(arg_uuid))
3453 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3454 "Short write while sending machine ID.");
3455
3456 l = send_one_fd(notify_socket, fd, 0);
3457 if (l < 0)
3458 return log_error_errno(errno, "Failed to send notify fd: %m");
3459
3460 pid_socket = safe_close(pid_socket);
3461 uuid_socket = safe_close(uuid_socket);
3462 notify_socket = safe_close(notify_socket);
3463 kmsg_socket = safe_close(kmsg_socket);
3464 rtnl_socket = safe_close(rtnl_socket);
3465 netns_fd = safe_close(netns_fd);
3466
3467 return 0;
3468 }
3469
3470 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
3471 bool tried_hashed = false;
3472 unsigned n_tries = 100;
3473 uid_t candidate;
3474 int r;
3475
3476 assert(shift);
3477 assert(ret_lock_file);
3478 assert(arg_userns_mode == USER_NAMESPACE_PICK);
3479 assert(arg_uid_range == 0x10000U);
3480
3481 candidate = *shift;
3482
3483 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3484
3485 for (;;) {
3486 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
3487 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
3488
3489 if (--n_tries <= 0)
3490 return -EBUSY;
3491
3492 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
3493 goto next;
3494 if ((candidate & UINT32_C(0xFFFF)) != 0)
3495 goto next;
3496
3497 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3498 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3499 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3500 goto next;
3501 if (r < 0)
3502 return r;
3503
3504 /* Make some superficial checks whether the range is currently known in the user database */
3505 if (getpwuid(candidate))
3506 goto next;
3507 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3508 goto next;
3509 if (getgrgid(candidate))
3510 goto next;
3511 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3512 goto next;
3513
3514 *ret_lock_file = lf;
3515 lf = (struct LockFile) LOCK_FILE_INIT;
3516 *shift = candidate;
3517 return 0;
3518
3519 next:
3520 if (arg_machine && !tried_hashed) {
3521 /* Try to hash the base from the container name */
3522
3523 static const uint8_t hash_key[] = {
3524 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3525 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3526 };
3527
3528 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3529
3530 tried_hashed = true;
3531 } else
3532 random_bytes(&candidate, sizeof(candidate));
3533
3534 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
3535 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3536 }
3537 }
3538
3539 static int setup_uid_map(pid_t pid) {
3540 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3541 int r;
3542
3543 assert(pid > 1);
3544
3545 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3546 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3547 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3548 if (r < 0)
3549 return log_error_errno(r, "Failed to write UID map: %m");
3550
3551 /* We always assign the same UID and GID ranges */
3552 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3553 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3554 if (r < 0)
3555 return log_error_errno(r, "Failed to write GID map: %m");
3556
3557 return 0;
3558 }
3559
3560 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3561 char buf[NOTIFY_BUFFER_MAX+1];
3562 char *p = NULL;
3563 struct iovec iovec = {
3564 .iov_base = buf,
3565 .iov_len = sizeof(buf)-1,
3566 };
3567 union {
3568 struct cmsghdr cmsghdr;
3569 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
3570 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
3571 } control = {};
3572 struct msghdr msghdr = {
3573 .msg_iov = &iovec,
3574 .msg_iovlen = 1,
3575 .msg_control = &control,
3576 .msg_controllen = sizeof(control),
3577 };
3578 struct cmsghdr *cmsg;
3579 struct ucred *ucred = NULL;
3580 ssize_t n;
3581 pid_t inner_child_pid;
3582 _cleanup_strv_free_ char **tags = NULL;
3583
3584 assert(userdata);
3585
3586 inner_child_pid = PTR_TO_PID(userdata);
3587
3588 if (revents != EPOLLIN) {
3589 log_warning("Got unexpected poll event for notify fd.");
3590 return 0;
3591 }
3592
3593 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3594 if (n < 0) {
3595 if (IN_SET(errno, EAGAIN, EINTR))
3596 return 0;
3597
3598 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3599 }
3600 cmsg_close_all(&msghdr);
3601
3602 CMSG_FOREACH(cmsg, &msghdr) {
3603 if (cmsg->cmsg_level == SOL_SOCKET &&
3604 cmsg->cmsg_type == SCM_CREDENTIALS &&
3605 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3606
3607 ucred = (struct ucred*) CMSG_DATA(cmsg);
3608 }
3609 }
3610
3611 if (!ucred || ucred->pid != inner_child_pid) {
3612 log_debug("Received notify message without valid credentials. Ignoring.");
3613 return 0;
3614 }
3615
3616 if ((size_t) n >= sizeof(buf)) {
3617 log_warning("Received notify message exceeded maximum size. Ignoring.");
3618 return 0;
3619 }
3620
3621 buf[n] = 0;
3622 tags = strv_split(buf, "\n\r");
3623 if (!tags)
3624 return log_oom();
3625
3626 if (strv_find(tags, "READY=1"))
3627 (void) sd_notifyf(false, "READY=1\n");
3628
3629 p = strv_find_startswith(tags, "STATUS=");
3630 if (p)
3631 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
3632
3633 return 0;
3634 }
3635
3636 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3637 int r;
3638
3639 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3640 if (r < 0)
3641 return log_error_errno(r, "Failed to allocate notify event source: %m");
3642
3643 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3644
3645 return 0;
3646 }
3647
3648 static int merge_settings(Settings *settings, const char *path) {
3649 int rl;
3650
3651 assert(settings);
3652 assert(path);
3653
3654 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3655 * that this steals the fields of the Settings* structure, and hence modifies it. */
3656
3657 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3658 settings->start_mode >= 0) {
3659 arg_start_mode = settings->start_mode;
3660 strv_free_and_replace(arg_parameters, settings->parameters);
3661 }
3662
3663 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
3664 arg_ephemeral = settings->ephemeral;
3665
3666 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
3667 settings->root) {
3668
3669 if (!arg_settings_trusted)
3670 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
3671 else
3672 free_and_replace(arg_directory, settings->root);
3673 }
3674
3675 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3676 settings->pivot_root_new) {
3677 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3678 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3679 }
3680
3681 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3682 settings->working_directory)
3683 free_and_replace(arg_chdir, settings->working_directory);
3684
3685 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3686 settings->environment)
3687 strv_free_and_replace(arg_setenv, settings->environment);
3688
3689 if ((arg_settings_mask & SETTING_USER) == 0) {
3690
3691 if (settings->user)
3692 free_and_replace(arg_user, settings->user);
3693
3694 if (uid_is_valid(settings->uid))
3695 arg_uid = settings->uid;
3696 if (gid_is_valid(settings->gid))
3697 arg_gid = settings->gid;
3698 if (settings->n_supplementary_gids > 0) {
3699 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
3700 arg_n_supplementary_gids = settings->n_supplementary_gids;
3701 }
3702 }
3703
3704 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3705 uint64_t plus, minus;
3706
3707 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3708 * Settings structure */
3709
3710 plus = settings->capability;
3711 minus = settings->drop_capability;
3712
3713 if ((arg_settings_mask & SETTING_NETWORK) == 0) {
3714 if (settings_private_network(settings))
3715 plus |= UINT64_C(1) << CAP_NET_ADMIN;
3716 else
3717 minus |= UINT64_C(1) << CAP_NET_ADMIN;
3718 }
3719
3720 if (!arg_settings_trusted && plus != 0) {
3721 if (settings->capability != 0)
3722 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
3723 } else
3724 arg_caps_retain |= plus;
3725
3726 arg_caps_retain &= ~minus;
3727
3728 /* Copy the full capabilities over too */
3729 if (capability_quintet_is_set(&settings->full_capabilities)) {
3730 if (!arg_settings_trusted)
3731 log_warning("Ignoring capabilitiy settings, file %s is not trusted.", path);
3732 else
3733 arg_full_capabilities = settings->full_capabilities;
3734 }
3735 }
3736
3737 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3738 settings->kill_signal > 0)
3739 arg_kill_signal = settings->kill_signal;
3740
3741 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3742 settings->personality != PERSONALITY_INVALID)
3743 arg_personality = settings->personality;
3744
3745 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3746 !sd_id128_is_null(settings->machine_id)) {
3747
3748 if (!arg_settings_trusted)
3749 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
3750 else
3751 arg_uuid = settings->machine_id;
3752 }
3753
3754 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3755 settings->read_only >= 0)
3756 arg_read_only = settings->read_only;
3757
3758 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3759 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3760 arg_volatile_mode = settings->volatile_mode;
3761
3762 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3763 settings->n_custom_mounts > 0) {
3764
3765 if (!arg_settings_trusted)
3766 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
3767 else {
3768 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3769 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3770 arg_n_custom_mounts = settings->n_custom_mounts;
3771 settings->n_custom_mounts = 0;
3772 }
3773 }
3774
3775 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3776 (settings->private_network >= 0 ||
3777 settings->network_veth >= 0 ||
3778 settings->network_bridge ||
3779 settings->network_zone ||
3780 settings->network_interfaces ||
3781 settings->network_macvlan ||
3782 settings->network_ipvlan ||
3783 settings->network_veth_extra ||
3784 settings->network_namespace_path)) {
3785
3786 if (!arg_settings_trusted)
3787 log_warning("Ignoring network settings, file %s is not trusted.", path);
3788 else {
3789 arg_network_veth = settings_network_veth(settings);
3790 arg_private_network = settings_private_network(settings);
3791
3792 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
3793 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
3794 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
3795 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
3796
3797 free_and_replace(arg_network_bridge, settings->network_bridge);
3798 free_and_replace(arg_network_zone, settings->network_zone);
3799
3800 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
3801 }
3802 }
3803
3804 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3805 settings->expose_ports) {
3806
3807 if (!arg_settings_trusted)
3808 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
3809 else {
3810 expose_port_free_all(arg_expose_ports);
3811 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3812 }
3813 }
3814
3815 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3816 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3817
3818 if (!arg_settings_trusted)
3819 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
3820 else {
3821 arg_userns_mode = settings->userns_mode;
3822 arg_uid_shift = settings->uid_shift;
3823 arg_uid_range = settings->uid_range;
3824 arg_userns_chown = settings->userns_chown;
3825 }
3826 }
3827
3828 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3829 arg_notify_ready = settings->notify_ready;
3830
3831 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3832
3833 if (!arg_settings_trusted && !strv_isempty(settings->syscall_whitelist))
3834 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
3835 else {
3836 strv_free_and_replace(arg_syscall_whitelist, settings->syscall_whitelist);
3837 strv_free_and_replace(arg_syscall_blacklist, settings->syscall_blacklist);
3838 }
3839
3840 #if HAVE_SECCOMP
3841 if (!arg_settings_trusted && settings->seccomp)
3842 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
3843 else {
3844 seccomp_release(arg_seccomp);
3845 arg_seccomp = TAKE_PTR(settings->seccomp);
3846 }
3847 #endif
3848 }
3849
3850 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
3851 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
3852 continue;
3853
3854 if (!settings->rlimit[rl])
3855 continue;
3856
3857 if (!arg_settings_trusted) {
3858 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
3859 continue;
3860 }
3861
3862 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
3863 }
3864
3865 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
3866 settings->hostname)
3867 free_and_replace(arg_hostname, settings->hostname);
3868
3869 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
3870 settings->no_new_privileges >= 0)
3871 arg_no_new_privileges = settings->no_new_privileges;
3872
3873 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
3874 settings->oom_score_adjust_set) {
3875
3876 if (!arg_settings_trusted)
3877 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
3878 else {
3879 arg_oom_score_adjust = settings->oom_score_adjust;
3880 arg_oom_score_adjust_set = true;
3881 }
3882 }
3883
3884 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
3885 settings->cpuset) {
3886
3887 if (!arg_settings_trusted)
3888 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
3889 else {
3890 if (arg_cpuset)
3891 CPU_FREE(arg_cpuset);
3892 arg_cpuset = TAKE_PTR(settings->cpuset);
3893 arg_cpuset_ncpus = settings->cpuset_ncpus;
3894 }
3895 }
3896
3897 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
3898 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
3899 arg_resolv_conf = settings->resolv_conf;
3900
3901 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
3902 settings->link_journal != _LINK_JOURNAL_INVALID) {
3903
3904 if (!arg_settings_trusted)
3905 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
3906 else {
3907 arg_link_journal = settings->link_journal;
3908 arg_link_journal_try = settings->link_journal_try;
3909 }
3910 }
3911
3912 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
3913 settings->timezone != _TIMEZONE_MODE_INVALID)
3914 arg_timezone = settings->timezone;
3915
3916 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
3917 settings->slice) {
3918
3919 if (!arg_settings_trusted)
3920 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
3921 else
3922 free_and_replace(arg_slice, settings->slice);
3923 }
3924
3925 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
3926 settings->use_cgns >= 0) {
3927
3928 if (!arg_settings_trusted)
3929 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
3930 else
3931 arg_use_cgns = settings->use_cgns;
3932 }
3933
3934 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
3935 settings->clone_ns_flags != (unsigned long) -1) {
3936
3937 if (!arg_settings_trusted)
3938 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
3939 else
3940 arg_clone_ns_flags = settings->clone_ns_flags;
3941 }
3942
3943 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
3944 settings->console_mode >= 0) {
3945
3946 if (!arg_settings_trusted)
3947 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
3948 else
3949 arg_console_mode = settings->console_mode;
3950 }
3951
3952 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
3953 * don't consult arg_settings_mask for them. */
3954
3955 sd_bus_message_unref(arg_property_message);
3956 arg_property_message = TAKE_PTR(settings->properties);
3957
3958 arg_console_width = settings->console_width;
3959 arg_console_height = settings->console_height;
3960
3961 device_node_free_many(arg_extra_nodes, arg_n_extra_nodes);
3962 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
3963 arg_n_extra_nodes = settings->n_extra_nodes;
3964
3965 return 0;
3966 }
3967
3968 static int load_settings(void) {
3969 _cleanup_(settings_freep) Settings *settings = NULL;
3970 _cleanup_fclose_ FILE *f = NULL;
3971 _cleanup_free_ char *p = NULL;
3972 const char *fn, *i;
3973 int r;
3974
3975 if (arg_oci_bundle)
3976 return 0;
3977
3978 /* If all settings are masked, there's no point in looking for
3979 * the settings file */
3980 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3981 return 0;
3982
3983 fn = strjoina(arg_machine, ".nspawn");
3984
3985 /* We first look in the admin's directories in /etc and /run */
3986 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3987 _cleanup_free_ char *j = NULL;
3988
3989 j = strjoin(i, "/", fn);
3990 if (!j)
3991 return log_oom();
3992
3993 f = fopen(j, "re");
3994 if (f) {
3995 p = TAKE_PTR(j);
3996
3997 /* By default, we trust configuration from /etc and /run */
3998 if (arg_settings_trusted < 0)
3999 arg_settings_trusted = true;
4000
4001 break;
4002 }
4003
4004 if (errno != ENOENT)
4005 return log_error_errno(errno, "Failed to open %s: %m", j);
4006 }
4007
4008 if (!f) {
4009 /* After that, let's look for a file next to the
4010 * actual image we shall boot. */
4011
4012 if (arg_image) {
4013 p = file_in_same_dir(arg_image, fn);
4014 if (!p)
4015 return log_oom();
4016 } else if (arg_directory) {
4017 p = file_in_same_dir(arg_directory, fn);
4018 if (!p)
4019 return log_oom();
4020 }
4021
4022 if (p) {
4023 f = fopen(p, "re");
4024 if (!f && errno != ENOENT)
4025 return log_error_errno(errno, "Failed to open %s: %m", p);
4026
4027 /* By default, we do not trust configuration from /var/lib/machines */
4028 if (arg_settings_trusted < 0)
4029 arg_settings_trusted = false;
4030 }
4031 }
4032
4033 if (!f)
4034 return 0;
4035
4036 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4037
4038 r = settings_load(f, p, &settings);
4039 if (r < 0)
4040 return r;
4041
4042 return merge_settings(settings, p);
4043 }
4044
4045 static int load_oci_bundle(void) {
4046 _cleanup_(settings_freep) Settings *settings = NULL;
4047 int r;
4048
4049 if (!arg_oci_bundle)
4050 return 0;
4051
4052 /* By default let's trust OCI bundles */
4053 if (arg_settings_trusted < 0)
4054 arg_settings_trusted = true;
4055
4056 r = oci_load(NULL, arg_oci_bundle, &settings);
4057 if (r < 0)
4058 return r;
4059
4060 return merge_settings(settings, arg_oci_bundle);
4061 }
4062
4063 static int run_container(int master,
4064 const char* console,
4065 DissectedImage *dissected_image,
4066 bool secondary,
4067 FDSet *fds,
4068 char veth_name[IFNAMSIZ], bool *veth_created,
4069 union in_addr_union *exposed,
4070 pid_t *pid, int *ret) {
4071
4072 static const struct sigaction sa = {
4073 .sa_handler = nop_signal_handler,
4074 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4075 };
4076
4077 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4078 _cleanup_close_ int etc_passwd_lock = -1;
4079 _cleanup_close_pair_ int
4080 kmsg_socket_pair[2] = { -1, -1 },
4081 rtnl_socket_pair[2] = { -1, -1 },
4082 pid_socket_pair[2] = { -1, -1 },
4083 uuid_socket_pair[2] = { -1, -1 },
4084 notify_socket_pair[2] = { -1, -1 },
4085 uid_shift_socket_pair[2] = { -1, -1 },
4086 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4087
4088 _cleanup_close_ int notify_socket= -1;
4089 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4090 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4091 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4092 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4093 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4094 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4095 ContainerStatus container_status = 0;
4096 int ifi = 0, r;
4097 ssize_t l;
4098 sigset_t mask_chld;
4099 _cleanup_close_ int netns_fd = -1;
4100
4101 assert_se(sigemptyset(&mask_chld) == 0);
4102 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4103
4104 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4105 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4106 * check with getpwuid() if the specific user already exists. Note that /etc might be
4107 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4108 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4109 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4110 * really ours. */
4111
4112 etc_passwd_lock = take_etc_passwd_lock(NULL);
4113 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4114 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4115 }
4116
4117 r = barrier_create(&barrier);
4118 if (r < 0)
4119 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4120
4121 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4122 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4123
4124 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4125 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4126
4127 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4128 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4129
4130 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4131 return log_error_errno(errno, "Failed to create id socket pair: %m");
4132
4133 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4134 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4135
4136 if (arg_userns_mode != USER_NAMESPACE_NO)
4137 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4138 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4139
4140 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4141 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4142 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4143
4144 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4145 * parent's blocking calls and give it a chance to call wait() and terminate. */
4146 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4147 if (r < 0)
4148 return log_error_errno(errno, "Failed to change the signal mask: %m");
4149
4150 r = sigaction(SIGCHLD, &sa, NULL);
4151 if (r < 0)
4152 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4153
4154 if (arg_network_namespace_path) {
4155 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4156 if (netns_fd < 0)
4157 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4158
4159 r = fd_is_network_ns(netns_fd);
4160 if (r == -EUCLEAN)
4161 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);
4162 else if (r < 0)
4163 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4164 else if (r == 0) {
4165 log_error("Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4166 return -EINVAL;
4167 }
4168 }
4169
4170 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4171 if (*pid < 0)
4172 return log_error_errno(errno, "clone() failed%s: %m",
4173 errno == EINVAL ?
4174 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4175
4176 if (*pid == 0) {
4177 /* The outer child only has a file system namespace. */
4178 barrier_set_role(&barrier, BARRIER_CHILD);
4179
4180 master = safe_close(master);
4181
4182 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4183 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4184 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4185 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4186 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
4187 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4188 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
4189
4190 (void) reset_all_signal_handlers();
4191 (void) reset_signal_mask();
4192
4193 r = outer_child(&barrier,
4194 arg_directory,
4195 console,
4196 dissected_image,
4197 secondary,
4198 pid_socket_pair[1],
4199 uuid_socket_pair[1],
4200 notify_socket_pair[1],
4201 kmsg_socket_pair[1],
4202 rtnl_socket_pair[1],
4203 uid_shift_socket_pair[1],
4204 unified_cgroup_hierarchy_socket_pair[1],
4205 fds,
4206 netns_fd);
4207 if (r < 0)
4208 _exit(EXIT_FAILURE);
4209
4210 _exit(EXIT_SUCCESS);
4211 }
4212
4213 barrier_set_role(&barrier, BARRIER_PARENT);
4214
4215 fds = fdset_free(fds);
4216
4217 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4218 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4219 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4220 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4221 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
4222 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
4223 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
4224
4225 if (arg_userns_mode != USER_NAMESPACE_NO) {
4226 /* The child just let us know the UID shift it might have read from the image. */
4227 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4228 if (l < 0)
4229 return log_error_errno(errno, "Failed to read UID shift: %m");
4230 if (l != sizeof arg_uid_shift) {
4231 log_error("Short read while reading UID shift.");
4232 return -EIO;
4233 }
4234
4235 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4236 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4237 * image, but if that's already in use, pick a new one, and report back to the child,
4238 * which one we now picked. */
4239
4240 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4241 if (r < 0)
4242 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4243
4244 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4245 if (l < 0)
4246 return log_error_errno(errno, "Failed to send UID shift: %m");
4247 if (l != sizeof arg_uid_shift) {
4248 log_error("Short write while writing UID shift.");
4249 return -EIO;
4250 }
4251 }
4252 }
4253
4254 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4255 /* The child let us know the support cgroup mode it might have read from the image. */
4256 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4257 if (l < 0)
4258 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4259 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
4260 log_error("Short read while reading cgroup mode (%zu bytes).%s",
4261 l, l == 0 ? " The child is most likely dead." : "");
4262 return -EIO;
4263 }
4264 }
4265
4266 /* Wait for the outer child. */
4267 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4268 if (r < 0)
4269 return r;
4270 if (r != EXIT_SUCCESS)
4271 return -EIO;
4272
4273 /* And now retrieve the PID of the inner child. */
4274 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4275 if (l < 0)
4276 return log_error_errno(errno, "Failed to read inner child PID: %m");
4277 if (l != sizeof *pid) {
4278 log_error("Short read while reading inner child PID.");
4279 return -EIO;
4280 }
4281
4282 /* We also retrieve container UUID in case it was generated by outer child */
4283 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4284 if (l < 0)
4285 return log_error_errno(errno, "Failed to read container machine ID: %m");
4286 if (l != sizeof(arg_uuid)) {
4287 log_error("Short read while reading container machined ID.");
4288 return -EIO;
4289 }
4290
4291 /* We also retrieve the socket used for notifications generated by outer child */
4292 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4293 if (notify_socket < 0)
4294 return log_error_errno(notify_socket,
4295 "Failed to receive notification socket from the outer child: %m");
4296
4297 log_debug("Init process invoked as PID "PID_FMT, *pid);
4298
4299 if (arg_userns_mode != USER_NAMESPACE_NO) {
4300 if (!barrier_place_and_sync(&barrier)) { /* #1 */
4301 log_error("Child died too early.");
4302 return -ESRCH;
4303 }
4304
4305 r = setup_uid_map(*pid);
4306 if (r < 0)
4307 return r;
4308
4309 (void) barrier_place(&barrier); /* #2 */
4310 }
4311
4312 if (arg_private_network) {
4313 if (!arg_network_namespace_path) {
4314 /* Wait until the child has unshared its network namespace. */
4315 if (!barrier_place_and_sync(&barrier)) { /* #3 */
4316 log_error("Child died too early");
4317 return -ESRCH;
4318 }
4319 }
4320
4321 r = move_network_interfaces(*pid, arg_network_interfaces);
4322 if (r < 0)
4323 return r;
4324
4325 if (arg_network_veth) {
4326 r = setup_veth(arg_machine, *pid, veth_name,
4327 arg_network_bridge || arg_network_zone);
4328 if (r < 0)
4329 return r;
4330 else if (r > 0)
4331 ifi = r;
4332
4333 if (arg_network_bridge) {
4334 /* Add the interface to a bridge */
4335 r = setup_bridge(veth_name, arg_network_bridge, false);
4336 if (r < 0)
4337 return r;
4338 if (r > 0)
4339 ifi = r;
4340 } else if (arg_network_zone) {
4341 /* Add the interface to a bridge, possibly creating it */
4342 r = setup_bridge(veth_name, arg_network_zone, true);
4343 if (r < 0)
4344 return r;
4345 if (r > 0)
4346 ifi = r;
4347 }
4348 }
4349
4350 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
4351 if (r < 0)
4352 return r;
4353
4354 /* We created the primary and extra veth links now; let's remember this, so that we know to
4355 remove them later on. Note that we don't bother with removing veth links that were created
4356 here when their setup failed half-way, because in that case the kernel should be able to
4357 remove them on its own, since they cannot be referenced by anything yet. */
4358 *veth_created = true;
4359
4360 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
4361 if (r < 0)
4362 return r;
4363
4364 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
4365 if (r < 0)
4366 return r;
4367 }
4368
4369 if (arg_register || !arg_keep_unit) {
4370 r = sd_bus_default_system(&bus);
4371 if (r < 0)
4372 return log_error_errno(r, "Failed to open system bus: %m");
4373
4374 r = sd_bus_set_close_on_exit(bus, false);
4375 if (r < 0)
4376 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
4377 }
4378
4379 if (!arg_keep_unit) {
4380 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4381 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4382 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4383
4384 r = sd_bus_match_signal_async(
4385 bus,
4386 NULL,
4387 "org.freedesktop.systemd1",
4388 NULL,
4389 "org.freedesktop.systemd1.Scope",
4390 "RequestStop",
4391 on_request_stop, NULL, PID_TO_PTR(*pid));
4392 if (r < 0)
4393 return log_error_errno(r, "Failed to request RequestStop match: %m");
4394 }
4395
4396 if (arg_register) {
4397 r = register_machine(
4398 bus,
4399 arg_machine,
4400 *pid,
4401 arg_directory,
4402 arg_uuid,
4403 ifi,
4404 arg_slice,
4405 arg_custom_mounts, arg_n_custom_mounts,
4406 arg_kill_signal,
4407 arg_property,
4408 arg_property_message,
4409 arg_keep_unit,
4410 arg_container_service_name);
4411 if (r < 0)
4412 return r;
4413
4414 } else if (!arg_keep_unit) {
4415 r = allocate_scope(
4416 bus,
4417 arg_machine,
4418 *pid,
4419 arg_slice,
4420 arg_custom_mounts, arg_n_custom_mounts,
4421 arg_kill_signal,
4422 arg_property,
4423 arg_property_message);
4424 if (r < 0)
4425 return r;
4426
4427 } else if (arg_slice || arg_property)
4428 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4429
4430 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
4431 if (r < 0)
4432 return r;
4433
4434 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4435 if (r < 0)
4436 return r;
4437
4438 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4439 if (r < 0)
4440 return r;
4441
4442 /* Notify the child that the parent is ready with all
4443 * its setup (including cgroup-ification), and that
4444 * the child can now hand over control to the code to
4445 * run inside the container. */
4446 (void) barrier_place(&barrier); /* #4 */
4447
4448 /* Block SIGCHLD here, before notifying child.
4449 * process_pty() will handle it with the other signals. */
4450 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4451
4452 /* Reset signal to default */
4453 r = default_signals(SIGCHLD, -1);
4454 if (r < 0)
4455 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4456
4457 r = sd_event_new(&event);
4458 if (r < 0)
4459 return log_error_errno(r, "Failed to get default event source: %m");
4460
4461 (void) sd_event_set_watchdog(event, true);
4462
4463 if (bus) {
4464 r = sd_bus_attach_event(bus, event, 0);
4465 if (r < 0)
4466 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4467 }
4468
4469 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
4470 if (r < 0)
4471 return r;
4472
4473 /* Let the child know that we are ready and wait that the child is completely ready now. */
4474 if (!barrier_place_and_sync(&barrier)) { /* #5 */
4475 log_error("Child died too early.");
4476 return -ESRCH;
4477 }
4478
4479 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4480 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4481 etc_passwd_lock = safe_close(etc_passwd_lock);
4482
4483 (void) sd_notifyf(false,
4484 "STATUS=Container running.\n"
4485 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
4486 if (!arg_notify_ready)
4487 (void) sd_notify(false, "READY=1\n");
4488
4489 if (arg_kill_signal > 0) {
4490 /* Try to kill the init system on SIGINT or SIGTERM */
4491 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4492 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
4493 } else {
4494 /* Immediately exit */
4495 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4496 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4497 }
4498
4499 /* Exit when the child exits */
4500 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
4501
4502 if (arg_expose_ports) {
4503 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
4504 if (r < 0)
4505 return r;
4506
4507 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
4508 }
4509
4510 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4511
4512 if (IN_SET(arg_console_mode, CONSOLE_INTERACTIVE, CONSOLE_READ_ONLY)) {
4513 assert(master >= 0);
4514
4515 r = pty_forward_new(event, master,
4516 PTY_FORWARD_IGNORE_VHANGUP | (arg_console_mode == CONSOLE_READ_ONLY ? PTY_FORWARD_READ_ONLY : 0),
4517 &forward);
4518 if (r < 0)
4519 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4520
4521 if (arg_console_width != (unsigned) -1 || arg_console_height != (unsigned) -1)
4522 (void) pty_forward_set_width_height(forward, arg_console_width, arg_console_height);
4523 }
4524
4525 r = sd_event_loop(event);
4526 if (r < 0)
4527 return log_error_errno(r, "Failed to run event loop: %m");
4528
4529 if (forward) {
4530 char last_char = 0;
4531
4532 (void) pty_forward_get_last_char(forward, &last_char);
4533 forward = pty_forward_free(forward);
4534
4535 if (!arg_quiet && last_char != '\n')
4536 putc('\n', stdout);
4537 }
4538
4539 /* Kill if it is not dead yet anyway */
4540 if (bus) {
4541 if (arg_register)
4542 terminate_machine(bus, arg_machine);
4543 else if (!arg_keep_unit)
4544 terminate_scope(bus, arg_machine);
4545 }
4546
4547 /* Normally redundant, but better safe than sorry */
4548 (void) kill(*pid, SIGKILL);
4549
4550 r = wait_for_container(*pid, &container_status);
4551 *pid = 0;
4552
4553 if (r < 0)
4554 /* We failed to wait for the container, or the container exited abnormally. */
4555 return r;
4556 if (r > 0 || container_status == CONTAINER_TERMINATED) {
4557 /* r > 0 → The container exited with a non-zero status.
4558 * As a special case, we need to replace 133 with a different value,
4559 * because 133 is special-cased in the service file to reboot the container.
4560 * otherwise → The container exited with zero status and a reboot was not requested.
4561 */
4562 if (r == EXIT_FORCE_RESTART)
4563 r = EXIT_FAILURE; /* replace 133 with the general failure code */
4564 *ret = r;
4565 return 0; /* finito */
4566 }
4567
4568 /* CONTAINER_REBOOTED, loop again */
4569
4570 if (arg_keep_unit) {
4571 /* Special handling if we are running as a service: instead of simply
4572 * restarting the machine we want to restart the entire service, so let's
4573 * inform systemd about this with the special exit code 133. The service
4574 * file uses RestartForceExitStatus=133 so that this results in a full
4575 * nspawn restart. This is necessary since we might have cgroup parameters
4576 * set we want to have flushed out. */
4577 *ret = EXIT_FORCE_RESTART;
4578 return 0; /* finito */
4579 }
4580
4581 expose_port_flush(arg_expose_ports, exposed);
4582
4583 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4584 *veth_created = false;
4585 return 1; /* loop again */
4586 }
4587
4588 static int initialize_rlimits(void) {
4589 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4590 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4591 * container execution environments. */
4592
4593 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
4594 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
4595 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
4596 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
4597 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
4598 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
4599 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
4600 [RLIMIT_MEMLOCK] = { 65536, 65536 },
4601 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
4602 [RLIMIT_NICE] = { 0, 0 },
4603 [RLIMIT_NOFILE] = { 1024, 4096 },
4604 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
4605 [RLIMIT_RTPRIO] = { 0, 0 },
4606 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
4607 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
4608
4609 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4610 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4611 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4612 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4613 * that PID 1 changes a number of other resource limits during early initialization which is why we
4614 * don't read the other limits from PID 1 but prefer the static table above. */
4615 };
4616
4617 int rl;
4618
4619 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
4620 /* Let's only fill in what the user hasn't explicitly configured anyway */
4621 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
4622 const struct rlimit *v;
4623 struct rlimit buffer;
4624
4625 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
4626 /* For these two let's read the limits off PID 1. See above for an explanation. */
4627
4628 if (prlimit(1, rl, NULL, &buffer) < 0)
4629 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
4630
4631 v = &buffer;
4632 } else
4633 v = kernel_defaults + rl;
4634
4635 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
4636 if (!arg_rlimit[rl])
4637 return log_oom();
4638 }
4639
4640 if (DEBUG_LOGGING) {
4641 _cleanup_free_ char *k = NULL;
4642
4643 (void) rlimit_format(arg_rlimit[rl], &k);
4644 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
4645 }
4646 }
4647
4648 return 0;
4649 }
4650
4651 static int run(int argc, char *argv[]) {
4652 _cleanup_free_ char *console = NULL;
4653 _cleanup_close_ int master = -1;
4654 _cleanup_fdset_free_ FDSet *fds = NULL;
4655 int r, n_fd_passed, ret = EXIT_SUCCESS;
4656 char veth_name[IFNAMSIZ] = "";
4657 bool secondary = false, remove_directory = false, remove_image = false;
4658 pid_t pid = 0;
4659 union in_addr_union exposed = {};
4660 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4661 bool veth_created = false, remove_tmprootdir = false;
4662 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
4663 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
4664 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4665 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
4666
4667 log_parse_environment();
4668 log_open();
4669
4670 /* Make sure rename_process() in the stub init process can work */
4671 save_argc_argv(argc, argv);
4672
4673 r = parse_argv(argc, argv);
4674 if (r <= 0)
4675 goto finish;
4676
4677 r = must_be_root();
4678 if (r < 0)
4679 goto finish;
4680
4681 r = initialize_rlimits();
4682 if (r < 0)
4683 goto finish;
4684
4685 r = load_oci_bundle();
4686 if (r < 0)
4687 goto finish;
4688
4689 r = determine_names();
4690 if (r < 0)
4691 goto finish;
4692
4693 r = load_settings();
4694 if (r < 0)
4695 goto finish;
4696
4697 r = cg_unified_flush();
4698 if (r < 0) {
4699 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4700 goto finish;
4701 }
4702
4703 r = verify_arguments();
4704 if (r < 0)
4705 goto finish;
4706
4707 r = detect_unified_cgroup_hierarchy_from_environment();
4708 if (r < 0)
4709 goto finish;
4710
4711 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4712 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4713 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4714 (void) ignore_signals(SIGPIPE, -1);
4715
4716 n_fd_passed = sd_listen_fds(false);
4717 if (n_fd_passed > 0) {
4718 r = fdset_new_listen_fds(&fds, false);
4719 if (r < 0) {
4720 log_error_errno(r, "Failed to collect file descriptors: %m");
4721 goto finish;
4722 }
4723 }
4724
4725 /* The "default" umask. This is appropriate for most file and directory
4726 * operations performed by nspawn, and is the umask that will be used for
4727 * the child. Functions like copy_devnodes() change the umask temporarily. */
4728 umask(0022);
4729
4730 if (arg_directory) {
4731 assert(!arg_image);
4732
4733 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
4734 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4735 r = -EINVAL;
4736 goto finish;
4737 }
4738
4739 if (arg_ephemeral) {
4740 _cleanup_free_ char *np = NULL;
4741
4742 r = chase_symlinks_and_update(&arg_directory, 0);
4743 if (r < 0)
4744 goto finish;
4745
4746 /* If the specified path is a mount point we
4747 * generate the new snapshot immediately
4748 * inside it under a random name. However if
4749 * the specified is not a mount point we
4750 * create the new snapshot in the parent
4751 * directory, just next to it. */
4752 r = path_is_mount_point(arg_directory, NULL, 0);
4753 if (r < 0) {
4754 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4755 goto finish;
4756 }
4757 if (r > 0)
4758 r = tempfn_random_child(arg_directory, "machine.", &np);
4759 else
4760 r = tempfn_random(arg_directory, "machine.", &np);
4761 if (r < 0) {
4762 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
4763 goto finish;
4764 }
4765
4766 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4767 if (r < 0) {
4768 log_error_errno(r, "Failed to lock %s: %m", np);
4769 goto finish;
4770 }
4771
4772 r = btrfs_subvol_snapshot(arg_directory, np,
4773 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4774 BTRFS_SNAPSHOT_FALLBACK_COPY |
4775 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4776 BTRFS_SNAPSHOT_RECURSIVE |
4777 BTRFS_SNAPSHOT_QUOTA);
4778 if (r < 0) {
4779 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4780 goto finish;
4781 }
4782
4783 free_and_replace(arg_directory, np);
4784
4785 remove_directory = true;
4786
4787 } else {
4788 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
4789 if (r < 0)
4790 goto finish;
4791
4792 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4793 if (r == -EBUSY) {
4794 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4795 goto finish;
4796 }
4797 if (r < 0) {
4798 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4799 goto finish;
4800 }
4801
4802 if (arg_template) {
4803 r = chase_symlinks_and_update(&arg_template, 0);
4804 if (r < 0)
4805 goto finish;
4806
4807 r = btrfs_subvol_snapshot(arg_template, arg_directory,
4808 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4809 BTRFS_SNAPSHOT_FALLBACK_COPY |
4810 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4811 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
4812 BTRFS_SNAPSHOT_RECURSIVE |
4813 BTRFS_SNAPSHOT_QUOTA);
4814 if (r == -EEXIST)
4815 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4816 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4817 else if (r < 0) {
4818 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4819 goto finish;
4820 } else
4821 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4822 "Populated %s from template %s.", arg_directory, arg_template);
4823 }
4824 }
4825
4826 if (arg_start_mode == START_BOOT) {
4827 const char *p;
4828
4829 if (arg_pivot_root_new)
4830 p = prefix_roota(arg_directory, arg_pivot_root_new);
4831 else
4832 p = arg_directory;
4833
4834 if (path_is_os_tree(p) <= 0) {
4835 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
4836 r = -EINVAL;
4837 goto finish;
4838 }
4839 } else {
4840 const char *p, *q;
4841
4842 if (arg_pivot_root_new)
4843 p = prefix_roota(arg_directory, arg_pivot_root_new);
4844 else
4845 p = arg_directory;
4846
4847 q = strjoina(p, "/usr/");
4848
4849 if (laccess(q, F_OK) < 0) {
4850 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
4851 r = -EINVAL;
4852 goto finish;
4853 }
4854 }
4855
4856 } else {
4857 assert(arg_image);
4858 assert(!arg_template);
4859
4860 r = chase_symlinks_and_update(&arg_image, 0);
4861 if (r < 0)
4862 goto finish;
4863
4864 if (arg_ephemeral) {
4865 _cleanup_free_ char *np = NULL;
4866
4867 r = tempfn_random(arg_image, "machine.", &np);
4868 if (r < 0) {
4869 log_error_errno(r, "Failed to generate name for image snapshot: %m");
4870 goto finish;
4871 }
4872
4873 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4874 if (r < 0) {
4875 r = log_error_errno(r, "Failed to create image lock: %m");
4876 goto finish;
4877 }
4878
4879 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
4880 if (r < 0) {
4881 r = log_error_errno(r, "Failed to copy image file: %m");
4882 goto finish;
4883 }
4884
4885 free_and_replace(arg_image, np);
4886
4887 remove_image = true;
4888 } else {
4889 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4890 if (r == -EBUSY) {
4891 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4892 goto finish;
4893 }
4894 if (r < 0) {
4895 r = log_error_errno(r, "Failed to create image lock: %m");
4896 goto finish;
4897 }
4898
4899 if (!arg_root_hash) {
4900 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4901 if (r < 0) {
4902 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4903 goto finish;
4904 }
4905 }
4906 }
4907
4908 if (!mkdtemp(tmprootdir)) {
4909 r = log_error_errno(errno, "Failed to create temporary directory: %m");
4910 goto finish;
4911 }
4912
4913 remove_tmprootdir = true;
4914
4915 arg_directory = strdup(tmprootdir);
4916 if (!arg_directory) {
4917 r = log_oom();
4918 goto finish;
4919 }
4920
4921 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4922 if (r < 0) {
4923 log_error_errno(r, "Failed to set up loopback block device: %m");
4924 goto finish;
4925 }
4926
4927 r = dissect_image_and_warn(
4928 loop->fd,
4929 arg_image,
4930 arg_root_hash, arg_root_hash_size,
4931 DISSECT_IMAGE_REQUIRE_ROOT,
4932 &dissected_image);
4933 if (r == -ENOPKG) {
4934 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4935 log_notice("Note that the disk image needs to\n"
4936 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4937 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4938 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4939 " d) or contain a file system without a partition table\n"
4940 "in order to be bootable with systemd-nspawn.");
4941 goto finish;
4942 }
4943 if (r < 0)
4944 goto finish;
4945
4946 if (!arg_root_hash && dissected_image->can_verity)
4947 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4948
4949 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
4950 if (r < 0)
4951 goto finish;
4952
4953 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4954 if (remove_image && unlink(arg_image) >= 0)
4955 remove_image = false;
4956 }
4957
4958 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
4959 if (r < 0)
4960 goto finish;
4961
4962 if (arg_console_mode < 0)
4963 arg_console_mode =
4964 isatty(STDIN_FILENO) > 0 &&
4965 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
4966
4967 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
4968 arg_quiet = true;
4969
4970 if (arg_console_mode != CONSOLE_PIPE) {
4971 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NONBLOCK);
4972 if (master < 0) {
4973 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4974 goto finish;
4975 }
4976
4977 r = ptsname_malloc(master, &console);
4978 if (r < 0) {
4979 r = log_error_errno(r, "Failed to determine tty name: %m");
4980 goto finish;
4981 }
4982
4983 if (arg_selinux_apifs_context) {
4984 r = mac_selinux_apply(console, arg_selinux_apifs_context);
4985 if (r < 0)
4986 goto finish;
4987 }
4988
4989 if (unlockpt(master) < 0) {
4990 r = log_error_errno(errno, "Failed to unlock tty: %m");
4991 goto finish;
4992 }
4993 }
4994
4995 if (!arg_quiet)
4996 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4997 arg_machine, arg_image ?: arg_directory);
4998
4999 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
5000
5001 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
5002 r = log_error_errno(errno, "Failed to become subreaper: %m");
5003 goto finish;
5004 }
5005
5006 for (;;) {
5007 r = run_container(master,
5008 console,
5009 dissected_image,
5010 secondary,
5011 fds,
5012 veth_name, &veth_created,
5013 &exposed,
5014 &pid, &ret);
5015 if (r <= 0)
5016 break;
5017 }
5018
5019 finish:
5020 (void) sd_notify(false,
5021 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5022 "STOPPING=1\nSTATUS=Terminating...");
5023
5024 if (pid > 0)
5025 (void) kill(pid, SIGKILL);
5026
5027 /* Try to flush whatever is still queued in the pty */
5028 if (master >= 0) {
5029 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
5030 master = safe_close(master);
5031 }
5032
5033 if (pid > 0)
5034 (void) wait_for_terminate(pid, NULL);
5035
5036 pager_close();
5037
5038 if (remove_directory && arg_directory) {
5039 int k;
5040
5041 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5042 if (k < 0)
5043 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5044 }
5045
5046 if (remove_image && arg_image) {
5047 if (unlink(arg_image) < 0)
5048 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5049 }
5050
5051 if (remove_tmprootdir) {
5052 if (rmdir(tmprootdir) < 0)
5053 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5054 }
5055
5056 if (arg_machine) {
5057 const char *p;
5058
5059 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5060 (void) rm_rf(p, REMOVE_ROOT);
5061 }
5062
5063 expose_port_flush(arg_expose_ports, &exposed);
5064
5065 if (veth_created)
5066 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5067 (void) remove_bridge(arg_network_zone);
5068
5069 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5070 expose_port_free_all(arg_expose_ports);
5071 rlimit_free_all(arg_rlimit);
5072 device_node_free_many(arg_extra_nodes, arg_n_extra_nodes);
5073
5074 if (r < 0)
5075 return r;
5076
5077 return ret;
5078 }
5079
5080 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.