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