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