]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #28132 from rpigott/dhcp-captive-portal
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #if HAVE_BLKID
4 #endif
5 #include <errno.h>
6 #include <getopt.h>
7 #include <linux/fs.h>
8 #include <linux/loop.h>
9 #if HAVE_SELINUX
10 #include <selinux/selinux.h>
11 #endif
12 #include <stdlib.h>
13 #include <sys/file.h>
14 #include <sys/ioctl.h>
15 #include <sys/personality.h>
16 #include <sys/prctl.h>
17 #include <sys/types.h>
18 #include <sys/wait.h>
19 #include <termios.h>
20 #include <unistd.h>
21
22 #include "sd-bus.h"
23 #include "sd-daemon.h"
24 #include "sd-id128.h"
25
26 #include "alloc-util.h"
27 #include "barrier.h"
28 #include "base-filesystem.h"
29 #include "blkid-util.h"
30 #include "btrfs-util.h"
31 #include "build.h"
32 #include "bus-error.h"
33 #include "bus-locator.h"
34 #include "bus-util.h"
35 #include "cap-list.h"
36 #include "capability-util.h"
37 #include "cgroup-util.h"
38 #include "chase.h"
39 #include "common-signal.h"
40 #include "copy.h"
41 #include "cpu-set-util.h"
42 #include "creds-util.h"
43 #include "dev-setup.h"
44 #include "discover-image.h"
45 #include "dissect-image.h"
46 #include "env-util.h"
47 #include "escape.h"
48 #include "fd-util.h"
49 #include "fdset.h"
50 #include "fileio.h"
51 #include "format-util.h"
52 #include "fs-util.h"
53 #include "gpt.h"
54 #include "hexdecoct.h"
55 #include "hostname-setup.h"
56 #include "hostname-util.h"
57 #include "id128-util.h"
58 #include "io-util.h"
59 #include "log.h"
60 #include "loop-util.h"
61 #include "loopback-setup.h"
62 #include "macro.h"
63 #include "main-func.h"
64 #include "missing_sched.h"
65 #include "mkdir.h"
66 #include "mount-util.h"
67 #include "mountpoint-util.h"
68 #include "namespace-util.h"
69 #include "netlink-util.h"
70 #include "nspawn-bind-user.h"
71 #include "nspawn-cgroup.h"
72 #include "nspawn-creds.h"
73 #include "nspawn-def.h"
74 #include "nspawn-expose-ports.h"
75 #include "nspawn-mount.h"
76 #include "nspawn-network.h"
77 #include "nspawn-oci.h"
78 #include "nspawn-patch-uid.h"
79 #include "nspawn-register.h"
80 #include "nspawn-seccomp.h"
81 #include "nspawn-settings.h"
82 #include "nspawn-setuid.h"
83 #include "nspawn-stub-pid1.h"
84 #include "nspawn-util.h"
85 #include "nspawn.h"
86 #include "nulstr-util.h"
87 #include "os-util.h"
88 #include "pager.h"
89 #include "parse-argument.h"
90 #include "parse-util.h"
91 #include "pretty-print.h"
92 #include "process-util.h"
93 #include "ptyfwd.h"
94 #include "random-util.h"
95 #include "raw-clone.h"
96 #include "resolve-util.h"
97 #include "rlimit-util.h"
98 #include "rm-rf.h"
99 #if HAVE_SECCOMP
100 #include "seccomp-util.h"
101 #endif
102 #include "selinux-util.h"
103 #include "signal-util.h"
104 #include "socket-util.h"
105 #include "stat-util.h"
106 #include "stdio-util.h"
107 #include "string-table.h"
108 #include "string-util.h"
109 #include "strv.h"
110 #include "sysctl-util.h"
111 #include "terminal-util.h"
112 #include "tmpfile-util.h"
113 #include "umask-util.h"
114 #include "unit-name.h"
115 #include "user-util.h"
116
117 /* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
119 #define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
120
121 #define EXIT_FORCE_RESTART 133
122
123 typedef enum ContainerStatus {
124 CONTAINER_TERMINATED,
125 CONTAINER_REBOOTED,
126 } ContainerStatus;
127
128 static char *arg_directory = NULL;
129 static char *arg_template = NULL;
130 static char *arg_chdir = NULL;
131 static char *arg_pivot_root_new = NULL;
132 static char *arg_pivot_root_old = NULL;
133 static char *arg_user = NULL;
134 static uid_t arg_uid = UID_INVALID;
135 static gid_t arg_gid = GID_INVALID;
136 static gid_t* arg_supplementary_gids = NULL;
137 static size_t arg_n_supplementary_gids = 0;
138 static sd_id128_t arg_uuid = {};
139 static char *arg_machine = NULL; /* The name used by the host to refer to this */
140 static char *arg_hostname = NULL; /* The name the payload sees by default */
141 static const char *arg_selinux_context = NULL;
142 static const char *arg_selinux_apifs_context = NULL;
143 static char *arg_slice = NULL;
144 static bool arg_private_network = false;
145 static bool arg_read_only = false;
146 static StartMode arg_start_mode = START_PID1;
147 static bool arg_ephemeral = false;
148 static LinkJournal arg_link_journal = LINK_AUTO;
149 static bool arg_link_journal_try = false;
150 static uint64_t arg_caps_retain =
151 (1ULL << CAP_AUDIT_CONTROL) |
152 (1ULL << CAP_AUDIT_WRITE) |
153 (1ULL << CAP_CHOWN) |
154 (1ULL << CAP_DAC_OVERRIDE) |
155 (1ULL << CAP_DAC_READ_SEARCH) |
156 (1ULL << CAP_FOWNER) |
157 (1ULL << CAP_FSETID) |
158 (1ULL << CAP_IPC_OWNER) |
159 (1ULL << CAP_KILL) |
160 (1ULL << CAP_LEASE) |
161 (1ULL << CAP_LINUX_IMMUTABLE) |
162 (1ULL << CAP_MKNOD) |
163 (1ULL << CAP_NET_BIND_SERVICE) |
164 (1ULL << CAP_NET_BROADCAST) |
165 (1ULL << CAP_NET_RAW) |
166 (1ULL << CAP_SETFCAP) |
167 (1ULL << CAP_SETGID) |
168 (1ULL << CAP_SETPCAP) |
169 (1ULL << CAP_SETUID) |
170 (1ULL << CAP_SYS_ADMIN) |
171 (1ULL << CAP_SYS_BOOT) |
172 (1ULL << CAP_SYS_CHROOT) |
173 (1ULL << CAP_SYS_NICE) |
174 (1ULL << CAP_SYS_PTRACE) |
175 (1ULL << CAP_SYS_RESOURCE) |
176 (1ULL << CAP_SYS_TTY_CONFIG);
177 static uint64_t arg_caps_ambient = 0;
178 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
179 static CustomMount *arg_custom_mounts = NULL;
180 static size_t arg_n_custom_mounts = 0;
181 static char **arg_setenv = NULL;
182 static bool arg_quiet = false;
183 static bool arg_register = true;
184 static bool arg_keep_unit = false;
185 static char **arg_network_interfaces = NULL;
186 static char **arg_network_macvlan = NULL;
187 static char **arg_network_ipvlan = NULL;
188 static bool arg_network_veth = false;
189 static char **arg_network_veth_extra = NULL;
190 static char *arg_network_bridge = NULL;
191 static char *arg_network_zone = NULL;
192 static char *arg_network_namespace_path = NULL;
193 static PagerFlags arg_pager_flags = 0;
194 static unsigned long arg_personality = PERSONALITY_INVALID;
195 static char *arg_image = NULL;
196 static char *arg_oci_bundle = NULL;
197 static VolatileMode arg_volatile_mode = VOLATILE_NO;
198 static ExposePort *arg_expose_ports = NULL;
199 static char **arg_property = NULL;
200 static sd_bus_message *arg_property_message = NULL;
201 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
202 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
203 static UserNamespaceOwnership arg_userns_ownership = _USER_NAMESPACE_OWNERSHIP_INVALID;
204 static int arg_kill_signal = 0;
205 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
206 static SettingsMask arg_settings_mask = 0;
207 static int arg_settings_trusted = -1;
208 static char **arg_parameters = NULL;
209 static const char *arg_container_service_name = "systemd-nspawn";
210 static bool arg_notify_ready = false;
211 static bool arg_use_cgns = true;
212 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
213 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
214 static VeritySettings arg_verity_settings = VERITY_SETTINGS_DEFAULT;
215 static char **arg_syscall_allow_list = NULL;
216 static char **arg_syscall_deny_list = NULL;
217 #if HAVE_SECCOMP
218 static scmp_filter_ctx arg_seccomp = NULL;
219 #endif
220 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
221 static bool arg_no_new_privileges = false;
222 static int arg_oom_score_adjust = 0;
223 static bool arg_oom_score_adjust_set = false;
224 static CPUSet arg_cpu_set = {};
225 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
226 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
227 static unsigned arg_console_width = UINT_MAX, arg_console_height = UINT_MAX;
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 static Credential *arg_credentials = NULL;
233 static size_t arg_n_credentials = 0;
234 static char **arg_bind_user = NULL;
235 static bool arg_suppress_sync = false;
236 static char *arg_settings_filename = NULL;
237 static Architecture arg_architecture = _ARCHITECTURE_INVALID;
238 static ImagePolicy *arg_image_policy = NULL;
239
240 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
241 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
248 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
250 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
252 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
253 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
254 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
255 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
256 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
257 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
258 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
259 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
260 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
261 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
262 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
263 STATIC_DESTRUCTOR_REGISTER(arg_verity_settings, verity_settings_done);
264 STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list, strv_freep);
265 STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list, strv_freep);
266 #if HAVE_SECCOMP
267 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
268 #endif
269 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
270 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
271 STATIC_DESTRUCTOR_REGISTER(arg_bind_user, strv_freep);
272 STATIC_DESTRUCTOR_REGISTER(arg_settings_filename, freep);
273 STATIC_DESTRUCTOR_REGISTER(arg_image_policy, image_policy_freep);
274
275 static int handle_arg_console(const char *arg) {
276 if (streq(arg, "help")) {
277 puts("autopipe\n"
278 "interactive\n"
279 "passive\n"
280 "pipe\n"
281 "read-only");
282 return 0;
283 }
284
285 if (streq(arg, "interactive"))
286 arg_console_mode = CONSOLE_INTERACTIVE;
287 else if (streq(arg, "read-only"))
288 arg_console_mode = CONSOLE_READ_ONLY;
289 else if (streq(arg, "passive"))
290 arg_console_mode = CONSOLE_PASSIVE;
291 else if (streq(arg, "pipe")) {
292 if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
293 log_full(arg_quiet ? LOG_DEBUG : LOG_NOTICE,
294 "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
295 "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
296 "Proceeding anyway.");
297
298 arg_console_mode = CONSOLE_PIPE;
299 } else if (streq(arg, "autopipe")) {
300 if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
301 arg_console_mode = CONSOLE_INTERACTIVE;
302 else
303 arg_console_mode = CONSOLE_PIPE;
304 } else
305 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
306
307 arg_settings_mask |= SETTING_CONSOLE_MODE;
308 return 1;
309 }
310
311 static int help(void) {
312 _cleanup_free_ char *link = NULL;
313 int r;
314
315 pager_open(arg_pager_flags);
316
317 r = terminal_urlify_man("systemd-nspawn", "1", &link);
318 if (r < 0)
319 return log_oom();
320
321 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
322 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
323 " -h --help Show this help\n"
324 " --version Print version string\n"
325 " -q --quiet Do not show status information\n"
326 " --no-pager Do not pipe output into a pager\n"
327 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
328 "%3$sImage:%4$s\n"
329 " -D --directory=PATH Root directory for the container\n"
330 " --template=PATH Initialize root directory from template directory,\n"
331 " if missing\n"
332 " -x --ephemeral Run container with snapshot of root directory, and\n"
333 " remove it after exit\n"
334 " -i --image=PATH Root file system disk image (or device node) for\n"
335 " the container\n"
336 " --image-policy=POLICY Specify disk image dissection policy\n"
337 " --oci-bundle=PATH OCI bundle directory\n"
338 " --read-only Mount the root directory read-only\n"
339 " --volatile[=MODE] Run the system in volatile mode\n"
340 " --root-hash=HASH Specify verity root hash for root disk image\n"
341 " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
342 " as a DER encoded PKCS7, either as a path to a file\n"
343 " or as an ASCII base64 encoded string prefixed by\n"
344 " 'base64:'\n"
345 " --verity-data=PATH Specify hash device for verity\n"
346 " --pivot-root=PATH[:PATH]\n"
347 " Pivot root to given directory in the container\n\n"
348 "%3$sExecution:%4$s\n"
349 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
350 " -b --boot Boot up full system (i.e. invoke init)\n"
351 " --chdir=PATH Set working directory in the container\n"
352 " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
353 " -u --user=USER Run the command under specified user or UID\n"
354 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
355 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
356 " --suppress-sync=BOOLEAN\n"
357 " Suppress any form of disk data synchronization\n\n"
358 "%3$sSystem Identity:%4$s\n"
359 " -M --machine=NAME Set the machine name for the container\n"
360 " --hostname=NAME Override the hostname for the container\n"
361 " --uuid=UUID Set a specific machine UUID for the container\n\n"
362 "%3$sProperties:%4$s\n"
363 " -S --slice=SLICE Place the container in the specified slice\n"
364 " --property=NAME=VALUE Set scope unit property\n"
365 " --register=BOOLEAN Register container as machine\n"
366 " --keep-unit Do not register a scope for the machine, reuse\n"
367 " the service unit nspawn is running in\n\n"
368 "%3$sUser Namespacing:%4$s\n"
369 " --private-users=no Run without user namespacing\n"
370 " --private-users=yes|pick|identity\n"
371 " Run within user namespace, autoselect UID/GID range\n"
372 " --private-users=UIDBASE[:NUIDS]\n"
373 " Similar, but with user configured UID/GID range\n"
374 " --private-users-ownership=MODE\n"
375 " Adjust ('chown') or map ('map') OS tree ownership\n"
376 " to private UID/GID range\n"
377 " -U Equivalent to --private-users=pick and\n"
378 " --private-users-ownership=auto\n\n"
379 "%3$sNetworking:%4$s\n"
380 " --private-network Disable network in container\n"
381 " --network-interface=HOSTIF[:CONTAINERIF]\n"
382 " Assign an existing network interface to the\n"
383 " container\n"
384 " --network-macvlan=HOSTIF[:CONTAINERIF]\n"
385 " Create a macvlan network interface based on an\n"
386 " existing network interface to the container\n"
387 " --network-ipvlan=HOSTIF[:CONTAINERIF]\n"
388 " Create an ipvlan network interface based on an\n"
389 " existing network interface to the container\n"
390 " -n --network-veth Add a virtual Ethernet connection between host\n"
391 " and container\n"
392 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
393 " Add an additional virtual Ethernet link between\n"
394 " host and container\n"
395 " --network-bridge=INTERFACE\n"
396 " Add a virtual Ethernet connection to the container\n"
397 " and attach it to an existing bridge on the host\n"
398 " --network-zone=NAME Similar, but attach the new interface to an\n"
399 " an automatically managed bridge interface\n"
400 " --network-namespace-path=PATH\n"
401 " Set network namespace to the one represented by\n"
402 " the specified kernel namespace file node\n"
403 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
404 " Expose a container IP port on the host\n\n"
405 "%3$sSecurity:%4$s\n"
406 " --capability=CAP In addition to the default, retain specified\n"
407 " capability\n"
408 " --drop-capability=CAP Drop the specified capability from the default set\n"
409 " --ambient-capability=CAP\n"
410 " Sets the specified capability for the started\n"
411 " process. Not useful if booting a machine.\n"
412 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
413 " --system-call-filter=LIST|~LIST\n"
414 " Permit/prohibit specific system calls\n"
415 " -Z --selinux-context=SECLABEL\n"
416 " Set the SELinux security context to be used by\n"
417 " processes in the container\n"
418 " -L --selinux-apifs-context=SECLABEL\n"
419 " Set the SELinux security context to be used by\n"
420 " API/tmpfs file systems in the container\n\n"
421 "%3$sResources:%4$s\n"
422 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
423 " --oom-score-adjust=VALUE\n"
424 " Adjust the OOM score value for the payload\n"
425 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
426 " --personality=ARCH Pick personality for this container\n\n"
427 "%3$sIntegration:%4$s\n"
428 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
429 " --timezone=MODE Select mode of /etc/localtime initialization\n"
430 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
431 " host, try-guest, try-host\n"
432 " -j Equivalent to --link-journal=try-guest\n\n"
433 "%3$sMounts:%4$s\n"
434 " --bind=PATH[:PATH[:OPTIONS]]\n"
435 " Bind mount a file or directory from the host into\n"
436 " the container\n"
437 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
438 " Similar, but creates a read-only bind mount\n"
439 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
440 " it\n"
441 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
442 " --overlay=PATH[:PATH...]:PATH\n"
443 " Create an overlay mount from the host to \n"
444 " the container\n"
445 " --overlay-ro=PATH[:PATH...]:PATH\n"
446 " Similar, but creates a read-only overlay mount\n"
447 " --bind-user=NAME Bind user from host to container\n\n"
448 "%3$sInput/Output:%4$s\n"
449 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
450 " set up for the container.\n"
451 " -P --pipe Equivalent to --console=pipe\n\n"
452 "%3$sCredentials:%4$s\n"
453 " --set-credential=ID:VALUE\n"
454 " Pass a credential with literal value to container.\n"
455 " --load-credential=ID:PATH\n"
456 " Load credential to pass to container from file or\n"
457 " AF_UNIX stream socket.\n"
458 "\nSee the %2$s for details.\n",
459 program_invocation_short_name,
460 link,
461 ansi_underline(),
462 ansi_normal(),
463 ansi_highlight(),
464 ansi_normal());
465
466 return 0;
467 }
468
469 static int custom_mount_check_all(void) {
470 size_t i;
471
472 for (i = 0; i < arg_n_custom_mounts; i++) {
473 CustomMount *m = &arg_custom_mounts[i];
474
475 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
476 if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_OFF)
477 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
478 "--private-users-ownership=own may not be combined with custom root mounts.");
479 if (arg_uid_shift == UID_INVALID)
480 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
481 "--private-users with automatic UID shift may not be combined with custom root mounts.");
482 }
483 }
484
485 return 0;
486 }
487
488 static int detect_unified_cgroup_hierarchy_from_environment(void) {
489 const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
490 int r;
491
492 /* Allow the user to control whether the unified hierarchy is used */
493
494 e = getenv(var);
495 if (!e) {
496 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
497 var = "UNIFIED_CGROUP_HIERARCHY";
498 e = getenv(var);
499 }
500
501 if (!isempty(e)) {
502 r = parse_boolean(e);
503 if (r < 0)
504 return log_error_errno(r, "Failed to parse $%s: %m", var);
505 if (r > 0)
506 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
507 else
508 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
509 }
510
511 return 0;
512 }
513
514 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
515 int r;
516
517 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
518 * in the image actually supports. */
519 r = cg_all_unified();
520 if (r < 0)
521 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
522 if (r > 0) {
523 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
524 * routine only detects 231, so we'll have a false negative here for 230. */
525 r = systemd_installation_has_version(directory, "230");
526 if (r < 0)
527 return log_error_errno(r, "Failed to determine systemd version in container: %m");
528 if (r > 0)
529 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
530 else
531 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
532 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
533 /* Mixed cgroup hierarchy support was added in 233 */
534 r = systemd_installation_has_version(directory, "233");
535 if (r < 0)
536 return log_error_errno(r, "Failed to determine systemd version in container: %m");
537 if (r > 0)
538 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
539 else
540 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
541 } else
542 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
543
544 log_debug("Using %s hierarchy for container.",
545 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
546 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
547
548 return 0;
549 }
550
551 static int parse_capability_spec(const char *spec, uint64_t *ret_mask) {
552 uint64_t mask = 0;
553 int r;
554
555 for (;;) {
556 _cleanup_free_ char *t = NULL;
557
558 r = extract_first_word(&spec, &t, ",", 0);
559 if (r < 0)
560 return log_error_errno(r, "Failed to parse capability %s.", t);
561 if (r == 0)
562 break;
563
564 if (streq(t, "help")) {
565 for (int i = 0; i < capability_list_length(); i++) {
566 const char *name;
567
568 name = capability_to_name(i);
569 if (name)
570 puts(name);
571 }
572
573 return 0; /* quit */
574 }
575
576 if (streq(t, "all"))
577 mask = UINT64_MAX;
578 else {
579 r = capability_from_name(t);
580 if (r < 0)
581 return log_error_errno(r, "Failed to parse capability %s.", t);
582
583 mask |= 1ULL << r;
584 }
585 }
586
587 *ret_mask = mask;
588 return 1; /* continue */
589 }
590
591 static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
592 int r;
593
594 r = getenv_bool(name);
595 if (r == -ENXIO)
596 return 0;
597 if (r < 0)
598 return log_error_errno(r, "Failed to parse $%s: %m", name);
599
600 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
601 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
602 return 0;
603 }
604
605 static int parse_mount_settings_env(void) {
606 const char *e;
607 int r;
608
609 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
610 if (r < 0 && r != -ENXIO)
611 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
612 if (r >= 0)
613 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
614
615 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
616 if (streq_ptr(e, "network"))
617 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
618
619 else if (e) {
620 r = parse_boolean(e);
621 if (r < 0)
622 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
623
624 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
625 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
626 }
627
628 return 0;
629 }
630
631 static int parse_environment(void) {
632 const char *e;
633 int r;
634
635 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
636 if (r < 0)
637 return r;
638 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
639 if (r < 0)
640 return r;
641 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
642 if (r < 0)
643 return r;
644 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
645 if (r < 0)
646 return r;
647
648 r = parse_mount_settings_env();
649 if (r < 0)
650 return r;
651
652 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
653 * even if it is supported. If not supported, it has no effect. */
654 if (!cg_ns_supported())
655 arg_use_cgns = false;
656 else {
657 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
658 if (r < 0) {
659 if (r != -ENXIO)
660 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
661
662 arg_use_cgns = true;
663 } else {
664 arg_use_cgns = r > 0;
665 arg_settings_mask |= SETTING_USE_CGNS;
666 }
667 }
668
669 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
670 if (e)
671 arg_container_service_name = e;
672
673 r = getenv_bool("SYSTEMD_SUPPRESS_SYNC");
674 if (r >= 0)
675 arg_suppress_sync = r;
676 else if (r != -ENXIO)
677 log_debug_errno(r, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
678
679 return detect_unified_cgroup_hierarchy_from_environment();
680 }
681
682 static int parse_argv(int argc, char *argv[]) {
683 enum {
684 ARG_VERSION = 0x100,
685 ARG_PRIVATE_NETWORK,
686 ARG_UUID,
687 ARG_READ_ONLY,
688 ARG_CAPABILITY,
689 ARG_AMBIENT_CAPABILITY,
690 ARG_DROP_CAPABILITY,
691 ARG_LINK_JOURNAL,
692 ARG_BIND,
693 ARG_BIND_RO,
694 ARG_TMPFS,
695 ARG_OVERLAY,
696 ARG_OVERLAY_RO,
697 ARG_INACCESSIBLE,
698 ARG_SHARE_SYSTEM,
699 ARG_REGISTER,
700 ARG_KEEP_UNIT,
701 ARG_NETWORK_INTERFACE,
702 ARG_NETWORK_MACVLAN,
703 ARG_NETWORK_IPVLAN,
704 ARG_NETWORK_BRIDGE,
705 ARG_NETWORK_ZONE,
706 ARG_NETWORK_VETH_EXTRA,
707 ARG_NETWORK_NAMESPACE_PATH,
708 ARG_PERSONALITY,
709 ARG_VOLATILE,
710 ARG_TEMPLATE,
711 ARG_PROPERTY,
712 ARG_PRIVATE_USERS,
713 ARG_KILL_SIGNAL,
714 ARG_SETTINGS,
715 ARG_CHDIR,
716 ARG_PIVOT_ROOT,
717 ARG_PRIVATE_USERS_CHOWN,
718 ARG_PRIVATE_USERS_OWNERSHIP,
719 ARG_NOTIFY_READY,
720 ARG_ROOT_HASH,
721 ARG_ROOT_HASH_SIG,
722 ARG_VERITY_DATA,
723 ARG_SYSTEM_CALL_FILTER,
724 ARG_RLIMIT,
725 ARG_HOSTNAME,
726 ARG_NO_NEW_PRIVILEGES,
727 ARG_OOM_SCORE_ADJUST,
728 ARG_CPU_AFFINITY,
729 ARG_RESOLV_CONF,
730 ARG_TIMEZONE,
731 ARG_CONSOLE,
732 ARG_PIPE,
733 ARG_OCI_BUNDLE,
734 ARG_NO_PAGER,
735 ARG_SET_CREDENTIAL,
736 ARG_LOAD_CREDENTIAL,
737 ARG_BIND_USER,
738 ARG_SUPPRESS_SYNC,
739 ARG_IMAGE_POLICY,
740 };
741
742 static const struct option options[] = {
743 { "help", no_argument, NULL, 'h' },
744 { "version", no_argument, NULL, ARG_VERSION },
745 { "directory", required_argument, NULL, 'D' },
746 { "template", required_argument, NULL, ARG_TEMPLATE },
747 { "ephemeral", no_argument, NULL, 'x' },
748 { "user", required_argument, NULL, 'u' },
749 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
750 { "as-pid2", no_argument, NULL, 'a' },
751 { "boot", no_argument, NULL, 'b' },
752 { "uuid", required_argument, NULL, ARG_UUID },
753 { "read-only", no_argument, NULL, ARG_READ_ONLY },
754 { "capability", required_argument, NULL, ARG_CAPABILITY },
755 { "ambient-capability", required_argument, NULL, ARG_AMBIENT_CAPABILITY },
756 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
757 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
758 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
759 { "bind", required_argument, NULL, ARG_BIND },
760 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
761 { "tmpfs", required_argument, NULL, ARG_TMPFS },
762 { "overlay", required_argument, NULL, ARG_OVERLAY },
763 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
764 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
765 { "machine", required_argument, NULL, 'M' },
766 { "hostname", required_argument, NULL, ARG_HOSTNAME },
767 { "slice", required_argument, NULL, 'S' },
768 { "setenv", required_argument, NULL, 'E' },
769 { "selinux-context", required_argument, NULL, 'Z' },
770 { "selinux-apifs-context", required_argument, NULL, 'L' },
771 { "quiet", no_argument, NULL, 'q' },
772 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
773 { "register", required_argument, NULL, ARG_REGISTER },
774 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
775 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
776 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
777 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
778 { "network-veth", no_argument, NULL, 'n' },
779 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
780 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
781 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
782 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
783 { "personality", required_argument, NULL, ARG_PERSONALITY },
784 { "image", required_argument, NULL, 'i' },
785 { "volatile", optional_argument, NULL, ARG_VOLATILE },
786 { "port", required_argument, NULL, 'p' },
787 { "property", required_argument, NULL, ARG_PROPERTY },
788 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
789 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN }, /* obsolete */
790 { "private-users-ownership",required_argument, NULL, ARG_PRIVATE_USERS_OWNERSHIP},
791 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
792 { "settings", required_argument, NULL, ARG_SETTINGS },
793 { "chdir", required_argument, NULL, ARG_CHDIR },
794 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
795 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
796 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
797 { "root-hash-sig", required_argument, NULL, ARG_ROOT_HASH_SIG },
798 { "verity-data", required_argument, NULL, ARG_VERITY_DATA },
799 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
800 { "rlimit", required_argument, NULL, ARG_RLIMIT },
801 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
802 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
803 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
804 { "timezone", required_argument, NULL, ARG_TIMEZONE },
805 { "console", required_argument, NULL, ARG_CONSOLE },
806 { "pipe", no_argument, NULL, ARG_PIPE },
807 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
808 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
809 { "set-credential", required_argument, NULL, ARG_SET_CREDENTIAL },
810 { "load-credential", required_argument, NULL, ARG_LOAD_CREDENTIAL },
811 { "bind-user", required_argument, NULL, ARG_BIND_USER },
812 { "suppress-sync", required_argument, NULL, ARG_SUPPRESS_SYNC },
813 { "image-policy", required_argument, NULL, ARG_IMAGE_POLICY },
814 {}
815 };
816
817 int c, r;
818 uint64_t plus = 0, minus = 0;
819 bool mask_all_settings = false, mask_no_settings = false;
820
821 assert(argc >= 0);
822 assert(argv);
823
824 /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
825 * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
826 optind = 0;
827 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
828 switch (c) {
829
830 case 'h':
831 return help();
832
833 case ARG_VERSION:
834 return version();
835
836 case 'D':
837 r = parse_path_argument(optarg, false, &arg_directory);
838 if (r < 0)
839 return r;
840
841 arg_settings_mask |= SETTING_DIRECTORY;
842 break;
843
844 case ARG_TEMPLATE:
845 r = parse_path_argument(optarg, false, &arg_template);
846 if (r < 0)
847 return r;
848
849 arg_settings_mask |= SETTING_DIRECTORY;
850 break;
851
852 case 'i':
853 r = parse_path_argument(optarg, false, &arg_image);
854 if (r < 0)
855 return r;
856
857 arg_settings_mask |= SETTING_DIRECTORY;
858 break;
859
860 case ARG_OCI_BUNDLE:
861 r = parse_path_argument(optarg, false, &arg_oci_bundle);
862 if (r < 0)
863 return r;
864
865 break;
866
867 case 'x':
868 arg_ephemeral = true;
869 arg_settings_mask |= SETTING_EPHEMERAL;
870 break;
871
872 case 'u':
873 r = free_and_strdup(&arg_user, optarg);
874 if (r < 0)
875 return log_oom();
876
877 arg_settings_mask |= SETTING_USER;
878 break;
879
880 case ARG_NETWORK_ZONE: {
881 _cleanup_free_ char *j = NULL;
882
883 j = strjoin("vz-", optarg);
884 if (!j)
885 return log_oom();
886
887 if (!ifname_valid(j))
888 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
889 "Network zone name not valid: %s", j);
890
891 free_and_replace(arg_network_zone, j);
892
893 arg_network_veth = true;
894 arg_private_network = true;
895 arg_settings_mask |= SETTING_NETWORK;
896 break;
897 }
898
899 case ARG_NETWORK_BRIDGE:
900
901 if (!ifname_valid(optarg))
902 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
903 "Bridge interface name not valid: %s", optarg);
904
905 r = free_and_strdup(&arg_network_bridge, optarg);
906 if (r < 0)
907 return log_oom();
908
909 _fallthrough_;
910 case 'n':
911 arg_network_veth = true;
912 arg_private_network = true;
913 arg_settings_mask |= SETTING_NETWORK;
914 break;
915
916 case ARG_NETWORK_VETH_EXTRA:
917 r = veth_extra_parse(&arg_network_veth_extra, optarg);
918 if (r < 0)
919 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
920
921 arg_private_network = true;
922 arg_settings_mask |= SETTING_NETWORK;
923 break;
924
925 case ARG_NETWORK_INTERFACE:
926 r = interface_pair_parse(&arg_network_interfaces, optarg);
927 if (r < 0)
928 return r;
929
930 arg_private_network = true;
931 arg_settings_mask |= SETTING_NETWORK;
932 break;
933
934 case ARG_NETWORK_MACVLAN:
935 r = macvlan_pair_parse(&arg_network_macvlan, optarg);
936 if (r < 0)
937 return r;
938
939 arg_private_network = true;
940 arg_settings_mask |= SETTING_NETWORK;
941 break;
942
943 case ARG_NETWORK_IPVLAN:
944 r = ipvlan_pair_parse(&arg_network_ipvlan, optarg);
945 if (r < 0)
946 return r;
947
948 _fallthrough_;
949 case ARG_PRIVATE_NETWORK:
950 arg_private_network = true;
951 arg_settings_mask |= SETTING_NETWORK;
952 break;
953
954 case ARG_NETWORK_NAMESPACE_PATH:
955 r = parse_path_argument(optarg, false, &arg_network_namespace_path);
956 if (r < 0)
957 return r;
958
959 arg_settings_mask |= SETTING_NETWORK;
960 break;
961
962 case 'b':
963 if (arg_start_mode == START_PID2)
964 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
965 "--boot and --as-pid2 may not be combined.");
966
967 arg_start_mode = START_BOOT;
968 arg_settings_mask |= SETTING_START_MODE;
969 break;
970
971 case 'a':
972 if (arg_start_mode == START_BOOT)
973 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
974 "--boot and --as-pid2 may not be combined.");
975
976 arg_start_mode = START_PID2;
977 arg_settings_mask |= SETTING_START_MODE;
978 break;
979
980 case ARG_UUID:
981 r = sd_id128_from_string(optarg, &arg_uuid);
982 if (r < 0)
983 return log_error_errno(r, "Invalid UUID: %s", optarg);
984
985 if (sd_id128_is_null(arg_uuid))
986 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
987 "Machine UUID may not be all zeroes.");
988
989 arg_settings_mask |= SETTING_MACHINE_ID;
990 break;
991
992 case 'S': {
993 _cleanup_free_ char *mangled = NULL;
994
995 r = unit_name_mangle_with_suffix(optarg, NULL, UNIT_NAME_MANGLE_WARN, ".slice", &mangled);
996 if (r < 0)
997 return log_oom();
998
999 free_and_replace(arg_slice, mangled);
1000 arg_settings_mask |= SETTING_SLICE;
1001 break;
1002 }
1003
1004 case 'M':
1005 if (isempty(optarg))
1006 arg_machine = mfree(arg_machine);
1007 else {
1008 if (!hostname_is_valid(optarg, 0))
1009 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1010 "Invalid machine name: %s", optarg);
1011
1012 r = free_and_strdup(&arg_machine, optarg);
1013 if (r < 0)
1014 return log_oom();
1015 }
1016 break;
1017
1018 case ARG_HOSTNAME:
1019 if (isempty(optarg))
1020 arg_hostname = mfree(arg_hostname);
1021 else {
1022 if (!hostname_is_valid(optarg, 0))
1023 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1024 "Invalid hostname: %s", optarg);
1025
1026 r = free_and_strdup(&arg_hostname, optarg);
1027 if (r < 0)
1028 return log_oom();
1029 }
1030
1031 arg_settings_mask |= SETTING_HOSTNAME;
1032 break;
1033
1034 case 'Z':
1035 arg_selinux_context = optarg;
1036 break;
1037
1038 case 'L':
1039 arg_selinux_apifs_context = optarg;
1040 break;
1041
1042 case ARG_READ_ONLY:
1043 arg_read_only = true;
1044 arg_settings_mask |= SETTING_READ_ONLY;
1045 break;
1046
1047 case ARG_AMBIENT_CAPABILITY: {
1048 uint64_t m;
1049 r = parse_capability_spec(optarg, &m);
1050 if (r <= 0)
1051 return r;
1052 arg_caps_ambient |= m;
1053 arg_settings_mask |= SETTING_CAPABILITY;
1054 break;
1055 }
1056 case ARG_CAPABILITY:
1057 case ARG_DROP_CAPABILITY: {
1058 uint64_t m;
1059 r = parse_capability_spec(optarg, &m);
1060 if (r <= 0)
1061 return r;
1062
1063 if (c == ARG_CAPABILITY)
1064 plus |= m;
1065 else
1066 minus |= m;
1067 arg_settings_mask |= SETTING_CAPABILITY;
1068 break;
1069 }
1070 case ARG_NO_NEW_PRIVILEGES:
1071 r = parse_boolean(optarg);
1072 if (r < 0)
1073 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
1074
1075 arg_no_new_privileges = r;
1076 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
1077 break;
1078
1079 case 'j':
1080 arg_link_journal = LINK_GUEST;
1081 arg_link_journal_try = true;
1082 arg_settings_mask |= SETTING_LINK_JOURNAL;
1083 break;
1084
1085 case ARG_LINK_JOURNAL:
1086 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
1087 if (r < 0)
1088 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
1089
1090 arg_settings_mask |= SETTING_LINK_JOURNAL;
1091 break;
1092
1093 case ARG_BIND:
1094 case ARG_BIND_RO:
1095 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
1096 if (r < 0)
1097 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
1098
1099 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1100 break;
1101
1102 case ARG_TMPFS:
1103 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1104 if (r < 0)
1105 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
1106
1107 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1108 break;
1109
1110 case ARG_OVERLAY:
1111 case ARG_OVERLAY_RO:
1112 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
1113 if (r == -EADDRNOTAVAIL)
1114 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1115 if (r < 0)
1116 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
1117
1118 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1119 break;
1120
1121 case ARG_INACCESSIBLE:
1122 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1123 if (r < 0)
1124 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
1125
1126 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1127 break;
1128
1129 case 'E':
1130 r = strv_env_replace_strdup_passthrough(&arg_setenv, optarg);
1131 if (r < 0)
1132 return log_error_errno(r, "Cannot assign environment variable %s: %m", optarg);
1133
1134 arg_settings_mask |= SETTING_ENVIRONMENT;
1135 break;
1136
1137 case 'q':
1138 arg_quiet = true;
1139 break;
1140
1141 case ARG_SHARE_SYSTEM:
1142 /* We don't officially support this anymore, except for compat reasons. People should use the
1143 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1144 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1145 arg_clone_ns_flags = 0;
1146 break;
1147
1148 case ARG_REGISTER:
1149 r = parse_boolean(optarg);
1150 if (r < 0) {
1151 log_error("Failed to parse --register= argument: %s", optarg);
1152 return r;
1153 }
1154
1155 arg_register = r;
1156 break;
1157
1158 case ARG_KEEP_UNIT:
1159 arg_keep_unit = true;
1160 break;
1161
1162 case ARG_PERSONALITY:
1163
1164 arg_personality = personality_from_string(optarg);
1165 if (arg_personality == PERSONALITY_INVALID)
1166 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1167 "Unknown or unsupported personality '%s'.", optarg);
1168
1169 arg_settings_mask |= SETTING_PERSONALITY;
1170 break;
1171
1172 case ARG_VOLATILE:
1173
1174 if (!optarg)
1175 arg_volatile_mode = VOLATILE_YES;
1176 else if (streq(optarg, "help")) {
1177 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1178 return 0;
1179 } else {
1180 VolatileMode m;
1181
1182 m = volatile_mode_from_string(optarg);
1183 if (m < 0)
1184 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1185 "Failed to parse --volatile= argument: %s", optarg);
1186 else
1187 arg_volatile_mode = m;
1188 }
1189
1190 arg_settings_mask |= SETTING_VOLATILE_MODE;
1191 break;
1192
1193 case 'p':
1194 r = expose_port_parse(&arg_expose_ports, optarg);
1195 if (r == -EEXIST)
1196 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1197 if (r < 0)
1198 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1199
1200 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1201 break;
1202
1203 case ARG_PROPERTY:
1204 if (strv_extend(&arg_property, optarg) < 0)
1205 return log_oom();
1206
1207 break;
1208
1209 case ARG_PRIVATE_USERS: {
1210 int boolean;
1211
1212 if (!optarg)
1213 boolean = true;
1214 else if (!in_charset(optarg, DIGITS))
1215 /* do *not* parse numbers as booleans */
1216 boolean = parse_boolean(optarg);
1217 else
1218 boolean = -1;
1219
1220 if (boolean == 0) {
1221 /* no: User namespacing off */
1222 arg_userns_mode = USER_NAMESPACE_NO;
1223 arg_uid_shift = UID_INVALID;
1224 arg_uid_range = UINT32_C(0x10000);
1225 } else if (boolean > 0) {
1226 /* yes: User namespacing on, UID range is read from root dir */
1227 arg_userns_mode = USER_NAMESPACE_FIXED;
1228 arg_uid_shift = UID_INVALID;
1229 arg_uid_range = UINT32_C(0x10000);
1230 } else if (streq(optarg, "pick")) {
1231 /* pick: User namespacing on, UID range is picked randomly */
1232 arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
1233 * implied by USER_NAMESPACE_PICK
1234 * further down. */
1235 arg_uid_shift = UID_INVALID;
1236 arg_uid_range = UINT32_C(0x10000);
1237
1238 } else if (streq(optarg, "identity")) {
1239 /* identity: User namespaces on, UID range is map the 0…0xFFFF range to
1240 * itself, i.e. we don't actually map anything, but do take benefit of
1241 * isolation of capability sets. */
1242 arg_userns_mode = USER_NAMESPACE_FIXED;
1243 arg_uid_shift = 0;
1244 arg_uid_range = UINT32_C(0x10000);
1245 } else {
1246 _cleanup_free_ char *buffer = NULL;
1247 const char *range, *shift;
1248
1249 /* anything else: User namespacing on, UID range is explicitly configured */
1250
1251 range = strchr(optarg, ':');
1252 if (range) {
1253 buffer = strndup(optarg, range - optarg);
1254 if (!buffer)
1255 return log_oom();
1256 shift = buffer;
1257
1258 range++;
1259 r = safe_atou32(range, &arg_uid_range);
1260 if (r < 0)
1261 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
1262 } else
1263 shift = optarg;
1264
1265 r = parse_uid(shift, &arg_uid_shift);
1266 if (r < 0)
1267 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
1268
1269 arg_userns_mode = USER_NAMESPACE_FIXED;
1270
1271 if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
1272 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID range cannot be empty or go beyond " UID_FMT ".", UID_INVALID);
1273 }
1274
1275 arg_settings_mask |= SETTING_USERNS;
1276 break;
1277 }
1278
1279 case 'U':
1280 if (userns_supported()) {
1281 arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
1282 * implied by USER_NAMESPACE_PICK
1283 * further down. */
1284 arg_uid_shift = UID_INVALID;
1285 arg_uid_range = UINT32_C(0x10000);
1286
1287 arg_settings_mask |= SETTING_USERNS;
1288 }
1289
1290 break;
1291
1292 case ARG_PRIVATE_USERS_CHOWN:
1293 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
1294
1295 arg_settings_mask |= SETTING_USERNS;
1296 break;
1297
1298 case ARG_PRIVATE_USERS_OWNERSHIP:
1299 if (streq(optarg, "help")) {
1300 DUMP_STRING_TABLE(user_namespace_ownership, UserNamespaceOwnership, _USER_NAMESPACE_OWNERSHIP_MAX);
1301 return 0;
1302 }
1303
1304 arg_userns_ownership = user_namespace_ownership_from_string(optarg);
1305 if (arg_userns_ownership < 0)
1306 return log_error_errno(arg_userns_ownership, "Cannot parse --user-namespace-ownership= value: %s", optarg);
1307
1308 arg_settings_mask |= SETTING_USERNS;
1309 break;
1310
1311 case ARG_KILL_SIGNAL:
1312 if (streq(optarg, "help")) {
1313 DUMP_STRING_TABLE(signal, int, _NSIG);
1314 return 0;
1315 }
1316
1317 arg_kill_signal = signal_from_string(optarg);
1318 if (arg_kill_signal < 0)
1319 return log_error_errno(arg_kill_signal, "Cannot parse signal: %s", optarg);
1320
1321 arg_settings_mask |= SETTING_KILL_SIGNAL;
1322 break;
1323
1324 case ARG_SETTINGS:
1325
1326 /* no → do not read files
1327 * yes → read files, do not override cmdline, trust only subset
1328 * override → read files, override cmdline, trust only subset
1329 * trusted → read files, do not override cmdline, trust all
1330 */
1331
1332 r = parse_boolean(optarg);
1333 if (r < 0) {
1334 if (streq(optarg, "trusted")) {
1335 mask_all_settings = false;
1336 mask_no_settings = false;
1337 arg_settings_trusted = true;
1338
1339 } else if (streq(optarg, "override")) {
1340 mask_all_settings = false;
1341 mask_no_settings = true;
1342 arg_settings_trusted = -1;
1343 } else
1344 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1345 } else if (r > 0) {
1346 /* yes */
1347 mask_all_settings = false;
1348 mask_no_settings = false;
1349 arg_settings_trusted = -1;
1350 } else {
1351 /* no */
1352 mask_all_settings = true;
1353 mask_no_settings = false;
1354 arg_settings_trusted = false;
1355 }
1356
1357 break;
1358
1359 case ARG_CHDIR:
1360 if (!path_is_absolute(optarg))
1361 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1362 "Working directory %s is not an absolute path.", optarg);
1363
1364 r = free_and_strdup(&arg_chdir, optarg);
1365 if (r < 0)
1366 return log_oom();
1367
1368 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1369 break;
1370
1371 case ARG_PIVOT_ROOT:
1372 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1373 if (r < 0)
1374 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1375
1376 arg_settings_mask |= SETTING_PIVOT_ROOT;
1377 break;
1378
1379 case ARG_NOTIFY_READY:
1380 r = parse_boolean(optarg);
1381 if (r < 0)
1382 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1383 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1384 arg_notify_ready = r;
1385 arg_settings_mask |= SETTING_NOTIFY_READY;
1386 break;
1387
1388 case ARG_ROOT_HASH: {
1389 _cleanup_free_ void *k = NULL;
1390 size_t l;
1391
1392 r = unhexmem(optarg, strlen(optarg), &k, &l);
1393 if (r < 0)
1394 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1395 if (l < sizeof(sd_id128_t))
1396 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128-bit long: %s", optarg);
1397
1398 free_and_replace(arg_verity_settings.root_hash, k);
1399 arg_verity_settings.root_hash_size = l;
1400 break;
1401 }
1402
1403 case ARG_ROOT_HASH_SIG: {
1404 char *value;
1405 size_t l;
1406 void *p;
1407
1408 if ((value = startswith(optarg, "base64:"))) {
1409 r = unbase64mem(value, strlen(value), &p, &l);
1410 if (r < 0)
1411 return log_error_errno(r, "Failed to parse root hash signature '%s': %m", optarg);
1412
1413 } else {
1414 r = read_full_file(optarg, (char**) &p, &l);
1415 if (r < 0)
1416 return log_error_errno(r, "Failed parse root hash signature file '%s': %m", optarg);
1417 }
1418
1419 free_and_replace(arg_verity_settings.root_hash_sig, p);
1420 arg_verity_settings.root_hash_sig_size = l;
1421 break;
1422 }
1423
1424 case ARG_VERITY_DATA:
1425 r = parse_path_argument(optarg, false, &arg_verity_settings.data_path);
1426 if (r < 0)
1427 return r;
1428 break;
1429
1430 case ARG_SYSTEM_CALL_FILTER: {
1431 bool negative;
1432 const char *items;
1433
1434 negative = optarg[0] == '~';
1435 items = negative ? optarg + 1 : optarg;
1436
1437 for (;;) {
1438 _cleanup_free_ char *word = NULL;
1439
1440 r = extract_first_word(&items, &word, NULL, 0);
1441 if (r == 0)
1442 break;
1443 if (r == -ENOMEM)
1444 return log_oom();
1445 if (r < 0)
1446 return log_error_errno(r, "Failed to parse system call filter: %m");
1447
1448 if (negative)
1449 r = strv_extend(&arg_syscall_deny_list, word);
1450 else
1451 r = strv_extend(&arg_syscall_allow_list, word);
1452 if (r < 0)
1453 return log_oom();
1454 }
1455
1456 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1457 break;
1458 }
1459
1460 case ARG_RLIMIT: {
1461 const char *eq;
1462 _cleanup_free_ char *name = NULL;
1463 int rl;
1464
1465 if (streq(optarg, "help")) {
1466 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1467 return 0;
1468 }
1469
1470 eq = strchr(optarg, '=');
1471 if (!eq)
1472 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1473 "--rlimit= expects an '=' assignment.");
1474
1475 name = strndup(optarg, eq - optarg);
1476 if (!name)
1477 return log_oom();
1478
1479 rl = rlimit_from_string_harder(name);
1480 if (rl < 0)
1481 return log_error_errno(rl, "Unknown resource limit: %s", name);
1482
1483 if (!arg_rlimit[rl]) {
1484 arg_rlimit[rl] = new0(struct rlimit, 1);
1485 if (!arg_rlimit[rl])
1486 return log_oom();
1487 }
1488
1489 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1490 if (r < 0)
1491 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1492
1493 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1494 break;
1495 }
1496
1497 case ARG_OOM_SCORE_ADJUST:
1498 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1499 if (r < 0)
1500 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1501
1502 arg_oom_score_adjust_set = true;
1503 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1504 break;
1505
1506 case ARG_CPU_AFFINITY: {
1507 CPUSet cpuset;
1508
1509 r = parse_cpu_set(optarg, &cpuset);
1510 if (r < 0)
1511 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
1512
1513 cpu_set_reset(&arg_cpu_set);
1514 arg_cpu_set = cpuset;
1515 arg_settings_mask |= SETTING_CPU_AFFINITY;
1516 break;
1517 }
1518
1519 case ARG_RESOLV_CONF:
1520 if (streq(optarg, "help")) {
1521 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1522 return 0;
1523 }
1524
1525 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1526 if (arg_resolv_conf < 0)
1527 return log_error_errno(arg_resolv_conf,
1528 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1529
1530 arg_settings_mask |= SETTING_RESOLV_CONF;
1531 break;
1532
1533 case ARG_TIMEZONE:
1534 if (streq(optarg, "help")) {
1535 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1536 return 0;
1537 }
1538
1539 arg_timezone = timezone_mode_from_string(optarg);
1540 if (arg_timezone < 0)
1541 return log_error_errno(arg_timezone,
1542 "Failed to parse /etc/localtime mode: %s", optarg);
1543
1544 arg_settings_mask |= SETTING_TIMEZONE;
1545 break;
1546
1547 case ARG_CONSOLE:
1548 r = handle_arg_console(optarg);
1549 if (r <= 0)
1550 return r;
1551 break;
1552
1553 case 'P':
1554 case ARG_PIPE:
1555 r = handle_arg_console("pipe");
1556 if (r <= 0)
1557 return r;
1558 break;
1559
1560 case ARG_NO_PAGER:
1561 arg_pager_flags |= PAGER_DISABLE;
1562 break;
1563
1564 case ARG_SET_CREDENTIAL: {
1565 _cleanup_free_ char *word = NULL, *data = NULL;
1566 const char *p = optarg;
1567 Credential *a;
1568 ssize_t l;
1569
1570 r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1571 if (r == -ENOMEM)
1572 return log_oom();
1573 if (r < 0)
1574 return log_error_errno(r, "Failed to parse --set-credential= parameter: %m");
1575 if (r == 0 || !p)
1576 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --set-credential=: %s", optarg);
1577
1578 if (!credential_name_valid(word))
1579 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
1580
1581 for (size_t i = 0; i < arg_n_credentials; i++)
1582 if (streq(arg_credentials[i].id, word))
1583 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
1584
1585 l = cunescape(p, UNESCAPE_ACCEPT_NUL, &data);
1586 if (l < 0)
1587 return log_error_errno(l, "Failed to unescape credential data: %s", p);
1588
1589 a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
1590 if (!a)
1591 return log_oom();
1592
1593 a[arg_n_credentials++] = (Credential) {
1594 .id = TAKE_PTR(word),
1595 .data = TAKE_PTR(data),
1596 .size = l,
1597 };
1598
1599 arg_credentials = a;
1600
1601 arg_settings_mask |= SETTING_CREDENTIALS;
1602 break;
1603 }
1604
1605 case ARG_LOAD_CREDENTIAL: {
1606 ReadFullFileFlags flags = READ_FULL_FILE_SECURE;
1607 _cleanup_(erase_and_freep) char *data = NULL;
1608 _cleanup_free_ char *word = NULL, *j = NULL;
1609 const char *p = optarg;
1610 Credential *a;
1611 size_t size, i;
1612
1613 r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1614 if (r == -ENOMEM)
1615 return log_oom();
1616 if (r < 0)
1617 return log_error_errno(r, "Failed to parse --load-credential= parameter: %m");
1618 if (r == 0 || !p)
1619 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --load-credential=: %s", optarg);
1620
1621 if (!credential_name_valid(word))
1622 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
1623
1624 for (i = 0; i < arg_n_credentials; i++)
1625 if (streq(arg_credentials[i].id, word))
1626 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
1627
1628 if (path_is_absolute(p))
1629 flags |= READ_FULL_FILE_CONNECT_SOCKET;
1630 else {
1631 const char *e;
1632
1633 r = get_credentials_dir(&e);
1634 if (r < 0)
1635 return log_error_errno(r, "Credential not available (no credentials passed at all): %s", word);
1636
1637 j = path_join(e, p);
1638 if (!j)
1639 return log_oom();
1640 }
1641
1642 r = read_full_file_full(AT_FDCWD, j ?: p, UINT64_MAX, SIZE_MAX,
1643 flags,
1644 NULL,
1645 &data, &size);
1646 if (r < 0)
1647 return log_error_errno(r, "Failed to read credential '%s': %m", j ?: p);
1648
1649 a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
1650 if (!a)
1651 return log_oom();
1652
1653 a[arg_n_credentials++] = (Credential) {
1654 .id = TAKE_PTR(word),
1655 .data = TAKE_PTR(data),
1656 .size = size,
1657 };
1658
1659 arg_credentials = a;
1660
1661 arg_settings_mask |= SETTING_CREDENTIALS;
1662 break;
1663 }
1664
1665 case ARG_BIND_USER:
1666 if (!valid_user_group_name(optarg, 0))
1667 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid user name to bind: %s", optarg);
1668
1669 if (strv_extend(&arg_bind_user, optarg) < 0)
1670 return log_oom();
1671
1672 arg_settings_mask |= SETTING_BIND_USER;
1673 break;
1674
1675 case ARG_SUPPRESS_SYNC:
1676 r = parse_boolean_argument("--suppress-sync=", optarg, &arg_suppress_sync);
1677 if (r < 0)
1678 return r;
1679
1680 arg_settings_mask |= SETTING_SUPPRESS_SYNC;
1681 break;
1682
1683 case ARG_IMAGE_POLICY:
1684 r = parse_image_policy_argument(optarg, &arg_image_policy);
1685 if (r < 0)
1686 return r;
1687 break;
1688
1689 case '?':
1690 return -EINVAL;
1691
1692 default:
1693 assert_not_reached();
1694 }
1695
1696 if (argc > optind) {
1697 strv_free(arg_parameters);
1698 arg_parameters = strv_copy(argv + optind);
1699 if (!arg_parameters)
1700 return log_oom();
1701
1702 arg_settings_mask |= SETTING_START_MODE;
1703 }
1704
1705 if (arg_ephemeral && arg_template && !arg_directory)
1706 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1707 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1708 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1709 * --directory=". */
1710 arg_directory = TAKE_PTR(arg_template);
1711
1712 arg_caps_retain |= plus;
1713 arg_caps_retain |= arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0;
1714
1715 /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
1716 * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
1717 * indicate that. */
1718 if (!arg_private_network && arg_userns_mode != USER_NAMESPACE_NO && arg_uid_shift > 0)
1719 arg_caps_retain &= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE);
1720
1721 arg_caps_retain &= ~minus;
1722
1723 /* Make sure to parse environment before we reset the settings mask below */
1724 r = parse_environment();
1725 if (r < 0)
1726 return r;
1727
1728 /* Load all settings from .nspawn files */
1729 if (mask_no_settings)
1730 arg_settings_mask = 0;
1731
1732 /* Don't load any settings from .nspawn files */
1733 if (mask_all_settings)
1734 arg_settings_mask = _SETTINGS_MASK_ALL;
1735
1736 return 1;
1737 }
1738
1739 static int verify_arguments(void) {
1740 int r;
1741
1742 if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
1743 /* If we are running the stub init in the container, we don't need to look at what the init
1744 * in the container supports, because we are not using it. Let's immediately pick the right
1745 * setting based on the host system configuration.
1746 *
1747 * We only do this, if the user didn't use an environment variable to override the detection.
1748 */
1749
1750 r = cg_all_unified();
1751 if (r < 0)
1752 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
1753 if (r > 0)
1754 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
1755 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
1756 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
1757 else
1758 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
1759 }
1760
1761 if (arg_userns_mode != USER_NAMESPACE_NO)
1762 arg_mount_settings |= MOUNT_USE_USERNS;
1763
1764 if (arg_private_network)
1765 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1766
1767 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1768 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1769 arg_register = false;
1770 if (arg_start_mode != START_PID1)
1771 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1772 }
1773
1774 if (arg_userns_ownership < 0)
1775 arg_userns_ownership =
1776 arg_userns_mode == USER_NAMESPACE_PICK ? USER_NAMESPACE_OWNERSHIP_AUTO :
1777 USER_NAMESPACE_OWNERSHIP_OFF;
1778
1779 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1780 arg_kill_signal = SIGRTMIN+3;
1781
1782 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 */
1783 arg_read_only = true;
1784
1785 if (has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts))
1786 arg_read_only = true;
1787
1788 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1789 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1790 * The latter is not technically a user session, but we don't need to labour the point. */
1791 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1792
1793 if (arg_directory && arg_image)
1794 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1795
1796 if (arg_template && arg_image)
1797 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1798
1799 if (arg_template && !(arg_directory || arg_machine))
1800 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1801
1802 if (arg_ephemeral && arg_template)
1803 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1804
1805 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
1806 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
1807
1808 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1809 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1810
1811 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_read_only)
1812 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1813 "--read-only and --private-users-ownership=chown may not be combined.");
1814
1815 /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
1816 * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
1817 * massive copy-up (in case of overlay) making the entire exercise pointless. */
1818 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_volatile_mode != VOLATILE_NO)
1819 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-ownership=chown may not be combined.");
1820
1821 /* If --network-namespace-path is given with any other network-related option (except --private-network),
1822 * we need to error out, to avoid conflicts between different network options. */
1823 if (arg_network_namespace_path &&
1824 (arg_network_interfaces || arg_network_macvlan ||
1825 arg_network_ipvlan || arg_network_veth_extra ||
1826 arg_network_bridge || arg_network_zone ||
1827 arg_network_veth))
1828 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1829
1830 if (arg_network_bridge && arg_network_zone)
1831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1832 "--network-bridge= and --network-zone= may not be combined.");
1833
1834 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1835 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1836
1837 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1838 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1839
1840 if (arg_expose_ports && !arg_private_network)
1841 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1842
1843 if (arg_caps_ambient) {
1844 if (arg_caps_ambient == UINT64_MAX)
1845 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= does not support the value all.");
1846
1847 if ((arg_caps_ambient & arg_caps_retain) != arg_caps_ambient)
1848 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not fully covered by Capability= setting.");
1849
1850 if (arg_start_mode == START_BOOT)
1851 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not useful for boot mode.");
1852 }
1853
1854 if (arg_userns_mode == USER_NAMESPACE_NO && !strv_isempty(arg_bind_user))
1855 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--bind-user= requires --private-users");
1856
1857 /* Drop duplicate --bind-user= entries */
1858 strv_uniq(arg_bind_user);
1859
1860 r = custom_mount_check_all();
1861 if (r < 0)
1862 return r;
1863
1864 return 0;
1865 }
1866
1867 static int verify_network_interfaces_initialized(void) {
1868 int r;
1869 r = test_network_interfaces_initialized(arg_network_interfaces);
1870 if (r < 0)
1871 return r;
1872
1873 r = test_network_interfaces_initialized(arg_network_macvlan);
1874 if (r < 0)
1875 return r;
1876
1877 r = test_network_interfaces_initialized(arg_network_ipvlan);
1878 if (r < 0)
1879 return r;
1880
1881 return 0;
1882 }
1883
1884 int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1885 assert(p);
1886
1887 if (arg_userns_mode == USER_NAMESPACE_NO)
1888 return 0;
1889
1890 if (uid == UID_INVALID && gid == GID_INVALID)
1891 return 0;
1892
1893 if (uid != UID_INVALID) {
1894 uid += arg_uid_shift;
1895
1896 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1897 return -EOVERFLOW;
1898 }
1899
1900 if (gid != GID_INVALID) {
1901 gid += (gid_t) arg_uid_shift;
1902
1903 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1904 return -EOVERFLOW;
1905 }
1906
1907 return RET_NERRNO(lchown(p, uid, gid));
1908 }
1909
1910 int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1911 const char *q;
1912 int r;
1913
1914 q = prefix_roota(root, path);
1915 r = RET_NERRNO(mkdir(q, mode));
1916 if (r == -EEXIST)
1917 return 0;
1918 if (r < 0)
1919 return r;
1920
1921 return userns_lchown(q, uid, gid);
1922 }
1923
1924 static const char *timezone_from_path(const char *path) {
1925 return PATH_STARTSWITH_SET(
1926 path,
1927 "../usr/share/zoneinfo/",
1928 "/usr/share/zoneinfo/");
1929 }
1930
1931 static bool etc_writable(void) {
1932 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1933 }
1934
1935 static int setup_timezone(const char *dest) {
1936 _cleanup_free_ char *p = NULL, *etc = NULL;
1937 const char *where, *check;
1938 TimezoneMode m;
1939 int r;
1940
1941 assert(dest);
1942
1943 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1944 r = readlink_malloc("/etc/localtime", &p);
1945 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1946 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1947 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1948 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1949 else if (r < 0) {
1950 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1951 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1952 * file.
1953 *
1954 * Example:
1955 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1956 */
1957 return 0;
1958 } else if (arg_timezone == TIMEZONE_AUTO)
1959 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1960 else
1961 m = arg_timezone;
1962 } else
1963 m = arg_timezone;
1964
1965 if (m == TIMEZONE_OFF)
1966 return 0;
1967
1968 r = chase("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1969 if (r < 0) {
1970 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1971 return 0;
1972 }
1973
1974 where = strjoina(etc, "/localtime");
1975
1976 switch (m) {
1977
1978 case TIMEZONE_DELETE:
1979 if (unlink(where) < 0)
1980 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1981
1982 return 0;
1983
1984 case TIMEZONE_SYMLINK: {
1985 _cleanup_free_ char *q = NULL;
1986 const char *z, *what;
1987
1988 z = timezone_from_path(p);
1989 if (!z) {
1990 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1991 return 0;
1992 }
1993
1994 r = readlink_malloc(where, &q);
1995 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1996 return 0; /* Already pointing to the right place? Then do nothing .. */
1997
1998 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1999 r = chase(check, dest, 0, NULL, NULL);
2000 if (r < 0)
2001 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
2002 else {
2003 if (unlink(where) < 0 && errno != ENOENT) {
2004 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
2005 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
2006 return 0;
2007 }
2008
2009 what = strjoina("../usr/share/zoneinfo/", z);
2010 if (symlink(what, where) < 0) {
2011 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
2012 errno, "Failed to correct timezone of container, ignoring: %m");
2013 return 0;
2014 }
2015
2016 break;
2017 }
2018
2019 _fallthrough_;
2020 }
2021
2022 case TIMEZONE_BIND: {
2023 _cleanup_free_ char *resolved = NULL;
2024 int found;
2025
2026 found = chase(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
2027 if (found < 0) {
2028 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
2029 return 0;
2030 }
2031
2032 if (found == 0) /* missing? */
2033 (void) touch(resolved);
2034
2035 r = mount_nofollow_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
2036 if (r >= 0)
2037 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
2038
2039 _fallthrough_;
2040 }
2041
2042 case TIMEZONE_COPY:
2043 /* If mounting failed, try to copy */
2044 r = copy_file_atomic("/etc/localtime", where, 0644, COPY_REFLINK|COPY_REPLACE);
2045 if (r < 0) {
2046 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
2047 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
2048 return 0;
2049 }
2050
2051 break;
2052
2053 default:
2054 assert_not_reached();
2055 }
2056
2057 /* Fix permissions of the symlink or file copy we just created */
2058 r = userns_lchown(where, 0, 0);
2059 if (r < 0)
2060 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
2061
2062 return 0;
2063 }
2064
2065 static int have_resolv_conf(const char *path) {
2066 assert(path);
2067
2068 if (access(path, F_OK) < 0) {
2069 if (errno == ENOENT)
2070 return 0;
2071
2072 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
2073 }
2074
2075 return 1;
2076 }
2077
2078 static int resolved_listening(void) {
2079 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2080 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
2081 _cleanup_free_ char *dns_stub_listener_mode = NULL;
2082 int r;
2083
2084 /* Check if resolved is listening */
2085
2086 r = sd_bus_open_system(&bus);
2087 if (r < 0)
2088 return log_debug_errno(r, "Failed to open system bus: %m");
2089
2090 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
2091 if (r < 0)
2092 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
2093 if (r == 0)
2094 return 0;
2095
2096 r = bus_get_property_string(bus, bus_resolve_mgr, "DNSStubListener", &error, &dns_stub_listener_mode);
2097 if (r < 0)
2098 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
2099
2100 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
2101 }
2102
2103 static int setup_resolv_conf(const char *dest) {
2104 _cleanup_free_ char *etc = NULL;
2105 const char *where, *what;
2106 ResolvConfMode m;
2107 int r;
2108
2109 assert(dest);
2110
2111 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
2112 if (arg_private_network)
2113 m = RESOLV_CONF_OFF;
2114 else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF) > 0 && resolved_listening() > 0)
2115 m = etc_writable() ? RESOLV_CONF_COPY_STUB : RESOLV_CONF_BIND_STUB;
2116 else if (have_resolv_conf("/etc/resolv.conf") > 0)
2117 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
2118 else
2119 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
2120
2121 } else
2122 m = arg_resolv_conf;
2123
2124 if (m == RESOLV_CONF_OFF)
2125 return 0;
2126
2127 r = chase("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
2128 if (r < 0) {
2129 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
2130 return 0;
2131 }
2132
2133 where = strjoina(etc, "/resolv.conf");
2134
2135 if (m == RESOLV_CONF_DELETE) {
2136 if (unlink(where) < 0)
2137 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
2138
2139 return 0;
2140 }
2141
2142 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_COPY_STATIC))
2143 what = PRIVATE_STATIC_RESOLV_CONF;
2144 else if (IN_SET(m, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_COPY_UPLINK))
2145 what = PRIVATE_UPLINK_RESOLV_CONF;
2146 else if (IN_SET(m, RESOLV_CONF_BIND_STUB, RESOLV_CONF_REPLACE_STUB, RESOLV_CONF_COPY_STUB))
2147 what = PRIVATE_STUB_RESOLV_CONF;
2148 else
2149 what = "/etc/resolv.conf";
2150
2151 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_BIND_STUB)) {
2152 _cleanup_free_ char *resolved = NULL;
2153 int found;
2154
2155 found = chase(where, dest, CHASE_NONEXISTENT|CHASE_NOFOLLOW, &resolved, NULL);
2156 if (found < 0) {
2157 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
2158 return 0;
2159 }
2160
2161 if (found == 0) /* missing? */
2162 (void) touch(resolved);
2163
2164 r = mount_nofollow_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
2165 if (r >= 0)
2166 return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
2167
2168 /* If that didn't work, let's copy the file */
2169 }
2170
2171 if (IN_SET(m, RESOLV_CONF_REPLACE_HOST, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_REPLACE_STUB))
2172 r = copy_file_atomic(what, where, 0644, COPY_REFLINK|COPY_REPLACE);
2173 else
2174 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, COPY_REFLINK);
2175 if (r < 0) {
2176 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
2177 * resolved or something similar runs inside and the symlink points there.
2178 *
2179 * If the disk image is read-only, there's also no point in complaining.
2180 */
2181 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC, RESOLV_CONF_COPY_UPLINK, RESOLV_CONF_COPY_STUB) &&
2182 IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
2183 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
2184 return 0;
2185 }
2186
2187 r = userns_lchown(where, 0, 0);
2188 if (r < 0)
2189 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
2190
2191 return 0;
2192 }
2193
2194 static int setup_boot_id(void) {
2195 _cleanup_(unlink_and_freep) char *from = NULL;
2196 _cleanup_free_ char *path = NULL;
2197 sd_id128_t rnd = SD_ID128_NULL;
2198 const char *to;
2199 int r;
2200
2201 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
2202
2203 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
2204 if (r < 0)
2205 return log_error_errno(r, "Failed to generate random boot ID path: %m");
2206
2207 r = sd_id128_randomize(&rnd);
2208 if (r < 0)
2209 return log_error_errno(r, "Failed to generate random boot id: %m");
2210
2211 r = id128_write(path, ID128_FORMAT_UUID, rnd);
2212 if (r < 0)
2213 return log_error_errno(r, "Failed to write boot id: %m");
2214
2215 from = TAKE_PTR(path);
2216 to = "/proc/sys/kernel/random/boot_id";
2217
2218 r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
2219 if (r < 0)
2220 return r;
2221
2222 return mount_nofollow_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2223 }
2224
2225 static int copy_devnodes(const char *dest) {
2226 static const char devnodes[] =
2227 "null\0"
2228 "zero\0"
2229 "full\0"
2230 "random\0"
2231 "urandom\0"
2232 "tty\0"
2233 "net/tun\0";
2234
2235 int r = 0;
2236
2237 assert(dest);
2238
2239 BLOCK_WITH_UMASK(0000);
2240
2241 /* Create /dev/net, so that we can create /dev/net/tun in it */
2242 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
2243 return log_error_errno(r, "Failed to create /dev/net directory: %m");
2244
2245 NULSTR_FOREACH(d, devnodes) {
2246 _cleanup_free_ char *from = NULL, *to = NULL;
2247 struct stat st;
2248
2249 from = path_join("/dev/", d);
2250 if (!from)
2251 return log_oom();
2252
2253 to = path_join(dest, from);
2254 if (!to)
2255 return log_oom();
2256
2257 if (stat(from, &st) < 0) {
2258
2259 if (errno != ENOENT)
2260 return log_error_errno(errno, "Failed to stat %s: %m", from);
2261
2262 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
2263 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2264 "%s is not a char or block device, cannot copy.", from);
2265 else {
2266 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
2267
2268 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
2269 /* Explicitly warn the user when /dev is already populated. */
2270 if (errno == EEXIST)
2271 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);
2272 if (errno != EPERM)
2273 return log_error_errno(errno, "mknod(%s) failed: %m", to);
2274
2275 /* Some systems abusively restrict mknod but allow bind mounts. */
2276 r = touch(to);
2277 if (r < 0)
2278 return log_error_errno(r, "touch (%s) failed: %m", to);
2279 r = mount_nofollow_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
2280 if (r < 0)
2281 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
2282 }
2283
2284 r = userns_lchown(to, 0, 0);
2285 if (r < 0)
2286 return log_error_errno(r, "chown() of device node %s failed: %m", to);
2287
2288 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
2289 if (!dn)
2290 return log_oom();
2291
2292 r = userns_mkdir(dest, dn, 0755, 0, 0);
2293 if (r < 0)
2294 return log_error_errno(r, "Failed to create '%s': %m", dn);
2295
2296 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
2297 return log_oom();
2298
2299 prefixed = path_join(dest, sl);
2300 if (!prefixed)
2301 return log_oom();
2302
2303 t = path_join("..", d);
2304 if (!t)
2305 return log_oom();
2306
2307 if (symlink(t, prefixed) < 0)
2308 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
2309 }
2310 }
2311
2312 return r;
2313 }
2314
2315 static int make_extra_nodes(const char *dest) {
2316 size_t i;
2317 int r;
2318
2319 BLOCK_WITH_UMASK(0000);
2320
2321 for (i = 0; i < arg_n_extra_nodes; i++) {
2322 _cleanup_free_ char *path = NULL;
2323 DeviceNode *n = arg_extra_nodes + i;
2324
2325 path = path_join(dest, n->path);
2326 if (!path)
2327 return log_oom();
2328
2329 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
2330 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
2331
2332 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
2333 if (r < 0)
2334 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
2335 }
2336
2337 return 0;
2338 }
2339
2340 static int setup_pts(const char *dest) {
2341 _cleanup_free_ char *options = NULL;
2342 const char *p;
2343 int r;
2344
2345 #if HAVE_SELINUX
2346 if (arg_selinux_apifs_context)
2347 (void) asprintf(&options,
2348 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
2349 arg_uid_shift + TTY_GID,
2350 arg_selinux_apifs_context);
2351 else
2352 #endif
2353 (void) asprintf(&options,
2354 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
2355 arg_uid_shift + TTY_GID);
2356
2357 if (!options)
2358 return log_oom();
2359
2360 /* Mount /dev/pts itself */
2361 p = prefix_roota(dest, "/dev/pts");
2362 r = RET_NERRNO(mkdir(p, 0755));
2363 if (r < 0)
2364 return log_error_errno(r, "Failed to create /dev/pts: %m");
2365
2366 r = mount_nofollow_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2367 if (r < 0)
2368 return r;
2369 r = userns_lchown(p, 0, 0);
2370 if (r < 0)
2371 return log_error_errno(r, "Failed to chown /dev/pts: %m");
2372
2373 /* Create /dev/ptmx symlink */
2374 p = prefix_roota(dest, "/dev/ptmx");
2375 if (symlink("pts/ptmx", p) < 0)
2376 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
2377 r = userns_lchown(p, 0, 0);
2378 if (r < 0)
2379 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
2380
2381 /* And fix /dev/pts/ptmx ownership */
2382 p = prefix_roota(dest, "/dev/pts/ptmx");
2383 r = userns_lchown(p, 0, 0);
2384 if (r < 0)
2385 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
2386
2387 return 0;
2388 }
2389
2390 static int setup_stdio_as_dev_console(void) {
2391 _cleanup_close_ int terminal = -EBADF;
2392 int r;
2393
2394 /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
2395 * explicitly, if we are configured to. */
2396 terminal = open_terminal("/dev/console", O_RDWR|O_NOCTTY);
2397 if (terminal < 0)
2398 return log_error_errno(terminal, "Failed to open console: %m");
2399
2400 /* Make sure we can continue logging to the original stderr, even if
2401 * stderr points elsewhere now */
2402 r = log_dup_console();
2403 if (r < 0)
2404 return log_error_errno(r, "Failed to duplicate stderr: %m");
2405
2406 /* invalidates 'terminal' on success and failure */
2407 r = rearrange_stdio(terminal, terminal, terminal);
2408 TAKE_FD(terminal);
2409 if (r < 0)
2410 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2411
2412 return 0;
2413 }
2414
2415 static int setup_dev_console(const char *console) {
2416 _cleanup_free_ char *p = NULL;
2417 int r;
2418
2419 /* Create /dev/console symlink */
2420 r = path_make_relative("/dev", console, &p);
2421 if (r < 0)
2422 return log_error_errno(r, "Failed to create relative path: %m");
2423
2424 if (symlink(p, "/dev/console") < 0)
2425 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
2426
2427 return 0;
2428 }
2429
2430 static int setup_keyring(void) {
2431 key_serial_t keyring;
2432
2433 /* Allocate a new session keyring for the container. This makes sure the keyring of the session
2434 * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
2435 * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
2436 * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
2437 * into the container. */
2438
2439 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2440 if (keyring == -1) {
2441 if (errno == ENOSYS)
2442 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2443 else if (ERRNO_IS_PRIVILEGE(errno))
2444 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2445 else
2446 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2447 }
2448
2449 return 0;
2450 }
2451
2452 static int setup_credentials(const char *root) {
2453 const char *q;
2454 int r;
2455
2456 if (arg_n_credentials <= 0)
2457 return 0;
2458
2459 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2460 if (r < 0)
2461 return log_error_errno(r, "Failed to create /run/host: %m");
2462
2463 r = userns_mkdir(root, "/run/host/credentials", 0700, 0, 0);
2464 if (r < 0)
2465 return log_error_errno(r, "Failed to create /run/host/credentials: %m");
2466
2467 q = prefix_roota(root, "/run/host/credentials");
2468 r = mount_nofollow_verbose(LOG_ERR, NULL, q, "ramfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0700");
2469 if (r < 0)
2470 return r;
2471
2472 for (size_t i = 0; i < arg_n_credentials; i++) {
2473 _cleanup_free_ char *j = NULL;
2474 _cleanup_close_ int fd = -EBADF;
2475
2476 j = path_join(q, arg_credentials[i].id);
2477 if (!j)
2478 return log_oom();
2479
2480 fd = open(j, O_CREAT|O_EXCL|O_WRONLY|O_CLOEXEC|O_NOFOLLOW, 0600);
2481 if (fd < 0)
2482 return log_error_errno(errno, "Failed to create credential file %s: %m", j);
2483
2484 r = loop_write(fd, arg_credentials[i].data, arg_credentials[i].size, /* do_poll= */ false);
2485 if (r < 0)
2486 return log_error_errno(r, "Failed to write credential to file %s: %m", j);
2487
2488 if (fchmod(fd, 0400) < 0)
2489 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", j);
2490
2491 if (arg_userns_mode != USER_NAMESPACE_NO) {
2492 if (fchown(fd, arg_uid_shift, arg_uid_shift) < 0)
2493 return log_error_errno(errno, "Failed to adjust ownership of %s: %m", j);
2494 }
2495 }
2496
2497 if (chmod(q, 0500) < 0)
2498 return log_error_errno(errno, "Failed to adjust access mode of %s: %m", q);
2499
2500 r = userns_lchown(q, 0, 0);
2501 if (r < 0)
2502 return r;
2503
2504 /* Make both mount and superblock read-only now */
2505 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
2506 if (r < 0)
2507 return r;
2508
2509 return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0500");
2510 }
2511
2512 static int setup_kmsg(int fd_inner_socket) {
2513 _cleanup_(unlink_and_freep) char *from = NULL;
2514 _cleanup_free_ char *fifo = NULL;
2515 _cleanup_close_ int fd = -EBADF;
2516 int r;
2517
2518 assert(fd_inner_socket >= 0);
2519
2520 BLOCK_WITH_UMASK(0000);
2521
2522 /* We create the kmsg FIFO as a temporary file in /run, but immediately delete it after bind mounting it to
2523 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2524 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2525 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2526
2527 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2528 if (r < 0)
2529 return log_error_errno(r, "Failed to generate kmsg path: %m");
2530
2531 if (mkfifo(fifo, 0600) < 0)
2532 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2533
2534 from = TAKE_PTR(fifo);
2535
2536 r = mount_nofollow_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2537 if (r < 0)
2538 return r;
2539
2540 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2541 if (fd < 0)
2542 return log_error_errno(errno, "Failed to open fifo: %m");
2543
2544 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2545 r = send_one_fd(fd_inner_socket, fd, 0);
2546 if (r < 0)
2547 return log_error_errno(r, "Failed to send FIFO fd: %m");
2548
2549 return 0;
2550 }
2551
2552 struct ExposeArgs {
2553 union in_addr_union address4;
2554 union in_addr_union address6;
2555 struct FirewallContext *fw_ctx;
2556 };
2557
2558 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2559 struct ExposeArgs *args = ASSERT_PTR(userdata);
2560
2561 assert(rtnl);
2562 assert(m);
2563
2564 (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET, &args->address4);
2565 (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET6, &args->address6);
2566 return 0;
2567 }
2568
2569 static int setup_hostname(void) {
2570 int r;
2571
2572 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2573 return 0;
2574
2575 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2576 if (r < 0)
2577 return log_error_errno(r, "Failed to set hostname: %m");
2578
2579 return 0;
2580 }
2581
2582 static int setup_journal(const char *directory) {
2583 _cleanup_free_ char *d = NULL;
2584 const char *p, *q;
2585 sd_id128_t this_id;
2586 bool try;
2587 int r;
2588
2589 /* Don't link journals in ephemeral mode */
2590 if (arg_ephemeral)
2591 return 0;
2592
2593 if (arg_link_journal == LINK_NO)
2594 return 0;
2595
2596 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2597
2598 r = sd_id128_get_machine(&this_id);
2599 if (r < 0)
2600 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2601
2602 if (sd_id128_equal(arg_uuid, this_id)) {
2603 log_full(try ? LOG_WARNING : LOG_ERR,
2604 "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid));
2605 if (try)
2606 return 0;
2607 return -EEXIST;
2608 }
2609
2610 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2611 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2612 if (r < 0) {
2613 bool ignore = r == -EROFS && try;
2614 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2615 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2616 return ignore ? 0 : r;
2617 }
2618 }
2619
2620 p = strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid));
2621 q = prefix_roota(directory, p);
2622
2623 if (path_is_mount_point(p, NULL, 0) > 0) {
2624 if (try)
2625 return 0;
2626
2627 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2628 "%s: already a mount point, refusing to use for journal", p);
2629 }
2630
2631 if (path_is_mount_point(q, NULL, 0) > 0) {
2632 if (try)
2633 return 0;
2634
2635 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2636 "%s: already a mount point, refusing to use for journal", q);
2637 }
2638
2639 r = readlink_and_make_absolute(p, &d);
2640 if (r >= 0) {
2641 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2642 path_equal(d, q)) {
2643
2644 r = userns_mkdir(directory, p, 0755, 0, 0);
2645 if (r < 0)
2646 log_warning_errno(r, "Failed to create directory %s: %m", q);
2647 return 0;
2648 }
2649
2650 if (unlink(p) < 0)
2651 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2652 } else if (r == -EINVAL) {
2653
2654 if (arg_link_journal == LINK_GUEST &&
2655 rmdir(p) < 0) {
2656
2657 if (errno == ENOTDIR) {
2658 log_error("%s already exists and is neither a symlink nor a directory", p);
2659 return r;
2660 } else
2661 return log_error_errno(errno, "Failed to remove %s: %m", p);
2662 }
2663 } else if (r != -ENOENT)
2664 return log_error_errno(r, "readlink(%s) failed: %m", p);
2665
2666 if (arg_link_journal == LINK_GUEST) {
2667
2668 if (symlink(q, p) < 0) {
2669 if (try) {
2670 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2671 return 0;
2672 } else
2673 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2674 }
2675
2676 r = userns_mkdir(directory, p, 0755, 0, 0);
2677 if (r < 0)
2678 log_warning_errno(r, "Failed to create directory %s: %m", q);
2679 return 0;
2680 }
2681
2682 if (arg_link_journal == LINK_HOST) {
2683 /* don't create parents here — if the host doesn't have
2684 * permanent journal set up, don't force it here */
2685
2686 r = RET_NERRNO(mkdir(p, 0755));
2687 if (r < 0 && r != -EEXIST) {
2688 if (try) {
2689 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2690 return 0;
2691 } else
2692 return log_error_errno(r, "Failed to create %s: %m", p);
2693 }
2694
2695 } else if (access(p, F_OK) < 0)
2696 return 0;
2697
2698 if (dir_is_empty(q, /* ignore_hidden_or_backup= */ false) == 0)
2699 log_warning("%s is not empty, proceeding anyway.", q);
2700
2701 r = userns_mkdir(directory, p, 0755, 0, 0);
2702 if (r < 0)
2703 return log_error_errno(r, "Failed to create %s: %m", q);
2704
2705 r = mount_nofollow_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2706 if (r < 0)
2707 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2708
2709 return 0;
2710 }
2711
2712 static int drop_capabilities(uid_t uid) {
2713 CapabilityQuintet q;
2714
2715 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2716 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2717 * arg_caps_retain. */
2718
2719 if (capability_quintet_is_set(&arg_full_capabilities)) {
2720 q = arg_full_capabilities;
2721
2722 if (q.bounding == UINT64_MAX)
2723 q.bounding = uid == 0 ? arg_caps_retain : 0;
2724
2725 if (q.effective == UINT64_MAX)
2726 q.effective = uid == 0 ? q.bounding : 0;
2727
2728 if (q.inheritable == UINT64_MAX)
2729 q.inheritable = uid == 0 ? q.bounding : arg_caps_ambient;
2730
2731 if (q.permitted == UINT64_MAX)
2732 q.permitted = uid == 0 ? q.bounding : arg_caps_ambient;
2733
2734 if (q.ambient == UINT64_MAX && ambient_capabilities_supported())
2735 q.ambient = arg_caps_ambient;
2736
2737 if (capability_quintet_mangle(&q))
2738 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2739
2740 } else {
2741 q = (CapabilityQuintet) {
2742 .bounding = arg_caps_retain,
2743 .effective = uid == 0 ? arg_caps_retain : 0,
2744 .inheritable = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2745 .permitted = uid == 0 ? arg_caps_retain : arg_caps_ambient,
2746 .ambient = ambient_capabilities_supported() ? arg_caps_ambient : UINT64_MAX,
2747 };
2748
2749 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2750 * in order to maintain the same behavior as systemd < 242. */
2751 if (capability_quintet_mangle(&q))
2752 log_full(arg_quiet ? LOG_DEBUG : LOG_WARNING,
2753 "Some capabilities will not be set because they are not in the current bounding set.");
2754
2755 }
2756
2757 return capability_quintet_enforce(&q);
2758 }
2759
2760 static int reset_audit_loginuid(void) {
2761 _cleanup_free_ char *p = NULL;
2762 int r;
2763
2764 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2765 return 0;
2766
2767 r = read_one_line_file("/proc/self/loginuid", &p);
2768 if (r == -ENOENT)
2769 return 0;
2770 if (r < 0)
2771 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2772
2773 /* Already reset? */
2774 if (streq(p, "4294967295"))
2775 return 0;
2776
2777 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2778 if (r < 0) {
2779 log_error_errno(r,
2780 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2781 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2782 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2783 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2784 "using systemd-nspawn. Sleeping for 5s... (%m)");
2785
2786 sleep(5);
2787 }
2788
2789 return 0;
2790 }
2791
2792 static int mount_tunnel_dig(const char *root) {
2793 const char *p, *q;
2794 int r;
2795
2796 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2797 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2798 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2799 (void) mkdir_p(p, 0600);
2800
2801 r = userns_mkdir(root, "/run/host", 0755, 0, 0);
2802 if (r < 0)
2803 return log_error_errno(r, "Failed to create /run/host: %m");
2804
2805 r = userns_mkdir(root, NSPAWN_MOUNT_TUNNEL, 0600, 0, 0);
2806 if (r < 0)
2807 return log_error_errno(r, "Failed to create "NSPAWN_MOUNT_TUNNEL": %m");
2808
2809 q = prefix_roota(root, NSPAWN_MOUNT_TUNNEL);
2810 r = mount_nofollow_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2811 if (r < 0)
2812 return r;
2813
2814 r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2815 if (r < 0)
2816 return r;
2817
2818 return 0;
2819 }
2820
2821 static int mount_tunnel_open(void) {
2822 int r;
2823
2824 r = mount_follow_verbose(LOG_ERR, NULL, NSPAWN_MOUNT_TUNNEL, NULL, MS_SLAVE, NULL);
2825 if (r < 0)
2826 return r;
2827
2828 return 0;
2829 }
2830
2831 static int setup_machine_id(const char *directory) {
2832 int r;
2833
2834 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2835 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2836 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2837 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2838 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2839 * container behaves nicely). */
2840
2841 r = id128_get_machine(directory, &arg_uuid);
2842 if (r < 0) {
2843 if (!ERRNO_IS_MACHINE_ID_UNSET(r)) /* If the file is missing, empty, or uninitialized, we don't mind */
2844 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2845
2846 if (sd_id128_is_null(arg_uuid)) {
2847 r = sd_id128_randomize(&arg_uuid);
2848 if (r < 0)
2849 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2850 }
2851 }
2852
2853 return 0;
2854 }
2855
2856 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2857 int r;
2858
2859 assert(directory);
2860
2861 if (arg_userns_mode == USER_NAMESPACE_NO || arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_CHOWN)
2862 return 0;
2863
2864 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2865 if (r == -EOPNOTSUPP)
2866 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.");
2867 if (r == -EBADE)
2868 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2869 if (r < 0)
2870 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2871 if (r == 0)
2872 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2873 else
2874 log_debug("Patched directory tree to match UID/GID range.");
2875
2876 return r;
2877 }
2878
2879 /*
2880 * Return values:
2881 * < 0 : wait_for_terminate() failed to get the state of the
2882 * container, the container was terminated by a signal, or
2883 * failed for an unknown reason. No change is made to the
2884 * container argument.
2885 * > 0 : The program executed in the container terminated with an
2886 * error. The exit code of the program executed in the
2887 * container is returned. The container argument has been set
2888 * to CONTAINER_TERMINATED.
2889 * 0 : The container is being rebooted, has been shut down or exited
2890 * successfully. The container argument has been set to either
2891 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2892 *
2893 * That is, success is indicated by a return value of zero, and an
2894 * error is indicated by a non-zero value.
2895 */
2896 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2897 siginfo_t status;
2898 int r;
2899
2900 r = wait_for_terminate(pid, &status);
2901 if (r < 0)
2902 return log_warning_errno(r, "Failed to wait for container: %m");
2903
2904 switch (status.si_code) {
2905
2906 case CLD_EXITED:
2907 if (status.si_status == 0)
2908 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2909 else
2910 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2911
2912 *container = CONTAINER_TERMINATED;
2913 return status.si_status;
2914
2915 case CLD_KILLED:
2916 if (status.si_status == SIGINT) {
2917 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2918 *container = CONTAINER_TERMINATED;
2919 return 0;
2920
2921 } else if (status.si_status == SIGHUP) {
2922 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2923 *container = CONTAINER_REBOOTED;
2924 return 0;
2925 }
2926
2927 _fallthrough_;
2928 case CLD_DUMPED:
2929 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2930 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2931
2932 default:
2933 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2934 "Container %s failed due to unknown reason.", arg_machine);
2935 }
2936 }
2937
2938 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2939 pid_t pid;
2940
2941 pid = PTR_TO_PID(userdata);
2942 if (pid > 0) {
2943 if (kill(pid, arg_kill_signal) >= 0) {
2944 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2945 sd_event_source_set_userdata(s, NULL);
2946 return 0;
2947 }
2948 }
2949
2950 sd_event_exit(sd_event_source_get_event(s), 0);
2951 return 0;
2952 }
2953
2954 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2955 pid_t pid;
2956
2957 assert(s);
2958 assert(ssi);
2959
2960 pid = PTR_TO_PID(userdata);
2961
2962 for (;;) {
2963 siginfo_t si = {};
2964
2965 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2966 return log_error_errno(errno, "Failed to waitid(): %m");
2967 if (si.si_pid == 0) /* No pending children. */
2968 break;
2969 if (si.si_pid == pid) {
2970 /* The main process we care for has exited. Return from
2971 * signal handler but leave the zombie. */
2972 sd_event_exit(sd_event_source_get_event(s), 0);
2973 break;
2974 }
2975
2976 /* Reap all other children. */
2977 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2978 }
2979
2980 return 0;
2981 }
2982
2983 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2984 pid_t pid;
2985
2986 assert(m);
2987
2988 pid = PTR_TO_PID(userdata);
2989
2990 if (arg_kill_signal > 0) {
2991 log_info("Container termination requested. Attempting to halt container.");
2992 (void) kill(pid, arg_kill_signal);
2993 } else {
2994 log_info("Container termination requested. Exiting.");
2995 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2996 }
2997
2998 return 0;
2999 }
3000
3001 static int determine_names(void) {
3002 int r;
3003
3004 if (arg_template && !arg_directory && arg_machine) {
3005
3006 /* If --template= was specified then we should not
3007 * search for a machine, but instead create a new one
3008 * in /var/lib/machine. */
3009
3010 arg_directory = path_join("/var/lib/machines", arg_machine);
3011 if (!arg_directory)
3012 return log_oom();
3013 }
3014
3015 if (!arg_image && !arg_directory) {
3016 if (arg_machine) {
3017 _cleanup_(image_unrefp) Image *i = NULL;
3018
3019 r = image_find(IMAGE_MACHINE, arg_machine, NULL, &i);
3020 if (r == -ENOENT)
3021 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
3022 if (r < 0)
3023 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
3024
3025 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
3026 r = free_and_strdup(&arg_image, i->path);
3027 else
3028 r = free_and_strdup(&arg_directory, i->path);
3029 if (r < 0)
3030 return log_oom();
3031
3032 if (!arg_ephemeral)
3033 arg_read_only = arg_read_only || i->read_only;
3034 } else {
3035 r = safe_getcwd(&arg_directory);
3036 if (r < 0)
3037 return log_error_errno(r, "Failed to determine current directory: %m");
3038 }
3039
3040 if (!arg_directory && !arg_image)
3041 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
3042 }
3043
3044 if (!arg_machine) {
3045 if (arg_directory && path_equal(arg_directory, "/"))
3046 arg_machine = gethostname_malloc();
3047 else if (arg_image) {
3048 char *e;
3049
3050 r = path_extract_filename(arg_image, &arg_machine);
3051 if (r < 0)
3052 return log_error_errno(r, "Failed to extract file name from '%s': %m", arg_image);
3053
3054 /* Truncate suffix if there is one */
3055 e = endswith(arg_machine, ".raw");
3056 if (e)
3057 *e = 0;
3058 } else {
3059 r = path_extract_filename(arg_directory, &arg_machine);
3060 if (r < 0)
3061 return log_error_errno(r, "Failed to extract file name from '%s': %m", arg_directory);
3062 }
3063
3064 hostname_cleanup(arg_machine);
3065 if (!hostname_is_valid(arg_machine, 0))
3066 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
3067
3068 /* Copy the machine name before the random suffix is added below, otherwise we won't be able
3069 * to match fixed config file names. */
3070 arg_settings_filename = strjoin(arg_machine, ".nspawn");
3071 if (!arg_settings_filename)
3072 return log_oom();
3073
3074 /* Add a random suffix when this is an ephemeral machine, so that we can run many
3075 * instances at once without manually having to specify -M each time. */
3076 if (arg_ephemeral)
3077 if (strextendf(&arg_machine, "-%016" PRIx64, random_u64()) < 0)
3078 return log_oom();
3079 } else {
3080 arg_settings_filename = strjoin(arg_machine, ".nspawn");
3081 if (!arg_settings_filename)
3082 return log_oom();
3083 }
3084
3085 return 0;
3086 }
3087
3088 static int chase_and_update(char **p, unsigned flags) {
3089 char *chased;
3090 int r;
3091
3092 assert(p);
3093
3094 if (!*p)
3095 return 0;
3096
3097 r = chase(*p, NULL, flags, &chased, NULL);
3098 if (r < 0)
3099 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
3100
3101 return free_and_replace(*p, chased);
3102 }
3103
3104 static int determine_uid_shift(const char *directory) {
3105
3106 if (arg_userns_mode == USER_NAMESPACE_NO) {
3107 arg_uid_shift = 0;
3108 return 0;
3109 }
3110
3111 if (arg_uid_shift == UID_INVALID) {
3112 struct stat st;
3113
3114 /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
3115
3116 if (stat(directory, &st) < 0)
3117 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
3118
3119 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
3120
3121 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
3122 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3123 "UID and GID base of %s don't match.", directory);
3124
3125 arg_uid_range = UINT32_C(0x10000);
3126
3127 if (arg_uid_shift != 0) {
3128 /* If the image is shifted already, then we'll fall back to classic chowning, for
3129 * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
3130
3131 if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_AUTO) {
3132 log_debug("UID base of %s is non-zero, not using UID mapping.", directory);
3133 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
3134 } else if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_MAP)
3135 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
3136 "UID base of %s is not zero, UID mapping not supported.", directory);
3137 }
3138 }
3139
3140 if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
3141 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID base too high for UID range.");
3142
3143 return 0;
3144 }
3145
3146 static unsigned long effective_clone_ns_flags(void) {
3147 unsigned long flags = arg_clone_ns_flags;
3148
3149 if (arg_private_network)
3150 flags |= CLONE_NEWNET;
3151 if (arg_use_cgns)
3152 flags |= CLONE_NEWCGROUP;
3153 if (arg_userns_mode != USER_NAMESPACE_NO)
3154 flags |= CLONE_NEWUSER;
3155
3156 return flags;
3157 }
3158
3159 static int patch_sysctl(void) {
3160
3161 /* This table is inspired by runc's sysctl() function */
3162 static const struct {
3163 const char *key;
3164 bool prefix;
3165 unsigned long clone_flags;
3166 } safe_sysctl[] = {
3167 { "kernel.hostname", false, CLONE_NEWUTS },
3168 { "kernel.domainname", false, CLONE_NEWUTS },
3169 { "kernel.msgmax", false, CLONE_NEWIPC },
3170 { "kernel.msgmnb", false, CLONE_NEWIPC },
3171 { "kernel.msgmni", false, CLONE_NEWIPC },
3172 { "kernel.sem", false, CLONE_NEWIPC },
3173 { "kernel.shmall", false, CLONE_NEWIPC },
3174 { "kernel.shmmax", false, CLONE_NEWIPC },
3175 { "kernel.shmmni", false, CLONE_NEWIPC },
3176 { "fs.mqueue.", true, CLONE_NEWIPC },
3177 { "net.", true, CLONE_NEWNET },
3178 };
3179
3180 unsigned long flags;
3181 int r;
3182
3183 flags = effective_clone_ns_flags();
3184
3185 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
3186 bool good = false;
3187 size_t i;
3188
3189 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
3190
3191 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
3192 continue;
3193
3194 if (safe_sysctl[i].prefix)
3195 good = startswith(*k, safe_sysctl[i].key);
3196 else
3197 good = streq(*k, safe_sysctl[i].key);
3198
3199 if (good)
3200 break;
3201 }
3202
3203 if (!good)
3204 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
3205
3206 r = sysctl_write(*k, *v);
3207 if (r < 0)
3208 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
3209 }
3210
3211 return 0;
3212 }
3213
3214 static int inner_child(
3215 Barrier *barrier,
3216 int fd_inner_socket,
3217 FDSet *fds,
3218 char **os_release_pairs) {
3219
3220 _cleanup_free_ char *home = NULL;
3221 size_t n_env = 1;
3222 char *envp[] = {
3223 (char*) "PATH=" DEFAULT_PATH_COMPAT,
3224 NULL, /* container */
3225 NULL, /* TERM */
3226 NULL, /* HOME */
3227 NULL, /* USER */
3228 NULL, /* LOGNAME */
3229 NULL, /* container_uuid */
3230 NULL, /* LISTEN_FDS */
3231 NULL, /* LISTEN_PID */
3232 NULL, /* NOTIFY_SOCKET */
3233 NULL, /* CREDENTIALS_DIRECTORY */
3234 NULL, /* LANG */
3235 NULL
3236 };
3237 const char *exec_target;
3238 _cleanup_strv_free_ char **env_use = NULL;
3239 int r, which_failed;
3240
3241 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
3242 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
3243 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
3244 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
3245 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
3246 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
3247 * namespace.
3248 *
3249 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
3250 * unshare(). See below. */
3251
3252 assert(barrier);
3253 assert(fd_inner_socket >= 0);
3254
3255 log_debug("Inner child is initializing.");
3256
3257 if (arg_userns_mode != USER_NAMESPACE_NO) {
3258 /* Tell the parent, that it now can write the UID map. */
3259 (void) barrier_place(barrier); /* #1 */
3260
3261 /* Wait until the parent wrote the UID map */
3262 if (!barrier_place_and_sync(barrier)) /* #2 */
3263 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3264
3265 /* Become the new root user inside our namespace */
3266 r = reset_uid_gid();
3267 if (r < 0)
3268 return log_error_errno(r, "Couldn't become new root: %m");
3269
3270 /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
3271 * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
3272 * propagation, but simply create new peer groups for all our mounts). */
3273 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SHARED|MS_REC, NULL);
3274 if (r < 0)
3275 return r;
3276 }
3277
3278 r = mount_all(NULL,
3279 arg_mount_settings | MOUNT_IN_USERNS,
3280 arg_uid_shift,
3281 arg_selinux_apifs_context);
3282 if (r < 0)
3283 return r;
3284
3285 if (!arg_network_namespace_path && arg_private_network) {
3286 r = unshare(CLONE_NEWNET);
3287 if (r < 0)
3288 return log_error_errno(errno, "Failed to unshare network namespace: %m");
3289
3290 /* Tell the parent that it can setup network interfaces. */
3291 (void) barrier_place(barrier); /* #3 */
3292 }
3293
3294 r = mount_sysfs(NULL, arg_mount_settings);
3295 if (r < 0)
3296 return r;
3297
3298 /* Wait until we are cgroup-ified, so that we
3299 * can mount the right cgroup path writable */
3300 if (!barrier_place_and_sync(barrier)) /* #4 */
3301 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3302 "Parent died too early");
3303
3304 if (arg_use_cgns) {
3305 r = unshare(CLONE_NEWCGROUP);
3306 if (r < 0)
3307 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
3308 r = mount_cgroups(
3309 "",
3310 arg_unified_cgroup_hierarchy,
3311 arg_userns_mode != USER_NAMESPACE_NO,
3312 arg_uid_shift,
3313 arg_uid_range,
3314 arg_selinux_apifs_context,
3315 true);
3316 } else
3317 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
3318 if (r < 0)
3319 return r;
3320
3321 r = setup_boot_id();
3322 if (r < 0)
3323 return r;
3324
3325 r = setup_kmsg(fd_inner_socket);
3326 if (r < 0)
3327 return r;
3328
3329 r = mount_custom(
3330 "/",
3331 arg_custom_mounts,
3332 arg_n_custom_mounts,
3333 0,
3334 0,
3335 arg_selinux_apifs_context,
3336 MOUNT_NON_ROOT_ONLY | MOUNT_IN_USERNS);
3337 if (r < 0)
3338 return r;
3339
3340 if (setsid() < 0)
3341 return log_error_errno(errno, "setsid() failed: %m");
3342
3343 if (arg_private_network)
3344 (void) loopback_setup();
3345
3346 if (arg_expose_ports) {
3347 r = expose_port_send_rtnl(fd_inner_socket);
3348 if (r < 0)
3349 return r;
3350 }
3351
3352 if (arg_console_mode != CONSOLE_PIPE) {
3353 _cleanup_close_ int master = -EBADF;
3354 _cleanup_free_ char *console = NULL;
3355
3356 /* Allocate a pty and make it available as /dev/console. */
3357 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
3358 if (master < 0)
3359 return log_error_errno(master, "Failed to allocate a pty: %m");
3360
3361 r = setup_dev_console(console);
3362 if (r < 0)
3363 return log_error_errno(r, "Failed to set up /dev/console: %m");
3364
3365 r = send_one_fd(fd_inner_socket, master, 0);
3366 if (r < 0)
3367 return log_error_errno(r, "Failed to send master fd: %m");
3368
3369 r = setup_stdio_as_dev_console();
3370 if (r < 0)
3371 return r;
3372 }
3373
3374 r = patch_sysctl();
3375 if (r < 0)
3376 return r;
3377
3378 if (arg_oom_score_adjust_set) {
3379 r = set_oom_score_adjust(arg_oom_score_adjust);
3380 if (r < 0)
3381 return log_error_errno(r, "Failed to adjust OOM score: %m");
3382 }
3383
3384 if (arg_cpu_set.set)
3385 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
3386 return log_error_errno(errno, "Failed to set CPU affinity: %m");
3387
3388 (void) setup_hostname();
3389
3390 if (arg_personality != PERSONALITY_INVALID) {
3391 r = safe_personality(arg_personality);
3392 if (r < 0)
3393 return log_error_errno(r, "personality() failed: %m");
3394 #ifdef ARCHITECTURE_SECONDARY
3395 } else if (arg_architecture == ARCHITECTURE_SECONDARY) {
3396 r = safe_personality(PER_LINUX32);
3397 if (r < 0)
3398 return log_error_errno(r, "personality() failed: %m");
3399 #endif
3400 } else if (!arg_quiet && arg_architecture >= 0 && arg_architecture != native_architecture())
3401 log_notice("Selected architecture '%s' not supported natively on the local CPU, assuming "
3402 "invocation with qemu userspace emulator (or equivalent) in effect.",
3403 architecture_to_string(arg_architecture));
3404
3405 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3406 if (r < 0)
3407 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3408
3409 #if HAVE_SECCOMP
3410 if (arg_seccomp) {
3411
3412 if (is_seccomp_available()) {
3413
3414 r = seccomp_load(arg_seccomp);
3415 if (ERRNO_IS_SECCOMP_FATAL(r))
3416 return log_error_errno(r, "Failed to install seccomp filter: %m");
3417 if (r < 0)
3418 log_debug_errno(r, "Failed to install seccomp filter: %m");
3419 }
3420 } else
3421 #endif
3422 {
3423 r = setup_seccomp(arg_caps_retain, arg_syscall_allow_list, arg_syscall_deny_list);
3424 if (r < 0)
3425 return r;
3426 }
3427
3428 if (arg_suppress_sync) {
3429 #if HAVE_SECCOMP
3430 r = seccomp_suppress_sync();
3431 if (r < 0)
3432 log_debug_errno(r, "Failed to install sync() suppression seccomp filter, ignoring: %m");
3433 #else
3434 log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
3435 #endif
3436 }
3437
3438 #if HAVE_SELINUX
3439 if (arg_selinux_context)
3440 if (setexeccon(arg_selinux_context) < 0)
3441 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3442 #endif
3443
3444 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3445 * if we need to later on. */
3446 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3447 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3448
3449 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3450 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids, arg_console_mode != CONSOLE_PIPE);
3451 else
3452 r = change_uid_gid(arg_user, arg_console_mode != CONSOLE_PIPE, &home);
3453 if (r < 0)
3454 return r;
3455
3456 r = drop_capabilities(getuid());
3457 if (r < 0)
3458 return log_error_errno(r, "Dropping capabilities failed: %m");
3459
3460 if (arg_no_new_privileges)
3461 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3462 return log_error_errno(errno, "Failed to disable new privileges: %m");
3463
3464 /* LXC sets container=lxc, so follow the scheme here */
3465 envp[n_env++] = strjoina("container=", arg_container_service_name);
3466
3467 envp[n_env] = strv_find_prefix(environ, "TERM=");
3468 if (envp[n_env])
3469 n_env++;
3470
3471 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3472 if (asprintf(envp + n_env++, "HOME=%s", home ?: "/root") < 0)
3473 return log_oom();
3474
3475 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3476 if (asprintf(envp + n_env++, "USER=%s", arg_user ?: "root") < 0 ||
3477 asprintf(envp + n_env++, "LOGNAME=%s", arg_user ?: "root") < 0)
3478 return log_oom();
3479
3480 assert(!sd_id128_is_null(arg_uuid));
3481
3482 if (asprintf(envp + n_env++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid)) < 0)
3483 return log_oom();
3484
3485 if (fdset_size(fds) > 0) {
3486 r = fdset_cloexec(fds, false);
3487 if (r < 0)
3488 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3489
3490 if ((asprintf(envp + n_env++, "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3491 (asprintf(envp + n_env++, "LISTEN_PID=1") < 0))
3492 return log_oom();
3493 }
3494 if (asprintf(envp + n_env++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3495 return log_oom();
3496
3497 if (arg_n_credentials > 0) {
3498 envp[n_env] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
3499 if (!envp[n_env])
3500 return log_oom();
3501 n_env++;
3502 }
3503
3504 if (arg_start_mode != START_BOOT) {
3505 envp[n_env] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE);
3506 if (!envp[n_env])
3507 return log_oom();
3508 n_env++;
3509 }
3510
3511 env_use = strv_env_merge(envp, os_release_pairs, arg_setenv);
3512 if (!env_use)
3513 return log_oom();
3514
3515 /* Let the parent know that we are ready and
3516 * wait until the parent is ready with the
3517 * setup, too... */
3518 if (!barrier_place_and_sync(barrier)) /* #5 */
3519 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
3520
3521 if (arg_chdir)
3522 if (chdir(arg_chdir) < 0)
3523 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3524
3525 if (arg_start_mode == START_PID2) {
3526 r = stub_pid1(arg_uuid);
3527 if (r < 0)
3528 return r;
3529 }
3530
3531 if (arg_console_mode != CONSOLE_PIPE) {
3532 /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
3533 * are configured for that. Acquire it as controlling tty. */
3534 if (ioctl(STDIN_FILENO, TIOCSCTTY) < 0)
3535 return log_error_errno(errno, "Failed to acquire controlling TTY: %m");
3536 }
3537
3538 log_debug("Inner child completed, invoking payload.");
3539
3540 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3541 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3542 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3543 log_close();
3544 log_set_open_when_needed(true);
3545 log_settle_target();
3546
3547 (void) fdset_close_others(fds);
3548
3549 if (arg_start_mode == START_BOOT) {
3550 char **a;
3551 size_t m;
3552
3553 /* Automatically search for the init system */
3554
3555 m = strv_length(arg_parameters);
3556 a = newa(char*, m + 2);
3557 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3558 a[1 + m] = NULL;
3559
3560 FOREACH_STRING(init,
3561 "/usr/lib/systemd/systemd",
3562 "/lib/systemd/systemd",
3563 "/sbin/init") {
3564 a[0] = (char*) init;
3565 execve(a[0], a, env_use);
3566 }
3567
3568 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3569 } else if (!strv_isempty(arg_parameters)) {
3570 const char *dollar_path;
3571
3572 exec_target = arg_parameters[0];
3573
3574 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3575 * binary. */
3576 dollar_path = strv_env_get(env_use, "PATH");
3577 if (dollar_path) {
3578 if (setenv("PATH", dollar_path, 1) < 0)
3579 return log_error_errno(errno, "Failed to update $PATH: %m");
3580 }
3581
3582 execvpe(arg_parameters[0], arg_parameters, env_use);
3583 } else {
3584 if (!arg_chdir)
3585 /* If we cannot change the directory, we'll end up in /, that is expected. */
3586 (void) chdir(home ?: "/root");
3587
3588 execle(DEFAULT_USER_SHELL, "-" DEFAULT_USER_SHELL_NAME, NULL, env_use);
3589 if (!streq(DEFAULT_USER_SHELL, "/bin/bash"))
3590 execle("/bin/bash", "-bash", NULL, env_use);
3591 if (!streq(DEFAULT_USER_SHELL, "/bin/sh"))
3592 execle("/bin/sh", "-sh", NULL, env_use);
3593
3594 exec_target = DEFAULT_USER_SHELL ", /bin/bash, /bin/sh";
3595 }
3596
3597 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3598 }
3599
3600 static int setup_notify_child(void) {
3601 _cleanup_close_ int fd = -EBADF;
3602 static const union sockaddr_union sa = {
3603 .un.sun_family = AF_UNIX,
3604 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3605 };
3606 int r;
3607
3608 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3609 if (fd < 0)
3610 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3611
3612 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3613 (void) sockaddr_un_unlink(&sa.un);
3614
3615 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3616 if (r < 0)
3617 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3618
3619 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3620 if (r < 0)
3621 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3622
3623 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3624 if (r < 0)
3625 return log_error_errno(r, "SO_PASSCRED failed: %m");
3626
3627 return TAKE_FD(fd);
3628 }
3629
3630 static int outer_child(
3631 Barrier *barrier,
3632 const char *directory,
3633 DissectedImage *dissected_image,
3634 int fd_outer_socket,
3635 int fd_inner_socket,
3636 FDSet *fds,
3637 int netns_fd) {
3638
3639 _cleanup_(bind_user_context_freep) BindUserContext *bind_user_context = NULL;
3640 _cleanup_strv_free_ char **os_release_pairs = NULL;
3641 _cleanup_close_ int fd = -EBADF, mntns_fd = -EBADF;
3642 bool idmap = false;
3643 const char *p;
3644 pid_t pid;
3645 ssize_t l;
3646 int r;
3647
3648 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
3649 * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
3650 * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
3651 * namespaces. After it completed a number of initializations a second child (the "inner" one) is
3652 * forked off it, and it exits. */
3653
3654 assert(barrier);
3655 assert(directory);
3656 assert(fd_outer_socket >= 0);
3657 assert(fd_inner_socket >= 0);
3658
3659 log_debug("Outer child is initializing.");
3660
3661 r = load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs);
3662 if (r < 0)
3663 log_debug_errno(r, "Failed to read os-release from host for container, ignoring: %m");
3664
3665 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3666 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3667
3668 r = reset_audit_loginuid();
3669 if (r < 0)
3670 return r;
3671
3672 /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
3673 * mounts to the real root. */
3674 r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3675 if (r < 0)
3676 return r;
3677
3678 if (dissected_image) {
3679 /* If we are operating on a disk image, then mount its root directory now, but leave out the
3680 * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
3681 * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
3682 * right place right away. This makes sure ESP partitions and userns are compatible. */
3683
3684 r = dissected_image_mount_and_warn(
3685 dissected_image,
3686 directory,
3687 arg_uid_shift,
3688 arg_uid_range,
3689 DISSECT_IMAGE_MOUNT_ROOT_ONLY|
3690 DISSECT_IMAGE_DISCARD_ON_LOOP|
3691 DISSECT_IMAGE_USR_NO_ROOT|
3692 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
3693 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3694 if (r < 0)
3695 return r;
3696 }
3697
3698 r = determine_uid_shift(directory);
3699 if (r < 0)
3700 return r;
3701
3702 if (arg_userns_mode != USER_NAMESPACE_NO) {
3703 r = namespace_open(0, NULL, &mntns_fd, NULL, NULL, NULL);
3704 if (r < 0)
3705 return log_error_errno(r, "Failed to pin outer mount namespace: %m");
3706
3707 l = send_one_fd(fd_outer_socket, mntns_fd, 0);
3708 if (l < 0)
3709 return log_error_errno(l, "Failed to send outer mount namespace fd: %m");
3710 mntns_fd = safe_close(mntns_fd);
3711
3712 /* Let the parent know which UID shift we read from the image */
3713 l = send(fd_outer_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3714 if (l < 0)
3715 return log_error_errno(errno, "Failed to send UID shift: %m");
3716 if (l != sizeof(arg_uid_shift))
3717 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3718 "Short write while sending UID shift.");
3719
3720 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3721 /* When we are supposed to pick the UID shift, the parent will check now whether the
3722 * UID shift we just read from the image is available. If yes, it will send the UID
3723 * shift back to us, if not it will pick a different one, and send it back to us. */
3724
3725 l = recv(fd_outer_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3726 if (l < 0)
3727 return log_error_errno(errno, "Failed to recv UID shift: %m");
3728 if (l != sizeof(arg_uid_shift))
3729 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3730 "Short read while receiving UID shift.");
3731 }
3732
3733 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3734 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3735 }
3736
3737 if (path_equal(directory, "/")) {
3738 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3739 * place, so that we can make changes to its mount structure (for example, to implement
3740 * --volatile=) without this interfering with our ability to access files such as
3741 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3742 * (instead of a temporary directory, since we are living in our own mount namespace here
3743 * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
3744 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3745
3746 r = mount_nofollow_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
3747 if (r < 0)
3748 return r;
3749
3750 directory = "/run/systemd/nspawn-root";
3751 }
3752
3753 /* Make sure we always have a mount that we can move to root later on. */
3754 r = make_mount_point(directory);
3755 if (r < 0)
3756 return r;
3757
3758 /* So the whole tree is now MS_SLAVE, i.e. we'll still receive mount/umount events from the host
3759 * mount namespace. For the directory we are going to run our container let's turn this off, so that
3760 * we'll live in our own little world from now on, and propagation from the host may only happen via
3761 * the mount tunnel dir, or not at all. */
3762 r = mount_follow_verbose(LOG_ERR, NULL, directory, NULL, MS_PRIVATE|MS_REC, NULL);
3763 if (r < 0)
3764 return r;
3765
3766 r = setup_pivot_root(
3767 directory,
3768 arg_pivot_root_new,
3769 arg_pivot_root_old);
3770 if (r < 0)
3771 return r;
3772
3773 r = setup_volatile_mode(
3774 directory,
3775 arg_volatile_mode,
3776 arg_uid_shift,
3777 arg_selinux_apifs_context);
3778 if (r < 0)
3779 return r;
3780
3781 r = bind_user_prepare(
3782 directory,
3783 arg_bind_user,
3784 arg_uid_shift,
3785 arg_uid_range,
3786 &arg_custom_mounts, &arg_n_custom_mounts,
3787 &bind_user_context);
3788 if (r < 0)
3789 return r;
3790
3791 if (arg_userns_mode != USER_NAMESPACE_NO && bind_user_context) {
3792 /* Send the user maps we determined to the parent, so that it installs it in our user
3793 * namespace UID map table */
3794
3795 for (size_t i = 0; i < bind_user_context->n_data; i++) {
3796 uid_t map[] = {
3797 bind_user_context->data[i].payload_user->uid,
3798 bind_user_context->data[i].host_user->uid,
3799 (uid_t) bind_user_context->data[i].payload_group->gid,
3800 (uid_t) bind_user_context->data[i].host_group->gid,
3801 };
3802
3803 l = send(fd_outer_socket, map, sizeof(map), MSG_NOSIGNAL);
3804 if (l < 0)
3805 return log_error_errno(errno, "Failed to send user UID map: %m");
3806 if (l != sizeof(map))
3807 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3808 "Short write while sending user UID map.");
3809 }
3810 }
3811
3812 r = mount_custom(
3813 directory,
3814 arg_custom_mounts,
3815 arg_n_custom_mounts,
3816 arg_uid_shift,
3817 arg_uid_range,
3818 arg_selinux_apifs_context,
3819 MOUNT_ROOT_ONLY);
3820 if (r < 0)
3821 return r;
3822
3823 if (arg_userns_mode != USER_NAMESPACE_NO &&
3824 IN_SET(arg_userns_ownership, USER_NAMESPACE_OWNERSHIP_MAP, USER_NAMESPACE_OWNERSHIP_AUTO) &&
3825 arg_uid_shift != 0) {
3826
3827 r = remount_idmap(directory, arg_uid_shift, arg_uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
3828 if (r == -EINVAL || ERRNO_IS_NOT_SUPPORTED(r)) {
3829 /* This might fail because the kernel or file system doesn't support idmapping. We
3830 * can't really distinguish this nicely, nor do we have any guarantees about the
3831 * error codes we see, could be EOPNOTSUPP or EINVAL. */
3832 if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_AUTO)
3833 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
3834 "ID mapped mounts are apparently not available, sorry.");
3835
3836 log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
3837 arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
3838 } else if (r < 0)
3839 return log_error_errno(r, "Failed to set up ID mapped mounts: %m");
3840 else {
3841 log_debug("ID mapped mounts available, making use of them.");
3842 idmap = true;
3843 }
3844 }
3845
3846 if (dissected_image) {
3847 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3848 r = dissected_image_mount(
3849 dissected_image,
3850 directory,
3851 arg_uid_shift,
3852 arg_uid_range,
3853 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|
3854 DISSECT_IMAGE_DISCARD_ON_LOOP|
3855 DISSECT_IMAGE_USR_NO_ROOT|
3856 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
3857 (idmap ? DISSECT_IMAGE_MOUNT_IDMAPPED : 0));
3858 if (r == -EUCLEAN)
3859 return log_error_errno(r, "File system check for image failed: %m");
3860 if (r < 0)
3861 return log_error_errno(r, "Failed to mount image file system: %m");
3862 }
3863
3864 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3865 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3866
3867 r = detect_unified_cgroup_hierarchy_from_image(directory);
3868 if (r < 0)
3869 return r;
3870
3871 l = send(fd_outer_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3872 if (l < 0)
3873 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3874 if (l != sizeof(arg_unified_cgroup_hierarchy))
3875 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3876 "Short write while sending cgroup mode.");
3877 }
3878
3879 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3880 if (r < 0)
3881 return r;
3882
3883 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3884 if (r < 0)
3885 return r;
3886
3887 if (arg_read_only && arg_volatile_mode == VOLATILE_NO &&
3888 !has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts)) {
3889 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3890 if (r < 0)
3891 return log_error_errno(r, "Failed to make tree read-only: %m");
3892 }
3893
3894 r = mount_all(directory,
3895 arg_mount_settings,
3896 arg_uid_shift,
3897 arg_selinux_apifs_context);
3898 if (r < 0)
3899 return r;
3900
3901 r = copy_devnodes(directory);
3902 if (r < 0)
3903 return r;
3904
3905 r = make_extra_nodes(directory);
3906 if (r < 0)
3907 return r;
3908
3909 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3910
3911 p = prefix_roota(directory, "/run/host");
3912 (void) make_inaccessible_nodes(p, arg_uid_shift, arg_uid_shift);
3913
3914 r = setup_pts(directory);
3915 if (r < 0)
3916 return r;
3917
3918 r = mount_tunnel_dig(directory);
3919 if (r < 0)
3920 return r;
3921
3922 r = setup_keyring();
3923 if (r < 0)
3924 return r;
3925
3926 r = setup_credentials(directory);
3927 if (r < 0)
3928 return r;
3929
3930 r = bind_user_setup(bind_user_context, directory);
3931 if (r < 0)
3932 return r;
3933
3934 r = mount_custom(
3935 directory,
3936 arg_custom_mounts,
3937 arg_n_custom_mounts,
3938 arg_uid_shift,
3939 arg_uid_range,
3940 arg_selinux_apifs_context,
3941 MOUNT_NON_ROOT_ONLY);
3942 if (r < 0)
3943 return r;
3944
3945 r = setup_timezone(directory);
3946 if (r < 0)
3947 return r;
3948
3949 r = setup_resolv_conf(directory);
3950 if (r < 0)
3951 return r;
3952
3953 r = setup_machine_id(directory);
3954 if (r < 0)
3955 return r;
3956
3957 r = setup_journal(directory);
3958 if (r < 0)
3959 return r;
3960
3961 /* The same stuff as the $container env var, but nicely readable for the entire payload */
3962 p = prefix_roota(directory, "/run/host/container-manager");
3963 (void) write_string_file(p, arg_container_service_name, WRITE_STRING_FILE_CREATE);
3964
3965 /* The same stuff as the $container_uuid env var */
3966 p = prefix_roota(directory, "/run/host/container-uuid");
3967 (void) write_string_filef(p, WRITE_STRING_FILE_CREATE, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid));
3968
3969 if (!arg_use_cgns) {
3970 r = mount_cgroups(
3971 directory,
3972 arg_unified_cgroup_hierarchy,
3973 arg_userns_mode != USER_NAMESPACE_NO,
3974 arg_uid_shift,
3975 arg_uid_range,
3976 arg_selinux_apifs_context,
3977 false);
3978 if (r < 0)
3979 return r;
3980 }
3981
3982 /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
3983 * mounts available in systemd services inside the container that create a new mount namespace. See
3984 * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
3985 * will inherit the shared propagation mode.
3986 *
3987 * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
3988 * directory mount to root later on.
3989 * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
3990 */
3991 r = mount_switch_root(directory, MS_SHARED);
3992 if (r < 0)
3993 return log_error_errno(r, "Failed to move root directory: %m");
3994
3995 /* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
3996 * dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
3997 * the container. */
3998 r = mount_tunnel_open();
3999 if (r < 0)
4000 return r;
4001
4002 if (arg_userns_mode != USER_NAMESPACE_NO) {
4003 /* In order to mount procfs and sysfs in an unprivileged container the kernel
4004 * requires that a fully visible instance is already present in the target mount
4005 * namespace. Mount one here so the inner child can mount its own instances. Later
4006 * we umount the temporary instances created here before we actually exec the
4007 * payload. Since the rootfs is shared the umount will propagate into the container.
4008 * Note, the inner child wouldn't be able to unmount the instances on its own since
4009 * it doesn't own the originating mount namespace. IOW, the outer child needs to do
4010 * this. */
4011 r = pin_fully_visible_fs();
4012 if (r < 0)
4013 return r;
4014 }
4015
4016 fd = setup_notify_child();
4017 if (fd < 0)
4018 return fd;
4019
4020 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
4021 arg_clone_ns_flags |
4022 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
4023 if (pid < 0)
4024 return log_error_errno(errno, "Failed to fork inner child: %m");
4025 if (pid == 0) {
4026 fd_outer_socket = safe_close(fd_outer_socket);
4027
4028 /* The inner child has all namespaces that are requested, so that we all are owned by the
4029 * user if user namespaces are turned on. */
4030
4031 if (arg_network_namespace_path) {
4032 r = namespace_enter(-1, -1, netns_fd, -1, -1);
4033 if (r < 0)
4034 return log_error_errno(r, "Failed to join network namespace: %m");
4035 }
4036
4037 r = inner_child(barrier, fd_inner_socket, fds, os_release_pairs);
4038 if (r < 0)
4039 _exit(EXIT_FAILURE);
4040
4041 _exit(EXIT_SUCCESS);
4042 }
4043
4044 l = send(fd_outer_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
4045 if (l < 0)
4046 return log_error_errno(errno, "Failed to send PID: %m");
4047 if (l != sizeof(pid))
4048 return log_error_errno(SYNTHETIC_ERRNO(EIO),
4049 "Short write while sending PID.");
4050
4051 l = send(fd_outer_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
4052 if (l < 0)
4053 return log_error_errno(errno, "Failed to send machine ID: %m");
4054 if (l != sizeof(arg_uuid))
4055 return log_error_errno(SYNTHETIC_ERRNO(EIO),
4056 "Short write while sending machine ID.");
4057
4058 l = send_one_fd(fd_outer_socket, fd, 0);
4059 if (l < 0)
4060 return log_error_errno(l, "Failed to send notify fd: %m");
4061
4062 fd_outer_socket = safe_close(fd_outer_socket);
4063 fd_inner_socket = safe_close(fd_inner_socket);
4064 netns_fd = safe_close(netns_fd);
4065
4066 return 0;
4067 }
4068
4069 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
4070 bool tried_hashed = false;
4071 unsigned n_tries = 100;
4072 uid_t candidate;
4073 int r;
4074
4075 assert(shift);
4076 assert(ret_lock_file);
4077 assert(arg_userns_mode == USER_NAMESPACE_PICK);
4078 assert(arg_uid_range == 0x10000U);
4079
4080 candidate = *shift;
4081
4082 (void) mkdir("/run/systemd/nspawn-uid", 0755);
4083
4084 for (;;) {
4085 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
4086 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
4087
4088 if (--n_tries <= 0)
4089 return -EBUSY;
4090
4091 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
4092 goto next;
4093 if ((candidate & UINT32_C(0xFFFF)) != 0)
4094 goto next;
4095
4096 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
4097 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
4098 if (r == -EBUSY) /* Range already taken by another nspawn instance */
4099 goto next;
4100 if (r < 0)
4101 return r;
4102
4103 /* Make some superficial checks whether the range is currently known in the user database */
4104 if (getpwuid(candidate))
4105 goto next;
4106 if (getpwuid(candidate + UINT32_C(0xFFFE)))
4107 goto next;
4108 if (getgrgid(candidate))
4109 goto next;
4110 if (getgrgid(candidate + UINT32_C(0xFFFE)))
4111 goto next;
4112
4113 *ret_lock_file = lf;
4114 lf = (struct LockFile) LOCK_FILE_INIT;
4115 *shift = candidate;
4116 return 0;
4117
4118 next:
4119 if (arg_machine && !tried_hashed) {
4120 /* Try to hash the base from the container name */
4121
4122 static const uint8_t hash_key[] = {
4123 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
4124 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
4125 };
4126
4127 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
4128
4129 tried_hashed = true;
4130 } else
4131 random_bytes(&candidate, sizeof(candidate));
4132
4133 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
4134 candidate &= (uid_t) UINT32_C(0xFFFF0000);
4135 }
4136 }
4137
4138 static int add_one_uid_map(
4139 char **p,
4140 uid_t container_uid,
4141 uid_t host_uid,
4142 uid_t range) {
4143
4144 return strextendf(p,
4145 UID_FMT " " UID_FMT " " UID_FMT "\n",
4146 container_uid, host_uid, range);
4147 }
4148
4149 static int make_uid_map_string(
4150 const uid_t bind_user_uid[],
4151 size_t n_bind_user_uid,
4152 size_t offset,
4153 char **ret) {
4154
4155 _cleanup_free_ char *s = NULL;
4156 uid_t previous_uid = 0;
4157 int r;
4158
4159 assert(n_bind_user_uid == 0 || bind_user_uid);
4160 assert(IN_SET(offset, 0, 2)); /* used to switch between UID and GID map */
4161 assert(ret);
4162
4163 /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
4164 * quadruplet, consisting of host and container UID + GID. */
4165
4166 for (size_t i = 0; i < n_bind_user_uid; i++) {
4167 uid_t payload_uid = bind_user_uid[i*4+offset],
4168 host_uid = bind_user_uid[i*4+offset+1];
4169
4170 assert(previous_uid <= payload_uid);
4171 assert(payload_uid < arg_uid_range);
4172
4173 /* Add a range to close the gap to previous entry */
4174 if (payload_uid > previous_uid) {
4175 r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, payload_uid - previous_uid);
4176 if (r < 0)
4177 return r;
4178 }
4179
4180 /* Map this specific user */
4181 r = add_one_uid_map(&s, payload_uid, host_uid, 1);
4182 if (r < 0)
4183 return r;
4184
4185 previous_uid = payload_uid + 1;
4186 }
4187
4188 /* And add a range to close the gap to finish the range */
4189 if (arg_uid_range > previous_uid) {
4190 r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, arg_uid_range - previous_uid);
4191 if (r < 0)
4192 return r;
4193 }
4194
4195 assert(s);
4196
4197 *ret = TAKE_PTR(s);
4198 return 0;
4199 }
4200
4201 static int setup_uid_map(
4202 pid_t pid,
4203 const uid_t bind_user_uid[],
4204 size_t n_bind_user_uid) {
4205
4206 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1];
4207 _cleanup_free_ char *s = NULL;
4208 int r;
4209
4210 assert(pid > 1);
4211
4212 /* Build the UID map string */
4213 if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 0, &s) < 0) /* offset=0 contains the UID pair */
4214 return log_oom();
4215
4216 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
4217 r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
4218 if (r < 0)
4219 return log_error_errno(r, "Failed to write UID map: %m");
4220
4221 /* And now build the GID map string */
4222 s = mfree(s);
4223 if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 2, &s) < 0) /* offset=2 contains the GID pair */
4224 return log_oom();
4225
4226 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
4227 r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
4228 if (r < 0)
4229 return log_error_errno(r, "Failed to write GID map: %m");
4230
4231 return 0;
4232 }
4233
4234 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4235 char buf[NOTIFY_BUFFER_MAX+1];
4236 char *p = NULL;
4237 struct iovec iovec = {
4238 .iov_base = buf,
4239 .iov_len = sizeof(buf)-1,
4240 };
4241 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
4242 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
4243 struct msghdr msghdr = {
4244 .msg_iov = &iovec,
4245 .msg_iovlen = 1,
4246 .msg_control = &control,
4247 .msg_controllen = sizeof(control),
4248 };
4249 struct ucred *ucred;
4250 ssize_t n;
4251 pid_t inner_child_pid;
4252 _cleanup_strv_free_ char **tags = NULL;
4253 int r;
4254
4255 assert(userdata);
4256
4257 inner_child_pid = PTR_TO_PID(userdata);
4258
4259 if (revents != EPOLLIN) {
4260 log_warning("Got unexpected poll event for notify fd.");
4261 return 0;
4262 }
4263
4264 n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
4265 if (n < 0) {
4266 if (ERRNO_IS_TRANSIENT(n))
4267 return 0;
4268 if (n == -EXFULL) {
4269 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
4270 return 0;
4271 }
4272 return log_warning_errno(n, "Couldn't read notification socket: %m");
4273 }
4274
4275 cmsg_close_all(&msghdr);
4276
4277 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
4278 if (!ucred || ucred->pid != inner_child_pid) {
4279 log_debug("Received notify message without valid credentials. Ignoring.");
4280 return 0;
4281 }
4282
4283 if ((size_t) n >= sizeof(buf)) {
4284 log_warning("Received notify message exceeded maximum size. Ignoring.");
4285 return 0;
4286 }
4287
4288 buf[n] = 0;
4289 tags = strv_split(buf, "\n\r");
4290 if (!tags)
4291 return log_oom();
4292
4293 if (strv_contains(tags, "READY=1")) {
4294 r = sd_notify(false, "READY=1\n");
4295 if (r < 0)
4296 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
4297 }
4298
4299 p = strv_find_startswith(tags, "STATUS=");
4300 if (p)
4301 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
4302
4303 return 0;
4304 }
4305
4306 static int setup_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
4307 int r;
4308
4309 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
4310 if (r < 0)
4311 return log_error_errno(r, "Failed to allocate notify event source: %m");
4312
4313 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
4314
4315 return 0;
4316 }
4317
4318 static int merge_settings(Settings *settings, const char *path) {
4319 int rl;
4320
4321 assert(settings);
4322 assert(path);
4323
4324 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
4325 * that this steals the fields of the Settings* structure, and hence modifies it. */
4326
4327 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
4328 settings->start_mode >= 0) {
4329 arg_start_mode = settings->start_mode;
4330 strv_free_and_replace(arg_parameters, settings->parameters);
4331 }
4332
4333 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0 &&
4334 settings->ephemeral >= 0)
4335 arg_ephemeral = settings->ephemeral;
4336
4337 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
4338 settings->root) {
4339
4340 if (!arg_settings_trusted)
4341 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
4342 else
4343 free_and_replace(arg_directory, settings->root);
4344 }
4345
4346 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
4347 settings->pivot_root_new) {
4348 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
4349 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
4350 }
4351
4352 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
4353 settings->working_directory)
4354 free_and_replace(arg_chdir, settings->working_directory);
4355
4356 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
4357 settings->environment)
4358 strv_free_and_replace(arg_setenv, settings->environment);
4359
4360 if ((arg_settings_mask & SETTING_USER) == 0) {
4361
4362 if (settings->user)
4363 free_and_replace(arg_user, settings->user);
4364
4365 if (uid_is_valid(settings->uid))
4366 arg_uid = settings->uid;
4367 if (gid_is_valid(settings->gid))
4368 arg_gid = settings->gid;
4369 if (settings->n_supplementary_gids > 0) {
4370 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
4371 arg_n_supplementary_gids = settings->n_supplementary_gids;
4372 }
4373 }
4374
4375 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
4376 uint64_t plus, minus;
4377 uint64_t network_minus = 0;
4378 uint64_t ambient;
4379
4380 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
4381 * Settings structure */
4382
4383 plus = settings->capability;
4384 minus = settings->drop_capability;
4385
4386 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4387 settings_network_configured(settings)) {
4388 if (settings_private_network(settings))
4389 plus |= UINT64_C(1) << CAP_NET_ADMIN;
4390 else
4391 network_minus |= UINT64_C(1) << CAP_NET_ADMIN;
4392 }
4393
4394 if (!arg_settings_trusted && plus != 0) {
4395 if (settings->capability != 0)
4396 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
4397 } else {
4398 arg_caps_retain &= ~network_minus;
4399 arg_caps_retain |= plus;
4400 }
4401
4402 arg_caps_retain &= ~minus;
4403
4404 /* Copy the full capabilities over too */
4405 if (capability_quintet_is_set(&settings->full_capabilities)) {
4406 if (!arg_settings_trusted)
4407 log_warning("Ignoring capability settings, file %s is not trusted.", path);
4408 else
4409 arg_full_capabilities = settings->full_capabilities;
4410 }
4411
4412 ambient = settings->ambient_capability;
4413 if (!arg_settings_trusted && ambient != 0)
4414 log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path);
4415 else
4416 arg_caps_ambient |= ambient;
4417 }
4418
4419 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
4420 settings->kill_signal > 0)
4421 arg_kill_signal = settings->kill_signal;
4422
4423 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
4424 settings->personality != PERSONALITY_INVALID)
4425 arg_personality = settings->personality;
4426
4427 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
4428 !sd_id128_is_null(settings->machine_id)) {
4429
4430 if (!arg_settings_trusted)
4431 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
4432 else
4433 arg_uuid = settings->machine_id;
4434 }
4435
4436 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
4437 settings->read_only >= 0)
4438 arg_read_only = settings->read_only;
4439
4440 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
4441 settings->volatile_mode != _VOLATILE_MODE_INVALID)
4442 arg_volatile_mode = settings->volatile_mode;
4443
4444 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
4445 settings->n_custom_mounts > 0) {
4446
4447 if (!arg_settings_trusted)
4448 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
4449 else {
4450 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4451 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
4452 arg_n_custom_mounts = settings->n_custom_mounts;
4453 settings->n_custom_mounts = 0;
4454 }
4455 }
4456
4457 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4458 settings_network_configured(settings)) {
4459
4460 if (!arg_settings_trusted)
4461 log_warning("Ignoring network settings, file %s is not trusted.", path);
4462 else {
4463 arg_network_veth = settings_network_veth(settings);
4464 arg_private_network = settings_private_network(settings);
4465
4466 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
4467 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
4468 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
4469 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
4470
4471 free_and_replace(arg_network_bridge, settings->network_bridge);
4472 free_and_replace(arg_network_zone, settings->network_zone);
4473
4474 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
4475 }
4476 }
4477
4478 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
4479 settings->expose_ports) {
4480
4481 if (!arg_settings_trusted)
4482 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
4483 else {
4484 expose_port_free_all(arg_expose_ports);
4485 arg_expose_ports = TAKE_PTR(settings->expose_ports);
4486 }
4487 }
4488
4489 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
4490 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
4491
4492 if (!arg_settings_trusted)
4493 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
4494 else {
4495 arg_userns_mode = settings->userns_mode;
4496 arg_uid_shift = settings->uid_shift;
4497 arg_uid_range = settings->uid_range;
4498 arg_userns_ownership = settings->userns_ownership;
4499 }
4500 }
4501
4502 if ((arg_settings_mask & SETTING_BIND_USER) == 0 &&
4503 !strv_isempty(settings->bind_user))
4504 strv_free_and_replace(arg_bind_user, settings->bind_user);
4505
4506 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0 &&
4507 settings->notify_ready >= 0)
4508 arg_notify_ready = settings->notify_ready;
4509
4510 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
4511
4512 if (!strv_isempty(settings->syscall_allow_list) || !strv_isempty(settings->syscall_deny_list)) {
4513 if (!arg_settings_trusted && !strv_isempty(settings->syscall_allow_list))
4514 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
4515 else {
4516 strv_free_and_replace(arg_syscall_allow_list, settings->syscall_allow_list);
4517 strv_free_and_replace(arg_syscall_deny_list, settings->syscall_deny_list);
4518 }
4519 }
4520
4521 #if HAVE_SECCOMP
4522 if (settings->seccomp) {
4523 if (!arg_settings_trusted)
4524 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
4525 else {
4526 seccomp_release(arg_seccomp);
4527 arg_seccomp = TAKE_PTR(settings->seccomp);
4528 }
4529 }
4530 #endif
4531 }
4532
4533 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
4534 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
4535 continue;
4536
4537 if (!settings->rlimit[rl])
4538 continue;
4539
4540 if (!arg_settings_trusted) {
4541 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
4542 continue;
4543 }
4544
4545 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
4546 }
4547
4548 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
4549 settings->hostname)
4550 free_and_replace(arg_hostname, settings->hostname);
4551
4552 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
4553 settings->no_new_privileges >= 0)
4554 arg_no_new_privileges = settings->no_new_privileges;
4555
4556 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
4557 settings->oom_score_adjust_set) {
4558
4559 if (!arg_settings_trusted)
4560 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
4561 else {
4562 arg_oom_score_adjust = settings->oom_score_adjust;
4563 arg_oom_score_adjust_set = true;
4564 }
4565 }
4566
4567 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
4568 settings->cpu_set.set) {
4569
4570 if (!arg_settings_trusted)
4571 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
4572 else {
4573 cpu_set_reset(&arg_cpu_set);
4574 arg_cpu_set = TAKE_STRUCT(settings->cpu_set);
4575 }
4576 }
4577
4578 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
4579 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
4580 arg_resolv_conf = settings->resolv_conf;
4581
4582 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
4583 settings->link_journal != _LINK_JOURNAL_INVALID) {
4584
4585 if (!arg_settings_trusted)
4586 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
4587 else {
4588 arg_link_journal = settings->link_journal;
4589 arg_link_journal_try = settings->link_journal_try;
4590 }
4591 }
4592
4593 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
4594 settings->timezone != _TIMEZONE_MODE_INVALID)
4595 arg_timezone = settings->timezone;
4596
4597 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
4598 settings->slice) {
4599
4600 if (!arg_settings_trusted)
4601 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
4602 else
4603 free_and_replace(arg_slice, settings->slice);
4604 }
4605
4606 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
4607 settings->use_cgns >= 0) {
4608
4609 if (!arg_settings_trusted)
4610 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
4611 else
4612 arg_use_cgns = settings->use_cgns;
4613 }
4614
4615 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
4616 settings->clone_ns_flags != ULONG_MAX) {
4617
4618 if (!arg_settings_trusted)
4619 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
4620 else
4621 arg_clone_ns_flags = settings->clone_ns_flags;
4622 }
4623
4624 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
4625 settings->console_mode >= 0) {
4626
4627 if (!arg_settings_trusted)
4628 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
4629 else
4630 arg_console_mode = settings->console_mode;
4631 }
4632
4633 if ((arg_settings_mask & SETTING_SUPPRESS_SYNC) == 0 &&
4634 settings->suppress_sync >= 0)
4635 arg_suppress_sync = settings->suppress_sync;
4636
4637 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4638 * don't consult arg_settings_mask for them. */
4639
4640 sd_bus_message_unref(arg_property_message);
4641 arg_property_message = TAKE_PTR(settings->properties);
4642
4643 arg_console_width = settings->console_width;
4644 arg_console_height = settings->console_height;
4645
4646 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
4647 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
4648 arg_n_extra_nodes = settings->n_extra_nodes;
4649 settings->n_extra_nodes = 0;
4650
4651 return 0;
4652 }
4653
4654 static int load_settings(void) {
4655 _cleanup_(settings_freep) Settings *settings = NULL;
4656 _cleanup_fclose_ FILE *f = NULL;
4657 _cleanup_free_ char *p = NULL;
4658 int r;
4659
4660 if (arg_oci_bundle)
4661 return 0;
4662
4663 /* If all settings are masked, there's no point in looking for
4664 * the settings file */
4665 if (FLAGS_SET(arg_settings_mask, _SETTINGS_MASK_ALL))
4666 return 0;
4667
4668 /* We first look in the admin's directories in /etc and /run */
4669 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4670 _cleanup_free_ char *j = NULL;
4671
4672 j = path_join(i, arg_settings_filename);
4673 if (!j)
4674 return log_oom();
4675
4676 f = fopen(j, "re");
4677 if (f) {
4678 p = TAKE_PTR(j);
4679
4680 /* By default, we trust configuration from /etc and /run */
4681 if (arg_settings_trusted < 0)
4682 arg_settings_trusted = true;
4683
4684 break;
4685 }
4686
4687 if (errno != ENOENT)
4688 return log_error_errno(errno, "Failed to open %s: %m", j);
4689 }
4690
4691 if (!f) {
4692 /* After that, let's look for a file next to the
4693 * actual image we shall boot. */
4694
4695 if (arg_image) {
4696 r = file_in_same_dir(arg_image, arg_settings_filename, &p);
4697 if (r < 0)
4698 return log_error_errno(r, "Failed to generate settings path from image path: %m");
4699 } else if (arg_directory) {
4700 r = file_in_same_dir(arg_directory, arg_settings_filename, &p);
4701 if (r < 0 && r != -EADDRNOTAVAIL) /* if directory is root fs, don't complain */
4702 return log_error_errno(r, "Failed to generate settings path from directory path: %m");
4703 }
4704
4705 if (p) {
4706 f = fopen(p, "re");
4707 if (!f && errno != ENOENT)
4708 return log_error_errno(errno, "Failed to open %s: %m", p);
4709
4710 /* By default, we do not trust configuration from /var/lib/machines */
4711 if (arg_settings_trusted < 0)
4712 arg_settings_trusted = false;
4713 }
4714 }
4715
4716 if (!f)
4717 return 0;
4718
4719 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4720
4721 r = settings_load(f, p, &settings);
4722 if (r < 0)
4723 return r;
4724
4725 return merge_settings(settings, p);
4726 }
4727
4728 static int load_oci_bundle(void) {
4729 _cleanup_(settings_freep) Settings *settings = NULL;
4730 int r;
4731
4732 if (!arg_oci_bundle)
4733 return 0;
4734
4735 /* By default let's trust OCI bundles */
4736 if (arg_settings_trusted < 0)
4737 arg_settings_trusted = true;
4738
4739 r = oci_load(NULL, arg_oci_bundle, &settings);
4740 if (r < 0)
4741 return r;
4742
4743 return merge_settings(settings, arg_oci_bundle);
4744 }
4745
4746 static int run_container(
4747 DissectedImage *dissected_image,
4748 FDSet *fds,
4749 char veth_name[IFNAMSIZ], bool *veth_created,
4750 struct ExposeArgs *expose_args,
4751 int *master, pid_t *pid, int *ret) {
4752
4753 static const struct sigaction sa = {
4754 .sa_handler = nop_signal_handler,
4755 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4756 };
4757
4758 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4759 _cleanup_close_ int etc_passwd_lock = -EBADF;
4760 _cleanup_close_pair_ int
4761 fd_inner_socket_pair[2] = PIPE_EBADF,
4762 fd_outer_socket_pair[2] = PIPE_EBADF;
4763
4764 _cleanup_close_ int notify_socket = -EBADF, mntns_fd = -EBADF, fd_kmsg_fifo = -EBADF;
4765 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4766 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4767 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4768 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4769 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4770 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4771 _cleanup_free_ uid_t *bind_user_uid = NULL;
4772 size_t n_bind_user_uid = 0;
4773 ContainerStatus container_status = 0;
4774 int ifi = 0, r;
4775 ssize_t l;
4776 sigset_t mask_chld;
4777 _cleanup_close_ int child_netns_fd = -EBADF;
4778
4779 assert_se(sigemptyset(&mask_chld) == 0);
4780 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4781
4782 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4783 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4784 * check with getpwuid() if the specific user already exists. Note that /etc might be
4785 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4786 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4787 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4788 * really ours. */
4789
4790 etc_passwd_lock = take_etc_passwd_lock(NULL);
4791 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4792 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4793 }
4794
4795 r = barrier_create(&barrier);
4796 if (r < 0)
4797 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4798
4799 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, fd_inner_socket_pair) < 0)
4800 return log_error_errno(errno, "Failed to create inner socket pair: %m");
4801
4802 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, fd_outer_socket_pair) < 0)
4803 return log_error_errno(errno, "Failed to create outer socket pair: %m");
4804
4805 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4806 * parent's blocking calls and give it a chance to call wait() and terminate. */
4807 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4808 if (r < 0)
4809 return log_error_errno(errno, "Failed to change the signal mask: %m");
4810
4811 r = sigaction(SIGCHLD, &sa, NULL);
4812 if (r < 0)
4813 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4814
4815 if (arg_network_namespace_path) {
4816 child_netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4817 if (child_netns_fd < 0)
4818 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4819
4820 r = fd_is_ns(child_netns_fd, CLONE_NEWNET);
4821 if (r == -EUCLEAN)
4822 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);
4823 else if (r < 0)
4824 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4825 else if (r == 0)
4826 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4827 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4828 }
4829
4830 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4831 if (*pid < 0)
4832 return log_error_errno(errno, "clone() failed%s: %m",
4833 errno == EINVAL ?
4834 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4835
4836 if (*pid == 0) {
4837 /* The outer child only has a file system namespace. */
4838 barrier_set_role(&barrier, BARRIER_CHILD);
4839
4840 fd_inner_socket_pair[0] = safe_close(fd_inner_socket_pair[0]);
4841 fd_outer_socket_pair[0] = safe_close(fd_outer_socket_pair[0]);
4842
4843 (void) reset_all_signal_handlers();
4844 (void) reset_signal_mask();
4845
4846 r = outer_child(&barrier,
4847 arg_directory,
4848 dissected_image,
4849 fd_outer_socket_pair[1],
4850 fd_inner_socket_pair[1],
4851 fds,
4852 child_netns_fd);
4853 if (r < 0)
4854 _exit(EXIT_FAILURE);
4855
4856 _exit(EXIT_SUCCESS);
4857 }
4858
4859 barrier_set_role(&barrier, BARRIER_PARENT);
4860
4861 fdset_close(fds);
4862
4863 fd_inner_socket_pair[1] = safe_close(fd_inner_socket_pair[1]);
4864 fd_outer_socket_pair[1] = safe_close(fd_outer_socket_pair[1]);
4865
4866 if (arg_userns_mode != USER_NAMESPACE_NO) {
4867 mntns_fd = receive_one_fd(fd_outer_socket_pair[0], 0);
4868 if (mntns_fd < 0)
4869 return log_error_errno(mntns_fd, "Failed to receive mount namespace fd from outer child: %m");
4870
4871 /* The child just let us know the UID shift it might have read from the image. */
4872 l = recv(fd_outer_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4873 if (l < 0)
4874 return log_error_errno(errno, "Failed to read UID shift: %m");
4875 if (l != sizeof arg_uid_shift)
4876 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
4877
4878 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4879 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4880 * image, but if that's already in use, pick a new one, and report back to the child,
4881 * which one we now picked. */
4882
4883 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4884 if (r < 0)
4885 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4886
4887 l = send(fd_outer_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4888 if (l < 0)
4889 return log_error_errno(errno, "Failed to send UID shift: %m");
4890 if (l != sizeof arg_uid_shift)
4891 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
4892 }
4893
4894 n_bind_user_uid = strv_length(arg_bind_user);
4895 if (n_bind_user_uid > 0) {
4896 /* Right after the UID shift, we'll receive the list of UID mappings for the
4897 * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
4898
4899 bind_user_uid = new(uid_t, n_bind_user_uid*4);
4900 if (!bind_user_uid)
4901 return log_oom();
4902
4903 for (size_t i = 0; i < n_bind_user_uid; i++) {
4904 l = recv(fd_outer_socket_pair[0], bind_user_uid + i*4, sizeof(uid_t)*4, 0);
4905 if (l < 0)
4906 return log_error_errno(errno, "Failed to read user UID map pair: %m");
4907 if (l != sizeof(uid_t)*4)
4908 return log_full_errno(l == 0 ? LOG_DEBUG : LOG_WARNING,
4909 SYNTHETIC_ERRNO(EIO),
4910 "Short read while reading bind user UID pairs.");
4911 }
4912 }
4913 }
4914
4915 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4916 /* The child let us know the support cgroup mode it might have read from the image. */
4917 l = recv(fd_outer_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4918 if (l < 0)
4919 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4920 if (l != sizeof(arg_unified_cgroup_hierarchy))
4921 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zi bytes).%s",
4922 l, l == 0 ? " The child is most likely dead." : "");
4923 }
4924
4925 /* Wait for the outer child. */
4926 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4927 if (r < 0)
4928 return r;
4929 if (r != EXIT_SUCCESS)
4930 return -EIO;
4931
4932 /* And now retrieve the PID of the inner child. */
4933 l = recv(fd_outer_socket_pair[0], pid, sizeof *pid, 0);
4934 if (l < 0)
4935 return log_error_errno(errno, "Failed to read inner child PID: %m");
4936 if (l != sizeof *pid)
4937 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
4938
4939 /* We also retrieve container UUID in case it was generated by outer child */
4940 l = recv(fd_outer_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4941 if (l < 0)
4942 return log_error_errno(errno, "Failed to read container machine ID: %m");
4943 if (l != sizeof(arg_uuid))
4944 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
4945
4946 /* We also retrieve the socket used for notifications generated by outer child */
4947 notify_socket = receive_one_fd(fd_outer_socket_pair[0], 0);
4948 if (notify_socket < 0)
4949 return log_error_errno(notify_socket,
4950 "Failed to receive notification socket from the outer child: %m");
4951
4952 log_debug("Init process invoked as PID "PID_FMT, *pid);
4953
4954 if (arg_userns_mode != USER_NAMESPACE_NO) {
4955 if (!barrier_place_and_sync(&barrier)) /* #1 */
4956 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4957
4958 r = setup_uid_map(*pid, bind_user_uid, n_bind_user_uid);
4959 if (r < 0)
4960 return r;
4961
4962 (void) barrier_place(&barrier); /* #2 */
4963 }
4964
4965 if (arg_private_network) {
4966 if (!arg_network_namespace_path) {
4967 /* Wait until the child has unshared its network namespace. */
4968 if (!barrier_place_and_sync(&barrier)) /* #3 */
4969 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
4970 }
4971
4972 if (child_netns_fd < 0) {
4973 /* Make sure we have an open file descriptor to the child's network
4974 * namespace so it stays alive even if the child exits. */
4975 r = namespace_open(*pid, NULL, NULL, &child_netns_fd, NULL, NULL);
4976 if (r < 0)
4977 return log_error_errno(r, "Failed to open child network namespace: %m");
4978 }
4979
4980 r = move_network_interfaces(child_netns_fd, arg_network_interfaces);
4981 if (r < 0)
4982 return r;
4983
4984 if (arg_network_veth) {
4985 r = setup_veth(arg_machine, *pid, veth_name,
4986 arg_network_bridge || arg_network_zone);
4987 if (r < 0)
4988 return r;
4989 else if (r > 0)
4990 ifi = r;
4991
4992 if (arg_network_bridge) {
4993 /* Add the interface to a bridge */
4994 r = setup_bridge(veth_name, arg_network_bridge, false);
4995 if (r < 0)
4996 return r;
4997 if (r > 0)
4998 ifi = r;
4999 } else if (arg_network_zone) {
5000 /* Add the interface to a bridge, possibly creating it */
5001 r = setup_bridge(veth_name, arg_network_zone, true);
5002 if (r < 0)
5003 return r;
5004 if (r > 0)
5005 ifi = r;
5006 }
5007 }
5008
5009 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
5010 if (r < 0)
5011 return r;
5012
5013 /* We created the primary and extra veth links now; let's remember this, so that we know to
5014 remove them later on. Note that we don't bother with removing veth links that were created
5015 here when their setup failed half-way, because in that case the kernel should be able to
5016 remove them on its own, since they cannot be referenced by anything yet. */
5017 *veth_created = true;
5018
5019 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
5020 if (r < 0)
5021 return r;
5022
5023 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
5024 if (r < 0)
5025 return r;
5026 }
5027
5028 if (arg_register || !arg_keep_unit) {
5029 r = sd_bus_default_system(&bus);
5030 if (r < 0)
5031 return log_error_errno(r, "Failed to open system bus: %m");
5032
5033 r = sd_bus_set_close_on_exit(bus, false);
5034 if (r < 0)
5035 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
5036 }
5037
5038 if (!arg_keep_unit) {
5039 /* When a new scope is created for this container, then we'll be registered as its controller, in which
5040 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
5041 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
5042
5043 r = sd_bus_match_signal_async(
5044 bus,
5045 NULL,
5046 "org.freedesktop.systemd1",
5047 NULL,
5048 "org.freedesktop.systemd1.Scope",
5049 "RequestStop",
5050 on_request_stop, NULL, PID_TO_PTR(*pid));
5051 if (r < 0)
5052 return log_error_errno(r, "Failed to request RequestStop match: %m");
5053 }
5054
5055 if (arg_register) {
5056 r = register_machine(
5057 bus,
5058 arg_machine,
5059 *pid,
5060 arg_directory,
5061 arg_uuid,
5062 ifi,
5063 arg_slice,
5064 arg_custom_mounts, arg_n_custom_mounts,
5065 arg_kill_signal,
5066 arg_property,
5067 arg_property_message,
5068 arg_keep_unit,
5069 arg_container_service_name);
5070 if (r < 0)
5071 return r;
5072
5073 } else if (!arg_keep_unit) {
5074 r = allocate_scope(
5075 bus,
5076 arg_machine,
5077 *pid,
5078 arg_slice,
5079 arg_custom_mounts, arg_n_custom_mounts,
5080 arg_kill_signal,
5081 arg_property,
5082 arg_property_message);
5083 if (r < 0)
5084 return r;
5085
5086 } else if (arg_slice || arg_property)
5087 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
5088
5089 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
5090 if (r < 0)
5091 return r;
5092
5093 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
5094 if (r < 0)
5095 return r;
5096
5097 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
5098 if (r < 0)
5099 return r;
5100
5101 /* Notify the child that the parent is ready with all
5102 * its setup (including cgroup-ification), and that
5103 * the child can now hand over control to the code to
5104 * run inside the container. */
5105 (void) barrier_place(&barrier); /* #4 */
5106
5107 /* Block SIGCHLD here, before notifying child.
5108 * process_pty() will handle it with the other signals. */
5109 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
5110
5111 /* Reset signal to default */
5112 r = default_signals(SIGCHLD);
5113 if (r < 0)
5114 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
5115
5116 r = sd_event_new(&event);
5117 if (r < 0)
5118 return log_error_errno(r, "Failed to get default event source: %m");
5119
5120 (void) sd_event_set_watchdog(event, true);
5121
5122 if (bus) {
5123 r = sd_bus_attach_event(bus, event, 0);
5124 if (r < 0)
5125 return log_error_errno(r, "Failed to attach bus to event loop: %m");
5126 }
5127
5128 r = setup_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
5129 if (r < 0)
5130 return r;
5131
5132 if (arg_userns_mode != USER_NAMESPACE_NO) {
5133 r = wipe_fully_visible_fs(mntns_fd);
5134 if (r < 0)
5135 return r;
5136 mntns_fd = safe_close(mntns_fd);
5137 }
5138
5139 /* Let the child know that we are ready and wait that the child is completely ready now. */
5140 if (!barrier_place_and_sync(&barrier)) /* #5 */
5141 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
5142
5143 /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
5144 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
5145 etc_passwd_lock = safe_close(etc_passwd_lock);
5146
5147 (void) sd_notifyf(false,
5148 "STATUS=Container running.\n"
5149 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
5150 if (!arg_notify_ready) {
5151 r = sd_notify(false, "READY=1\n");
5152 if (r < 0)
5153 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
5154 }
5155
5156 if (arg_kill_signal > 0) {
5157 /* Try to kill the init system on SIGINT or SIGTERM */
5158 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
5159 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
5160 } else {
5161 /* Immediately exit */
5162 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
5163 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
5164 }
5165
5166 (void) sd_event_add_signal(event, NULL, SIGRTMIN+18, sigrtmin18_handler, NULL);
5167
5168 r = sd_event_add_memory_pressure(event, NULL, NULL, NULL);
5169 if (r < 0)
5170 log_debug_errno(r, "Failed allocate memory pressure event source, ignoring: %m");
5171
5172 /* Exit when the child exits */
5173 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
5174
5175 /* Retrieve the kmsg fifo allocated by inner child */
5176 fd_kmsg_fifo = receive_one_fd(fd_inner_socket_pair[0], 0);
5177 if (fd_kmsg_fifo < 0)
5178 return log_error_errno(fd_kmsg_fifo, "Failed to receive kmsg fifo from inner child: %m");
5179
5180 if (arg_expose_ports) {
5181 r = expose_port_watch_rtnl(event, fd_inner_socket_pair[0], on_address_change, expose_args, &rtnl);
5182 if (r < 0)
5183 return r;
5184
5185 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
5186 (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
5187 }
5188
5189 if (arg_console_mode != CONSOLE_PIPE) {
5190 _cleanup_close_ int fd = -EBADF;
5191 PTYForwardFlags flags = 0;
5192
5193 /* Retrieve the master pty allocated by inner child */
5194 fd = receive_one_fd(fd_inner_socket_pair[0], 0);
5195 if (fd < 0)
5196 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
5197
5198 switch (arg_console_mode) {
5199
5200 case CONSOLE_READ_ONLY:
5201 flags |= PTY_FORWARD_READ_ONLY;
5202
5203 _fallthrough_;
5204
5205 case CONSOLE_INTERACTIVE:
5206 flags |= PTY_FORWARD_IGNORE_VHANGUP;
5207
5208 r = pty_forward_new(event, fd, flags, &forward);
5209 if (r < 0)
5210 return log_error_errno(r, "Failed to create PTY forwarder: %m");
5211
5212 if (arg_console_width != UINT_MAX || arg_console_height != UINT_MAX)
5213 (void) pty_forward_set_width_height(forward,
5214 arg_console_width,
5215 arg_console_height);
5216 break;
5217
5218 default:
5219 assert(arg_console_mode == CONSOLE_PASSIVE);
5220 }
5221
5222 *master = TAKE_FD(fd);
5223 }
5224
5225 fd_inner_socket_pair[0] = safe_close(fd_inner_socket_pair[0]);
5226
5227 r = sd_event_loop(event);
5228 if (r < 0)
5229 return log_error_errno(r, "Failed to run event loop: %m");
5230
5231 if (forward) {
5232 char last_char = 0;
5233
5234 (void) pty_forward_get_last_char(forward, &last_char);
5235 forward = pty_forward_free(forward);
5236
5237 if (!arg_quiet && last_char != '\n')
5238 putc('\n', stdout);
5239 }
5240
5241 /* Kill if it is not dead yet anyway */
5242 if (!arg_register && !arg_keep_unit && bus)
5243 terminate_scope(bus, arg_machine);
5244
5245 /* Normally redundant, but better safe than sorry */
5246 (void) kill(*pid, SIGKILL);
5247
5248 fd_kmsg_fifo = safe_close(fd_kmsg_fifo);
5249
5250 if (arg_private_network) {
5251 /* Move network interfaces back to the parent network namespace. We use `safe_fork`
5252 * to avoid having to move the parent to the child network namespace. */
5253 r = safe_fork(NULL, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_WAIT|FORK_LOG, NULL);
5254 if (r < 0)
5255 return r;
5256
5257 if (r == 0) {
5258 _cleanup_close_ int parent_netns_fd = -EBADF;
5259
5260 r = namespace_open(getpid_cached(), NULL, NULL, &parent_netns_fd, NULL, NULL);
5261 if (r < 0) {
5262 log_error_errno(r, "Failed to open parent network namespace: %m");
5263 _exit(EXIT_FAILURE);
5264 }
5265
5266 r = namespace_enter(-1, -1, child_netns_fd, -1, -1);
5267 if (r < 0) {
5268 log_error_errno(r, "Failed to enter child network namespace: %m");
5269 _exit(EXIT_FAILURE);
5270 }
5271
5272 /* Reverse network interfaces pair list so that interfaces get their initial name back.
5273 * This is about ensuring interfaces get their old name back when being moved back. */
5274 arg_network_interfaces = strv_reverse(arg_network_interfaces);
5275
5276 r = move_network_interfaces(parent_netns_fd, arg_network_interfaces);
5277 if (r < 0)
5278 log_error_errno(r, "Failed to move network interfaces back to parent network namespace: %m");
5279
5280 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
5281 }
5282 }
5283
5284 r = wait_for_container(TAKE_PID(*pid), &container_status);
5285
5286 /* Tell machined that we are gone. */
5287 if (bus)
5288 (void) unregister_machine(bus, arg_machine);
5289
5290 if (r < 0)
5291 /* We failed to wait for the container, or the container exited abnormally. */
5292 return r;
5293 if (r > 0 || container_status == CONTAINER_TERMINATED) {
5294 /* r > 0 → The container exited with a non-zero status.
5295 * As a special case, we need to replace 133 with a different value,
5296 * because 133 is special-cased in the service file to reboot the container.
5297 * otherwise → The container exited with zero status and a reboot was not requested.
5298 */
5299 if (r == EXIT_FORCE_RESTART)
5300 r = EXIT_FAILURE; /* replace 133 with the general failure code */
5301 *ret = r;
5302 return 0; /* finito */
5303 }
5304
5305 /* CONTAINER_REBOOTED, loop again */
5306
5307 if (arg_keep_unit) {
5308 /* Special handling if we are running as a service: instead of simply
5309 * restarting the machine we want to restart the entire service, so let's
5310 * inform systemd about this with the special exit code 133. The service
5311 * file uses RestartForceExitStatus=133 so that this results in a full
5312 * nspawn restart. This is necessary since we might have cgroup parameters
5313 * set we want to have flushed out. */
5314 *ret = EXIT_FORCE_RESTART;
5315 return 0; /* finito */
5316 }
5317
5318 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
5319 expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
5320
5321 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5322 *veth_created = false;
5323 return 1; /* loop again */
5324 }
5325
5326 static int initialize_rlimits(void) {
5327 /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
5328 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
5329 * container execution environments. */
5330
5331 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
5332 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
5333 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
5334 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
5335 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
5336 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
5337 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
5338 [RLIMIT_MEMLOCK] = { DEFAULT_RLIMIT_MEMLOCK, DEFAULT_RLIMIT_MEMLOCK },
5339 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
5340 [RLIMIT_NICE] = { 0, 0 },
5341 [RLIMIT_NOFILE] = { 1024, 4096 },
5342 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
5343 [RLIMIT_RTPRIO] = { 0, 0 },
5344 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
5345 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
5346
5347 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
5348 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
5349 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
5350 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
5351 * that PID 1 changes a number of other resource limits during early initialization which is why we
5352 * don't read the other limits from PID 1 but prefer the static table above. */
5353 };
5354
5355 int rl;
5356
5357 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
5358 /* Let's only fill in what the user hasn't explicitly configured anyway */
5359 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
5360 const struct rlimit *v;
5361 struct rlimit buffer;
5362
5363 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
5364 /* For these two let's read the limits off PID 1. See above for an explanation. */
5365
5366 if (prlimit(1, rl, NULL, &buffer) < 0)
5367 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
5368
5369 v = &buffer;
5370 } else if (rl == RLIMIT_NOFILE) {
5371 /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
5372 * userspace. Given that nspawn containers are often run without our PID 1,
5373 * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
5374 * so that container userspace gets similar resources as host userspace
5375 * gets. */
5376 buffer = kernel_defaults[rl];
5377 buffer.rlim_max = MIN((rlim_t) read_nr_open(), (rlim_t) HIGH_RLIMIT_NOFILE);
5378 v = &buffer;
5379 } else
5380 v = kernel_defaults + rl;
5381
5382 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
5383 if (!arg_rlimit[rl])
5384 return log_oom();
5385 }
5386
5387 if (DEBUG_LOGGING) {
5388 _cleanup_free_ char *k = NULL;
5389
5390 (void) rlimit_format(arg_rlimit[rl], &k);
5391 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
5392 }
5393 }
5394
5395 return 0;
5396 }
5397
5398 static int cant_be_in_netns(void) {
5399 _cleanup_close_ int fd = -EBADF;
5400 struct ucred ucred;
5401 int r;
5402
5403 /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
5404 * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
5405 * nice message. */
5406
5407 if (!arg_image) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
5408 return 0;
5409
5410 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_NONBLOCK|SOCK_CLOEXEC, 0);
5411 if (fd < 0)
5412 return log_error_errno(errno, "Failed to allocate udev control socket: %m");
5413
5414 r = connect_unix_path(fd, AT_FDCWD, "/run/udev/control");
5415 if (r < 0) {
5416 if (r == -ENOENT || ERRNO_IS_DISCONNECT(r))
5417 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5418 "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
5419
5420 return log_error_errno(r, "Failed to connect socket to udev control socket: %m");
5421 }
5422
5423 r = getpeercred(fd, &ucred);
5424 if (r < 0)
5425 return log_error_errno(r, "Failed to determine peer of udev control socket: %m");
5426
5427 r = in_same_namespace(ucred.pid, 0, NAMESPACE_NET);
5428 if (r < 0)
5429 return log_error_errno(r, "Failed to determine network namespace of udev: %m");
5430 if (r == 0)
5431 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
5432 "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
5433 return 0;
5434 }
5435
5436 static int run(int argc, char *argv[]) {
5437 bool remove_directory = false, remove_image = false, veth_created = false, remove_tmprootdir = false;
5438 _cleanup_close_ int master = -EBADF;
5439 _cleanup_fdset_free_ FDSet *fds = NULL;
5440 int r, n_fd_passed, ret = EXIT_SUCCESS;
5441 char veth_name[IFNAMSIZ] = "";
5442 struct ExposeArgs expose_args = {};
5443 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
5444 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
5445 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
5446 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
5447 _cleanup_(fw_ctx_freep) FirewallContext *fw_ctx = NULL;
5448 pid_t pid = 0;
5449
5450 log_parse_environment();
5451 log_open();
5452
5453 r = parse_argv(argc, argv);
5454 if (r <= 0)
5455 goto finish;
5456
5457 if (geteuid() != 0) {
5458 r = log_warning_errno(SYNTHETIC_ERRNO(EPERM),
5459 argc >= 2 ? "Need to be root." :
5460 "Need to be root (and some arguments are usually required).\nHint: try --help");
5461 goto finish;
5462 }
5463
5464 r = cant_be_in_netns();
5465 if (r < 0)
5466 goto finish;
5467
5468 r = initialize_rlimits();
5469 if (r < 0)
5470 goto finish;
5471
5472 r = load_oci_bundle();
5473 if (r < 0)
5474 goto finish;
5475
5476 r = determine_names();
5477 if (r < 0)
5478 goto finish;
5479
5480 r = load_settings();
5481 if (r < 0)
5482 goto finish;
5483
5484 r = cg_unified();
5485 if (r < 0) {
5486 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
5487 goto finish;
5488 }
5489
5490 r = verify_arguments();
5491 if (r < 0)
5492 goto finish;
5493
5494 r = verify_network_interfaces_initialized();
5495 if (r < 0)
5496 goto finish;
5497
5498 /* Reapply environment settings. */
5499 (void) detect_unified_cgroup_hierarchy_from_environment();
5500
5501 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
5502 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
5503 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
5504 (void) ignore_signals(SIGPIPE);
5505
5506 n_fd_passed = sd_listen_fds(false);
5507 if (n_fd_passed > 0) {
5508 r = fdset_new_listen_fds(&fds, false);
5509 if (r < 0) {
5510 log_error_errno(r, "Failed to collect file descriptors: %m");
5511 goto finish;
5512 }
5513 }
5514
5515 /* The "default" umask. This is appropriate for most file and directory
5516 * operations performed by nspawn, and is the umask that will be used for
5517 * the child. Functions like copy_devnodes() change the umask temporarily. */
5518 umask(0022);
5519
5520 if (arg_directory) {
5521 assert(!arg_image);
5522
5523 /* Safety precaution: let's not allow running images from the live host OS image, as long as
5524 * /var from the host will propagate into container dynamically (because bad things happen if
5525 * two systems write to the same /var). Let's allow it for the special cases where /var is
5526 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
5527 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
5528 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5529 "Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
5530 goto finish;
5531 }
5532
5533 if (arg_ephemeral) {
5534 _cleanup_free_ char *np = NULL;
5535
5536 r = chase_and_update(&arg_directory, 0);
5537 if (r < 0)
5538 goto finish;
5539
5540 /* If the specified path is a mount point we generate the new snapshot immediately
5541 * inside it under a random name. However if the specified is not a mount point we
5542 * create the new snapshot in the parent directory, just next to it. */
5543 r = path_is_mount_point(arg_directory, NULL, 0);
5544 if (r < 0) {
5545 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
5546 goto finish;
5547 }
5548 if (r > 0)
5549 r = tempfn_random_child(arg_directory, "machine.", &np);
5550 else
5551 r = tempfn_random(arg_directory, "machine.", &np);
5552 if (r < 0) {
5553 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
5554 goto finish;
5555 }
5556
5557 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
5558 * only owned by us and no one else. */
5559 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5560 if (r < 0) {
5561 log_error_errno(r, "Failed to lock %s: %m", np);
5562 goto finish;
5563 }
5564
5565 {
5566 BLOCK_SIGNALS(SIGINT);
5567 r = btrfs_subvol_snapshot_at(AT_FDCWD, arg_directory, AT_FDCWD, np,
5568 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5569 BTRFS_SNAPSHOT_FALLBACK_COPY |
5570 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5571 BTRFS_SNAPSHOT_RECURSIVE |
5572 BTRFS_SNAPSHOT_QUOTA |
5573 BTRFS_SNAPSHOT_SIGINT);
5574 }
5575 if (r == -EINTR) {
5576 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
5577 goto finish;
5578 }
5579 if (r < 0) {
5580 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
5581 goto finish;
5582 }
5583
5584 free_and_replace(arg_directory, np);
5585 remove_directory = true;
5586 } else {
5587 r = chase_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
5588 if (r < 0)
5589 goto finish;
5590
5591 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5592 if (r == -EBUSY) {
5593 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
5594 goto finish;
5595 }
5596 if (r < 0) {
5597 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
5598 goto finish;
5599 }
5600
5601 if (arg_template) {
5602 r = chase_and_update(&arg_template, 0);
5603 if (r < 0)
5604 goto finish;
5605
5606 {
5607 BLOCK_SIGNALS(SIGINT);
5608 r = btrfs_subvol_snapshot_at(AT_FDCWD, arg_template, AT_FDCWD, arg_directory,
5609 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
5610 BTRFS_SNAPSHOT_FALLBACK_COPY |
5611 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
5612 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
5613 BTRFS_SNAPSHOT_RECURSIVE |
5614 BTRFS_SNAPSHOT_QUOTA |
5615 BTRFS_SNAPSHOT_SIGINT);
5616 }
5617 if (r == -EEXIST)
5618 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5619 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
5620 else if (r == -EINTR) {
5621 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
5622 goto finish;
5623 } else if (r < 0) {
5624 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
5625 goto finish;
5626 } else
5627 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
5628 "Populated %s from template %s.", arg_directory, arg_template);
5629 }
5630 }
5631
5632 if (arg_start_mode == START_BOOT) {
5633 _cleanup_free_ char *b = NULL;
5634 const char *p;
5635
5636 if (arg_pivot_root_new) {
5637 b = path_join(arg_directory, arg_pivot_root_new);
5638 if (!b)
5639 return log_oom();
5640
5641 p = b;
5642 } else
5643 p = arg_directory;
5644
5645 if (path_is_os_tree(p) <= 0) {
5646 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5647 "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
5648 goto finish;
5649 }
5650 } else {
5651 _cleanup_free_ char *p = NULL;
5652
5653 if (arg_pivot_root_new)
5654 p = path_join(arg_directory, arg_pivot_root_new, "/usr/");
5655 else
5656 p = path_join(arg_directory, "/usr/");
5657 if (!p)
5658 return log_oom();
5659
5660 if (laccess(p, F_OK) < 0) {
5661 r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
5662 "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory);
5663 goto finish;
5664 }
5665 }
5666
5667 } else {
5668 DissectImageFlags dissect_image_flags =
5669 DISSECT_IMAGE_GENERIC_ROOT |
5670 DISSECT_IMAGE_REQUIRE_ROOT |
5671 DISSECT_IMAGE_RELAX_VAR_CHECK |
5672 DISSECT_IMAGE_USR_NO_ROOT |
5673 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
5674 DISSECT_IMAGE_PIN_PARTITION_DEVICES;
5675 assert(arg_image);
5676 assert(!arg_template);
5677
5678 r = chase_and_update(&arg_image, 0);
5679 if (r < 0)
5680 goto finish;
5681
5682 if (arg_ephemeral) {
5683 _cleanup_free_ char *np = NULL;
5684
5685 r = tempfn_random(arg_image, "machine.", &np);
5686 if (r < 0) {
5687 log_error_errno(r, "Failed to generate name for image snapshot: %m");
5688 goto finish;
5689 }
5690
5691 /* Always take an exclusive lock on our own ephemeral copy. */
5692 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
5693 if (r < 0) {
5694 r = log_error_errno(r, "Failed to create image lock: %m");
5695 goto finish;
5696 }
5697
5698 {
5699 BLOCK_SIGNALS(SIGINT);
5700 r = copy_file_full(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600,
5701 FS_NOCOW_FL, FS_NOCOW_FL,
5702 COPY_REFLINK|COPY_CRTIME|COPY_SIGINT,
5703 NULL, NULL);
5704 }
5705 if (r == -EINTR) {
5706 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
5707 goto finish;
5708 }
5709 if (r < 0) {
5710 r = log_error_errno(r, "Failed to copy image file: %m");
5711 goto finish;
5712 }
5713
5714 free_and_replace(arg_image, np);
5715 remove_image = true;
5716 } else {
5717 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5718 if (r == -EBUSY) {
5719 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
5720 goto finish;
5721 }
5722 if (r < 0) {
5723 r = log_error_errno(r, "Failed to create image lock: %m");
5724 goto finish;
5725 }
5726
5727 r = verity_settings_load(
5728 &arg_verity_settings,
5729 arg_image, NULL, NULL);
5730 if (r < 0) {
5731 log_error_errno(r, "Failed to read verity artefacts for %s: %m", arg_image);
5732 goto finish;
5733 }
5734
5735 if (arg_verity_settings.data_path)
5736 dissect_image_flags |= DISSECT_IMAGE_NO_PARTITION_TABLE;
5737 }
5738
5739 if (!mkdtemp(tmprootdir)) {
5740 r = log_error_errno(errno, "Failed to create temporary directory: %m");
5741 goto finish;
5742 }
5743
5744 remove_tmprootdir = true;
5745
5746 arg_directory = strdup(tmprootdir);
5747 if (!arg_directory) {
5748 r = log_oom();
5749 goto finish;
5750 }
5751
5752 r = loop_device_make_by_path(
5753 arg_image,
5754 arg_read_only ? O_RDONLY : O_RDWR,
5755 /* sector_size= */ UINT32_MAX,
5756 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
5757 LOCK_SH,
5758 &loop);
5759 if (r < 0) {
5760 log_error_errno(r, "Failed to set up loopback block device: %m");
5761 goto finish;
5762 }
5763
5764 r = dissect_loop_device_and_warn(
5765 loop,
5766 &arg_verity_settings,
5767 /* mount_options=*/ NULL,
5768 arg_image_policy ?: &image_policy_container,
5769 dissect_image_flags,
5770 &dissected_image);
5771 if (r == -ENOPKG) {
5772 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5773 log_notice("Note that the disk image needs to\n"
5774 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5775 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5776 " c) or follow https://uapi-group.org/specifications/specs/discoverable_partitions_specification\n"
5777 " d) or contain a file system without a partition table\n"
5778 "in order to be bootable with systemd-nspawn.");
5779 goto finish;
5780 }
5781 if (r < 0)
5782 goto finish;
5783
5784 r = dissected_image_load_verity_sig_partition(
5785 dissected_image,
5786 loop->fd,
5787 &arg_verity_settings);
5788 if (r < 0)
5789 goto finish;
5790
5791 if (dissected_image->has_verity && !arg_verity_settings.root_hash && !dissected_image->has_verity_sig)
5792 log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
5793 "root hash signature found! Proceeding without integrity checking.", arg_image);
5794
5795 r = dissected_image_decrypt_interactively(
5796 dissected_image,
5797 NULL,
5798 &arg_verity_settings,
5799 0);
5800 if (r < 0)
5801 goto finish;
5802
5803 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5804 if (remove_image && unlink(arg_image) >= 0)
5805 remove_image = false;
5806
5807 if (arg_architecture < 0)
5808 arg_architecture = dissected_image_architecture(dissected_image);
5809 }
5810
5811 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5812 if (r < 0)
5813 goto finish;
5814
5815 if (arg_console_mode < 0)
5816 arg_console_mode =
5817 isatty(STDIN_FILENO) > 0 &&
5818 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
5819
5820 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5821 arg_quiet = true;
5822
5823 if (!arg_quiet)
5824 log_info("Spawning container %s on %s.\nPress Ctrl-] three times within 1s to kill container.",
5825 arg_machine, arg_image ?: arg_directory);
5826
5827 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, SIGRTMIN+18, -1) >= 0);
5828
5829 r = make_reaper_process(true);
5830 if (r < 0) {
5831 log_error_errno(r, "Failed to become subreaper: %m");
5832 goto finish;
5833 }
5834
5835 if (arg_expose_ports) {
5836 r = fw_ctx_new(&fw_ctx);
5837 if (r < 0) {
5838 log_error_errno(r, "Cannot expose configured ports, firewall initialization failed: %m");
5839 goto finish;
5840 }
5841 expose_args.fw_ctx = fw_ctx;
5842 }
5843 for (;;) {
5844 r = run_container(dissected_image,
5845 fds,
5846 veth_name, &veth_created,
5847 &expose_args, &master,
5848 &pid, &ret);
5849 if (r <= 0)
5850 break;
5851 }
5852
5853 finish:
5854 (void) sd_notify(false,
5855 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5856 "STOPPING=1\nSTATUS=Terminating...");
5857
5858 if (pid > 0)
5859 (void) kill(pid, SIGKILL);
5860
5861 /* Try to flush whatever is still queued in the pty */
5862 if (master >= 0) {
5863 (void) copy_bytes(master, STDOUT_FILENO, UINT64_MAX, 0);
5864 master = safe_close(master);
5865 }
5866
5867 if (pid > 0)
5868 (void) wait_for_terminate(pid, NULL);
5869
5870 pager_close();
5871
5872 if (remove_directory && arg_directory) {
5873 int k;
5874
5875 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5876 if (k < 0)
5877 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5878 }
5879
5880 if (remove_image && arg_image) {
5881 if (unlink(arg_image) < 0)
5882 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5883 }
5884
5885 if (remove_tmprootdir) {
5886 if (rmdir(tmprootdir) < 0)
5887 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5888 }
5889
5890 if (arg_machine) {
5891 const char *p;
5892
5893 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5894 (void) rm_rf(p, REMOVE_ROOT);
5895 }
5896
5897 expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET, &expose_args.address4);
5898 expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET6, &expose_args.address6);
5899
5900 if (veth_created)
5901 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5902 (void) remove_bridge(arg_network_zone);
5903
5904 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5905 expose_port_free_all(arg_expose_ports);
5906 rlimit_free_all(arg_rlimit);
5907 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
5908 credential_free_all(arg_credentials, arg_n_credentials);
5909
5910 if (r < 0)
5911 return r;
5912
5913 return ret;
5914 }
5915
5916 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.