]> 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
id128: introduce ID128_UUID_STRING_MAX for sizing UUID buffers
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
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/personality.h>
15 #include <sys/prctl.h>
16 #include <sys/types.h>
17 #include <sys/wait.h>
18 #include <unistd.h>
19
20 #include "sd-bus.h"
21 #include "sd-daemon.h"
22 #include "sd-id128.h"
23
24 #include "alloc-util.h"
25 #include "barrier.h"
26 #include "base-filesystem.h"
27 #include "blkid-util.h"
28 #include "btrfs-util.h"
29 #include "bus-error.h"
30 #include "bus-util.h"
31 #include "cap-list.h"
32 #include "capability-util.h"
33 #include "cgroup-util.h"
34 #include "copy.h"
35 #include "cpu-set-util.h"
36 #include "dev-setup.h"
37 #include "dissect-image.h"
38 #include "env-util.h"
39 #include "fd-util.h"
40 #include "fdset.h"
41 #include "fileio.h"
42 #include "format-util.h"
43 #include "fs-util.h"
44 #include "gpt.h"
45 #include "hexdecoct.h"
46 #include "hostname-util.h"
47 #include "id128-util.h"
48 #include "log.h"
49 #include "loop-util.h"
50 #include "loopback-setup.h"
51 #include "machine-image.h"
52 #include "macro.h"
53 #include "main-func.h"
54 #include "missing_sched.h"
55 #include "mkdir.h"
56 #include "mount-util.h"
57 #include "mountpoint-util.h"
58 #include "namespace-util.h"
59 #include "netlink-util.h"
60 #include "nspawn-cgroup.h"
61 #include "nspawn-def.h"
62 #include "nspawn-expose-ports.h"
63 #include "nspawn-mount.h"
64 #include "nspawn-network.h"
65 #include "nspawn-oci.h"
66 #include "nspawn-patch-uid.h"
67 #include "nspawn-register.h"
68 #include "nspawn-seccomp.h"
69 #include "nspawn-settings.h"
70 #include "nspawn-setuid.h"
71 #include "nspawn-stub-pid1.h"
72 #include "nulstr-util.h"
73 #include "os-util.h"
74 #include "pager.h"
75 #include "parse-util.h"
76 #include "path-util.h"
77 #include "pretty-print.h"
78 #include "process-util.h"
79 #include "ptyfwd.h"
80 #include "random-util.h"
81 #include "raw-clone.h"
82 #include "rlimit-util.h"
83 #include "rm-rf.h"
84 #if HAVE_SECCOMP
85 #include "seccomp-util.h"
86 #endif
87 #include "selinux-util.h"
88 #include "signal-util.h"
89 #include "socket-util.h"
90 #include "stat-util.h"
91 #include "stdio-util.h"
92 #include "string-table.h"
93 #include "string-util.h"
94 #include "strv.h"
95 #include "sysctl-util.h"
96 #include "terminal-util.h"
97 #include "tmpfile-util.h"
98 #include "umask-util.h"
99 #include "unit-name.h"
100 #include "user-util.h"
101 #include "util.h"
102
103 #if HAVE_SPLIT_USR
104 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
105 #else
106 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
107 #endif
108
109 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
110 * nspawn_notify_socket_path is relative to the container
111 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
112 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
113
114 #define EXIT_FORCE_RESTART 133
115
116 typedef enum ContainerStatus {
117 CONTAINER_TERMINATED,
118 CONTAINER_REBOOTED,
119 } ContainerStatus;
120
121 static char *arg_directory = NULL;
122 static char *arg_template = NULL;
123 static char *arg_chdir = NULL;
124 static char *arg_pivot_root_new = NULL;
125 static char *arg_pivot_root_old = NULL;
126 static char *arg_user = NULL;
127 static uid_t arg_uid = UID_INVALID;
128 static gid_t arg_gid = GID_INVALID;
129 static gid_t* arg_supplementary_gids = NULL;
130 static size_t arg_n_supplementary_gids = 0;
131 static sd_id128_t arg_uuid = {};
132 static char *arg_machine = NULL; /* The name used by the host to refer to this */
133 static char *arg_hostname = NULL; /* The name the payload sees by default */
134 static const char *arg_selinux_context = NULL;
135 static const char *arg_selinux_apifs_context = NULL;
136 static char *arg_slice = NULL;
137 static bool arg_private_network = false;
138 static bool arg_read_only = false;
139 static StartMode arg_start_mode = START_PID1;
140 static bool arg_ephemeral = false;
141 static LinkJournal arg_link_journal = LINK_AUTO;
142 static bool arg_link_journal_try = false;
143 static uint64_t arg_caps_retain =
144 (1ULL << CAP_AUDIT_CONTROL) |
145 (1ULL << CAP_AUDIT_WRITE) |
146 (1ULL << CAP_CHOWN) |
147 (1ULL << CAP_DAC_OVERRIDE) |
148 (1ULL << CAP_DAC_READ_SEARCH) |
149 (1ULL << CAP_FOWNER) |
150 (1ULL << CAP_FSETID) |
151 (1ULL << CAP_IPC_OWNER) |
152 (1ULL << CAP_KILL) |
153 (1ULL << CAP_LEASE) |
154 (1ULL << CAP_LINUX_IMMUTABLE) |
155 (1ULL << CAP_MKNOD) |
156 (1ULL << CAP_NET_BIND_SERVICE) |
157 (1ULL << CAP_NET_BROADCAST) |
158 (1ULL << CAP_NET_RAW) |
159 (1ULL << CAP_SETFCAP) |
160 (1ULL << CAP_SETGID) |
161 (1ULL << CAP_SETPCAP) |
162 (1ULL << CAP_SETUID) |
163 (1ULL << CAP_SYS_ADMIN) |
164 (1ULL << CAP_SYS_BOOT) |
165 (1ULL << CAP_SYS_CHROOT) |
166 (1ULL << CAP_SYS_NICE) |
167 (1ULL << CAP_SYS_PTRACE) |
168 (1ULL << CAP_SYS_RESOURCE) |
169 (1ULL << CAP_SYS_TTY_CONFIG);
170 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
171 static CustomMount *arg_custom_mounts = NULL;
172 static size_t arg_n_custom_mounts = 0;
173 static char **arg_setenv = NULL;
174 static bool arg_quiet = false;
175 static bool arg_register = true;
176 static bool arg_keep_unit = false;
177 static char **arg_network_interfaces = NULL;
178 static char **arg_network_macvlan = NULL;
179 static char **arg_network_ipvlan = NULL;
180 static bool arg_network_veth = false;
181 static char **arg_network_veth_extra = NULL;
182 static char *arg_network_bridge = NULL;
183 static char *arg_network_zone = NULL;
184 static char *arg_network_namespace_path = NULL;
185 static PagerFlags arg_pager_flags = 0;
186 static unsigned long arg_personality = PERSONALITY_INVALID;
187 static char *arg_image = NULL;
188 static char *arg_oci_bundle = NULL;
189 static VolatileMode arg_volatile_mode = VOLATILE_NO;
190 static ExposePort *arg_expose_ports = NULL;
191 static char **arg_property = NULL;
192 static sd_bus_message *arg_property_message = NULL;
193 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
194 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
195 static bool arg_userns_chown = false;
196 static int arg_kill_signal = 0;
197 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
198 static SettingsMask arg_settings_mask = 0;
199 static int arg_settings_trusted = -1;
200 static char **arg_parameters = NULL;
201 static const char *arg_container_service_name = "systemd-nspawn";
202 static bool arg_notify_ready = false;
203 static bool arg_use_cgns = true;
204 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
205 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
206 static void *arg_root_hash = NULL;
207 static size_t arg_root_hash_size = 0;
208 static char **arg_syscall_whitelist = NULL;
209 static char **arg_syscall_blacklist = NULL;
210 #if HAVE_SECCOMP
211 static scmp_filter_ctx arg_seccomp = NULL;
212 #endif
213 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
214 static bool arg_no_new_privileges = false;
215 static int arg_oom_score_adjust = 0;
216 static bool arg_oom_score_adjust_set = false;
217 static CPUSet arg_cpu_set = {};
218 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
219 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
220 static unsigned arg_console_width = (unsigned) -1, arg_console_height = (unsigned) -1;
221 static DeviceNode* arg_extra_nodes = NULL;
222 static size_t arg_n_extra_nodes = 0;
223 static char **arg_sysctl = NULL;
224 static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
225
226 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
227 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
228 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
229 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
230 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
231 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
232 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
233 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
234 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
235 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
236 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
237 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
238 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
239 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
240 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
241 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
248 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_root_hash, freep);
250 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist, strv_freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist, strv_freep);
252 #if HAVE_SECCOMP
253 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
254 #endif
255 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
256 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
257
258 static int handle_arg_console(const char *arg) {
259 if (streq(arg, "help")) {
260 puts("interactive\n"
261 "read-only\n"
262 "passive\n"
263 "pipe");
264 return 0;
265 }
266
267 if (streq(arg, "interactive"))
268 arg_console_mode = CONSOLE_INTERACTIVE;
269 else if (streq(arg, "read-only"))
270 arg_console_mode = CONSOLE_READ_ONLY;
271 else if (streq(arg, "passive"))
272 arg_console_mode = CONSOLE_PASSIVE;
273 else if (streq(arg, "pipe"))
274 arg_console_mode = CONSOLE_PIPE;
275 else
276 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
277
278 arg_settings_mask |= SETTING_CONSOLE_MODE;
279 return 1;
280 }
281
282 static int help(void) {
283 _cleanup_free_ char *link = NULL;
284 int r;
285
286 (void) pager_open(arg_pager_flags);
287
288 r = terminal_urlify_man("systemd-nspawn", "1", &link);
289 if (r < 0)
290 return log_oom();
291
292 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
293 "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
294 " -h --help Show this help\n"
295 " --version Print version string\n"
296 " -q --quiet Do not show status information\n"
297 " --no-pager Do not pipe output into a pager\n"
298 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
299 "%3$sImage:%4$s\n"
300 " -D --directory=PATH Root directory for the container\n"
301 " --template=PATH Initialize root directory from template directory,\n"
302 " if missing\n"
303 " -x --ephemeral Run container with snapshot of root directory, and\n"
304 " remove it after exit\n"
305 " -i --image=PATH Root file system disk image (or device node) for\n"
306 " the container\n"
307 " --oci-bundle=PATH OCI bundle directory\n"
308 " --read-only Mount the root directory read-only\n"
309 " --volatile[=MODE] Run the system in volatile mode\n"
310 " --root-hash=HASH Specify verity root hash for root disk image\n"
311 " --pivot-root=PATH[:PATH]\n"
312 " Pivot root to given directory in the container\n\n"
313 "%3$sExecution:%4$s\n"
314 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
315 " -b --boot Boot up full system (i.e. invoke init)\n"
316 " --chdir=PATH Set working directory in the container\n"
317 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
318 " -u --user=USER Run the command under specified user or UID\n"
319 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
320 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
321 "%3$sSystem Identity:%4$s\n"
322 " -M --machine=NAME Set the machine name for the container\n"
323 " --hostname=NAME Override the hostname for the container\n"
324 " --uuid=UUID Set a specific machine UUID for the container\n\n"
325 "%3$sProperties:%4$s\n"
326 " -S --slice=SLICE Place the container in the specified slice\n"
327 " --property=NAME=VALUE Set scope unit property\n"
328 " --register=BOOLEAN Register container as machine\n"
329 " --keep-unit Do not register a scope for the machine, reuse\n"
330 " the service unit nspawn is running in\n\n"
331 "%3$sUser Namespacing:%4$s\n"
332 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
333 " --private-users[=UIDBASE[:NUIDS]]\n"
334 " Similar, but with user configured UID/GID range\n"
335 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
336 "%3$sNetworking:%4$s\n"
337 " --private-network Disable network in container\n"
338 " --network-interface=INTERFACE\n"
339 " Assign an existing network interface to the\n"
340 " container\n"
341 " --network-macvlan=INTERFACE\n"
342 " Create a macvlan network interface based on an\n"
343 " existing network interface to the container\n"
344 " --network-ipvlan=INTERFACE\n"
345 " Create a ipvlan network interface based on an\n"
346 " existing network interface to the container\n"
347 " -n --network-veth Add a virtual Ethernet connection between host\n"
348 " and container\n"
349 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
350 " Add an additional virtual Ethernet link between\n"
351 " host and container\n"
352 " --network-bridge=INTERFACE\n"
353 " Add a virtual Ethernet connection to the container\n"
354 " and attach it to an existing bridge on the host\n"
355 " --network-zone=NAME Similar, but attach the new interface to an\n"
356 " an automatically managed bridge interface\n"
357 " --network-namespace-path=PATH\n"
358 " Set network namespace to the one represented by\n"
359 " the specified kernel namespace file node\n"
360 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
361 " Expose a container IP port on the host\n\n"
362 "%3$sSecurity:%4$s\n"
363 " --capability=CAP In addition to the default, retain specified\n"
364 " capability\n"
365 " --drop-capability=CAP Drop the specified capability from the default set\n"
366 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
367 " --system-call-filter=LIST|~LIST\n"
368 " Permit/prohibit specific system calls\n"
369 " -Z --selinux-context=SECLABEL\n"
370 " Set the SELinux security context to be used by\n"
371 " processes in the container\n"
372 " -L --selinux-apifs-context=SECLABEL\n"
373 " Set the SELinux security context to be used by\n"
374 " API/tmpfs file systems in the container\n\n"
375 "%3$sResources:%4$s\n"
376 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
377 " --oom-score-adjust=VALUE\n"
378 " Adjust the OOM score value for the payload\n"
379 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
380 " --personality=ARCH Pick personality for this container\n\n"
381 "%3$sIntegration:%4$s\n"
382 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
383 " --timezone=MODE Select mode of /etc/localtime initialization\n"
384 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
385 " host, try-guest, try-host\n"
386 " -j Equivalent to --link-journal=try-guest\n\n"
387 "%3$sMounts:%4$s\n"
388 " --bind=PATH[:PATH[:OPTIONS]]\n"
389 " Bind mount a file or directory from the host into\n"
390 " the container\n"
391 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
392 " Similar, but creates a read-only bind mount\n"
393 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
394 " it\n"
395 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
396 " --overlay=PATH[:PATH...]:PATH\n"
397 " Create an overlay mount from the host to \n"
398 " the container\n"
399 " --overlay-ro=PATH[:PATH...]:PATH\n"
400 " Similar, but creates a read-only overlay mount\n\n"
401 "%3$sInput/Output:%4$s\n"
402 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
403 " set up for the container.\n"
404 " -P --pipe Equivalent to --console=pipe\n"
405 "\nSee the %2$s for details.\n"
406 , program_invocation_short_name
407 , link
408 , ansi_underline(), ansi_normal()
409 , ansi_highlight(), ansi_normal()
410 );
411
412 return 0;
413 }
414
415 static int custom_mount_check_all(void) {
416 size_t i;
417
418 for (i = 0; i < arg_n_custom_mounts; i++) {
419 CustomMount *m = &arg_custom_mounts[i];
420
421 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
422 if (arg_userns_chown)
423 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
424 "--private-users-chown may not be combined with custom root mounts.");
425 else if (arg_uid_shift == UID_INVALID)
426 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
427 "--private-users with automatic UID shift may not be combined with custom root mounts.");
428 }
429 }
430
431 return 0;
432 }
433
434 static int detect_unified_cgroup_hierarchy_from_environment(void) {
435 const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
436 int r;
437
438 /* Allow the user to control whether the unified hierarchy is used */
439
440 e = getenv(var);
441 if (!e) {
442 /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
443 var = "UNIFIED_CGROUP_HIERARCHY";
444 e = getenv(var);
445 }
446
447 if (!isempty(e)) {
448 r = parse_boolean(e);
449 if (r < 0)
450 return log_error_errno(r, "Failed to parse $%s: %m", var);
451 if (r > 0)
452 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
453 else
454 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
455 }
456
457 return 0;
458 }
459
460 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
461 int r;
462
463 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
464 * in the image actually supports. */
465 r = cg_all_unified();
466 if (r < 0)
467 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
468 if (r > 0) {
469 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
470 * routine only detects 231, so we'll have a false negative here for 230. */
471 r = systemd_installation_has_version(directory, 230);
472 if (r < 0)
473 return log_error_errno(r, "Failed to determine systemd version in container: %m");
474 if (r > 0)
475 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
476 else
477 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
478 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
479 /* Mixed cgroup hierarchy support was added in 233 */
480 r = systemd_installation_has_version(directory, 233);
481 if (r < 0)
482 return log_error_errno(r, "Failed to determine systemd version in container: %m");
483 if (r > 0)
484 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
485 else
486 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
487 } else
488 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
489
490 log_debug("Using %s hierarchy for container.",
491 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
492 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
493
494 return 0;
495 }
496
497 static int parse_capability_spec(const char *spec, uint64_t *ret_mask) {
498 uint64_t mask = 0;
499 int r;
500
501 for (;;) {
502 _cleanup_free_ char *t = NULL;
503
504 r = extract_first_word(&spec, &t, ",", 0);
505 if (r < 0)
506 return log_error_errno(r, "Failed to parse capability %s.", t);
507 if (r == 0)
508 break;
509
510 if (streq(t, "help")) {
511 for (int i = 0; i < capability_list_length(); i++) {
512 const char *name;
513
514 name = capability_to_name(i);
515 if (name)
516 puts(name);
517 }
518
519 return 0; /* quit */
520 }
521
522 if (streq(t, "all"))
523 mask = (uint64_t) -1;
524 else {
525 r = capability_from_name(t);
526 if (r < 0)
527 return log_error_errno(r, "Failed to parse capability %s.", t);
528
529 mask |= 1ULL << r;
530 }
531 }
532
533 *ret_mask = mask;
534 return 1; /* continue */
535 }
536
537 static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
538 int r;
539
540 r = getenv_bool(name);
541 if (r == -ENXIO)
542 return 0;
543 if (r < 0)
544 return log_error_errno(r, "Failed to parse $%s: %m", name);
545
546 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
547 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
548 return 0;
549 }
550
551 static int parse_mount_settings_env(void) {
552 const char *e;
553 int r;
554
555 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
556 if (r < 0 && r != -ENXIO)
557 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
558 if (r >= 0)
559 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
560
561 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
562 if (streq_ptr(e, "network"))
563 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
564
565 else if (e) {
566 r = parse_boolean(e);
567 if (r < 0)
568 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
569
570 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
571 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
572 }
573
574 return 0;
575 }
576
577 static int parse_environment(void) {
578 const char *e;
579 int r;
580
581 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
582 if (r < 0)
583 return r;
584 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
585 if (r < 0)
586 return r;
587 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
588 if (r < 0)
589 return r;
590 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
591 if (r < 0)
592 return r;
593
594 r = parse_mount_settings_env();
595 if (r < 0)
596 return r;
597
598 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
599 * even if it is supported. If not supported, it has no effect. */
600 if (!cg_ns_supported())
601 arg_use_cgns = false;
602 else {
603 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
604 if (r < 0) {
605 if (r != -ENXIO)
606 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
607
608 arg_use_cgns = true;
609 } else {
610 arg_use_cgns = r > 0;
611 arg_settings_mask |= SETTING_USE_CGNS;
612 }
613 }
614
615 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
616 if (e)
617 arg_container_service_name = e;
618
619 return detect_unified_cgroup_hierarchy_from_environment();
620 }
621
622 static int parse_argv(int argc, char *argv[]) {
623 enum {
624 ARG_VERSION = 0x100,
625 ARG_PRIVATE_NETWORK,
626 ARG_UUID,
627 ARG_READ_ONLY,
628 ARG_CAPABILITY,
629 ARG_DROP_CAPABILITY,
630 ARG_LINK_JOURNAL,
631 ARG_BIND,
632 ARG_BIND_RO,
633 ARG_TMPFS,
634 ARG_OVERLAY,
635 ARG_OVERLAY_RO,
636 ARG_INACCESSIBLE,
637 ARG_SHARE_SYSTEM,
638 ARG_REGISTER,
639 ARG_KEEP_UNIT,
640 ARG_NETWORK_INTERFACE,
641 ARG_NETWORK_MACVLAN,
642 ARG_NETWORK_IPVLAN,
643 ARG_NETWORK_BRIDGE,
644 ARG_NETWORK_ZONE,
645 ARG_NETWORK_VETH_EXTRA,
646 ARG_NETWORK_NAMESPACE_PATH,
647 ARG_PERSONALITY,
648 ARG_VOLATILE,
649 ARG_TEMPLATE,
650 ARG_PROPERTY,
651 ARG_PRIVATE_USERS,
652 ARG_KILL_SIGNAL,
653 ARG_SETTINGS,
654 ARG_CHDIR,
655 ARG_PIVOT_ROOT,
656 ARG_PRIVATE_USERS_CHOWN,
657 ARG_NOTIFY_READY,
658 ARG_ROOT_HASH,
659 ARG_SYSTEM_CALL_FILTER,
660 ARG_RLIMIT,
661 ARG_HOSTNAME,
662 ARG_NO_NEW_PRIVILEGES,
663 ARG_OOM_SCORE_ADJUST,
664 ARG_CPU_AFFINITY,
665 ARG_RESOLV_CONF,
666 ARG_TIMEZONE,
667 ARG_CONSOLE,
668 ARG_PIPE,
669 ARG_OCI_BUNDLE,
670 ARG_NO_PAGER,
671 };
672
673 static const struct option options[] = {
674 { "help", no_argument, NULL, 'h' },
675 { "version", no_argument, NULL, ARG_VERSION },
676 { "directory", required_argument, NULL, 'D' },
677 { "template", required_argument, NULL, ARG_TEMPLATE },
678 { "ephemeral", no_argument, NULL, 'x' },
679 { "user", required_argument, NULL, 'u' },
680 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
681 { "as-pid2", no_argument, NULL, 'a' },
682 { "boot", no_argument, NULL, 'b' },
683 { "uuid", required_argument, NULL, ARG_UUID },
684 { "read-only", no_argument, NULL, ARG_READ_ONLY },
685 { "capability", required_argument, NULL, ARG_CAPABILITY },
686 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
687 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
688 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
689 { "bind", required_argument, NULL, ARG_BIND },
690 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
691 { "tmpfs", required_argument, NULL, ARG_TMPFS },
692 { "overlay", required_argument, NULL, ARG_OVERLAY },
693 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
694 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
695 { "machine", required_argument, NULL, 'M' },
696 { "hostname", required_argument, NULL, ARG_HOSTNAME },
697 { "slice", required_argument, NULL, 'S' },
698 { "setenv", required_argument, NULL, 'E' },
699 { "selinux-context", required_argument, NULL, 'Z' },
700 { "selinux-apifs-context", required_argument, NULL, 'L' },
701 { "quiet", no_argument, NULL, 'q' },
702 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
703 { "register", required_argument, NULL, ARG_REGISTER },
704 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
705 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
706 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
707 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
708 { "network-veth", no_argument, NULL, 'n' },
709 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
710 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
711 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
712 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
713 { "personality", required_argument, NULL, ARG_PERSONALITY },
714 { "image", required_argument, NULL, 'i' },
715 { "volatile", optional_argument, NULL, ARG_VOLATILE },
716 { "port", required_argument, NULL, 'p' },
717 { "property", required_argument, NULL, ARG_PROPERTY },
718 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
719 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
720 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
721 { "settings", required_argument, NULL, ARG_SETTINGS },
722 { "chdir", required_argument, NULL, ARG_CHDIR },
723 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
724 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
725 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
726 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
727 { "rlimit", required_argument, NULL, ARG_RLIMIT },
728 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
729 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
730 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
731 { "timezone", required_argument, NULL, ARG_TIMEZONE },
732 { "console", required_argument, NULL, ARG_CONSOLE },
733 { "pipe", no_argument, NULL, ARG_PIPE },
734 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
735 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
736 {}
737 };
738
739 int c, r;
740 uint64_t plus = 0, minus = 0;
741 bool mask_all_settings = false, mask_no_settings = false;
742
743 assert(argc >= 0);
744 assert(argv);
745
746 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
747 switch (c) {
748
749 case 'h':
750 return help();
751
752 case ARG_VERSION:
753 return version();
754
755 case 'D':
756 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
757 if (r < 0)
758 return r;
759
760 arg_settings_mask |= SETTING_DIRECTORY;
761 break;
762
763 case ARG_TEMPLATE:
764 r = parse_path_argument_and_warn(optarg, false, &arg_template);
765 if (r < 0)
766 return r;
767
768 arg_settings_mask |= SETTING_DIRECTORY;
769 break;
770
771 case 'i':
772 r = parse_path_argument_and_warn(optarg, false, &arg_image);
773 if (r < 0)
774 return r;
775
776 arg_settings_mask |= SETTING_DIRECTORY;
777 break;
778
779 case ARG_OCI_BUNDLE:
780 r = parse_path_argument_and_warn(optarg, false, &arg_oci_bundle);
781 if (r < 0)
782 return r;
783
784 break;
785
786 case 'x':
787 arg_ephemeral = true;
788 arg_settings_mask |= SETTING_EPHEMERAL;
789 break;
790
791 case 'u':
792 r = free_and_strdup(&arg_user, optarg);
793 if (r < 0)
794 return log_oom();
795
796 arg_settings_mask |= SETTING_USER;
797 break;
798
799 case ARG_NETWORK_ZONE: {
800 char *j;
801
802 j = strjoin("vz-", optarg);
803 if (!j)
804 return log_oom();
805
806 if (!ifname_valid(j)) {
807 log_error("Network zone name not valid: %s", j);
808 free(j);
809 return -EINVAL;
810 }
811
812 free_and_replace(arg_network_zone, j);
813
814 arg_network_veth = true;
815 arg_private_network = true;
816 arg_settings_mask |= SETTING_NETWORK;
817 break;
818 }
819
820 case ARG_NETWORK_BRIDGE:
821
822 if (!ifname_valid(optarg))
823 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
824 "Bridge interface name not valid: %s", optarg);
825
826 r = free_and_strdup(&arg_network_bridge, optarg);
827 if (r < 0)
828 return log_oom();
829
830 _fallthrough_;
831 case 'n':
832 arg_network_veth = true;
833 arg_private_network = true;
834 arg_settings_mask |= SETTING_NETWORK;
835 break;
836
837 case ARG_NETWORK_VETH_EXTRA:
838 r = veth_extra_parse(&arg_network_veth_extra, optarg);
839 if (r < 0)
840 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
841
842 arg_private_network = true;
843 arg_settings_mask |= SETTING_NETWORK;
844 break;
845
846 case ARG_NETWORK_INTERFACE:
847 if (!ifname_valid(optarg))
848 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
849 "Network interface name not valid: %s", optarg);
850
851 if (strv_extend(&arg_network_interfaces, optarg) < 0)
852 return log_oom();
853
854 arg_private_network = true;
855 arg_settings_mask |= SETTING_NETWORK;
856 break;
857
858 case ARG_NETWORK_MACVLAN:
859
860 if (!ifname_valid(optarg))
861 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
862 "MACVLAN network interface name not valid: %s", optarg);
863
864 if (strv_extend(&arg_network_macvlan, optarg) < 0)
865 return log_oom();
866
867 arg_private_network = true;
868 arg_settings_mask |= SETTING_NETWORK;
869 break;
870
871 case ARG_NETWORK_IPVLAN:
872
873 if (!ifname_valid(optarg))
874 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
875 "IPVLAN network interface name not valid: %s", optarg);
876
877 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
878 return log_oom();
879
880 _fallthrough_;
881 case ARG_PRIVATE_NETWORK:
882 arg_private_network = true;
883 arg_settings_mask |= SETTING_NETWORK;
884 break;
885
886 case ARG_NETWORK_NAMESPACE_PATH:
887 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
888 if (r < 0)
889 return r;
890
891 arg_settings_mask |= SETTING_NETWORK;
892 break;
893
894 case 'b':
895 if (arg_start_mode == START_PID2)
896 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
897 "--boot and --as-pid2 may not be combined.");
898
899 arg_start_mode = START_BOOT;
900 arg_settings_mask |= SETTING_START_MODE;
901 break;
902
903 case 'a':
904 if (arg_start_mode == START_BOOT)
905 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
906 "--boot and --as-pid2 may not be combined.");
907
908 arg_start_mode = START_PID2;
909 arg_settings_mask |= SETTING_START_MODE;
910 break;
911
912 case ARG_UUID:
913 r = sd_id128_from_string(optarg, &arg_uuid);
914 if (r < 0)
915 return log_error_errno(r, "Invalid UUID: %s", optarg);
916
917 if (sd_id128_is_null(arg_uuid))
918 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
919 "Machine UUID may not be all zeroes.");
920
921 arg_settings_mask |= SETTING_MACHINE_ID;
922 break;
923
924 case 'S': {
925 _cleanup_free_ char *mangled = NULL;
926
927 r = unit_name_mangle_with_suffix(optarg, NULL, UNIT_NAME_MANGLE_WARN, ".slice", &mangled);
928 if (r < 0)
929 return log_oom();
930
931 free_and_replace(arg_slice, mangled);
932 arg_settings_mask |= SETTING_SLICE;
933 break;
934 }
935
936 case 'M':
937 if (isempty(optarg))
938 arg_machine = mfree(arg_machine);
939 else {
940 if (!machine_name_is_valid(optarg))
941 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
942 "Invalid machine name: %s", optarg);
943
944 r = free_and_strdup(&arg_machine, optarg);
945 if (r < 0)
946 return log_oom();
947 }
948 break;
949
950 case ARG_HOSTNAME:
951 if (isempty(optarg))
952 arg_hostname = mfree(arg_hostname);
953 else {
954 if (!hostname_is_valid(optarg, false))
955 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
956 "Invalid hostname: %s", optarg);
957
958 r = free_and_strdup(&arg_hostname, optarg);
959 if (r < 0)
960 return log_oom();
961 }
962
963 arg_settings_mask |= SETTING_HOSTNAME;
964 break;
965
966 case 'Z':
967 arg_selinux_context = optarg;
968 break;
969
970 case 'L':
971 arg_selinux_apifs_context = optarg;
972 break;
973
974 case ARG_READ_ONLY:
975 arg_read_only = true;
976 arg_settings_mask |= SETTING_READ_ONLY;
977 break;
978
979 case ARG_CAPABILITY:
980 case ARG_DROP_CAPABILITY: {
981 uint64_t m;
982 r = parse_capability_spec(optarg, &m);
983 if (r <= 0)
984 return r;
985
986 if (c == ARG_CAPABILITY)
987 plus |= m;
988 else
989 minus |= m;
990 arg_settings_mask |= SETTING_CAPABILITY;
991 break;
992 }
993 case ARG_NO_NEW_PRIVILEGES:
994 r = parse_boolean(optarg);
995 if (r < 0)
996 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
997
998 arg_no_new_privileges = r;
999 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
1000 break;
1001
1002 case 'j':
1003 arg_link_journal = LINK_GUEST;
1004 arg_link_journal_try = true;
1005 arg_settings_mask |= SETTING_LINK_JOURNAL;
1006 break;
1007
1008 case ARG_LINK_JOURNAL:
1009 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
1010 if (r < 0)
1011 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
1012
1013 arg_settings_mask |= SETTING_LINK_JOURNAL;
1014 break;
1015
1016 case ARG_BIND:
1017 case ARG_BIND_RO:
1018 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
1019 if (r < 0)
1020 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
1021
1022 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1023 break;
1024
1025 case ARG_TMPFS:
1026 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1027 if (r < 0)
1028 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
1029
1030 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1031 break;
1032
1033 case ARG_OVERLAY:
1034 case ARG_OVERLAY_RO:
1035 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
1036 if (r == -EADDRNOTAVAIL)
1037 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1038 if (r < 0)
1039 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
1040
1041 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1042 break;
1043
1044 case ARG_INACCESSIBLE:
1045 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1046 if (r < 0)
1047 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
1048
1049 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1050 break;
1051
1052 case 'E': {
1053 char **n;
1054
1055 if (!env_assignment_is_valid(optarg))
1056 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1057 "Environment variable assignment '%s' is not valid.", optarg);
1058
1059 n = strv_env_set(arg_setenv, optarg);
1060 if (!n)
1061 return log_oom();
1062
1063 strv_free_and_replace(arg_setenv, n);
1064 arg_settings_mask |= SETTING_ENVIRONMENT;
1065 break;
1066 }
1067
1068 case 'q':
1069 arg_quiet = true;
1070 break;
1071
1072 case ARG_SHARE_SYSTEM:
1073 /* We don't officially support this anymore, except for compat reasons. People should use the
1074 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1075 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1076 arg_clone_ns_flags = 0;
1077 break;
1078
1079 case ARG_REGISTER:
1080 r = parse_boolean(optarg);
1081 if (r < 0) {
1082 log_error("Failed to parse --register= argument: %s", optarg);
1083 return r;
1084 }
1085
1086 arg_register = r;
1087 break;
1088
1089 case ARG_KEEP_UNIT:
1090 arg_keep_unit = true;
1091 break;
1092
1093 case ARG_PERSONALITY:
1094
1095 arg_personality = personality_from_string(optarg);
1096 if (arg_personality == PERSONALITY_INVALID)
1097 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1098 "Unknown or unsupported personality '%s'.", optarg);
1099
1100 arg_settings_mask |= SETTING_PERSONALITY;
1101 break;
1102
1103 case ARG_VOLATILE:
1104
1105 if (!optarg)
1106 arg_volatile_mode = VOLATILE_YES;
1107 else if (streq(optarg, "help")) {
1108 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1109 return 0;
1110 } else {
1111 VolatileMode m;
1112
1113 m = volatile_mode_from_string(optarg);
1114 if (m < 0)
1115 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1116 "Failed to parse --volatile= argument: %s", optarg);
1117 else
1118 arg_volatile_mode = m;
1119 }
1120
1121 arg_settings_mask |= SETTING_VOLATILE_MODE;
1122 break;
1123
1124 case 'p':
1125 r = expose_port_parse(&arg_expose_ports, optarg);
1126 if (r == -EEXIST)
1127 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1128 if (r < 0)
1129 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1130
1131 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1132 break;
1133
1134 case ARG_PROPERTY:
1135 if (strv_extend(&arg_property, optarg) < 0)
1136 return log_oom();
1137
1138 break;
1139
1140 case ARG_PRIVATE_USERS: {
1141 int boolean = -1;
1142
1143 if (!optarg)
1144 boolean = true;
1145 else if (!in_charset(optarg, DIGITS))
1146 /* do *not* parse numbers as booleans */
1147 boolean = parse_boolean(optarg);
1148
1149 if (boolean == false) {
1150 /* no: User namespacing off */
1151 arg_userns_mode = USER_NAMESPACE_NO;
1152 arg_uid_shift = UID_INVALID;
1153 arg_uid_range = UINT32_C(0x10000);
1154 } else if (boolean == true) {
1155 /* yes: User namespacing on, UID range is read from root dir */
1156 arg_userns_mode = USER_NAMESPACE_FIXED;
1157 arg_uid_shift = UID_INVALID;
1158 arg_uid_range = UINT32_C(0x10000);
1159 } else if (streq(optarg, "pick")) {
1160 /* pick: User namespacing on, UID range is picked randomly */
1161 arg_userns_mode = USER_NAMESPACE_PICK;
1162 arg_uid_shift = UID_INVALID;
1163 arg_uid_range = UINT32_C(0x10000);
1164 } else {
1165 _cleanup_free_ char *buffer = NULL;
1166 const char *range, *shift;
1167
1168 /* anything else: User namespacing on, UID range is explicitly configured */
1169
1170 range = strchr(optarg, ':');
1171 if (range) {
1172 buffer = strndup(optarg, range - optarg);
1173 if (!buffer)
1174 return log_oom();
1175 shift = buffer;
1176
1177 range++;
1178 r = safe_atou32(range, &arg_uid_range);
1179 if (r < 0)
1180 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
1181 } else
1182 shift = optarg;
1183
1184 r = parse_uid(shift, &arg_uid_shift);
1185 if (r < 0)
1186 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
1187
1188 arg_userns_mode = USER_NAMESPACE_FIXED;
1189 }
1190
1191 if (arg_uid_range <= 0)
1192 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1193 "UID range cannot be 0.");
1194
1195 arg_settings_mask |= SETTING_USERNS;
1196 break;
1197 }
1198
1199 case 'U':
1200 if (userns_supported()) {
1201 arg_userns_mode = USER_NAMESPACE_PICK;
1202 arg_uid_shift = UID_INVALID;
1203 arg_uid_range = UINT32_C(0x10000);
1204
1205 arg_settings_mask |= SETTING_USERNS;
1206 }
1207
1208 break;
1209
1210 case ARG_PRIVATE_USERS_CHOWN:
1211 arg_userns_chown = true;
1212
1213 arg_settings_mask |= SETTING_USERNS;
1214 break;
1215
1216 case ARG_KILL_SIGNAL:
1217 if (streq(optarg, "help")) {
1218 DUMP_STRING_TABLE(signal, int, _NSIG);
1219 return 0;
1220 }
1221
1222 arg_kill_signal = signal_from_string(optarg);
1223 if (arg_kill_signal < 0)
1224 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1225 "Cannot parse signal: %s", optarg);
1226
1227 arg_settings_mask |= SETTING_KILL_SIGNAL;
1228 break;
1229
1230 case ARG_SETTINGS:
1231
1232 /* no → do not read files
1233 * yes → read files, do not override cmdline, trust only subset
1234 * override → read files, override cmdline, trust only subset
1235 * trusted → read files, do not override cmdline, trust all
1236 */
1237
1238 r = parse_boolean(optarg);
1239 if (r < 0) {
1240 if (streq(optarg, "trusted")) {
1241 mask_all_settings = false;
1242 mask_no_settings = false;
1243 arg_settings_trusted = true;
1244
1245 } else if (streq(optarg, "override")) {
1246 mask_all_settings = false;
1247 mask_no_settings = true;
1248 arg_settings_trusted = -1;
1249 } else
1250 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1251 } else if (r > 0) {
1252 /* yes */
1253 mask_all_settings = false;
1254 mask_no_settings = false;
1255 arg_settings_trusted = -1;
1256 } else {
1257 /* no */
1258 mask_all_settings = true;
1259 mask_no_settings = false;
1260 arg_settings_trusted = false;
1261 }
1262
1263 break;
1264
1265 case ARG_CHDIR:
1266 if (!path_is_absolute(optarg))
1267 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1268 "Working directory %s is not an absolute path.", optarg);
1269
1270 r = free_and_strdup(&arg_chdir, optarg);
1271 if (r < 0)
1272 return log_oom();
1273
1274 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1275 break;
1276
1277 case ARG_PIVOT_ROOT:
1278 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1279 if (r < 0)
1280 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1281
1282 arg_settings_mask |= SETTING_PIVOT_ROOT;
1283 break;
1284
1285 case ARG_NOTIFY_READY:
1286 r = parse_boolean(optarg);
1287 if (r < 0)
1288 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1289 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1290 arg_notify_ready = r;
1291 arg_settings_mask |= SETTING_NOTIFY_READY;
1292 break;
1293
1294 case ARG_ROOT_HASH: {
1295 void *k;
1296 size_t l;
1297
1298 r = unhexmem(optarg, strlen(optarg), &k, &l);
1299 if (r < 0)
1300 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1301 if (l < sizeof(sd_id128_t)) {
1302 free(k);
1303 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128bit long: %s", optarg);
1304 }
1305
1306 free(arg_root_hash);
1307 arg_root_hash = k;
1308 arg_root_hash_size = l;
1309 break;
1310 }
1311
1312 case ARG_SYSTEM_CALL_FILTER: {
1313 bool negative;
1314 const char *items;
1315
1316 negative = optarg[0] == '~';
1317 items = negative ? optarg + 1 : optarg;
1318
1319 for (;;) {
1320 _cleanup_free_ char *word = NULL;
1321
1322 r = extract_first_word(&items, &word, NULL, 0);
1323 if (r == 0)
1324 break;
1325 if (r == -ENOMEM)
1326 return log_oom();
1327 if (r < 0)
1328 return log_error_errno(r, "Failed to parse system call filter: %m");
1329
1330 if (negative)
1331 r = strv_extend(&arg_syscall_blacklist, word);
1332 else
1333 r = strv_extend(&arg_syscall_whitelist, word);
1334 if (r < 0)
1335 return log_oom();
1336 }
1337
1338 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1339 break;
1340 }
1341
1342 case ARG_RLIMIT: {
1343 const char *eq;
1344 _cleanup_free_ char *name = NULL;
1345 int rl;
1346
1347 if (streq(optarg, "help")) {
1348 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1349 return 0;
1350 }
1351
1352 eq = strchr(optarg, '=');
1353 if (!eq)
1354 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1355 "--rlimit= expects an '=' assignment.");
1356
1357 name = strndup(optarg, eq - optarg);
1358 if (!name)
1359 return log_oom();
1360
1361 rl = rlimit_from_string_harder(name);
1362 if (rl < 0)
1363 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1364 "Unknown resource limit: %s", name);
1365
1366 if (!arg_rlimit[rl]) {
1367 arg_rlimit[rl] = new0(struct rlimit, 1);
1368 if (!arg_rlimit[rl])
1369 return log_oom();
1370 }
1371
1372 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1373 if (r < 0)
1374 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1375
1376 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1377 break;
1378 }
1379
1380 case ARG_OOM_SCORE_ADJUST:
1381 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1382 if (r < 0)
1383 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1384
1385 arg_oom_score_adjust_set = true;
1386 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1387 break;
1388
1389 case ARG_CPU_AFFINITY: {
1390 CPUSet cpuset;
1391
1392 r = parse_cpu_set(optarg, &cpuset);
1393 if (r < 0)
1394 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
1395
1396 cpu_set_reset(&arg_cpu_set);
1397 arg_cpu_set = cpuset;
1398 arg_settings_mask |= SETTING_CPU_AFFINITY;
1399 break;
1400 }
1401
1402 case ARG_RESOLV_CONF:
1403 if (streq(optarg, "help")) {
1404 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1405 return 0;
1406 }
1407
1408 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1409 if (arg_resolv_conf < 0)
1410 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1411 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1412
1413 arg_settings_mask |= SETTING_RESOLV_CONF;
1414 break;
1415
1416 case ARG_TIMEZONE:
1417 if (streq(optarg, "help")) {
1418 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1419 return 0;
1420 }
1421
1422 arg_timezone = timezone_mode_from_string(optarg);
1423 if (arg_timezone < 0)
1424 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1425 "Failed to parse /etc/localtime mode: %s", optarg);
1426
1427 arg_settings_mask |= SETTING_TIMEZONE;
1428 break;
1429
1430 case ARG_CONSOLE:
1431 r = handle_arg_console(optarg);
1432 if (r <= 0)
1433 return r;
1434 break;
1435
1436 case 'P':
1437 case ARG_PIPE:
1438 r = handle_arg_console("pipe");
1439 if (r <= 0)
1440 return r;
1441 break;
1442
1443 case ARG_NO_PAGER:
1444 arg_pager_flags |= PAGER_DISABLE;
1445 break;
1446
1447 case '?':
1448 return -EINVAL;
1449
1450 default:
1451 assert_not_reached("Unhandled option");
1452 }
1453
1454 if (argc > optind) {
1455 strv_free(arg_parameters);
1456 arg_parameters = strv_copy(argv + optind);
1457 if (!arg_parameters)
1458 return log_oom();
1459
1460 arg_settings_mask |= SETTING_START_MODE;
1461 }
1462
1463 if (arg_ephemeral && arg_template && !arg_directory)
1464 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1465 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1466 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1467 * --directory=". */
1468 arg_directory = TAKE_PTR(arg_template);
1469
1470 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0)) & ~minus;
1471
1472 /* Make sure to parse environment before we reset the settings mask below */
1473 r = parse_environment();
1474 if (r < 0)
1475 return r;
1476
1477 /* Load all settings from .nspawn files */
1478 if (mask_no_settings)
1479 arg_settings_mask = 0;
1480
1481 /* Don't load any settings from .nspawn files */
1482 if (mask_all_settings)
1483 arg_settings_mask = _SETTINGS_MASK_ALL;
1484
1485 return 1;
1486 }
1487
1488 static int verify_arguments(void) {
1489 int r;
1490
1491 if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
1492 /* If we are running the stub init in the container, we don't need to look at what the init
1493 * in the container supports, because we are not using it. Let's immediately pick the right
1494 * setting based on the host system configuration.
1495 *
1496 * We only do this, if the user didn't use an environment variable to override the detection.
1497 */
1498
1499 r = cg_all_unified();
1500 if (r < 0)
1501 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
1502 if (r > 0)
1503 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
1504 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
1505 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
1506 else
1507 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
1508 }
1509
1510 if (arg_userns_mode != USER_NAMESPACE_NO)
1511 arg_mount_settings |= MOUNT_USE_USERNS;
1512
1513 if (arg_private_network)
1514 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1515
1516 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1517 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1518 arg_register = false;
1519 if (arg_start_mode != START_PID1)
1520 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1521 }
1522
1523 if (arg_userns_mode == USER_NAMESPACE_PICK)
1524 arg_userns_chown = true;
1525
1526 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1527 arg_kill_signal = SIGRTMIN+3;
1528
1529 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 */
1530 arg_read_only = true;
1531
1532 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1533 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1534 * The latter is not technically a user session, but we don't need to labour the point. */
1535 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1536
1537 if (arg_directory && arg_image)
1538 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1539
1540 if (arg_template && arg_image)
1541 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1542
1543 if (arg_template && !(arg_directory || arg_machine))
1544 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1545
1546 if (arg_ephemeral && arg_template)
1547 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1548
1549 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
1550 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
1551
1552 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1553 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1554
1555 if (arg_userns_chown && arg_read_only)
1556 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1557 "--read-only and --private-users-chown may not be combined.");
1558
1559 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1560 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1561 * copy-up (in case of overlay) making the entire exercise pointless. */
1562 if (arg_userns_chown && arg_volatile_mode != VOLATILE_NO)
1563 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-chown may not be combined.");
1564
1565 /* If --network-namespace-path is given with any other network-related option, we need to error out,
1566 * to avoid conflicts between different network options. */
1567 if (arg_network_namespace_path &&
1568 (arg_network_interfaces || arg_network_macvlan ||
1569 arg_network_ipvlan || arg_network_veth_extra ||
1570 arg_network_bridge || arg_network_zone ||
1571 arg_network_veth || arg_private_network))
1572 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1573
1574 if (arg_network_bridge && arg_network_zone)
1575 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1576 "--network-bridge= and --network-zone= may not be combined.");
1577
1578 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1579 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1580
1581 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1582 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1583
1584 if (arg_expose_ports && !arg_private_network)
1585 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1586
1587 #if ! HAVE_LIBIPTC
1588 if (arg_expose_ports)
1589 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--port= is not supported, compiled without libiptc support.");
1590 #endif
1591
1592 r = custom_mount_check_all();
1593 if (r < 0)
1594 return r;
1595
1596 return 0;
1597 }
1598
1599 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1600 assert(p);
1601
1602 if (arg_userns_mode == USER_NAMESPACE_NO)
1603 return 0;
1604
1605 if (uid == UID_INVALID && gid == GID_INVALID)
1606 return 0;
1607
1608 if (uid != UID_INVALID) {
1609 uid += arg_uid_shift;
1610
1611 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1612 return -EOVERFLOW;
1613 }
1614
1615 if (gid != GID_INVALID) {
1616 gid += (gid_t) arg_uid_shift;
1617
1618 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1619 return -EOVERFLOW;
1620 }
1621
1622 if (lchown(p, uid, gid) < 0)
1623 return -errno;
1624
1625 return 0;
1626 }
1627
1628 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1629 const char *q;
1630 int r;
1631
1632 q = prefix_roota(root, path);
1633 r = mkdir_errno_wrapper(q, mode);
1634 if (r == -EEXIST)
1635 return 0;
1636 if (r < 0)
1637 return r;
1638
1639 return userns_lchown(q, uid, gid);
1640 }
1641
1642 static const char *timezone_from_path(const char *path) {
1643 return PATH_STARTSWITH_SET(
1644 path,
1645 "../usr/share/zoneinfo/",
1646 "/usr/share/zoneinfo/");
1647 }
1648
1649 static bool etc_writable(void) {
1650 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1651 }
1652
1653 static int setup_timezone(const char *dest) {
1654 _cleanup_free_ char *p = NULL, *etc = NULL;
1655 const char *where, *check;
1656 TimezoneMode m;
1657 int r;
1658
1659 assert(dest);
1660
1661 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1662 r = readlink_malloc("/etc/localtime", &p);
1663 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1664 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1665 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1666 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1667 else if (r < 0) {
1668 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1669 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1670 * file.
1671 *
1672 * Example:
1673 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1674 */
1675 return 0;
1676 } else if (arg_timezone == TIMEZONE_AUTO)
1677 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1678 else
1679 m = arg_timezone;
1680 } else
1681 m = arg_timezone;
1682
1683 if (m == TIMEZONE_OFF)
1684 return 0;
1685
1686 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1687 if (r < 0) {
1688 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1689 return 0;
1690 }
1691
1692 where = strjoina(etc, "/localtime");
1693
1694 switch (m) {
1695
1696 case TIMEZONE_DELETE:
1697 if (unlink(where) < 0)
1698 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1699
1700 return 0;
1701
1702 case TIMEZONE_SYMLINK: {
1703 _cleanup_free_ char *q = NULL;
1704 const char *z, *what;
1705
1706 z = timezone_from_path(p);
1707 if (!z) {
1708 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1709 return 0;
1710 }
1711
1712 r = readlink_malloc(where, &q);
1713 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1714 return 0; /* Already pointing to the right place? Then do nothing .. */
1715
1716 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1717 r = chase_symlinks(check, dest, 0, NULL, NULL);
1718 if (r < 0)
1719 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1720 else {
1721 if (unlink(where) < 0 && errno != ENOENT) {
1722 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1723 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1724 return 0;
1725 }
1726
1727 what = strjoina("../usr/share/zoneinfo/", z);
1728 if (symlink(what, where) < 0) {
1729 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1730 errno, "Failed to correct timezone of container, ignoring: %m");
1731 return 0;
1732 }
1733
1734 break;
1735 }
1736
1737 _fallthrough_;
1738 }
1739
1740 case TIMEZONE_BIND: {
1741 _cleanup_free_ char *resolved = NULL;
1742 int found;
1743
1744 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
1745 if (found < 0) {
1746 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1747 return 0;
1748 }
1749
1750 if (found == 0) /* missing? */
1751 (void) touch(resolved);
1752
1753 r = mount_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1754 if (r >= 0)
1755 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1756
1757 _fallthrough_;
1758 }
1759
1760 case TIMEZONE_COPY:
1761 /* If mounting failed, try to copy */
1762 r = copy_file_atomic("/etc/localtime", where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
1763 if (r < 0) {
1764 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1765 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
1766 return 0;
1767 }
1768
1769 break;
1770
1771 default:
1772 assert_not_reached("unexpected mode");
1773 }
1774
1775 /* Fix permissions of the symlink or file copy we just created */
1776 r = userns_lchown(where, 0, 0);
1777 if (r < 0)
1778 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
1779
1780 return 0;
1781 }
1782
1783 static int have_resolv_conf(const char *path) {
1784 assert(path);
1785
1786 if (access(path, F_OK) < 0) {
1787 if (errno == ENOENT)
1788 return 0;
1789
1790 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1791 }
1792
1793 return 1;
1794 }
1795
1796 static int resolved_listening(void) {
1797 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1798 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1799 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1800 int r;
1801
1802 /* Check if resolved is listening */
1803
1804 r = sd_bus_open_system(&bus);
1805 if (r < 0)
1806 return log_debug_errno(r, "Failed to open system bus: %m");
1807
1808 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1809 if (r < 0)
1810 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1811 if (r == 0)
1812 return 0;
1813
1814 r = sd_bus_get_property_string(bus,
1815 "org.freedesktop.resolve1",
1816 "/org/freedesktop/resolve1",
1817 "org.freedesktop.resolve1.Manager",
1818 "DNSStubListener",
1819 &error,
1820 &dns_stub_listener_mode);
1821 if (r < 0)
1822 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
1823
1824 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1825 }
1826
1827 static int setup_resolv_conf(const char *dest) {
1828 _cleanup_free_ char *etc = NULL;
1829 const char *where, *what;
1830 ResolvConfMode m;
1831 int r;
1832
1833 assert(dest);
1834
1835 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
1836 if (arg_private_network)
1837 m = RESOLV_CONF_OFF;
1838 else if (have_resolv_conf(STATIC_RESOLV_CONF) > 0 && resolved_listening() > 0)
1839 m = etc_writable() ? RESOLV_CONF_COPY_STATIC : RESOLV_CONF_BIND_STATIC;
1840 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1841 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
1842 else
1843 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
1844 } else
1845 m = arg_resolv_conf;
1846
1847 if (m == RESOLV_CONF_OFF)
1848 return 0;
1849
1850 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1851 if (r < 0) {
1852 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1853 return 0;
1854 }
1855
1856 where = strjoina(etc, "/resolv.conf");
1857
1858 if (m == RESOLV_CONF_DELETE) {
1859 if (unlink(where) < 0)
1860 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1861
1862 return 0;
1863 }
1864
1865 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_COPY_STATIC))
1866 what = STATIC_RESOLV_CONF;
1867 else
1868 what = "/etc/resolv.conf";
1869
1870 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC)) {
1871 _cleanup_free_ char *resolved = NULL;
1872 int found;
1873
1874 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
1875 if (found < 0) {
1876 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1877 return 0;
1878 }
1879
1880 if (found == 0) /* missing? */
1881 (void) touch(resolved);
1882
1883 r = mount_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
1884 if (r >= 0)
1885 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1886 }
1887
1888 /* If that didn't work, let's copy the file */
1889 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, 0, COPY_REFLINK);
1890 if (r < 0) {
1891 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1892 * resolved or something similar runs inside and the symlink points there.
1893 *
1894 * If the disk image is read-only, there's also no point in complaining.
1895 */
1896 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC) && IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1897 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1898 return 0;
1899 }
1900
1901 r = userns_lchown(where, 0, 0);
1902 if (r < 0)
1903 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1904
1905 return 0;
1906 }
1907
1908 static int setup_boot_id(void) {
1909 _cleanup_(unlink_and_freep) char *from = NULL;
1910 _cleanup_free_ char *path = NULL;
1911 sd_id128_t rnd = SD_ID128_NULL;
1912 const char *to;
1913 int r;
1914
1915 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1916
1917 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
1918 if (r < 0)
1919 return log_error_errno(r, "Failed to generate random boot ID path: %m");
1920
1921 r = sd_id128_randomize(&rnd);
1922 if (r < 0)
1923 return log_error_errno(r, "Failed to generate random boot id: %m");
1924
1925 r = id128_write(path, ID128_UUID, rnd, false);
1926 if (r < 0)
1927 return log_error_errno(r, "Failed to write boot id: %m");
1928
1929 from = TAKE_PTR(path);
1930 to = "/proc/sys/kernel/random/boot_id";
1931
1932 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1933 if (r < 0)
1934 return r;
1935
1936 return mount_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1937 }
1938
1939 static int copy_devnodes(const char *dest) {
1940 static const char devnodes[] =
1941 "null\0"
1942 "zero\0"
1943 "full\0"
1944 "random\0"
1945 "urandom\0"
1946 "tty\0"
1947 "net/tun\0";
1948
1949 _cleanup_umask_ mode_t u;
1950 const char *d;
1951 int r = 0;
1952
1953 assert(dest);
1954
1955 u = umask(0000);
1956
1957 /* Create /dev/net, so that we can create /dev/net/tun in it */
1958 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1959 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1960
1961 NULSTR_FOREACH(d, devnodes) {
1962 _cleanup_free_ char *from = NULL, *to = NULL;
1963 struct stat st;
1964
1965 from = path_join("/dev/", d);
1966 if (!from)
1967 return log_oom();
1968
1969 to = path_join(dest, from);
1970 if (!to)
1971 return log_oom();
1972
1973 if (stat(from, &st) < 0) {
1974
1975 if (errno != ENOENT)
1976 return log_error_errno(errno, "Failed to stat %s: %m", from);
1977
1978 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
1979 return log_error_errno(SYNTHETIC_ERRNO(EIO),
1980 "%s is not a char or block device, cannot copy.", from);
1981 else {
1982 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
1983
1984 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1985 /* Explicitly warn the user when /dev is already populated. */
1986 if (errno == EEXIST)
1987 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);
1988 if (errno != EPERM)
1989 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1990
1991 /* Some systems abusively restrict mknod but allow bind mounts. */
1992 r = touch(to);
1993 if (r < 0)
1994 return log_error_errno(r, "touch (%s) failed: %m", to);
1995 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1996 if (r < 0)
1997 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1998 }
1999
2000 r = userns_lchown(to, 0, 0);
2001 if (r < 0)
2002 return log_error_errno(r, "chown() of device node %s failed: %m", to);
2003
2004 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
2005 if (!dn)
2006 return log_oom();
2007
2008 r = userns_mkdir(dest, dn, 0755, 0, 0);
2009 if (r < 0)
2010 return log_error_errno(r, "Failed to create '%s': %m", dn);
2011
2012 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
2013 return log_oom();
2014
2015 prefixed = path_join(dest, sl);
2016 if (!prefixed)
2017 return log_oom();
2018
2019 t = path_join("..", d);
2020 if (!t)
2021 return log_oom();
2022
2023 if (symlink(t, prefixed) < 0)
2024 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
2025 }
2026 }
2027
2028 return r;
2029 }
2030
2031 static int make_extra_nodes(const char *dest) {
2032 _cleanup_umask_ mode_t u;
2033 size_t i;
2034 int r;
2035
2036 u = umask(0000);
2037
2038 for (i = 0; i < arg_n_extra_nodes; i++) {
2039 _cleanup_free_ char *path = NULL;
2040 DeviceNode *n = arg_extra_nodes + i;
2041
2042 path = path_join(dest, n->path);
2043 if (!path)
2044 return log_oom();
2045
2046 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
2047 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
2048
2049 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
2050 if (r < 0)
2051 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
2052 }
2053
2054 return 0;
2055 }
2056
2057 static int setup_pts(const char *dest) {
2058 _cleanup_free_ char *options = NULL;
2059 const char *p;
2060 int r;
2061
2062 #if HAVE_SELINUX
2063 if (arg_selinux_apifs_context)
2064 (void) asprintf(&options,
2065 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
2066 arg_uid_shift + TTY_GID,
2067 arg_selinux_apifs_context);
2068 else
2069 #endif
2070 (void) asprintf(&options,
2071 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
2072 arg_uid_shift + TTY_GID);
2073
2074 if (!options)
2075 return log_oom();
2076
2077 /* Mount /dev/pts itself */
2078 p = prefix_roota(dest, "/dev/pts");
2079 r = mkdir_errno_wrapper(p, 0755);
2080 if (r < 0)
2081 return log_error_errno(r, "Failed to create /dev/pts: %m");
2082
2083 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2084 if (r < 0)
2085 return r;
2086 r = userns_lchown(p, 0, 0);
2087 if (r < 0)
2088 return log_error_errno(r, "Failed to chown /dev/pts: %m");
2089
2090 /* Create /dev/ptmx symlink */
2091 p = prefix_roota(dest, "/dev/ptmx");
2092 if (symlink("pts/ptmx", p) < 0)
2093 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
2094 r = userns_lchown(p, 0, 0);
2095 if (r < 0)
2096 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
2097
2098 /* And fix /dev/pts/ptmx ownership */
2099 p = prefix_roota(dest, "/dev/pts/ptmx");
2100 r = userns_lchown(p, 0, 0);
2101 if (r < 0)
2102 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
2103
2104 return 0;
2105 }
2106
2107 static int setup_stdio_as_dev_console(void) {
2108 int terminal;
2109 int r;
2110
2111 terminal = open_terminal("/dev/console", O_RDWR);
2112 if (terminal < 0)
2113 return log_error_errno(terminal, "Failed to open console: %m");
2114
2115 /* Make sure we can continue logging to the original stderr, even if
2116 * stderr points elsewhere now */
2117 r = log_dup_console();
2118 if (r < 0)
2119 return log_error_errno(r, "Failed to duplicate stderr: %m");
2120
2121 /* invalidates 'terminal' on success and failure */
2122 r = rearrange_stdio(terminal, terminal, terminal);
2123 if (r < 0)
2124 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2125
2126 return 0;
2127 }
2128
2129 static int setup_dev_console(const char *console) {
2130 _cleanup_free_ char *p = NULL;
2131 int r;
2132
2133 /* Create /dev/console symlink */
2134 r = path_make_relative("/dev", console, &p);
2135 if (r < 0)
2136 return log_error_errno(r, "Failed to create relative path: %m");
2137
2138 if (symlink(p, "/dev/console") < 0)
2139 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
2140
2141 return 0;
2142 }
2143
2144 static int setup_keyring(void) {
2145 key_serial_t keyring;
2146
2147 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2148 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2149 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2150 * these system calls let's make sure we don't leak anything into the container. */
2151
2152 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2153 if (keyring == -1) {
2154 if (errno == ENOSYS)
2155 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2156 else if (IN_SET(errno, EACCES, EPERM))
2157 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2158 else
2159 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2160 }
2161
2162 return 0;
2163 }
2164
2165 static int setup_kmsg(int kmsg_socket) {
2166 _cleanup_(unlink_and_freep) char *from = NULL;
2167 _cleanup_free_ char *fifo = NULL;
2168 _cleanup_close_ int fd = -1;
2169 _cleanup_umask_ mode_t u;
2170 int r;
2171
2172 assert(kmsg_socket >= 0);
2173
2174 u = umask(0000);
2175
2176 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2177 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2178 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2179 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2180
2181 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2182 if (r < 0)
2183 return log_error_errno(r, "Failed to generate kmsg path: %m");
2184
2185 if (mkfifo(fifo, 0600) < 0)
2186 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2187
2188 from = TAKE_PTR(fifo);
2189
2190 r = mount_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2191 if (r < 0)
2192 return r;
2193
2194 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2195 if (fd < 0)
2196 return log_error_errno(errno, "Failed to open fifo: %m");
2197
2198 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2199 r = send_one_fd(kmsg_socket, fd, 0);
2200 if (r < 0)
2201 return log_error_errno(r, "Failed to send FIFO fd: %m");
2202
2203 return 0;
2204 }
2205
2206 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2207 union in_addr_union *exposed = userdata;
2208
2209 assert(rtnl);
2210 assert(m);
2211 assert(exposed);
2212
2213 expose_port_execute(rtnl, arg_expose_ports, exposed);
2214 return 0;
2215 }
2216
2217 static int setup_hostname(void) {
2218 int r;
2219
2220 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2221 return 0;
2222
2223 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2224 if (r < 0)
2225 return log_error_errno(r, "Failed to set hostname: %m");
2226
2227 return 0;
2228 }
2229
2230 static int setup_journal(const char *directory) {
2231 _cleanup_free_ char *d = NULL;
2232 const char *dirname, *p, *q;
2233 sd_id128_t this_id;
2234 char id[33];
2235 bool try;
2236 int r;
2237
2238 /* Don't link journals in ephemeral mode */
2239 if (arg_ephemeral)
2240 return 0;
2241
2242 if (arg_link_journal == LINK_NO)
2243 return 0;
2244
2245 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2246
2247 r = sd_id128_get_machine(&this_id);
2248 if (r < 0)
2249 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2250
2251 if (sd_id128_equal(arg_uuid, this_id)) {
2252 log_full(try ? LOG_WARNING : LOG_ERR,
2253 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
2254 if (try)
2255 return 0;
2256 return -EEXIST;
2257 }
2258
2259 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2260 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2261 if (r < 0) {
2262 bool ignore = r == -EROFS && try;
2263 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2264 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2265 return ignore ? 0 : r;
2266 }
2267 }
2268
2269 (void) sd_id128_to_string(arg_uuid, id);
2270
2271 p = strjoina("/var/log/journal/", id);
2272 q = prefix_roota(directory, p);
2273
2274 if (path_is_mount_point(p, NULL, 0) > 0) {
2275 if (try)
2276 return 0;
2277
2278 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2279 "%s: already a mount point, refusing to use for journal", p);
2280 }
2281
2282 if (path_is_mount_point(q, NULL, 0) > 0) {
2283 if (try)
2284 return 0;
2285
2286 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2287 "%s: already a mount point, refusing to use for journal", q);
2288 }
2289
2290 r = readlink_and_make_absolute(p, &d);
2291 if (r >= 0) {
2292 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2293 path_equal(d, q)) {
2294
2295 r = userns_mkdir(directory, p, 0755, 0, 0);
2296 if (r < 0)
2297 log_warning_errno(r, "Failed to create directory %s: %m", q);
2298 return 0;
2299 }
2300
2301 if (unlink(p) < 0)
2302 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2303 } else if (r == -EINVAL) {
2304
2305 if (arg_link_journal == LINK_GUEST &&
2306 rmdir(p) < 0) {
2307
2308 if (errno == ENOTDIR) {
2309 log_error("%s already exists and is neither a symlink nor a directory", p);
2310 return r;
2311 } else
2312 return log_error_errno(errno, "Failed to remove %s: %m", p);
2313 }
2314 } else if (r != -ENOENT)
2315 return log_error_errno(r, "readlink(%s) failed: %m", p);
2316
2317 if (arg_link_journal == LINK_GUEST) {
2318
2319 if (symlink(q, p) < 0) {
2320 if (try) {
2321 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2322 return 0;
2323 } else
2324 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2325 }
2326
2327 r = userns_mkdir(directory, p, 0755, 0, 0);
2328 if (r < 0)
2329 log_warning_errno(r, "Failed to create directory %s: %m", q);
2330 return 0;
2331 }
2332
2333 if (arg_link_journal == LINK_HOST) {
2334 /* don't create parents here — if the host doesn't have
2335 * permanent journal set up, don't force it here */
2336
2337 r = mkdir_errno_wrapper(p, 0755);
2338 if (r < 0 && r != -EEXIST) {
2339 if (try) {
2340 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2341 return 0;
2342 } else
2343 return log_error_errno(r, "Failed to create %s: %m", p);
2344 }
2345
2346 } else if (access(p, F_OK) < 0)
2347 return 0;
2348
2349 if (dir_is_empty(q) == 0)
2350 log_warning("%s is not empty, proceeding anyway.", q);
2351
2352 r = userns_mkdir(directory, p, 0755, 0, 0);
2353 if (r < 0)
2354 return log_error_errno(r, "Failed to create %s: %m", q);
2355
2356 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2357 if (r < 0)
2358 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2359
2360 return 0;
2361 }
2362
2363 static int drop_capabilities(uid_t uid) {
2364 CapabilityQuintet q;
2365
2366 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2367 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2368 * arg_caps_retain. */
2369
2370 if (capability_quintet_is_set(&arg_full_capabilities)) {
2371 q = arg_full_capabilities;
2372
2373 if (q.bounding == (uint64_t) -1)
2374 q.bounding = uid == 0 ? arg_caps_retain : 0;
2375
2376 if (q.effective == (uint64_t) -1)
2377 q.effective = uid == 0 ? q.bounding : 0;
2378
2379 if (q.inheritable == (uint64_t) -1)
2380 q.inheritable = uid == 0 ? q.bounding : 0;
2381
2382 if (q.permitted == (uint64_t) -1)
2383 q.permitted = uid == 0 ? q.bounding : 0;
2384
2385 if (q.ambient == (uint64_t) -1 && ambient_capabilities_supported())
2386 q.ambient = 0;
2387
2388 if (capability_quintet_mangle(&q))
2389 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2390
2391 } else {
2392 q = (CapabilityQuintet) {
2393 .bounding = arg_caps_retain,
2394 .effective = uid == 0 ? arg_caps_retain : 0,
2395 .inheritable = uid == 0 ? arg_caps_retain : 0,
2396 .permitted = uid == 0 ? arg_caps_retain : 0,
2397 .ambient = ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2398 };
2399
2400 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2401 * in order to maintain the same behavior as systemd < 242. */
2402 if (capability_quintet_mangle(&q))
2403 log_full(arg_quiet ? LOG_DEBUG : LOG_WARNING,
2404 "Some capabilities will not be set because they are not in the current bounding set.");
2405
2406 }
2407
2408 return capability_quintet_enforce(&q);
2409 }
2410
2411 static int reset_audit_loginuid(void) {
2412 _cleanup_free_ char *p = NULL;
2413 int r;
2414
2415 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2416 return 0;
2417
2418 r = read_one_line_file("/proc/self/loginuid", &p);
2419 if (r == -ENOENT)
2420 return 0;
2421 if (r < 0)
2422 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2423
2424 /* Already reset? */
2425 if (streq(p, "4294967295"))
2426 return 0;
2427
2428 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2429 if (r < 0) {
2430 log_error_errno(r,
2431 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2432 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2433 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2434 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2435 "using systemd-nspawn. Sleeping for 5s... (%m)");
2436
2437 sleep(5);
2438 }
2439
2440 return 0;
2441 }
2442
2443 static int setup_propagate(const char *root) {
2444 const char *p, *q;
2445 int r;
2446
2447 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2448 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2449 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2450 (void) mkdir_p(p, 0600);
2451
2452 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
2453 if (r < 0)
2454 return log_error_errno(r, "Failed to create /run/systemd: %m");
2455
2456 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
2457 if (r < 0)
2458 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
2459
2460 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2461 if (r < 0)
2462 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
2463
2464 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
2465 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2466 if (r < 0)
2467 return r;
2468
2469 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2470 if (r < 0)
2471 return r;
2472
2473 /* machined will MS_MOVE into that directory, and that's only
2474 * supported for non-shared mounts. */
2475 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2476 }
2477
2478 static int setup_machine_id(const char *directory) {
2479 const char *etc_machine_id;
2480 sd_id128_t id;
2481 int r;
2482
2483 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2484 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2485 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2486 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2487 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2488 * container behaves nicely). */
2489
2490 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2491
2492 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
2493 if (r < 0) {
2494 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2495 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2496
2497 if (sd_id128_is_null(arg_uuid)) {
2498 r = sd_id128_randomize(&arg_uuid);
2499 if (r < 0)
2500 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2501 }
2502 } else {
2503 if (sd_id128_is_null(id))
2504 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2505 "Machine ID in container image is zero, refusing.");
2506
2507 arg_uuid = id;
2508 }
2509
2510 return 0;
2511 }
2512
2513 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2514 int r;
2515
2516 assert(directory);
2517
2518 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
2519 return 0;
2520
2521 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2522 if (r == -EOPNOTSUPP)
2523 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.");
2524 if (r == -EBADE)
2525 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2526 if (r < 0)
2527 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2528 if (r == 0)
2529 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2530 else
2531 log_debug("Patched directory tree to match UID/GID range.");
2532
2533 return r;
2534 }
2535
2536 /*
2537 * Return values:
2538 * < 0 : wait_for_terminate() failed to get the state of the
2539 * container, the container was terminated by a signal, or
2540 * failed for an unknown reason. No change is made to the
2541 * container argument.
2542 * > 0 : The program executed in the container terminated with an
2543 * error. The exit code of the program executed in the
2544 * container is returned. The container argument has been set
2545 * to CONTAINER_TERMINATED.
2546 * 0 : The container is being rebooted, has been shut down or exited
2547 * successfully. The container argument has been set to either
2548 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2549 *
2550 * That is, success is indicated by a return value of zero, and an
2551 * error is indicated by a non-zero value.
2552 */
2553 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2554 siginfo_t status;
2555 int r;
2556
2557 r = wait_for_terminate(pid, &status);
2558 if (r < 0)
2559 return log_warning_errno(r, "Failed to wait for container: %m");
2560
2561 switch (status.si_code) {
2562
2563 case CLD_EXITED:
2564 if (status.si_status == 0)
2565 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2566 else
2567 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2568
2569 *container = CONTAINER_TERMINATED;
2570 return status.si_status;
2571
2572 case CLD_KILLED:
2573 if (status.si_status == SIGINT) {
2574 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2575 *container = CONTAINER_TERMINATED;
2576 return 0;
2577
2578 } else if (status.si_status == SIGHUP) {
2579 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2580 *container = CONTAINER_REBOOTED;
2581 return 0;
2582 }
2583
2584 _fallthrough_;
2585 case CLD_DUMPED:
2586 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2587 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2588
2589 default:
2590 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2591 "Container %s failed due to unknown reason.", arg_machine);
2592 }
2593 }
2594
2595 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2596 pid_t pid;
2597
2598 pid = PTR_TO_PID(userdata);
2599 if (pid > 0) {
2600 if (kill(pid, arg_kill_signal) >= 0) {
2601 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2602 sd_event_source_set_userdata(s, NULL);
2603 return 0;
2604 }
2605 }
2606
2607 sd_event_exit(sd_event_source_get_event(s), 0);
2608 return 0;
2609 }
2610
2611 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2612 pid_t pid;
2613
2614 assert(s);
2615 assert(ssi);
2616
2617 pid = PTR_TO_PID(userdata);
2618
2619 for (;;) {
2620 siginfo_t si = {};
2621
2622 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2623 return log_error_errno(errno, "Failed to waitid(): %m");
2624 if (si.si_pid == 0) /* No pending children. */
2625 break;
2626 if (si.si_pid == pid) {
2627 /* The main process we care for has exited. Return from
2628 * signal handler but leave the zombie. */
2629 sd_event_exit(sd_event_source_get_event(s), 0);
2630 break;
2631 }
2632
2633 /* Reap all other children. */
2634 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2635 }
2636
2637 return 0;
2638 }
2639
2640 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2641 pid_t pid;
2642
2643 assert(m);
2644
2645 pid = PTR_TO_PID(userdata);
2646
2647 if (arg_kill_signal > 0) {
2648 log_info("Container termination requested. Attempting to halt container.");
2649 (void) kill(pid, arg_kill_signal);
2650 } else {
2651 log_info("Container termination requested. Exiting.");
2652 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2653 }
2654
2655 return 0;
2656 }
2657
2658 static int determine_names(void) {
2659 int r;
2660
2661 if (arg_template && !arg_directory && arg_machine) {
2662
2663 /* If --template= was specified then we should not
2664 * search for a machine, but instead create a new one
2665 * in /var/lib/machine. */
2666
2667 arg_directory = path_join("/var/lib/machines", arg_machine);
2668 if (!arg_directory)
2669 return log_oom();
2670 }
2671
2672 if (!arg_image && !arg_directory) {
2673 if (arg_machine) {
2674 _cleanup_(image_unrefp) Image *i = NULL;
2675
2676 r = image_find(IMAGE_MACHINE, arg_machine, &i);
2677 if (r == -ENOENT)
2678 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
2679 if (r < 0)
2680 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2681
2682 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2683 r = free_and_strdup(&arg_image, i->path);
2684 else
2685 r = free_and_strdup(&arg_directory, i->path);
2686 if (r < 0)
2687 return log_oom();
2688
2689 if (!arg_ephemeral)
2690 arg_read_only = arg_read_only || i->read_only;
2691 } else {
2692 r = safe_getcwd(&arg_directory);
2693 if (r < 0)
2694 return log_error_errno(r, "Failed to determine current directory: %m");
2695 }
2696
2697 if (!arg_directory && !arg_image)
2698 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
2699 }
2700
2701 if (!arg_machine) {
2702 if (arg_directory && path_equal(arg_directory, "/"))
2703 arg_machine = gethostname_malloc();
2704 else {
2705 if (arg_image) {
2706 char *e;
2707
2708 arg_machine = strdup(basename(arg_image));
2709
2710 /* Truncate suffix if there is one */
2711 e = endswith(arg_machine, ".raw");
2712 if (e)
2713 *e = 0;
2714 } else
2715 arg_machine = strdup(basename(arg_directory));
2716 }
2717 if (!arg_machine)
2718 return log_oom();
2719
2720 hostname_cleanup(arg_machine);
2721 if (!machine_name_is_valid(arg_machine))
2722 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
2723
2724 if (arg_ephemeral) {
2725 char *b;
2726
2727 /* Add a random suffix when this is an
2728 * ephemeral machine, so that we can run many
2729 * instances at once without manually having
2730 * to specify -M each time. */
2731
2732 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2733 return log_oom();
2734
2735 free(arg_machine);
2736 arg_machine = b;
2737 }
2738 }
2739
2740 return 0;
2741 }
2742
2743 static int chase_symlinks_and_update(char **p, unsigned flags) {
2744 char *chased;
2745 int r;
2746
2747 assert(p);
2748
2749 if (!*p)
2750 return 0;
2751
2752 r = chase_symlinks(*p, NULL, flags, &chased, NULL);
2753 if (r < 0)
2754 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2755
2756 return free_and_replace(*p, chased);
2757 }
2758
2759 static int determine_uid_shift(const char *directory) {
2760 int r;
2761
2762 if (arg_userns_mode == USER_NAMESPACE_NO) {
2763 arg_uid_shift = 0;
2764 return 0;
2765 }
2766
2767 if (arg_uid_shift == UID_INVALID) {
2768 struct stat st;
2769
2770 r = stat(directory, &st);
2771 if (r < 0)
2772 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2773
2774 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2775
2776 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
2777 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2778 "UID and GID base of %s don't match.", directory);
2779
2780 arg_uid_range = UINT32_C(0x10000);
2781 }
2782
2783 if (arg_uid_shift > (uid_t) -1 - arg_uid_range)
2784 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2785 "UID base too high for UID range.");
2786
2787 return 0;
2788 }
2789
2790 static unsigned long effective_clone_ns_flags(void) {
2791 unsigned long flags = arg_clone_ns_flags;
2792
2793 if (arg_private_network)
2794 flags |= CLONE_NEWNET;
2795 if (arg_use_cgns)
2796 flags |= CLONE_NEWCGROUP;
2797 if (arg_userns_mode != USER_NAMESPACE_NO)
2798 flags |= CLONE_NEWUSER;
2799
2800 return flags;
2801 }
2802
2803 static int patch_sysctl(void) {
2804
2805 /* This table is inspired by runc's sysctl() function */
2806 static const struct {
2807 const char *key;
2808 bool prefix;
2809 unsigned long clone_flags;
2810 } safe_sysctl[] = {
2811 { "kernel.hostname", false, CLONE_NEWUTS },
2812 { "kernel.domainname", false, CLONE_NEWUTS },
2813 { "kernel.msgmax", false, CLONE_NEWIPC },
2814 { "kernel.msgmnb", false, CLONE_NEWIPC },
2815 { "kernel.msgmni", false, CLONE_NEWIPC },
2816 { "kernel.sem", false, CLONE_NEWIPC },
2817 { "kernel.shmall", false, CLONE_NEWIPC },
2818 { "kernel.shmmax", false, CLONE_NEWIPC },
2819 { "kernel.shmmni", false, CLONE_NEWIPC },
2820 { "fs.mqueue.", true, CLONE_NEWIPC },
2821 { "net.", true, CLONE_NEWNET },
2822 };
2823
2824 unsigned long flags;
2825 char **k, **v;
2826 int r;
2827
2828 flags = effective_clone_ns_flags();
2829
2830 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
2831 bool good = false;
2832 size_t i;
2833
2834 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
2835
2836 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
2837 continue;
2838
2839 if (safe_sysctl[i].prefix)
2840 good = startswith(*k, safe_sysctl[i].key);
2841 else
2842 good = streq(*k, safe_sysctl[i].key);
2843
2844 if (good)
2845 break;
2846 }
2847
2848 if (!good)
2849 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
2850
2851 r = sysctl_write(*k, *v);
2852 if (r < 0)
2853 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
2854 }
2855
2856 return 0;
2857 }
2858
2859 static int inner_child(
2860 Barrier *barrier,
2861 const char *directory,
2862 bool secondary,
2863 int kmsg_socket,
2864 int rtnl_socket,
2865 int master_pty_socket,
2866 FDSet *fds) {
2867
2868 _cleanup_free_ char *home = NULL;
2869 char as_uuid[ID128_UUID_STRING_MAX];
2870 size_t n_env = 1;
2871 const char *envp[] = {
2872 "PATH=" DEFAULT_PATH_COMPAT,
2873 NULL, /* container */
2874 NULL, /* TERM */
2875 NULL, /* HOME */
2876 NULL, /* USER */
2877 NULL, /* LOGNAME */
2878 NULL, /* container_uuid */
2879 NULL, /* LISTEN_FDS */
2880 NULL, /* LISTEN_PID */
2881 NULL, /* NOTIFY_SOCKET */
2882 NULL
2883 };
2884 const char *exec_target;
2885 _cleanup_strv_free_ char **env_use = NULL;
2886 int r, which_failed;
2887
2888 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2889 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2890 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2891 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2892 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2893 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2894 * namespace.
2895 *
2896 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2897 * unshare(). See below. */
2898
2899 assert(barrier);
2900 assert(directory);
2901 assert(kmsg_socket >= 0);
2902
2903 log_debug("Inner child is initializing.");
2904
2905 if (arg_userns_mode != USER_NAMESPACE_NO) {
2906 /* Tell the parent, that it now can write the UID map. */
2907 (void) barrier_place(barrier); /* #1 */
2908
2909 /* Wait until the parent wrote the UID map */
2910 if (!barrier_place_and_sync(barrier)) /* #2 */
2911 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2912 "Parent died too early");
2913 }
2914
2915 r = reset_uid_gid();
2916 if (r < 0)
2917 return log_error_errno(r, "Couldn't become new root: %m");
2918
2919 r = mount_all(NULL,
2920 arg_mount_settings | MOUNT_IN_USERNS,
2921 arg_uid_shift,
2922 arg_selinux_apifs_context);
2923 if (r < 0)
2924 return r;
2925
2926 if (!arg_network_namespace_path && arg_private_network) {
2927 r = unshare(CLONE_NEWNET);
2928 if (r < 0)
2929 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2930
2931 /* Tell the parent that it can setup network interfaces. */
2932 (void) barrier_place(barrier); /* #3 */
2933 }
2934
2935 r = mount_sysfs(NULL, arg_mount_settings);
2936 if (r < 0)
2937 return r;
2938
2939 /* Wait until we are cgroup-ified, so that we
2940 * can mount the right cgroup path writable */
2941 if (!barrier_place_and_sync(barrier)) /* #4 */
2942 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2943 "Parent died too early");
2944
2945 if (arg_use_cgns) {
2946 r = unshare(CLONE_NEWCGROUP);
2947 if (r < 0)
2948 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2949 r = mount_cgroups(
2950 "",
2951 arg_unified_cgroup_hierarchy,
2952 arg_userns_mode != USER_NAMESPACE_NO,
2953 arg_uid_shift,
2954 arg_uid_range,
2955 arg_selinux_apifs_context,
2956 true);
2957 if (r < 0)
2958 return r;
2959 } else {
2960 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2961 if (r < 0)
2962 return r;
2963 }
2964
2965 r = setup_boot_id();
2966 if (r < 0)
2967 return r;
2968
2969 r = setup_kmsg(kmsg_socket);
2970 if (r < 0)
2971 return r;
2972 kmsg_socket = safe_close(kmsg_socket);
2973
2974 r = mount_custom(
2975 "/",
2976 arg_custom_mounts,
2977 arg_n_custom_mounts,
2978 false,
2979 0,
2980 0,
2981 arg_selinux_apifs_context,
2982 true);
2983 if (r < 0)
2984 return r;
2985
2986 if (setsid() < 0)
2987 return log_error_errno(errno, "setsid() failed: %m");
2988
2989 if (arg_private_network)
2990 loopback_setup();
2991
2992 if (arg_expose_ports) {
2993 r = expose_port_send_rtnl(rtnl_socket);
2994 if (r < 0)
2995 return r;
2996 rtnl_socket = safe_close(rtnl_socket);
2997 }
2998
2999 if (arg_console_mode != CONSOLE_PIPE) {
3000 _cleanup_close_ int master = -1;
3001 _cleanup_free_ char *console = NULL;
3002
3003 /* Allocate a pty and make it available as /dev/console. */
3004 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
3005 if (master < 0)
3006 return log_error_errno(master, "Failed to allocate a pty: %m");
3007
3008 r = setup_dev_console(console);
3009 if (r < 0)
3010 return log_error_errno(r, "Failed to setup /dev/console: %m");
3011
3012 r = send_one_fd(master_pty_socket, master, 0);
3013 if (r < 0)
3014 return log_error_errno(r, "Failed to send master fd: %m");
3015 master_pty_socket = safe_close(master_pty_socket);
3016
3017 r = setup_stdio_as_dev_console();
3018 if (r < 0)
3019 return r;
3020 }
3021
3022 r = patch_sysctl();
3023 if (r < 0)
3024 return r;
3025
3026 if (arg_oom_score_adjust_set) {
3027 r = set_oom_score_adjust(arg_oom_score_adjust);
3028 if (r < 0)
3029 return log_error_errno(r, "Failed to adjust OOM score: %m");
3030 }
3031
3032 if (arg_cpu_set.set)
3033 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
3034 return log_error_errno(errno, "Failed to set CPU affinity: %m");
3035
3036 (void) setup_hostname();
3037
3038 if (arg_personality != PERSONALITY_INVALID) {
3039 r = safe_personality(arg_personality);
3040 if (r < 0)
3041 return log_error_errno(r, "personality() failed: %m");
3042 } else if (secondary) {
3043 r = safe_personality(PER_LINUX32);
3044 if (r < 0)
3045 return log_error_errno(r, "personality() failed: %m");
3046 }
3047
3048 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3049 if (r < 0)
3050 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3051
3052 #if HAVE_SECCOMP
3053 if (arg_seccomp) {
3054
3055 if (is_seccomp_available()) {
3056
3057 r = seccomp_load(arg_seccomp);
3058 if (ERRNO_IS_SECCOMP_FATAL(r))
3059 return log_error_errno(r, "Failed to install seccomp filter: %m");
3060 if (r < 0)
3061 log_debug_errno(r, "Failed to install seccomp filter: %m");
3062 }
3063 } else
3064 #endif
3065 {
3066 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
3067 if (r < 0)
3068 return r;
3069 }
3070
3071 #if HAVE_SELINUX
3072 if (arg_selinux_context)
3073 if (setexeccon(arg_selinux_context) < 0)
3074 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3075 #endif
3076
3077 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3078 * if we need to later on. */
3079 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3080 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3081
3082 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3083 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids);
3084 else
3085 r = change_uid_gid(arg_user, &home);
3086 if (r < 0)
3087 return r;
3088
3089 r = drop_capabilities(getuid());
3090 if (r < 0)
3091 return log_error_errno(r, "Dropping capabilities failed: %m");
3092
3093 if (arg_no_new_privileges)
3094 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3095 return log_error_errno(errno, "Failed to disable new privileges: %m");
3096
3097 /* LXC sets container=lxc, so follow the scheme here */
3098 envp[n_env++] = strjoina("container=", arg_container_service_name);
3099
3100 envp[n_env] = strv_find_prefix(environ, "TERM=");
3101 if (envp[n_env])
3102 n_env++;
3103
3104 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3105 if (asprintf((char**)(envp + n_env++), "HOME=%s", home ?: "/root") < 0)
3106 return log_oom();
3107
3108 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3109 if (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ?: "root") < 0 ||
3110 asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3111 return log_oom();
3112
3113 assert(!sd_id128_is_null(arg_uuid));
3114
3115 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
3116 return log_oom();
3117
3118 if (fdset_size(fds) > 0) {
3119 r = fdset_cloexec(fds, false);
3120 if (r < 0)
3121 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3122
3123 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3124 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3125 return log_oom();
3126 }
3127 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3128 return log_oom();
3129
3130 env_use = strv_env_merge(2, envp, arg_setenv);
3131 if (!env_use)
3132 return log_oom();
3133
3134 /* Let the parent know that we are ready and
3135 * wait until the parent is ready with the
3136 * setup, too... */
3137 if (!barrier_place_and_sync(barrier)) /* #5 */
3138 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3139 "Parent died too early");
3140
3141 if (arg_chdir)
3142 if (chdir(arg_chdir) < 0)
3143 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3144
3145 if (arg_start_mode == START_PID2) {
3146 r = stub_pid1(arg_uuid);
3147 if (r < 0)
3148 return r;
3149 }
3150
3151 log_debug("Inner child completed, invoking payload.");
3152
3153 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3154 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3155 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3156 log_close();
3157 log_set_open_when_needed(true);
3158
3159 (void) fdset_close_others(fds);
3160
3161 if (arg_start_mode == START_BOOT) {
3162 char **a;
3163 size_t m;
3164
3165 /* Automatically search for the init system */
3166
3167 m = strv_length(arg_parameters);
3168 a = newa(char*, m + 2);
3169 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3170 a[1 + m] = NULL;
3171
3172 a[0] = (char*) "/usr/lib/systemd/systemd";
3173 execve(a[0], a, env_use);
3174
3175 a[0] = (char*) "/lib/systemd/systemd";
3176 execve(a[0], a, env_use);
3177
3178 a[0] = (char*) "/sbin/init";
3179 execve(a[0], a, env_use);
3180
3181 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3182 } else if (!strv_isempty(arg_parameters)) {
3183 const char *dollar_path;
3184
3185 exec_target = arg_parameters[0];
3186
3187 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3188 * binary. */
3189 dollar_path = strv_env_get(env_use, "PATH");
3190 if (dollar_path) {
3191 if (putenv((char*) dollar_path) != 0)
3192 return log_error_errno(errno, "Failed to update $PATH: %m");
3193 }
3194
3195 execvpe(arg_parameters[0], arg_parameters, env_use);
3196 } else {
3197 if (!arg_chdir)
3198 /* If we cannot change the directory, we'll end up in /, that is expected. */
3199 (void) chdir(home ?: "/root");
3200
3201 execle("/bin/bash", "-bash", NULL, env_use);
3202 execle("/bin/sh", "-sh", NULL, env_use);
3203
3204 exec_target = "/bin/bash, /bin/sh";
3205 }
3206
3207 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3208 }
3209
3210 static int setup_sd_notify_child(void) {
3211 _cleanup_close_ int fd = -1;
3212 union sockaddr_union sa = {
3213 .un.sun_family = AF_UNIX,
3214 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3215 };
3216 int r;
3217
3218 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3219 if (fd < 0)
3220 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3221
3222 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3223 (void) sockaddr_un_unlink(&sa.un);
3224
3225 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3226 if (r < 0)
3227 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3228
3229 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3230 if (r < 0)
3231 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3232
3233 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3234 if (r < 0)
3235 return log_error_errno(r, "SO_PASSCRED failed: %m");
3236
3237 return TAKE_FD(fd);
3238 }
3239
3240 static int outer_child(
3241 Barrier *barrier,
3242 const char *directory,
3243 DissectedImage *dissected_image,
3244 bool secondary,
3245 int pid_socket,
3246 int uuid_socket,
3247 int notify_socket,
3248 int kmsg_socket,
3249 int rtnl_socket,
3250 int uid_shift_socket,
3251 int master_pty_socket,
3252 int unified_cgroup_hierarchy_socket,
3253 FDSet *fds,
3254 int netns_fd) {
3255
3256 _cleanup_close_ int fd = -1;
3257 pid_t pid;
3258 ssize_t l;
3259 int r;
3260
3261 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3262 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3263 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3264 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3265
3266 assert(barrier);
3267 assert(directory);
3268 assert(pid_socket >= 0);
3269 assert(uuid_socket >= 0);
3270 assert(notify_socket >= 0);
3271 assert(master_pty_socket >= 0);
3272 assert(kmsg_socket >= 0);
3273
3274 log_debug("Outer child is initializing.");
3275
3276 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3277 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3278
3279 r = reset_audit_loginuid();
3280 if (r < 0)
3281 return r;
3282
3283 /* Mark everything as slave, so that we still
3284 * receive mounts from the real root, but don't
3285 * propagate mounts to the real root. */
3286 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3287 if (r < 0)
3288 return r;
3289
3290 if (dissected_image) {
3291 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3292 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3293 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3294 * makes sure ESP partitions and userns are compatible. */
3295
3296 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3297 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
3298 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0)|
3299 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3300 if (r < 0)
3301 return r;
3302 }
3303
3304 r = determine_uid_shift(directory);
3305 if (r < 0)
3306 return r;
3307
3308 if (arg_userns_mode != USER_NAMESPACE_NO) {
3309 /* Let the parent know which UID shift we read from the image */
3310 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3311 if (l < 0)
3312 return log_error_errno(errno, "Failed to send UID shift: %m");
3313 if (l != sizeof(arg_uid_shift))
3314 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3315 "Short write while sending UID shift.");
3316
3317 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3318 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3319 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3320 * not it will pick a different one, and send it back to us. */
3321
3322 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3323 if (l < 0)
3324 return log_error_errno(errno, "Failed to recv UID shift: %m");
3325 if (l != sizeof(arg_uid_shift))
3326 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3327 "Short read while receiving UID shift.");
3328 }
3329
3330 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3331 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3332 }
3333
3334 if (path_equal(directory, "/")) {
3335 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3336 * place, so that we can make changes to its mount structure (for example, to implement
3337 * --volatile=) without this interfering with our ability to access files such as
3338 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3339 * (instead of a temporary directory, since we are living in our own mount namspace here
3340 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3341 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3342
3343 r = mount_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
3344 if (r < 0)
3345 return r;
3346
3347 directory = "/run/systemd/nspawn-root";
3348
3349 } else if (!dissected_image) {
3350 /* Turn directory into bind mount (we need that so that we can move the bind mount to root
3351 * later on). */
3352 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
3353 if (r < 0)
3354 return r;
3355 }
3356
3357 r = setup_pivot_root(
3358 directory,
3359 arg_pivot_root_new,
3360 arg_pivot_root_old);
3361 if (r < 0)
3362 return r;
3363
3364 r = setup_volatile_mode(
3365 directory,
3366 arg_volatile_mode,
3367 arg_userns_mode != USER_NAMESPACE_NO,
3368 arg_uid_shift,
3369 arg_uid_range,
3370 arg_selinux_apifs_context);
3371 if (r < 0)
3372 return r;
3373
3374 if (dissected_image) {
3375 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3376 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3377 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
3378 if (r < 0)
3379 return r;
3380 }
3381
3382 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3383 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3384
3385 r = detect_unified_cgroup_hierarchy_from_image(directory);
3386 if (r < 0)
3387 return r;
3388
3389 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3390 if (l < 0)
3391 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3392 if (l != sizeof(arg_unified_cgroup_hierarchy))
3393 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3394 "Short write while sending cgroup mode.");
3395
3396 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3397 }
3398
3399 /* Mark everything as shared so our mounts get propagated down. This is
3400 * required to make new bind mounts available in systemd services
3401 * inside the container that create a new mount namespace.
3402 * See https://github.com/systemd/systemd/issues/3860
3403 * Further submounts (such as /dev) done after this will inherit the
3404 * shared propagation mode. */
3405 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3406 if (r < 0)
3407 return r;
3408
3409 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3410 if (r < 0)
3411 return r;
3412
3413 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3414 if (r < 0)
3415 return r;
3416
3417 if (arg_read_only && arg_volatile_mode == VOLATILE_NO) {
3418 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3419 if (r < 0)
3420 return log_error_errno(r, "Failed to make tree read-only: %m");
3421 }
3422
3423 r = mount_all(directory,
3424 arg_mount_settings,
3425 arg_uid_shift,
3426 arg_selinux_apifs_context);
3427 if (r < 0)
3428 return r;
3429
3430 r = copy_devnodes(directory);
3431 if (r < 0)
3432 return r;
3433
3434 r = make_extra_nodes(directory);
3435 if (r < 0)
3436 return r;
3437
3438 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3439 (void) make_inaccessible_nodes(directory, arg_uid_shift, arg_uid_shift);
3440
3441 r = setup_pts(directory);
3442 if (r < 0)
3443 return r;
3444
3445 r = setup_propagate(directory);
3446 if (r < 0)
3447 return r;
3448
3449 r = setup_keyring();
3450 if (r < 0)
3451 return r;
3452
3453 r = setup_timezone(directory);
3454 if (r < 0)
3455 return r;
3456
3457 r = setup_resolv_conf(directory);
3458 if (r < 0)
3459 return r;
3460
3461 r = setup_machine_id(directory);
3462 if (r < 0)
3463 return r;
3464
3465 r = setup_journal(directory);
3466 if (r < 0)
3467 return r;
3468
3469 r = mount_custom(
3470 directory,
3471 arg_custom_mounts,
3472 arg_n_custom_mounts,
3473 arg_userns_mode != USER_NAMESPACE_NO,
3474 arg_uid_shift,
3475 arg_uid_range,
3476 arg_selinux_apifs_context,
3477 false);
3478 if (r < 0)
3479 return r;
3480
3481 if (!arg_use_cgns) {
3482 r = mount_cgroups(
3483 directory,
3484 arg_unified_cgroup_hierarchy,
3485 arg_userns_mode != USER_NAMESPACE_NO,
3486 arg_uid_shift,
3487 arg_uid_range,
3488 arg_selinux_apifs_context,
3489 false);
3490 if (r < 0)
3491 return r;
3492 }
3493
3494 r = mount_move_root(directory);
3495 if (r < 0)
3496 return log_error_errno(r, "Failed to move root directory: %m");
3497
3498 fd = setup_sd_notify_child();
3499 if (fd < 0)
3500 return fd;
3501
3502 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3503 arg_clone_ns_flags |
3504 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3505 if (pid < 0)
3506 return log_error_errno(errno, "Failed to fork inner child: %m");
3507 if (pid == 0) {
3508 pid_socket = safe_close(pid_socket);
3509 uuid_socket = safe_close(uuid_socket);
3510 notify_socket = safe_close(notify_socket);
3511 uid_shift_socket = safe_close(uid_shift_socket);
3512
3513 /* The inner child has all namespaces that are
3514 * requested, so that we all are owned by the user if
3515 * user namespaces are turned on. */
3516
3517 if (arg_network_namespace_path) {
3518 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3519 if (r < 0)
3520 return log_error_errno(r, "Failed to join network namespace: %m");
3521 }
3522
3523 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds);
3524 if (r < 0)
3525 _exit(EXIT_FAILURE);
3526
3527 _exit(EXIT_SUCCESS);
3528 }
3529
3530 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3531 if (l < 0)
3532 return log_error_errno(errno, "Failed to send PID: %m");
3533 if (l != sizeof(pid))
3534 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3535 "Short write while sending PID.");
3536
3537 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3538 if (l < 0)
3539 return log_error_errno(errno, "Failed to send machine ID: %m");
3540 if (l != sizeof(arg_uuid))
3541 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3542 "Short write while sending machine ID.");
3543
3544 l = send_one_fd(notify_socket, fd, 0);
3545 if (l < 0)
3546 return log_error_errno(l, "Failed to send notify fd: %m");
3547
3548 pid_socket = safe_close(pid_socket);
3549 uuid_socket = safe_close(uuid_socket);
3550 notify_socket = safe_close(notify_socket);
3551 master_pty_socket = safe_close(master_pty_socket);
3552 kmsg_socket = safe_close(kmsg_socket);
3553 rtnl_socket = safe_close(rtnl_socket);
3554 netns_fd = safe_close(netns_fd);
3555
3556 return 0;
3557 }
3558
3559 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
3560 bool tried_hashed = false;
3561 unsigned n_tries = 100;
3562 uid_t candidate;
3563 int r;
3564
3565 assert(shift);
3566 assert(ret_lock_file);
3567 assert(arg_userns_mode == USER_NAMESPACE_PICK);
3568 assert(arg_uid_range == 0x10000U);
3569
3570 candidate = *shift;
3571
3572 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3573
3574 for (;;) {
3575 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
3576 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
3577
3578 if (--n_tries <= 0)
3579 return -EBUSY;
3580
3581 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
3582 goto next;
3583 if ((candidate & UINT32_C(0xFFFF)) != 0)
3584 goto next;
3585
3586 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3587 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3588 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3589 goto next;
3590 if (r < 0)
3591 return r;
3592
3593 /* Make some superficial checks whether the range is currently known in the user database */
3594 if (getpwuid(candidate))
3595 goto next;
3596 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3597 goto next;
3598 if (getgrgid(candidate))
3599 goto next;
3600 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3601 goto next;
3602
3603 *ret_lock_file = lf;
3604 lf = (struct LockFile) LOCK_FILE_INIT;
3605 *shift = candidate;
3606 return 0;
3607
3608 next:
3609 if (arg_machine && !tried_hashed) {
3610 /* Try to hash the base from the container name */
3611
3612 static const uint8_t hash_key[] = {
3613 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3614 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3615 };
3616
3617 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3618
3619 tried_hashed = true;
3620 } else
3621 random_bytes(&candidate, sizeof(candidate));
3622
3623 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
3624 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3625 }
3626 }
3627
3628 static int setup_uid_map(pid_t pid) {
3629 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3630 int r;
3631
3632 assert(pid > 1);
3633
3634 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3635 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3636 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3637 if (r < 0)
3638 return log_error_errno(r, "Failed to write UID map: %m");
3639
3640 /* We always assign the same UID and GID ranges */
3641 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3642 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3643 if (r < 0)
3644 return log_error_errno(r, "Failed to write GID map: %m");
3645
3646 return 0;
3647 }
3648
3649 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3650 char buf[NOTIFY_BUFFER_MAX+1];
3651 char *p = NULL;
3652 struct iovec iovec = {
3653 .iov_base = buf,
3654 .iov_len = sizeof(buf)-1,
3655 };
3656 union {
3657 struct cmsghdr cmsghdr;
3658 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
3659 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
3660 } control = {};
3661 struct msghdr msghdr = {
3662 .msg_iov = &iovec,
3663 .msg_iovlen = 1,
3664 .msg_control = &control,
3665 .msg_controllen = sizeof(control),
3666 };
3667 struct cmsghdr *cmsg;
3668 struct ucred *ucred = NULL;
3669 ssize_t n;
3670 pid_t inner_child_pid;
3671 _cleanup_strv_free_ char **tags = NULL;
3672
3673 assert(userdata);
3674
3675 inner_child_pid = PTR_TO_PID(userdata);
3676
3677 if (revents != EPOLLIN) {
3678 log_warning("Got unexpected poll event for notify fd.");
3679 return 0;
3680 }
3681
3682 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3683 if (n < 0) {
3684 if (IN_SET(errno, EAGAIN, EINTR))
3685 return 0;
3686
3687 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3688 }
3689 cmsg_close_all(&msghdr);
3690
3691 CMSG_FOREACH(cmsg, &msghdr) {
3692 if (cmsg->cmsg_level == SOL_SOCKET &&
3693 cmsg->cmsg_type == SCM_CREDENTIALS &&
3694 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3695
3696 ucred = (struct ucred*) CMSG_DATA(cmsg);
3697 }
3698 }
3699
3700 if (!ucred || ucred->pid != inner_child_pid) {
3701 log_debug("Received notify message without valid credentials. Ignoring.");
3702 return 0;
3703 }
3704
3705 if ((size_t) n >= sizeof(buf)) {
3706 log_warning("Received notify message exceeded maximum size. Ignoring.");
3707 return 0;
3708 }
3709
3710 buf[n] = 0;
3711 tags = strv_split(buf, "\n\r");
3712 if (!tags)
3713 return log_oom();
3714
3715 if (strv_find(tags, "READY=1"))
3716 (void) sd_notifyf(false, "READY=1\n");
3717
3718 p = strv_find_startswith(tags, "STATUS=");
3719 if (p)
3720 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
3721
3722 return 0;
3723 }
3724
3725 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3726 int r;
3727
3728 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3729 if (r < 0)
3730 return log_error_errno(r, "Failed to allocate notify event source: %m");
3731
3732 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3733
3734 return 0;
3735 }
3736
3737 static int merge_settings(Settings *settings, const char *path) {
3738 int rl;
3739
3740 assert(settings);
3741 assert(path);
3742
3743 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3744 * that this steals the fields of the Settings* structure, and hence modifies it. */
3745
3746 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3747 settings->start_mode >= 0) {
3748 arg_start_mode = settings->start_mode;
3749 strv_free_and_replace(arg_parameters, settings->parameters);
3750 }
3751
3752 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
3753 arg_ephemeral = settings->ephemeral;
3754
3755 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
3756 settings->root) {
3757
3758 if (!arg_settings_trusted)
3759 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
3760 else
3761 free_and_replace(arg_directory, settings->root);
3762 }
3763
3764 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3765 settings->pivot_root_new) {
3766 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3767 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3768 }
3769
3770 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3771 settings->working_directory)
3772 free_and_replace(arg_chdir, settings->working_directory);
3773
3774 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3775 settings->environment)
3776 strv_free_and_replace(arg_setenv, settings->environment);
3777
3778 if ((arg_settings_mask & SETTING_USER) == 0) {
3779
3780 if (settings->user)
3781 free_and_replace(arg_user, settings->user);
3782
3783 if (uid_is_valid(settings->uid))
3784 arg_uid = settings->uid;
3785 if (gid_is_valid(settings->gid))
3786 arg_gid = settings->gid;
3787 if (settings->n_supplementary_gids > 0) {
3788 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
3789 arg_n_supplementary_gids = settings->n_supplementary_gids;
3790 }
3791 }
3792
3793 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3794 uint64_t plus, minus;
3795 uint64_t network_minus = 0;
3796
3797 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3798 * Settings structure */
3799
3800 plus = settings->capability;
3801 minus = settings->drop_capability;
3802
3803 if ((arg_settings_mask & SETTING_NETWORK) == 0) {
3804 if (settings_private_network(settings))
3805 plus |= UINT64_C(1) << CAP_NET_ADMIN;
3806 else
3807 network_minus |= UINT64_C(1) << CAP_NET_ADMIN;
3808 }
3809
3810 if (!arg_settings_trusted && plus != 0) {
3811 if (settings->capability != 0)
3812 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
3813 } else {
3814 arg_caps_retain &= ~network_minus;
3815 arg_caps_retain |= plus;
3816 }
3817
3818 arg_caps_retain &= ~minus;
3819
3820 /* Copy the full capabilities over too */
3821 if (capability_quintet_is_set(&settings->full_capabilities)) {
3822 if (!arg_settings_trusted)
3823 log_warning("Ignoring capability settings, file %s is not trusted.", path);
3824 else
3825 arg_full_capabilities = settings->full_capabilities;
3826 }
3827 }
3828
3829 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3830 settings->kill_signal > 0)
3831 arg_kill_signal = settings->kill_signal;
3832
3833 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3834 settings->personality != PERSONALITY_INVALID)
3835 arg_personality = settings->personality;
3836
3837 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3838 !sd_id128_is_null(settings->machine_id)) {
3839
3840 if (!arg_settings_trusted)
3841 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
3842 else
3843 arg_uuid = settings->machine_id;
3844 }
3845
3846 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3847 settings->read_only >= 0)
3848 arg_read_only = settings->read_only;
3849
3850 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3851 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3852 arg_volatile_mode = settings->volatile_mode;
3853
3854 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3855 settings->n_custom_mounts > 0) {
3856
3857 if (!arg_settings_trusted)
3858 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
3859 else {
3860 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3861 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3862 arg_n_custom_mounts = settings->n_custom_mounts;
3863 settings->n_custom_mounts = 0;
3864 }
3865 }
3866
3867 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3868 (settings->private_network >= 0 ||
3869 settings->network_veth >= 0 ||
3870 settings->network_bridge ||
3871 settings->network_zone ||
3872 settings->network_interfaces ||
3873 settings->network_macvlan ||
3874 settings->network_ipvlan ||
3875 settings->network_veth_extra ||
3876 settings->network_namespace_path)) {
3877
3878 if (!arg_settings_trusted)
3879 log_warning("Ignoring network settings, file %s is not trusted.", path);
3880 else {
3881 arg_network_veth = settings_network_veth(settings);
3882 arg_private_network = settings_private_network(settings);
3883
3884 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
3885 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
3886 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
3887 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
3888
3889 free_and_replace(arg_network_bridge, settings->network_bridge);
3890 free_and_replace(arg_network_zone, settings->network_zone);
3891
3892 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
3893 }
3894 }
3895
3896 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3897 settings->expose_ports) {
3898
3899 if (!arg_settings_trusted)
3900 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
3901 else {
3902 expose_port_free_all(arg_expose_ports);
3903 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3904 }
3905 }
3906
3907 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3908 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3909
3910 if (!arg_settings_trusted)
3911 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
3912 else {
3913 arg_userns_mode = settings->userns_mode;
3914 arg_uid_shift = settings->uid_shift;
3915 arg_uid_range = settings->uid_range;
3916 arg_userns_chown = settings->userns_chown;
3917 }
3918 }
3919
3920 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3921 arg_notify_ready = settings->notify_ready;
3922
3923 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3924
3925 if (!arg_settings_trusted && !strv_isempty(settings->syscall_whitelist))
3926 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
3927 else {
3928 strv_free_and_replace(arg_syscall_whitelist, settings->syscall_whitelist);
3929 strv_free_and_replace(arg_syscall_blacklist, settings->syscall_blacklist);
3930 }
3931
3932 #if HAVE_SECCOMP
3933 if (!arg_settings_trusted && settings->seccomp)
3934 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
3935 else {
3936 seccomp_release(arg_seccomp);
3937 arg_seccomp = TAKE_PTR(settings->seccomp);
3938 }
3939 #endif
3940 }
3941
3942 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
3943 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
3944 continue;
3945
3946 if (!settings->rlimit[rl])
3947 continue;
3948
3949 if (!arg_settings_trusted) {
3950 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
3951 continue;
3952 }
3953
3954 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
3955 }
3956
3957 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
3958 settings->hostname)
3959 free_and_replace(arg_hostname, settings->hostname);
3960
3961 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
3962 settings->no_new_privileges >= 0)
3963 arg_no_new_privileges = settings->no_new_privileges;
3964
3965 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
3966 settings->oom_score_adjust_set) {
3967
3968 if (!arg_settings_trusted)
3969 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
3970 else {
3971 arg_oom_score_adjust = settings->oom_score_adjust;
3972 arg_oom_score_adjust_set = true;
3973 }
3974 }
3975
3976 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
3977 settings->cpu_set.set) {
3978
3979 if (!arg_settings_trusted)
3980 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
3981 else {
3982 cpu_set_reset(&arg_cpu_set);
3983 arg_cpu_set = settings->cpu_set;
3984 settings->cpu_set = (CPUSet) {};
3985 }
3986 }
3987
3988 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
3989 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
3990 arg_resolv_conf = settings->resolv_conf;
3991
3992 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
3993 settings->link_journal != _LINK_JOURNAL_INVALID) {
3994
3995 if (!arg_settings_trusted)
3996 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
3997 else {
3998 arg_link_journal = settings->link_journal;
3999 arg_link_journal_try = settings->link_journal_try;
4000 }
4001 }
4002
4003 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
4004 settings->timezone != _TIMEZONE_MODE_INVALID)
4005 arg_timezone = settings->timezone;
4006
4007 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
4008 settings->slice) {
4009
4010 if (!arg_settings_trusted)
4011 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
4012 else
4013 free_and_replace(arg_slice, settings->slice);
4014 }
4015
4016 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
4017 settings->use_cgns >= 0) {
4018
4019 if (!arg_settings_trusted)
4020 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
4021 else
4022 arg_use_cgns = settings->use_cgns;
4023 }
4024
4025 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
4026 settings->clone_ns_flags != (unsigned long) -1) {
4027
4028 if (!arg_settings_trusted)
4029 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
4030 else
4031 arg_clone_ns_flags = settings->clone_ns_flags;
4032 }
4033
4034 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
4035 settings->console_mode >= 0) {
4036
4037 if (!arg_settings_trusted)
4038 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
4039 else
4040 arg_console_mode = settings->console_mode;
4041 }
4042
4043 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4044 * don't consult arg_settings_mask for them. */
4045
4046 sd_bus_message_unref(arg_property_message);
4047 arg_property_message = TAKE_PTR(settings->properties);
4048
4049 arg_console_width = settings->console_width;
4050 arg_console_height = settings->console_height;
4051
4052 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
4053 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
4054 arg_n_extra_nodes = settings->n_extra_nodes;
4055
4056 return 0;
4057 }
4058
4059 static int load_settings(void) {
4060 _cleanup_(settings_freep) Settings *settings = NULL;
4061 _cleanup_fclose_ FILE *f = NULL;
4062 _cleanup_free_ char *p = NULL;
4063 const char *fn, *i;
4064 int r;
4065
4066 if (arg_oci_bundle)
4067 return 0;
4068
4069 /* If all settings are masked, there's no point in looking for
4070 * the settings file */
4071 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
4072 return 0;
4073
4074 fn = strjoina(arg_machine, ".nspawn");
4075
4076 /* We first look in the admin's directories in /etc and /run */
4077 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4078 _cleanup_free_ char *j = NULL;
4079
4080 j = path_join(i, fn);
4081 if (!j)
4082 return log_oom();
4083
4084 f = fopen(j, "re");
4085 if (f) {
4086 p = TAKE_PTR(j);
4087
4088 /* By default, we trust configuration from /etc and /run */
4089 if (arg_settings_trusted < 0)
4090 arg_settings_trusted = true;
4091
4092 break;
4093 }
4094
4095 if (errno != ENOENT)
4096 return log_error_errno(errno, "Failed to open %s: %m", j);
4097 }
4098
4099 if (!f) {
4100 /* After that, let's look for a file next to the
4101 * actual image we shall boot. */
4102
4103 if (arg_image) {
4104 p = file_in_same_dir(arg_image, fn);
4105 if (!p)
4106 return log_oom();
4107 } else if (arg_directory && !path_equal(arg_directory, "/")) {
4108 p = file_in_same_dir(arg_directory, fn);
4109 if (!p)
4110 return log_oom();
4111 }
4112
4113 if (p) {
4114 f = fopen(p, "re");
4115 if (!f && errno != ENOENT)
4116 return log_error_errno(errno, "Failed to open %s: %m", p);
4117
4118 /* By default, we do not trust configuration from /var/lib/machines */
4119 if (arg_settings_trusted < 0)
4120 arg_settings_trusted = false;
4121 }
4122 }
4123
4124 if (!f)
4125 return 0;
4126
4127 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4128
4129 r = settings_load(f, p, &settings);
4130 if (r < 0)
4131 return r;
4132
4133 return merge_settings(settings, p);
4134 }
4135
4136 static int load_oci_bundle(void) {
4137 _cleanup_(settings_freep) Settings *settings = NULL;
4138 int r;
4139
4140 if (!arg_oci_bundle)
4141 return 0;
4142
4143 /* By default let's trust OCI bundles */
4144 if (arg_settings_trusted < 0)
4145 arg_settings_trusted = true;
4146
4147 r = oci_load(NULL, arg_oci_bundle, &settings);
4148 if (r < 0)
4149 return r;
4150
4151 return merge_settings(settings, arg_oci_bundle);
4152 }
4153
4154 static int run_container(
4155 DissectedImage *dissected_image,
4156 bool secondary,
4157 FDSet *fds,
4158 char veth_name[IFNAMSIZ], bool *veth_created,
4159 union in_addr_union *exposed,
4160 int *master, pid_t *pid, int *ret) {
4161
4162 static const struct sigaction sa = {
4163 .sa_handler = nop_signal_handler,
4164 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4165 };
4166
4167 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4168 _cleanup_close_ int etc_passwd_lock = -1;
4169 _cleanup_close_pair_ int
4170 kmsg_socket_pair[2] = { -1, -1 },
4171 rtnl_socket_pair[2] = { -1, -1 },
4172 pid_socket_pair[2] = { -1, -1 },
4173 uuid_socket_pair[2] = { -1, -1 },
4174 notify_socket_pair[2] = { -1, -1 },
4175 uid_shift_socket_pair[2] = { -1, -1 },
4176 master_pty_socket_pair[2] = { -1, -1 },
4177 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4178
4179 _cleanup_close_ int notify_socket = -1;
4180 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4181 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4182 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4183 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4184 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4185 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4186 ContainerStatus container_status = 0;
4187 int ifi = 0, r;
4188 ssize_t l;
4189 sigset_t mask_chld;
4190 _cleanup_close_ int netns_fd = -1;
4191
4192 assert_se(sigemptyset(&mask_chld) == 0);
4193 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4194
4195 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4196 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4197 * check with getpwuid() if the specific user already exists. Note that /etc might be
4198 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4199 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4200 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4201 * really ours. */
4202
4203 etc_passwd_lock = take_etc_passwd_lock(NULL);
4204 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4205 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4206 }
4207
4208 r = barrier_create(&barrier);
4209 if (r < 0)
4210 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4211
4212 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4213 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4214
4215 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4216 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4217
4218 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4219 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4220
4221 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4222 return log_error_errno(errno, "Failed to create id socket pair: %m");
4223
4224 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4225 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4226
4227 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
4228 return log_error_errno(errno, "Failed to create console socket pair: %m");
4229
4230 if (arg_userns_mode != USER_NAMESPACE_NO)
4231 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4232 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4233
4234 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4235 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4236 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4237
4238 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4239 * parent's blocking calls and give it a chance to call wait() and terminate. */
4240 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4241 if (r < 0)
4242 return log_error_errno(errno, "Failed to change the signal mask: %m");
4243
4244 r = sigaction(SIGCHLD, &sa, NULL);
4245 if (r < 0)
4246 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4247
4248 if (arg_network_namespace_path) {
4249 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4250 if (netns_fd < 0)
4251 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4252
4253 r = fd_is_network_ns(netns_fd);
4254 if (r == -EUCLEAN)
4255 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);
4256 else if (r < 0)
4257 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4258 else if (r == 0)
4259 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4260 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4261 }
4262
4263 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4264 if (*pid < 0)
4265 return log_error_errno(errno, "clone() failed%s: %m",
4266 errno == EINVAL ?
4267 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4268
4269 if (*pid == 0) {
4270 /* The outer child only has a file system namespace. */
4271 barrier_set_role(&barrier, BARRIER_CHILD);
4272
4273 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4274 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4275 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4276 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4277 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
4278 master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
4279 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4280 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
4281
4282 (void) reset_all_signal_handlers();
4283 (void) reset_signal_mask();
4284
4285 r = outer_child(&barrier,
4286 arg_directory,
4287 dissected_image,
4288 secondary,
4289 pid_socket_pair[1],
4290 uuid_socket_pair[1],
4291 notify_socket_pair[1],
4292 kmsg_socket_pair[1],
4293 rtnl_socket_pair[1],
4294 uid_shift_socket_pair[1],
4295 master_pty_socket_pair[1],
4296 unified_cgroup_hierarchy_socket_pair[1],
4297 fds,
4298 netns_fd);
4299 if (r < 0)
4300 _exit(EXIT_FAILURE);
4301
4302 _exit(EXIT_SUCCESS);
4303 }
4304
4305 barrier_set_role(&barrier, BARRIER_PARENT);
4306
4307 fdset_close(fds);
4308
4309 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4310 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4311 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4312 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4313 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
4314 master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
4315 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
4316 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
4317
4318 if (arg_userns_mode != USER_NAMESPACE_NO) {
4319 /* The child just let us know the UID shift it might have read from the image. */
4320 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4321 if (l < 0)
4322 return log_error_errno(errno, "Failed to read UID shift: %m");
4323 if (l != sizeof arg_uid_shift)
4324 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
4325
4326 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4327 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4328 * image, but if that's already in use, pick a new one, and report back to the child,
4329 * which one we now picked. */
4330
4331 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4332 if (r < 0)
4333 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4334
4335 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4336 if (l < 0)
4337 return log_error_errno(errno, "Failed to send UID shift: %m");
4338 if (l != sizeof arg_uid_shift)
4339 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
4340 }
4341 }
4342
4343 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4344 /* The child let us know the support cgroup mode it might have read from the image. */
4345 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4346 if (l < 0)
4347 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4348 if (l != sizeof(arg_unified_cgroup_hierarchy))
4349 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zu bytes).%s",
4350 l, l == 0 ? " The child is most likely dead." : "");
4351 }
4352
4353 /* Wait for the outer child. */
4354 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4355 if (r < 0)
4356 return r;
4357 if (r != EXIT_SUCCESS)
4358 return -EIO;
4359
4360 /* And now retrieve the PID of the inner child. */
4361 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4362 if (l < 0)
4363 return log_error_errno(errno, "Failed to read inner child PID: %m");
4364 if (l != sizeof *pid)
4365 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
4366
4367 /* We also retrieve container UUID in case it was generated by outer child */
4368 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4369 if (l < 0)
4370 return log_error_errno(errno, "Failed to read container machine ID: %m");
4371 if (l != sizeof(arg_uuid))
4372 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
4373
4374 /* We also retrieve the socket used for notifications generated by outer child */
4375 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4376 if (notify_socket < 0)
4377 return log_error_errno(notify_socket,
4378 "Failed to receive notification socket from the outer child: %m");
4379
4380 log_debug("Init process invoked as PID "PID_FMT, *pid);
4381
4382 if (arg_userns_mode != USER_NAMESPACE_NO) {
4383 if (!barrier_place_and_sync(&barrier)) /* #1 */
4384 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4385
4386 r = setup_uid_map(*pid);
4387 if (r < 0)
4388 return r;
4389
4390 (void) barrier_place(&barrier); /* #2 */
4391 }
4392
4393 if (arg_private_network) {
4394 if (!arg_network_namespace_path) {
4395 /* Wait until the child has unshared its network namespace. */
4396 if (!barrier_place_and_sync(&barrier)) /* #3 */
4397 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
4398 }
4399
4400 r = move_network_interfaces(*pid, arg_network_interfaces);
4401 if (r < 0)
4402 return r;
4403
4404 if (arg_network_veth) {
4405 r = setup_veth(arg_machine, *pid, veth_name,
4406 arg_network_bridge || arg_network_zone);
4407 if (r < 0)
4408 return r;
4409 else if (r > 0)
4410 ifi = r;
4411
4412 if (arg_network_bridge) {
4413 /* Add the interface to a bridge */
4414 r = setup_bridge(veth_name, arg_network_bridge, false);
4415 if (r < 0)
4416 return r;
4417 if (r > 0)
4418 ifi = r;
4419 } else if (arg_network_zone) {
4420 /* Add the interface to a bridge, possibly creating it */
4421 r = setup_bridge(veth_name, arg_network_zone, true);
4422 if (r < 0)
4423 return r;
4424 if (r > 0)
4425 ifi = r;
4426 }
4427 }
4428
4429 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
4430 if (r < 0)
4431 return r;
4432
4433 /* We created the primary and extra veth links now; let's remember this, so that we know to
4434 remove them later on. Note that we don't bother with removing veth links that were created
4435 here when their setup failed half-way, because in that case the kernel should be able to
4436 remove them on its own, since they cannot be referenced by anything yet. */
4437 *veth_created = true;
4438
4439 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
4440 if (r < 0)
4441 return r;
4442
4443 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
4444 if (r < 0)
4445 return r;
4446 }
4447
4448 if (arg_register || !arg_keep_unit) {
4449 r = sd_bus_default_system(&bus);
4450 if (r < 0)
4451 return log_error_errno(r, "Failed to open system bus: %m");
4452
4453 r = sd_bus_set_close_on_exit(bus, false);
4454 if (r < 0)
4455 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
4456 }
4457
4458 if (!arg_keep_unit) {
4459 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4460 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4461 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4462
4463 r = sd_bus_match_signal_async(
4464 bus,
4465 NULL,
4466 "org.freedesktop.systemd1",
4467 NULL,
4468 "org.freedesktop.systemd1.Scope",
4469 "RequestStop",
4470 on_request_stop, NULL, PID_TO_PTR(*pid));
4471 if (r < 0)
4472 return log_error_errno(r, "Failed to request RequestStop match: %m");
4473 }
4474
4475 if (arg_register) {
4476 r = register_machine(
4477 bus,
4478 arg_machine,
4479 *pid,
4480 arg_directory,
4481 arg_uuid,
4482 ifi,
4483 arg_slice,
4484 arg_custom_mounts, arg_n_custom_mounts,
4485 arg_kill_signal,
4486 arg_property,
4487 arg_property_message,
4488 arg_keep_unit,
4489 arg_container_service_name);
4490 if (r < 0)
4491 return r;
4492
4493 } else if (!arg_keep_unit) {
4494 r = allocate_scope(
4495 bus,
4496 arg_machine,
4497 *pid,
4498 arg_slice,
4499 arg_custom_mounts, arg_n_custom_mounts,
4500 arg_kill_signal,
4501 arg_property,
4502 arg_property_message);
4503 if (r < 0)
4504 return r;
4505
4506 } else if (arg_slice || arg_property)
4507 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4508
4509 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
4510 if (r < 0)
4511 return r;
4512
4513 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4514 if (r < 0)
4515 return r;
4516
4517 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4518 if (r < 0)
4519 return r;
4520
4521 /* Notify the child that the parent is ready with all
4522 * its setup (including cgroup-ification), and that
4523 * the child can now hand over control to the code to
4524 * run inside the container. */
4525 (void) barrier_place(&barrier); /* #4 */
4526
4527 /* Block SIGCHLD here, before notifying child.
4528 * process_pty() will handle it with the other signals. */
4529 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4530
4531 /* Reset signal to default */
4532 r = default_signals(SIGCHLD, -1);
4533 if (r < 0)
4534 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4535
4536 r = sd_event_new(&event);
4537 if (r < 0)
4538 return log_error_errno(r, "Failed to get default event source: %m");
4539
4540 (void) sd_event_set_watchdog(event, true);
4541
4542 if (bus) {
4543 r = sd_bus_attach_event(bus, event, 0);
4544 if (r < 0)
4545 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4546 }
4547
4548 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
4549 if (r < 0)
4550 return r;
4551
4552 /* Let the child know that we are ready and wait that the child is completely ready now. */
4553 if (!barrier_place_and_sync(&barrier)) /* #5 */
4554 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4555
4556 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4557 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4558 etc_passwd_lock = safe_close(etc_passwd_lock);
4559
4560 (void) sd_notifyf(false,
4561 "STATUS=Container running.\n"
4562 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
4563 if (!arg_notify_ready)
4564 (void) sd_notify(false, "READY=1\n");
4565
4566 if (arg_kill_signal > 0) {
4567 /* Try to kill the init system on SIGINT or SIGTERM */
4568 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4569 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
4570 } else {
4571 /* Immediately exit */
4572 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4573 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4574 }
4575
4576 /* Exit when the child exits */
4577 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
4578
4579 if (arg_expose_ports) {
4580 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
4581 if (r < 0)
4582 return r;
4583
4584 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
4585 }
4586
4587 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4588
4589 if (arg_console_mode != CONSOLE_PIPE) {
4590 _cleanup_close_ int fd = -1;
4591 PTYForwardFlags flags = 0;
4592
4593 /* Retrieve the master pty allocated by inner child */
4594 fd = receive_one_fd(master_pty_socket_pair[0], 0);
4595 if (fd < 0)
4596 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
4597
4598 switch (arg_console_mode) {
4599
4600 case CONSOLE_READ_ONLY:
4601 flags |= PTY_FORWARD_READ_ONLY;
4602
4603 _fallthrough_;
4604
4605 case CONSOLE_INTERACTIVE:
4606 flags |= PTY_FORWARD_IGNORE_VHANGUP;
4607
4608 r = pty_forward_new(event, fd, flags, &forward);
4609 if (r < 0)
4610 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4611
4612 if (arg_console_width != (unsigned) -1 || arg_console_height != (unsigned) -1)
4613 (void) pty_forward_set_width_height(forward,
4614 arg_console_width,
4615 arg_console_height);
4616 break;
4617
4618 default:
4619 assert(arg_console_mode == CONSOLE_PASSIVE);
4620 }
4621
4622 *master = TAKE_FD(fd);
4623 }
4624
4625 r = sd_event_loop(event);
4626 if (r < 0)
4627 return log_error_errno(r, "Failed to run event loop: %m");
4628
4629 if (forward) {
4630 char last_char = 0;
4631
4632 (void) pty_forward_get_last_char(forward, &last_char);
4633 forward = pty_forward_free(forward);
4634
4635 if (!arg_quiet && last_char != '\n')
4636 putc('\n', stdout);
4637 }
4638
4639 /* Kill if it is not dead yet anyway */
4640 if (!arg_register && !arg_keep_unit && bus)
4641 terminate_scope(bus, arg_machine);
4642
4643 /* Normally redundant, but better safe than sorry */
4644 (void) kill(*pid, SIGKILL);
4645
4646 r = wait_for_container(*pid, &container_status);
4647 *pid = 0;
4648
4649 /* Tell machined that we are gone. */
4650 if (bus)
4651 (void) unregister_machine(bus, arg_machine);
4652
4653 if (r < 0)
4654 /* We failed to wait for the container, or the container exited abnormally. */
4655 return r;
4656 if (r > 0 || container_status == CONTAINER_TERMINATED) {
4657 /* r > 0 → The container exited with a non-zero status.
4658 * As a special case, we need to replace 133 with a different value,
4659 * because 133 is special-cased in the service file to reboot the container.
4660 * otherwise → The container exited with zero status and a reboot was not requested.
4661 */
4662 if (r == EXIT_FORCE_RESTART)
4663 r = EXIT_FAILURE; /* replace 133 with the general failure code */
4664 *ret = r;
4665 return 0; /* finito */
4666 }
4667
4668 /* CONTAINER_REBOOTED, loop again */
4669
4670 if (arg_keep_unit) {
4671 /* Special handling if we are running as a service: instead of simply
4672 * restarting the machine we want to restart the entire service, so let's
4673 * inform systemd about this with the special exit code 133. The service
4674 * file uses RestartForceExitStatus=133 so that this results in a full
4675 * nspawn restart. This is necessary since we might have cgroup parameters
4676 * set we want to have flushed out. */
4677 *ret = EXIT_FORCE_RESTART;
4678 return 0; /* finito */
4679 }
4680
4681 expose_port_flush(arg_expose_ports, exposed);
4682
4683 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4684 *veth_created = false;
4685 return 1; /* loop again */
4686 }
4687
4688 static int initialize_rlimits(void) {
4689 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4690 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4691 * container execution environments. */
4692
4693 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
4694 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
4695 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
4696 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
4697 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
4698 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
4699 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
4700 [RLIMIT_MEMLOCK] = { 65536, 65536 },
4701 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
4702 [RLIMIT_NICE] = { 0, 0 },
4703 [RLIMIT_NOFILE] = { 1024, 4096 },
4704 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
4705 [RLIMIT_RTPRIO] = { 0, 0 },
4706 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
4707 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
4708
4709 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4710 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4711 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4712 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4713 * that PID 1 changes a number of other resource limits during early initialization which is why we
4714 * don't read the other limits from PID 1 but prefer the static table above. */
4715 };
4716
4717 int rl;
4718
4719 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
4720 /* Let's only fill in what the user hasn't explicitly configured anyway */
4721 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
4722 const struct rlimit *v;
4723 struct rlimit buffer;
4724
4725 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
4726 /* For these two let's read the limits off PID 1. See above for an explanation. */
4727
4728 if (prlimit(1, rl, NULL, &buffer) < 0)
4729 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
4730
4731 v = &buffer;
4732 } else
4733 v = kernel_defaults + rl;
4734
4735 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
4736 if (!arg_rlimit[rl])
4737 return log_oom();
4738 }
4739
4740 if (DEBUG_LOGGING) {
4741 _cleanup_free_ char *k = NULL;
4742
4743 (void) rlimit_format(arg_rlimit[rl], &k);
4744 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
4745 }
4746 }
4747
4748 return 0;
4749 }
4750
4751 static int run(int argc, char *argv[]) {
4752 bool secondary = false, remove_directory = false, remove_image = false,
4753 veth_created = false, remove_tmprootdir = false;
4754 _cleanup_close_ int master = -1;
4755 _cleanup_fdset_free_ FDSet *fds = NULL;
4756 int r, n_fd_passed, ret = EXIT_SUCCESS;
4757 char veth_name[IFNAMSIZ] = "";
4758 union in_addr_union exposed = {};
4759 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4760 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
4761 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
4762 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4763 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
4764 pid_t pid = 0;
4765
4766 log_parse_environment();
4767 log_open();
4768
4769 r = parse_argv(argc, argv);
4770 if (r <= 0)
4771 goto finish;
4772
4773 r = must_be_root();
4774 if (r < 0)
4775 goto finish;
4776
4777 r = initialize_rlimits();
4778 if (r < 0)
4779 goto finish;
4780
4781 r = load_oci_bundle();
4782 if (r < 0)
4783 goto finish;
4784
4785 r = determine_names();
4786 if (r < 0)
4787 goto finish;
4788
4789 r = load_settings();
4790 if (r < 0)
4791 goto finish;
4792
4793 r = cg_unified();
4794 if (r < 0) {
4795 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4796 goto finish;
4797 }
4798
4799 r = verify_arguments();
4800 if (r < 0)
4801 goto finish;
4802
4803 /* Reapply environment settings. */
4804 (void) detect_unified_cgroup_hierarchy_from_environment();
4805
4806 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4807 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4808 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4809 (void) ignore_signals(SIGPIPE, -1);
4810
4811 n_fd_passed = sd_listen_fds(false);
4812 if (n_fd_passed > 0) {
4813 r = fdset_new_listen_fds(&fds, false);
4814 if (r < 0) {
4815 log_error_errno(r, "Failed to collect file descriptors: %m");
4816 goto finish;
4817 }
4818 }
4819
4820 /* The "default" umask. This is appropriate for most file and directory
4821 * operations performed by nspawn, and is the umask that will be used for
4822 * the child. Functions like copy_devnodes() change the umask temporarily. */
4823 umask(0022);
4824
4825 if (arg_directory) {
4826 assert(!arg_image);
4827
4828 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4829 * /var from the host will propagate into container dynamically (because bad things happen if
4830 * two systems write to the same /var). Let's allow it for the special cases where /var is
4831 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4832 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
4833 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4834 r = -EINVAL;
4835 goto finish;
4836 }
4837
4838 if (arg_ephemeral) {
4839 _cleanup_free_ char *np = NULL;
4840
4841 r = chase_symlinks_and_update(&arg_directory, 0);
4842 if (r < 0)
4843 goto finish;
4844
4845 /* If the specified path is a mount point we generate the new snapshot immediately
4846 * inside it under a random name. However if the specified is not a mount point we
4847 * create the new snapshot in the parent directory, just next to it. */
4848 r = path_is_mount_point(arg_directory, NULL, 0);
4849 if (r < 0) {
4850 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4851 goto finish;
4852 }
4853 if (r > 0)
4854 r = tempfn_random_child(arg_directory, "machine.", &np);
4855 else
4856 r = tempfn_random(arg_directory, "machine.", &np);
4857 if (r < 0) {
4858 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
4859 goto finish;
4860 }
4861
4862 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4863 * only owned by us and noone else. */
4864 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
4865 if (r < 0) {
4866 log_error_errno(r, "Failed to lock %s: %m", np);
4867 goto finish;
4868 }
4869
4870 {
4871 BLOCK_SIGNALS(SIGINT);
4872 r = btrfs_subvol_snapshot(arg_directory, np,
4873 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4874 BTRFS_SNAPSHOT_FALLBACK_COPY |
4875 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4876 BTRFS_SNAPSHOT_RECURSIVE |
4877 BTRFS_SNAPSHOT_QUOTA |
4878 BTRFS_SNAPSHOT_SIGINT);
4879 }
4880 if (r == -EINTR) {
4881 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
4882 goto finish;
4883 }
4884 if (r < 0) {
4885 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4886 goto finish;
4887 }
4888
4889 free_and_replace(arg_directory, np);
4890 remove_directory = true;
4891 } else {
4892 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
4893 if (r < 0)
4894 goto finish;
4895
4896 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4897 if (r == -EBUSY) {
4898 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4899 goto finish;
4900 }
4901 if (r < 0) {
4902 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4903 goto finish;
4904 }
4905
4906 if (arg_template) {
4907 r = chase_symlinks_and_update(&arg_template, 0);
4908 if (r < 0)
4909 goto finish;
4910
4911 {
4912 BLOCK_SIGNALS(SIGINT);
4913 r = btrfs_subvol_snapshot(arg_template, arg_directory,
4914 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4915 BTRFS_SNAPSHOT_FALLBACK_COPY |
4916 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4917 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
4918 BTRFS_SNAPSHOT_RECURSIVE |
4919 BTRFS_SNAPSHOT_QUOTA |
4920 BTRFS_SNAPSHOT_SIGINT);
4921 }
4922 if (r == -EEXIST)
4923 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4924 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4925 else if (r == -EINTR) {
4926 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
4927 goto finish;
4928 } else if (r < 0) {
4929 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4930 goto finish;
4931 } else
4932 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4933 "Populated %s from template %s.", arg_directory, arg_template);
4934 }
4935 }
4936
4937 if (arg_start_mode == START_BOOT) {
4938 const char *p;
4939
4940 if (arg_pivot_root_new)
4941 p = prefix_roota(arg_directory, arg_pivot_root_new);
4942 else
4943 p = arg_directory;
4944
4945 if (path_is_os_tree(p) <= 0) {
4946 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
4947 r = -EINVAL;
4948 goto finish;
4949 }
4950 } else {
4951 const char *p, *q;
4952
4953 if (arg_pivot_root_new)
4954 p = prefix_roota(arg_directory, arg_pivot_root_new);
4955 else
4956 p = arg_directory;
4957
4958 q = strjoina(p, "/usr/");
4959
4960 if (laccess(q, F_OK) < 0) {
4961 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
4962 r = -EINVAL;
4963 goto finish;
4964 }
4965 }
4966
4967 } else {
4968 assert(arg_image);
4969 assert(!arg_template);
4970
4971 r = chase_symlinks_and_update(&arg_image, 0);
4972 if (r < 0)
4973 goto finish;
4974
4975 if (arg_ephemeral) {
4976 _cleanup_free_ char *np = NULL;
4977
4978 r = tempfn_random(arg_image, "machine.", &np);
4979 if (r < 0) {
4980 log_error_errno(r, "Failed to generate name for image snapshot: %m");
4981 goto finish;
4982 }
4983
4984 /* Always take an exclusive lock on our own ephemeral copy. */
4985 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
4986 if (r < 0) {
4987 r = log_error_errno(r, "Failed to create image lock: %m");
4988 goto finish;
4989 }
4990
4991 {
4992 BLOCK_SIGNALS(SIGINT);
4993 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME|COPY_SIGINT);
4994 }
4995 if (r == -EINTR) {
4996 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
4997 goto finish;
4998 }
4999 if (r < 0) {
5000 r = log_error_errno(r, "Failed to copy image file: %m");
5001 goto finish;
5002 }
5003
5004 free_and_replace(arg_image, np);
5005 remove_image = true;
5006 } else {
5007 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
5008 if (r == -EBUSY) {
5009 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
5010 goto finish;
5011 }
5012 if (r < 0) {
5013 r = log_error_errno(r, "Failed to create image lock: %m");
5014 goto finish;
5015 }
5016
5017 if (!arg_root_hash) {
5018 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
5019 if (r < 0) {
5020 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
5021 goto finish;
5022 }
5023 }
5024 }
5025
5026 if (!mkdtemp(tmprootdir)) {
5027 r = log_error_errno(errno, "Failed to create temporary directory: %m");
5028 goto finish;
5029 }
5030
5031 remove_tmprootdir = true;
5032
5033 arg_directory = strdup(tmprootdir);
5034 if (!arg_directory) {
5035 r = log_oom();
5036 goto finish;
5037 }
5038
5039 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, LO_FLAGS_PARTSCAN, &loop);
5040 if (r < 0) {
5041 log_error_errno(r, "Failed to set up loopback block device: %m");
5042 goto finish;
5043 }
5044
5045 r = dissect_image_and_warn(
5046 loop->fd,
5047 arg_image,
5048 arg_root_hash, arg_root_hash_size,
5049 DISSECT_IMAGE_REQUIRE_ROOT,
5050 &dissected_image);
5051 if (r == -ENOPKG) {
5052 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5053 log_notice("Note that the disk image needs to\n"
5054 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5055 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5056 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
5057 " d) or contain a file system without a partition table\n"
5058 "in order to be bootable with systemd-nspawn.");
5059 goto finish;
5060 }
5061 if (r < 0)
5062 goto finish;
5063
5064 if (!arg_root_hash && dissected_image->can_verity)
5065 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
5066
5067 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
5068 if (r < 0)
5069 goto finish;
5070
5071 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5072 if (remove_image && unlink(arg_image) >= 0)
5073 remove_image = false;
5074 }
5075
5076 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5077 if (r < 0)
5078 goto finish;
5079
5080 if (arg_console_mode < 0)
5081 arg_console_mode =
5082 isatty(STDIN_FILENO) > 0 &&
5083 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
5084
5085 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5086 arg_quiet = true;
5087
5088 if (!arg_quiet)
5089 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5090 arg_machine, arg_image ?: arg_directory);
5091
5092 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
5093
5094 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
5095 r = log_error_errno(errno, "Failed to become subreaper: %m");
5096 goto finish;
5097 }
5098
5099 for (;;) {
5100 r = run_container(dissected_image,
5101 secondary,
5102 fds,
5103 veth_name, &veth_created,
5104 &exposed, &master,
5105 &pid, &ret);
5106 if (r <= 0)
5107 break;
5108 }
5109
5110 finish:
5111 (void) sd_notify(false,
5112 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5113 "STOPPING=1\nSTATUS=Terminating...");
5114
5115 if (pid > 0)
5116 (void) kill(pid, SIGKILL);
5117
5118 /* Try to flush whatever is still queued in the pty */
5119 if (master >= 0) {
5120 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
5121 master = safe_close(master);
5122 }
5123
5124 if (pid > 0)
5125 (void) wait_for_terminate(pid, NULL);
5126
5127 pager_close();
5128
5129 if (remove_directory && arg_directory) {
5130 int k;
5131
5132 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5133 if (k < 0)
5134 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5135 }
5136
5137 if (remove_image && arg_image) {
5138 if (unlink(arg_image) < 0)
5139 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5140 }
5141
5142 if (remove_tmprootdir) {
5143 if (rmdir(tmprootdir) < 0)
5144 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5145 }
5146
5147 if (arg_machine) {
5148 const char *p;
5149
5150 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5151 (void) rm_rf(p, REMOVE_ROOT);
5152 }
5153
5154 expose_port_flush(arg_expose_ports, &exposed);
5155
5156 if (veth_created)
5157 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5158 (void) remove_bridge(arg_network_zone);
5159
5160 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5161 expose_port_free_all(arg_expose_ports);
5162 rlimit_free_all(arg_rlimit);
5163 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
5164
5165 if (r < 0)
5166 return r;
5167
5168 return ret;
5169 }
5170
5171 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.