]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:


fed5c651cedfa6c56fd72b23f0143cf98a085898
[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 = strjoin("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 _cleanup_free_ char *name = NULL;
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 = path_join("/dev/", d);
1899 if (!from)
1900 return log_oom();
1901
1902 to = path_join(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 = path_join("/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 = path_join(dest, sl);
1949 if (!prefixed)
1950 return log_oom();
1951
1952 t = path_join("..", 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 = path_join(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_stdio_as_dev_console(void) {
2041 int terminal;
2042 int r;
2043
2044 terminal = open_terminal("/dev/console", O_RDWR);
2045 if (terminal < 0)
2046 return log_error_errno(terminal, "Failed to open console: %m");
2047
2048 /* Make sure we can continue logging to the original stderr, even if
2049 * stderr points elsewhere now */
2050 r = log_dup_console();
2051 if (r < 0)
2052 return log_error_errno(r, "Failed to duplicate stderr: %m");
2053
2054 /* invalidates 'terminal' on success and failure */
2055 r = rearrange_stdio(terminal, terminal, terminal);
2056 if (r < 0)
2057 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2058
2059 return 0;
2060 }
2061
2062 static int setup_dev_console(const char *console) {
2063 _cleanup_free_ char *p = NULL;
2064 int r;
2065
2066 /* Create /dev/console symlink */
2067 r = path_make_relative("/dev", console, &p);
2068 if (r < 0)
2069 return log_error_errno(r, "Failed to create relative path: %m");
2070
2071 if (symlink(p, "/dev/console") < 0)
2072 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
2073
2074 return 0;
2075 }
2076
2077 static int setup_keyring(void) {
2078 key_serial_t keyring;
2079
2080 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2081 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2082 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2083 * these system calls let's make sure we don't leak anything into the container. */
2084
2085 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2086 if (keyring == -1) {
2087 if (errno == ENOSYS)
2088 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2089 else if (IN_SET(errno, EACCES, EPERM))
2090 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2091 else
2092 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2093 }
2094
2095 return 0;
2096 }
2097
2098 static int setup_kmsg(int kmsg_socket) {
2099 _cleanup_(unlink_and_freep) char *from = NULL;
2100 _cleanup_free_ char *fifo = NULL;
2101 _cleanup_close_ int fd = -1;
2102 _cleanup_umask_ mode_t u;
2103 int r;
2104
2105 assert(kmsg_socket >= 0);
2106
2107 u = umask(0000);
2108
2109 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2110 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2111 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2112 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2113
2114 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2115 if (r < 0)
2116 return log_error_errno(r, "Failed to generate kmsg path: %m");
2117
2118 if (mkfifo(fifo, 0600) < 0)
2119 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2120
2121 from = TAKE_PTR(fifo);
2122
2123 r = mount_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2124 if (r < 0)
2125 return r;
2126
2127 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2128 if (fd < 0)
2129 return log_error_errno(errno, "Failed to open fifo: %m");
2130
2131 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2132 r = send_one_fd(kmsg_socket, fd, 0);
2133 if (r < 0)
2134 return log_error_errno(r, "Failed to send FIFO fd: %m");
2135
2136 return 0;
2137 }
2138
2139 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2140 union in_addr_union *exposed = userdata;
2141
2142 assert(rtnl);
2143 assert(m);
2144 assert(exposed);
2145
2146 expose_port_execute(rtnl, arg_expose_ports, exposed);
2147 return 0;
2148 }
2149
2150 static int setup_hostname(void) {
2151 int r;
2152
2153 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2154 return 0;
2155
2156 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2157 if (r < 0)
2158 return log_error_errno(r, "Failed to set hostname: %m");
2159
2160 return 0;
2161 }
2162
2163 static int setup_journal(const char *directory) {
2164 _cleanup_free_ char *d = NULL;
2165 const char *dirname, *p, *q;
2166 sd_id128_t this_id;
2167 char id[33];
2168 bool try;
2169 int r;
2170
2171 /* Don't link journals in ephemeral mode */
2172 if (arg_ephemeral)
2173 return 0;
2174
2175 if (arg_link_journal == LINK_NO)
2176 return 0;
2177
2178 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2179
2180 r = sd_id128_get_machine(&this_id);
2181 if (r < 0)
2182 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2183
2184 if (sd_id128_equal(arg_uuid, this_id)) {
2185 log_full(try ? LOG_WARNING : LOG_ERR,
2186 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
2187 if (try)
2188 return 0;
2189 return -EEXIST;
2190 }
2191
2192 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2193 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2194 if (r < 0) {
2195 bool ignore = r == -EROFS && try;
2196 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2197 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2198 return ignore ? 0 : r;
2199 }
2200 }
2201
2202 (void) sd_id128_to_string(arg_uuid, id);
2203
2204 p = strjoina("/var/log/journal/", id);
2205 q = prefix_roota(directory, p);
2206
2207 if (path_is_mount_point(p, NULL, 0) > 0) {
2208 if (try)
2209 return 0;
2210
2211 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2212 "%s: already a mount point, refusing to use for journal", p);
2213 }
2214
2215 if (path_is_mount_point(q, NULL, 0) > 0) {
2216 if (try)
2217 return 0;
2218
2219 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2220 "%s: already a mount point, refusing to use for journal", q);
2221 }
2222
2223 r = readlink_and_make_absolute(p, &d);
2224 if (r >= 0) {
2225 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2226 path_equal(d, q)) {
2227
2228 r = userns_mkdir(directory, p, 0755, 0, 0);
2229 if (r < 0)
2230 log_warning_errno(r, "Failed to create directory %s: %m", q);
2231 return 0;
2232 }
2233
2234 if (unlink(p) < 0)
2235 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2236 } else if (r == -EINVAL) {
2237
2238 if (arg_link_journal == LINK_GUEST &&
2239 rmdir(p) < 0) {
2240
2241 if (errno == ENOTDIR) {
2242 log_error("%s already exists and is neither a symlink nor a directory", p);
2243 return r;
2244 } else
2245 return log_error_errno(errno, "Failed to remove %s: %m", p);
2246 }
2247 } else if (r != -ENOENT)
2248 return log_error_errno(r, "readlink(%s) failed: %m", p);
2249
2250 if (arg_link_journal == LINK_GUEST) {
2251
2252 if (symlink(q, p) < 0) {
2253 if (try) {
2254 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2255 return 0;
2256 } else
2257 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2258 }
2259
2260 r = userns_mkdir(directory, p, 0755, 0, 0);
2261 if (r < 0)
2262 log_warning_errno(r, "Failed to create directory %s: %m", q);
2263 return 0;
2264 }
2265
2266 if (arg_link_journal == LINK_HOST) {
2267 /* don't create parents here — if the host doesn't have
2268 * permanent journal set up, don't force it here */
2269
2270 r = mkdir_errno_wrapper(p, 0755);
2271 if (r < 0 && r != -EEXIST) {
2272 if (try) {
2273 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2274 return 0;
2275 } else
2276 return log_error_errno(r, "Failed to create %s: %m", p);
2277 }
2278
2279 } else if (access(p, F_OK) < 0)
2280 return 0;
2281
2282 if (dir_is_empty(q) == 0)
2283 log_warning("%s is not empty, proceeding anyway.", q);
2284
2285 r = userns_mkdir(directory, p, 0755, 0, 0);
2286 if (r < 0)
2287 return log_error_errno(r, "Failed to create %s: %m", q);
2288
2289 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2290 if (r < 0)
2291 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2292
2293 return 0;
2294 }
2295
2296 static int drop_capabilities(uid_t uid) {
2297 CapabilityQuintet q;
2298
2299 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2300 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2301 * arg_caps_retain. */
2302
2303 if (capability_quintet_is_set(&arg_full_capabilities)) {
2304 q = arg_full_capabilities;
2305
2306 if (q.bounding == (uint64_t) -1)
2307 q.bounding = uid == 0 ? arg_caps_retain : 0;
2308
2309 if (q.effective == (uint64_t) -1)
2310 q.effective = uid == 0 ? q.bounding : 0;
2311
2312 if (q.inheritable == (uint64_t) -1)
2313 q.inheritable = uid == 0 ? q.bounding : 0;
2314
2315 if (q.permitted == (uint64_t) -1)
2316 q.permitted = uid == 0 ? q.bounding : 0;
2317
2318 if (q.ambient == (uint64_t) -1 && ambient_capabilities_supported())
2319 q.ambient = 0;
2320
2321 if (capability_quintet_mangle(&q))
2322 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2323
2324 } else {
2325 q = (CapabilityQuintet) {
2326 .bounding = arg_caps_retain,
2327 .effective = uid == 0 ? arg_caps_retain : 0,
2328 .inheritable = uid == 0 ? arg_caps_retain : 0,
2329 .permitted = uid == 0 ? arg_caps_retain : 0,
2330 .ambient = ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2331 };
2332
2333 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2334 * in order to maintain the same behavior as systemd < 242. */
2335 if (capability_quintet_mangle(&q))
2336 log_warning("Some capabilities will not be set because they are not in the current bounding set.");
2337
2338 }
2339
2340 return capability_quintet_enforce(&q);
2341 }
2342
2343 static int reset_audit_loginuid(void) {
2344 _cleanup_free_ char *p = NULL;
2345 int r;
2346
2347 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2348 return 0;
2349
2350 r = read_one_line_file("/proc/self/loginuid", &p);
2351 if (r == -ENOENT)
2352 return 0;
2353 if (r < 0)
2354 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2355
2356 /* Already reset? */
2357 if (streq(p, "4294967295"))
2358 return 0;
2359
2360 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2361 if (r < 0) {
2362 log_error_errno(r,
2363 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2364 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2365 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2366 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2367 "using systemd-nspawn. Sleeping for 5s... (%m)");
2368
2369 sleep(5);
2370 }
2371
2372 return 0;
2373 }
2374
2375 static int setup_propagate(const char *root) {
2376 const char *p, *q;
2377 int r;
2378
2379 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2380 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2381 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2382 (void) mkdir_p(p, 0600);
2383
2384 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
2385 if (r < 0)
2386 return log_error_errno(r, "Failed to create /run/systemd: %m");
2387
2388 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
2389 if (r < 0)
2390 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
2391
2392 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2393 if (r < 0)
2394 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
2395
2396 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
2397 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2398 if (r < 0)
2399 return r;
2400
2401 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2402 if (r < 0)
2403 return r;
2404
2405 /* machined will MS_MOVE into that directory, and that's only
2406 * supported for non-shared mounts. */
2407 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2408 }
2409
2410 static int setup_machine_id(const char *directory) {
2411 const char *etc_machine_id;
2412 sd_id128_t id;
2413 int r;
2414
2415 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2416 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2417 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2418 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2419 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2420 * container behaves nicely). */
2421
2422 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2423
2424 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
2425 if (r < 0) {
2426 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2427 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2428
2429 if (sd_id128_is_null(arg_uuid)) {
2430 r = sd_id128_randomize(&arg_uuid);
2431 if (r < 0)
2432 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2433 }
2434 } else {
2435 if (sd_id128_is_null(id))
2436 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2437 "Machine ID in container image is zero, refusing.");
2438
2439 arg_uuid = id;
2440 }
2441
2442 return 0;
2443 }
2444
2445 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2446 int r;
2447
2448 assert(directory);
2449
2450 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
2451 return 0;
2452
2453 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2454 if (r == -EOPNOTSUPP)
2455 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.");
2456 if (r == -EBADE)
2457 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2458 if (r < 0)
2459 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2460 if (r == 0)
2461 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2462 else
2463 log_debug("Patched directory tree to match UID/GID range.");
2464
2465 return r;
2466 }
2467
2468 /*
2469 * Return values:
2470 * < 0 : wait_for_terminate() failed to get the state of the
2471 * container, the container was terminated by a signal, or
2472 * failed for an unknown reason. No change is made to the
2473 * container argument.
2474 * > 0 : The program executed in the container terminated with an
2475 * error. The exit code of the program executed in the
2476 * container is returned. The container argument has been set
2477 * to CONTAINER_TERMINATED.
2478 * 0 : The container is being rebooted, has been shut down or exited
2479 * successfully. The container argument has been set to either
2480 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2481 *
2482 * That is, success is indicated by a return value of zero, and an
2483 * error is indicated by a non-zero value.
2484 */
2485 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2486 siginfo_t status;
2487 int r;
2488
2489 r = wait_for_terminate(pid, &status);
2490 if (r < 0)
2491 return log_warning_errno(r, "Failed to wait for container: %m");
2492
2493 switch (status.si_code) {
2494
2495 case CLD_EXITED:
2496 if (status.si_status == 0)
2497 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2498 else
2499 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2500
2501 *container = CONTAINER_TERMINATED;
2502 return status.si_status;
2503
2504 case CLD_KILLED:
2505 if (status.si_status == SIGINT) {
2506 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2507 *container = CONTAINER_TERMINATED;
2508 return 0;
2509
2510 } else if (status.si_status == SIGHUP) {
2511 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2512 *container = CONTAINER_REBOOTED;
2513 return 0;
2514 }
2515
2516 _fallthrough_;
2517 case CLD_DUMPED:
2518 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2519 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2520
2521 default:
2522 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2523 "Container %s failed due to unknown reason.", arg_machine);
2524 }
2525 }
2526
2527 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2528 pid_t pid;
2529
2530 pid = PTR_TO_PID(userdata);
2531 if (pid > 0) {
2532 if (kill(pid, arg_kill_signal) >= 0) {
2533 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2534 sd_event_source_set_userdata(s, NULL);
2535 return 0;
2536 }
2537 }
2538
2539 sd_event_exit(sd_event_source_get_event(s), 0);
2540 return 0;
2541 }
2542
2543 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2544 pid_t pid;
2545
2546 assert(s);
2547 assert(ssi);
2548
2549 pid = PTR_TO_PID(userdata);
2550
2551 for (;;) {
2552 siginfo_t si = {};
2553
2554 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2555 return log_error_errno(errno, "Failed to waitid(): %m");
2556 if (si.si_pid == 0) /* No pending children. */
2557 break;
2558 if (si.si_pid == pid) {
2559 /* The main process we care for has exited. Return from
2560 * signal handler but leave the zombie. */
2561 sd_event_exit(sd_event_source_get_event(s), 0);
2562 break;
2563 }
2564
2565 /* Reap all other children. */
2566 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2567 }
2568
2569 return 0;
2570 }
2571
2572 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2573 pid_t pid;
2574
2575 assert(m);
2576
2577 pid = PTR_TO_PID(userdata);
2578
2579 if (arg_kill_signal > 0) {
2580 log_info("Container termination requested. Attempting to halt container.");
2581 (void) kill(pid, arg_kill_signal);
2582 } else {
2583 log_info("Container termination requested. Exiting.");
2584 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2585 }
2586
2587 return 0;
2588 }
2589
2590 static int determine_names(void) {
2591 int r;
2592
2593 if (arg_template && !arg_directory && arg_machine) {
2594
2595 /* If --template= was specified then we should not
2596 * search for a machine, but instead create a new one
2597 * in /var/lib/machine. */
2598
2599 arg_directory = path_join("/var/lib/machines", arg_machine);
2600 if (!arg_directory)
2601 return log_oom();
2602 }
2603
2604 if (!arg_image && !arg_directory) {
2605 if (arg_machine) {
2606 _cleanup_(image_unrefp) Image *i = NULL;
2607
2608 r = image_find(IMAGE_MACHINE, arg_machine, &i);
2609 if (r == -ENOENT)
2610 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
2611 if (r < 0)
2612 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2613
2614 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2615 r = free_and_strdup(&arg_image, i->path);
2616 else
2617 r = free_and_strdup(&arg_directory, i->path);
2618 if (r < 0)
2619 return log_oom();
2620
2621 if (!arg_ephemeral)
2622 arg_read_only = arg_read_only || i->read_only;
2623 } else {
2624 r = safe_getcwd(&arg_directory);
2625 if (r < 0)
2626 return log_error_errno(r, "Failed to determine current directory: %m");
2627 }
2628
2629 if (!arg_directory && !arg_image)
2630 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
2631 }
2632
2633 if (!arg_machine) {
2634 if (arg_directory && path_equal(arg_directory, "/"))
2635 arg_machine = gethostname_malloc();
2636 else {
2637 if (arg_image) {
2638 char *e;
2639
2640 arg_machine = strdup(basename(arg_image));
2641
2642 /* Truncate suffix if there is one */
2643 e = endswith(arg_machine, ".raw");
2644 if (e)
2645 *e = 0;
2646 } else
2647 arg_machine = strdup(basename(arg_directory));
2648 }
2649 if (!arg_machine)
2650 return log_oom();
2651
2652 hostname_cleanup(arg_machine);
2653 if (!machine_name_is_valid(arg_machine))
2654 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
2655
2656 if (arg_ephemeral) {
2657 char *b;
2658
2659 /* Add a random suffix when this is an
2660 * ephemeral machine, so that we can run many
2661 * instances at once without manually having
2662 * to specify -M each time. */
2663
2664 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2665 return log_oom();
2666
2667 free(arg_machine);
2668 arg_machine = b;
2669 }
2670 }
2671
2672 return 0;
2673 }
2674
2675 static int chase_symlinks_and_update(char **p, unsigned flags) {
2676 char *chased;
2677 int r;
2678
2679 assert(p);
2680
2681 if (!*p)
2682 return 0;
2683
2684 r = chase_symlinks(*p, NULL, flags, &chased);
2685 if (r < 0)
2686 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2687
2688 free_and_replace(*p, chased);
2689 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2690 }
2691
2692 static int determine_uid_shift(const char *directory) {
2693 int r;
2694
2695 if (arg_userns_mode == USER_NAMESPACE_NO) {
2696 arg_uid_shift = 0;
2697 return 0;
2698 }
2699
2700 if (arg_uid_shift == UID_INVALID) {
2701 struct stat st;
2702
2703 r = stat(directory, &st);
2704 if (r < 0)
2705 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2706
2707 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2708
2709 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
2710 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2711 "UID and GID base of %s don't match.", directory);
2712
2713 arg_uid_range = UINT32_C(0x10000);
2714 }
2715
2716 if (arg_uid_shift > (uid_t) -1 - arg_uid_range)
2717 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2718 "UID base too high for UID range.");
2719
2720 return 0;
2721 }
2722
2723 static unsigned long effective_clone_ns_flags(void) {
2724 unsigned long flags = arg_clone_ns_flags;
2725
2726 if (arg_private_network)
2727 flags |= CLONE_NEWNET;
2728 if (arg_use_cgns)
2729 flags |= CLONE_NEWCGROUP;
2730 if (arg_userns_mode != USER_NAMESPACE_NO)
2731 flags |= CLONE_NEWUSER;
2732
2733 return flags;
2734 }
2735
2736 static int patch_sysctl(void) {
2737
2738 /* This table is inspired by runc's sysctl() function */
2739 static const struct {
2740 const char *key;
2741 bool prefix;
2742 unsigned long clone_flags;
2743 } safe_sysctl[] = {
2744 { "kernel.hostname", false, CLONE_NEWUTS },
2745 { "kernel.domainname", false, CLONE_NEWUTS },
2746 { "kernel.msgmax", false, CLONE_NEWIPC },
2747 { "kernel.msgmnb", false, CLONE_NEWIPC },
2748 { "kernel.msgmni", false, CLONE_NEWIPC },
2749 { "kernel.sem", false, CLONE_NEWIPC },
2750 { "kernel.shmall", false, CLONE_NEWIPC },
2751 { "kernel.shmmax", false, CLONE_NEWIPC },
2752 { "kernel.shmmni", false, CLONE_NEWIPC },
2753 { "fs.mqueue.", true, CLONE_NEWIPC },
2754 { "net.", true, CLONE_NEWNET },
2755 };
2756
2757 unsigned long flags;
2758 char **k, **v;
2759 int r;
2760
2761 flags = effective_clone_ns_flags();
2762
2763 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
2764 bool good = false;
2765 size_t i;
2766
2767 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
2768
2769 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
2770 continue;
2771
2772 if (safe_sysctl[i].prefix)
2773 good = startswith(*k, safe_sysctl[i].key);
2774 else
2775 good = streq(*k, safe_sysctl[i].key);
2776
2777 if (good)
2778 break;
2779 }
2780
2781 if (!good)
2782 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
2783
2784 r = sysctl_write(*k, *v);
2785 if (r < 0)
2786 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
2787 }
2788
2789 return 0;
2790 }
2791
2792 static int inner_child(
2793 Barrier *barrier,
2794 const char *directory,
2795 bool secondary,
2796 int kmsg_socket,
2797 int rtnl_socket,
2798 int master_pty_socket,
2799 FDSet *fds) {
2800
2801 _cleanup_free_ char *home = NULL;
2802 char as_uuid[37];
2803 size_t n_env = 1;
2804 const char *envp[] = {
2805 "PATH=" DEFAULT_PATH_COMPAT,
2806 NULL, /* container */
2807 NULL, /* TERM */
2808 NULL, /* HOME */
2809 NULL, /* USER */
2810 NULL, /* LOGNAME */
2811 NULL, /* container_uuid */
2812 NULL, /* LISTEN_FDS */
2813 NULL, /* LISTEN_PID */
2814 NULL, /* NOTIFY_SOCKET */
2815 NULL
2816 };
2817 const char *exec_target;
2818 _cleanup_strv_free_ char **env_use = NULL;
2819 int r, which_failed;
2820
2821 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2822 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2823 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2824 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2825 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2826 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2827 * namespace.
2828 *
2829 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2830 * unshare(). See below. */
2831
2832 assert(barrier);
2833 assert(directory);
2834 assert(kmsg_socket >= 0);
2835
2836 log_debug("Inner child is initializing.");
2837
2838 if (arg_userns_mode != USER_NAMESPACE_NO) {
2839 /* Tell the parent, that it now can write the UID map. */
2840 (void) barrier_place(barrier); /* #1 */
2841
2842 /* Wait until the parent wrote the UID map */
2843 if (!barrier_place_and_sync(barrier)) /* #2 */
2844 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2845 "Parent died too early");
2846 }
2847
2848 r = reset_uid_gid();
2849 if (r < 0)
2850 return log_error_errno(r, "Couldn't become new root: %m");
2851
2852 r = mount_all(NULL,
2853 arg_mount_settings | MOUNT_IN_USERNS,
2854 arg_uid_shift,
2855 arg_selinux_apifs_context);
2856 if (r < 0)
2857 return r;
2858
2859 if (!arg_network_namespace_path && arg_private_network) {
2860 r = unshare(CLONE_NEWNET);
2861 if (r < 0)
2862 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2863
2864 /* Tell the parent that it can setup network interfaces. */
2865 (void) barrier_place(barrier); /* #3 */
2866 }
2867
2868 r = mount_sysfs(NULL, arg_mount_settings);
2869 if (r < 0)
2870 return r;
2871
2872 /* Wait until we are cgroup-ified, so that we
2873 * can mount the right cgroup path writable */
2874 if (!barrier_place_and_sync(barrier)) /* #4 */
2875 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2876 "Parent died too early");
2877
2878 if (arg_use_cgns) {
2879 r = unshare(CLONE_NEWCGROUP);
2880 if (r < 0)
2881 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2882 r = mount_cgroups(
2883 "",
2884 arg_unified_cgroup_hierarchy,
2885 arg_userns_mode != USER_NAMESPACE_NO,
2886 arg_uid_shift,
2887 arg_uid_range,
2888 arg_selinux_apifs_context,
2889 true);
2890 if (r < 0)
2891 return r;
2892 } else {
2893 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2894 if (r < 0)
2895 return r;
2896 }
2897
2898 r = setup_boot_id();
2899 if (r < 0)
2900 return r;
2901
2902 r = setup_kmsg(kmsg_socket);
2903 if (r < 0)
2904 return r;
2905 kmsg_socket = safe_close(kmsg_socket);
2906
2907 r = mount_custom(
2908 "/",
2909 arg_custom_mounts,
2910 arg_n_custom_mounts,
2911 false,
2912 0,
2913 0,
2914 arg_selinux_apifs_context,
2915 true);
2916 if (r < 0)
2917 return r;
2918
2919 if (setsid() < 0)
2920 return log_error_errno(errno, "setsid() failed: %m");
2921
2922 if (arg_private_network)
2923 loopback_setup();
2924
2925 if (arg_expose_ports) {
2926 r = expose_port_send_rtnl(rtnl_socket);
2927 if (r < 0)
2928 return r;
2929 rtnl_socket = safe_close(rtnl_socket);
2930 }
2931
2932 if (arg_console_mode != CONSOLE_PIPE) {
2933 _cleanup_close_ int master = -1;
2934 _cleanup_free_ char *console = NULL;
2935
2936 /* Allocate a pty and make it available as /dev/console. */
2937 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
2938 if (master < 0)
2939 return log_error_errno(master, "Failed to allocate a pty: %m");
2940
2941 r = setup_dev_console(console);
2942 if (r < 0)
2943 return log_error_errno(r, "Failed to setup /dev/console: %m");
2944
2945 r = send_one_fd(master_pty_socket, master, 0);
2946 if (r < 0)
2947 return log_error_errno(r, "Failed to send master fd: %m");
2948 master_pty_socket = safe_close(master_pty_socket);
2949
2950 r = setup_stdio_as_dev_console();
2951 if (r < 0)
2952 return r;
2953 }
2954
2955 r = patch_sysctl();
2956 if (r < 0)
2957 return r;
2958
2959 if (arg_oom_score_adjust_set) {
2960 r = set_oom_score_adjust(arg_oom_score_adjust);
2961 if (r < 0)
2962 return log_error_errno(r, "Failed to adjust OOM score: %m");
2963 }
2964
2965 if (arg_cpu_set.set)
2966 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
2967 return log_error_errno(errno, "Failed to set CPU affinity: %m");
2968
2969 (void) setup_hostname();
2970
2971 if (arg_personality != PERSONALITY_INVALID) {
2972 r = safe_personality(arg_personality);
2973 if (r < 0)
2974 return log_error_errno(r, "personality() failed: %m");
2975 } else if (secondary) {
2976 r = safe_personality(PER_LINUX32);
2977 if (r < 0)
2978 return log_error_errno(r, "personality() failed: %m");
2979 }
2980
2981 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
2982 if (r < 0)
2983 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
2984
2985 #if HAVE_SECCOMP
2986 if (arg_seccomp) {
2987
2988 if (is_seccomp_available()) {
2989
2990 r = seccomp_load(arg_seccomp);
2991 if (ERRNO_IS_SECCOMP_FATAL(r))
2992 return log_error_errno(r, "Failed to install seccomp filter: %m");
2993 if (r < 0)
2994 log_debug_errno(r, "Failed to install seccomp filter: %m");
2995 }
2996 } else
2997 #endif
2998 {
2999 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
3000 if (r < 0)
3001 return r;
3002 }
3003
3004 #if HAVE_SELINUX
3005 if (arg_selinux_context)
3006 if (setexeccon(arg_selinux_context) < 0)
3007 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3008 #endif
3009
3010 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3011 * if we need to later on. */
3012 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3013 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3014
3015 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3016 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids);
3017 else
3018 r = change_uid_gid(arg_user, &home);
3019 if (r < 0)
3020 return r;
3021
3022 r = drop_capabilities(getuid());
3023 if (r < 0)
3024 return log_error_errno(r, "Dropping capabilities failed: %m");
3025
3026 if (arg_no_new_privileges)
3027 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3028 return log_error_errno(errno, "Failed to disable new privileges: %m");
3029
3030 /* LXC sets container=lxc, so follow the scheme here */
3031 envp[n_env++] = strjoina("container=", arg_container_service_name);
3032
3033 envp[n_env] = strv_find_prefix(environ, "TERM=");
3034 if (envp[n_env])
3035 n_env++;
3036
3037 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3038 if (asprintf((char**)(envp + n_env++), "HOME=%s", home ?: "/root") < 0)
3039 return log_oom();
3040
3041 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3042 if (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ?: "root") < 0 ||
3043 asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3044 return log_oom();
3045
3046 assert(!sd_id128_is_null(arg_uuid));
3047
3048 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
3049 return log_oom();
3050
3051 if (fdset_size(fds) > 0) {
3052 r = fdset_cloexec(fds, false);
3053 if (r < 0)
3054 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3055
3056 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3057 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3058 return log_oom();
3059 }
3060 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3061 return log_oom();
3062
3063 env_use = strv_env_merge(2, envp, arg_setenv);
3064 if (!env_use)
3065 return log_oom();
3066
3067 /* Let the parent know that we are ready and
3068 * wait until the parent is ready with the
3069 * setup, too... */
3070 if (!barrier_place_and_sync(barrier)) /* #5 */
3071 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3072 "Parent died too early");
3073
3074 if (arg_chdir)
3075 if (chdir(arg_chdir) < 0)
3076 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3077
3078 if (arg_start_mode == START_PID2) {
3079 r = stub_pid1(arg_uuid);
3080 if (r < 0)
3081 return r;
3082 }
3083
3084 log_debug("Inner child completed, invoking payload.");
3085
3086 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3087 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3088 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3089 log_close();
3090 log_set_open_when_needed(true);
3091
3092 (void) fdset_close_others(fds);
3093
3094 if (arg_start_mode == START_BOOT) {
3095 char **a;
3096 size_t m;
3097
3098 /* Automatically search for the init system */
3099
3100 m = strv_length(arg_parameters);
3101 a = newa(char*, m + 2);
3102 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3103 a[1 + m] = NULL;
3104
3105 a[0] = (char*) "/usr/lib/systemd/systemd";
3106 execve(a[0], a, env_use);
3107
3108 a[0] = (char*) "/lib/systemd/systemd";
3109 execve(a[0], a, env_use);
3110
3111 a[0] = (char*) "/sbin/init";
3112 execve(a[0], a, env_use);
3113
3114 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3115 } else if (!strv_isempty(arg_parameters)) {
3116 const char *dollar_path;
3117
3118 exec_target = arg_parameters[0];
3119
3120 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3121 * binary. */
3122 dollar_path = strv_env_get(env_use, "PATH");
3123 if (dollar_path) {
3124 if (putenv((char*) dollar_path) != 0)
3125 return log_error_errno(errno, "Failed to update $PATH: %m");
3126 }
3127
3128 execvpe(arg_parameters[0], arg_parameters, env_use);
3129 } else {
3130 if (!arg_chdir)
3131 /* If we cannot change the directory, we'll end up in /, that is expected. */
3132 (void) chdir(home ?: "/root");
3133
3134 execle("/bin/bash", "-bash", NULL, env_use);
3135 execle("/bin/sh", "-sh", NULL, env_use);
3136
3137 exec_target = "/bin/bash, /bin/sh";
3138 }
3139
3140 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3141 }
3142
3143 static int setup_sd_notify_child(void) {
3144 _cleanup_close_ int fd = -1;
3145 union sockaddr_union sa = {
3146 .un.sun_family = AF_UNIX,
3147 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3148 };
3149 int r;
3150
3151 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3152 if (fd < 0)
3153 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3154
3155 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3156 (void) sockaddr_un_unlink(&sa.un);
3157
3158 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3159 if (r < 0)
3160 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3161
3162 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3163 if (r < 0)
3164 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3165
3166 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3167 if (r < 0)
3168 return log_error_errno(r, "SO_PASSCRED failed: %m");
3169
3170 return TAKE_FD(fd);
3171 }
3172
3173 static int outer_child(
3174 Barrier *barrier,
3175 const char *directory,
3176 DissectedImage *dissected_image,
3177 bool secondary,
3178 int pid_socket,
3179 int uuid_socket,
3180 int notify_socket,
3181 int kmsg_socket,
3182 int rtnl_socket,
3183 int uid_shift_socket,
3184 int master_pty_socket,
3185 int unified_cgroup_hierarchy_socket,
3186 FDSet *fds,
3187 int netns_fd) {
3188
3189 _cleanup_close_ int fd = -1;
3190 pid_t pid;
3191 ssize_t l;
3192 int r;
3193
3194 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3195 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3196 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3197 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3198
3199 assert(barrier);
3200 assert(directory);
3201 assert(pid_socket >= 0);
3202 assert(uuid_socket >= 0);
3203 assert(notify_socket >= 0);
3204 assert(master_pty_socket >= 0);
3205 assert(kmsg_socket >= 0);
3206
3207 log_debug("Outer child is initializing.");
3208
3209 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3210 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3211
3212 r = reset_audit_loginuid();
3213 if (r < 0)
3214 return r;
3215
3216 /* Mark everything as slave, so that we still
3217 * receive mounts from the real root, but don't
3218 * propagate mounts to the real root. */
3219 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3220 if (r < 0)
3221 return r;
3222
3223 if (dissected_image) {
3224 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3225 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3226 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3227 * makes sure ESP partitions and userns are compatible. */
3228
3229 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3230 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
3231 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0)|
3232 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3233 if (r < 0)
3234 return r;
3235 }
3236
3237 r = determine_uid_shift(directory);
3238 if (r < 0)
3239 return r;
3240
3241 if (arg_userns_mode != USER_NAMESPACE_NO) {
3242 /* Let the parent know which UID shift we read from the image */
3243 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3244 if (l < 0)
3245 return log_error_errno(errno, "Failed to send UID shift: %m");
3246 if (l != sizeof(arg_uid_shift))
3247 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3248 "Short write while sending UID shift.");
3249
3250 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3251 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3252 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3253 * not it will pick a different one, and send it back to us. */
3254
3255 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3256 if (l < 0)
3257 return log_error_errno(errno, "Failed to recv UID shift: %m");
3258 if (l != sizeof(arg_uid_shift))
3259 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3260 "Short read while receiving UID shift.");
3261 }
3262
3263 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3264 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3265 }
3266
3267 if (!dissected_image) {
3268 /* Turn directory into bind mount */
3269 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
3270 if (r < 0)
3271 return r;
3272 }
3273
3274 r = setup_pivot_root(
3275 directory,
3276 arg_pivot_root_new,
3277 arg_pivot_root_old);
3278 if (r < 0)
3279 return r;
3280
3281 r = setup_volatile_mode(
3282 directory,
3283 arg_volatile_mode,
3284 arg_userns_mode != USER_NAMESPACE_NO,
3285 arg_uid_shift,
3286 arg_uid_range,
3287 arg_selinux_apifs_context);
3288 if (r < 0)
3289 return r;
3290
3291 if (dissected_image) {
3292 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3293 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3294 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
3295 if (r < 0)
3296 return r;
3297 }
3298
3299 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3300 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3301
3302 r = detect_unified_cgroup_hierarchy_from_image(directory);
3303 if (r < 0)
3304 return r;
3305
3306 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3307 if (l < 0)
3308 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3309 if (l != sizeof(arg_unified_cgroup_hierarchy))
3310 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3311 "Short write while sending cgroup mode.");
3312
3313 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3314 }
3315
3316 /* Mark everything as shared so our mounts get propagated down. This is
3317 * required to make new bind mounts available in systemd services
3318 * inside the container that create a new mount namespace.
3319 * See https://github.com/systemd/systemd/issues/3860
3320 * Further submounts (such as /dev) done after this will inherit the
3321 * shared propagation mode. */
3322 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3323 if (r < 0)
3324 return r;
3325
3326 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3327 if (r < 0)
3328 return r;
3329
3330 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3331 if (r < 0)
3332 return r;
3333
3334 if (arg_read_only && arg_volatile_mode == VOLATILE_NO) {
3335 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3336 if (r < 0)
3337 return log_error_errno(r, "Failed to make tree read-only: %m");
3338 }
3339
3340 r = mount_all(directory,
3341 arg_mount_settings,
3342 arg_uid_shift,
3343 arg_selinux_apifs_context);
3344 if (r < 0)
3345 return r;
3346
3347 r = copy_devnodes(directory);
3348 if (r < 0)
3349 return r;
3350
3351 r = make_extra_nodes(directory);
3352 if (r < 0)
3353 return r;
3354
3355 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3356 (void) make_inaccessible_nodes(directory, arg_uid_shift, arg_uid_shift);
3357
3358 r = setup_pts(directory);
3359 if (r < 0)
3360 return r;
3361
3362 r = setup_propagate(directory);
3363 if (r < 0)
3364 return r;
3365
3366 r = setup_keyring();
3367 if (r < 0)
3368 return r;
3369
3370 r = setup_timezone(directory);
3371 if (r < 0)
3372 return r;
3373
3374 r = setup_resolv_conf(directory);
3375 if (r < 0)
3376 return r;
3377
3378 r = setup_machine_id(directory);
3379 if (r < 0)
3380 return r;
3381
3382 r = setup_journal(directory);
3383 if (r < 0)
3384 return r;
3385
3386 r = mount_custom(
3387 directory,
3388 arg_custom_mounts,
3389 arg_n_custom_mounts,
3390 arg_userns_mode != USER_NAMESPACE_NO,
3391 arg_uid_shift,
3392 arg_uid_range,
3393 arg_selinux_apifs_context,
3394 false);
3395 if (r < 0)
3396 return r;
3397
3398 if (!arg_use_cgns) {
3399 r = mount_cgroups(
3400 directory,
3401 arg_unified_cgroup_hierarchy,
3402 arg_userns_mode != USER_NAMESPACE_NO,
3403 arg_uid_shift,
3404 arg_uid_range,
3405 arg_selinux_apifs_context,
3406 false);
3407 if (r < 0)
3408 return r;
3409 }
3410
3411 r = mount_move_root(directory);
3412 if (r < 0)
3413 return log_error_errno(r, "Failed to move root directory: %m");
3414
3415 fd = setup_sd_notify_child();
3416 if (fd < 0)
3417 return fd;
3418
3419 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3420 arg_clone_ns_flags |
3421 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3422 if (pid < 0)
3423 return log_error_errno(errno, "Failed to fork inner child: %m");
3424 if (pid == 0) {
3425 pid_socket = safe_close(pid_socket);
3426 uuid_socket = safe_close(uuid_socket);
3427 notify_socket = safe_close(notify_socket);
3428 uid_shift_socket = safe_close(uid_shift_socket);
3429
3430 /* The inner child has all namespaces that are
3431 * requested, so that we all are owned by the user if
3432 * user namespaces are turned on. */
3433
3434 if (arg_network_namespace_path) {
3435 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3436 if (r < 0)
3437 return log_error_errno(r, "Failed to join network namespace: %m");
3438 }
3439
3440 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds);
3441 if (r < 0)
3442 _exit(EXIT_FAILURE);
3443
3444 _exit(EXIT_SUCCESS);
3445 }
3446
3447 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3448 if (l < 0)
3449 return log_error_errno(errno, "Failed to send PID: %m");
3450 if (l != sizeof(pid))
3451 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3452 "Short write while sending PID.");
3453
3454 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3455 if (l < 0)
3456 return log_error_errno(errno, "Failed to send machine ID: %m");
3457 if (l != sizeof(arg_uuid))
3458 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3459 "Short write while sending machine ID.");
3460
3461 l = send_one_fd(notify_socket, fd, 0);
3462 if (l < 0)
3463 return log_error_errno(l, "Failed to send notify fd: %m");
3464
3465 pid_socket = safe_close(pid_socket);
3466 uuid_socket = safe_close(uuid_socket);
3467 notify_socket = safe_close(notify_socket);
3468 master_pty_socket = safe_close(master_pty_socket);
3469 kmsg_socket = safe_close(kmsg_socket);
3470 rtnl_socket = safe_close(rtnl_socket);
3471 netns_fd = safe_close(netns_fd);
3472
3473 return 0;
3474 }
3475
3476 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
3477 bool tried_hashed = false;
3478 unsigned n_tries = 100;
3479 uid_t candidate;
3480 int r;
3481
3482 assert(shift);
3483 assert(ret_lock_file);
3484 assert(arg_userns_mode == USER_NAMESPACE_PICK);
3485 assert(arg_uid_range == 0x10000U);
3486
3487 candidate = *shift;
3488
3489 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3490
3491 for (;;) {
3492 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
3493 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
3494
3495 if (--n_tries <= 0)
3496 return -EBUSY;
3497
3498 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
3499 goto next;
3500 if ((candidate & UINT32_C(0xFFFF)) != 0)
3501 goto next;
3502
3503 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3504 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3505 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3506 goto next;
3507 if (r < 0)
3508 return r;
3509
3510 /* Make some superficial checks whether the range is currently known in the user database */
3511 if (getpwuid(candidate))
3512 goto next;
3513 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3514 goto next;
3515 if (getgrgid(candidate))
3516 goto next;
3517 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3518 goto next;
3519
3520 *ret_lock_file = lf;
3521 lf = (struct LockFile) LOCK_FILE_INIT;
3522 *shift = candidate;
3523 return 0;
3524
3525 next:
3526 if (arg_machine && !tried_hashed) {
3527 /* Try to hash the base from the container name */
3528
3529 static const uint8_t hash_key[] = {
3530 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3531 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3532 };
3533
3534 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3535
3536 tried_hashed = true;
3537 } else
3538 random_bytes(&candidate, sizeof(candidate));
3539
3540 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
3541 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3542 }
3543 }
3544
3545 static int setup_uid_map(pid_t pid) {
3546 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3547 int r;
3548
3549 assert(pid > 1);
3550
3551 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3552 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3553 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3554 if (r < 0)
3555 return log_error_errno(r, "Failed to write UID map: %m");
3556
3557 /* We always assign the same UID and GID ranges */
3558 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3559 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3560 if (r < 0)
3561 return log_error_errno(r, "Failed to write GID map: %m");
3562
3563 return 0;
3564 }
3565
3566 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3567 char buf[NOTIFY_BUFFER_MAX+1];
3568 char *p = NULL;
3569 struct iovec iovec = {
3570 .iov_base = buf,
3571 .iov_len = sizeof(buf)-1,
3572 };
3573 union {
3574 struct cmsghdr cmsghdr;
3575 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
3576 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
3577 } control = {};
3578 struct msghdr msghdr = {
3579 .msg_iov = &iovec,
3580 .msg_iovlen = 1,
3581 .msg_control = &control,
3582 .msg_controllen = sizeof(control),
3583 };
3584 struct cmsghdr *cmsg;
3585 struct ucred *ucred = NULL;
3586 ssize_t n;
3587 pid_t inner_child_pid;
3588 _cleanup_strv_free_ char **tags = NULL;
3589
3590 assert(userdata);
3591
3592 inner_child_pid = PTR_TO_PID(userdata);
3593
3594 if (revents != EPOLLIN) {
3595 log_warning("Got unexpected poll event for notify fd.");
3596 return 0;
3597 }
3598
3599 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3600 if (n < 0) {
3601 if (IN_SET(errno, EAGAIN, EINTR))
3602 return 0;
3603
3604 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3605 }
3606 cmsg_close_all(&msghdr);
3607
3608 CMSG_FOREACH(cmsg, &msghdr) {
3609 if (cmsg->cmsg_level == SOL_SOCKET &&
3610 cmsg->cmsg_type == SCM_CREDENTIALS &&
3611 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3612
3613 ucred = (struct ucred*) CMSG_DATA(cmsg);
3614 }
3615 }
3616
3617 if (!ucred || ucred->pid != inner_child_pid) {
3618 log_debug("Received notify message without valid credentials. Ignoring.");
3619 return 0;
3620 }
3621
3622 if ((size_t) n >= sizeof(buf)) {
3623 log_warning("Received notify message exceeded maximum size. Ignoring.");
3624 return 0;
3625 }
3626
3627 buf[n] = 0;
3628 tags = strv_split(buf, "\n\r");
3629 if (!tags)
3630 return log_oom();
3631
3632 if (strv_find(tags, "READY=1"))
3633 (void) sd_notifyf(false, "READY=1\n");
3634
3635 p = strv_find_startswith(tags, "STATUS=");
3636 if (p)
3637 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
3638
3639 return 0;
3640 }
3641
3642 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3643 int r;
3644
3645 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3646 if (r < 0)
3647 return log_error_errno(r, "Failed to allocate notify event source: %m");
3648
3649 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3650
3651 return 0;
3652 }
3653
3654 static int merge_settings(Settings *settings, const char *path) {
3655 int rl;
3656
3657 assert(settings);
3658 assert(path);
3659
3660 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3661 * that this steals the fields of the Settings* structure, and hence modifies it. */
3662
3663 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3664 settings->start_mode >= 0) {
3665 arg_start_mode = settings->start_mode;
3666 strv_free_and_replace(arg_parameters, settings->parameters);
3667 }
3668
3669 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
3670 arg_ephemeral = settings->ephemeral;
3671
3672 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
3673 settings->root) {
3674
3675 if (!arg_settings_trusted)
3676 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
3677 else
3678 free_and_replace(arg_directory, settings->root);
3679 }
3680
3681 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3682 settings->pivot_root_new) {
3683 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3684 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3685 }
3686
3687 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3688 settings->working_directory)
3689 free_and_replace(arg_chdir, settings->working_directory);
3690
3691 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3692 settings->environment)
3693 strv_free_and_replace(arg_setenv, settings->environment);
3694
3695 if ((arg_settings_mask & SETTING_USER) == 0) {
3696
3697 if (settings->user)
3698 free_and_replace(arg_user, settings->user);
3699
3700 if (uid_is_valid(settings->uid))
3701 arg_uid = settings->uid;
3702 if (gid_is_valid(settings->gid))
3703 arg_gid = settings->gid;
3704 if (settings->n_supplementary_gids > 0) {
3705 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
3706 arg_n_supplementary_gids = settings->n_supplementary_gids;
3707 }
3708 }
3709
3710 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3711 uint64_t plus, minus;
3712
3713 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3714 * Settings structure */
3715
3716 plus = settings->capability;
3717 minus = settings->drop_capability;
3718
3719 if ((arg_settings_mask & SETTING_NETWORK) == 0) {
3720 if (settings_private_network(settings))
3721 plus |= UINT64_C(1) << CAP_NET_ADMIN;
3722 else
3723 minus |= UINT64_C(1) << CAP_NET_ADMIN;
3724 }
3725
3726 if (!arg_settings_trusted && plus != 0) {
3727 if (settings->capability != 0)
3728 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
3729 } else
3730 arg_caps_retain |= plus;
3731
3732 arg_caps_retain &= ~minus;
3733
3734 /* Copy the full capabilities over too */
3735 if (capability_quintet_is_set(&settings->full_capabilities)) {
3736 if (!arg_settings_trusted)
3737 log_warning("Ignoring capability settings, file %s is not trusted.", path);
3738 else
3739 arg_full_capabilities = settings->full_capabilities;
3740 }
3741 }
3742
3743 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3744 settings->kill_signal > 0)
3745 arg_kill_signal = settings->kill_signal;
3746
3747 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3748 settings->personality != PERSONALITY_INVALID)
3749 arg_personality = settings->personality;
3750
3751 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3752 !sd_id128_is_null(settings->machine_id)) {
3753
3754 if (!arg_settings_trusted)
3755 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
3756 else
3757 arg_uuid = settings->machine_id;
3758 }
3759
3760 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3761 settings->read_only >= 0)
3762 arg_read_only = settings->read_only;
3763
3764 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3765 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3766 arg_volatile_mode = settings->volatile_mode;
3767
3768 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3769 settings->n_custom_mounts > 0) {
3770
3771 if (!arg_settings_trusted)
3772 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
3773 else {
3774 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3775 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3776 arg_n_custom_mounts = settings->n_custom_mounts;
3777 settings->n_custom_mounts = 0;
3778 }
3779 }
3780
3781 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3782 (settings->private_network >= 0 ||
3783 settings->network_veth >= 0 ||
3784 settings->network_bridge ||
3785 settings->network_zone ||
3786 settings->network_interfaces ||
3787 settings->network_macvlan ||
3788 settings->network_ipvlan ||
3789 settings->network_veth_extra ||
3790 settings->network_namespace_path)) {
3791
3792 if (!arg_settings_trusted)
3793 log_warning("Ignoring network settings, file %s is not trusted.", path);
3794 else {
3795 arg_network_veth = settings_network_veth(settings);
3796 arg_private_network = settings_private_network(settings);
3797
3798 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
3799 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
3800 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
3801 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
3802
3803 free_and_replace(arg_network_bridge, settings->network_bridge);
3804 free_and_replace(arg_network_zone, settings->network_zone);
3805
3806 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
3807 }
3808 }
3809
3810 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3811 settings->expose_ports) {
3812
3813 if (!arg_settings_trusted)
3814 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
3815 else {
3816 expose_port_free_all(arg_expose_ports);
3817 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3818 }
3819 }
3820
3821 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3822 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3823
3824 if (!arg_settings_trusted)
3825 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
3826 else {
3827 arg_userns_mode = settings->userns_mode;
3828 arg_uid_shift = settings->uid_shift;
3829 arg_uid_range = settings->uid_range;
3830 arg_userns_chown = settings->userns_chown;
3831 }
3832 }
3833
3834 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3835 arg_notify_ready = settings->notify_ready;
3836
3837 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3838
3839 if (!arg_settings_trusted && !strv_isempty(settings->syscall_whitelist))
3840 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
3841 else {
3842 strv_free_and_replace(arg_syscall_whitelist, settings->syscall_whitelist);
3843 strv_free_and_replace(arg_syscall_blacklist, settings->syscall_blacklist);
3844 }
3845
3846 #if HAVE_SECCOMP
3847 if (!arg_settings_trusted && settings->seccomp)
3848 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
3849 else {
3850 seccomp_release(arg_seccomp);
3851 arg_seccomp = TAKE_PTR(settings->seccomp);
3852 }
3853 #endif
3854 }
3855
3856 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
3857 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
3858 continue;
3859
3860 if (!settings->rlimit[rl])
3861 continue;
3862
3863 if (!arg_settings_trusted) {
3864 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
3865 continue;
3866 }
3867
3868 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
3869 }
3870
3871 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
3872 settings->hostname)
3873 free_and_replace(arg_hostname, settings->hostname);
3874
3875 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
3876 settings->no_new_privileges >= 0)
3877 arg_no_new_privileges = settings->no_new_privileges;
3878
3879 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
3880 settings->oom_score_adjust_set) {
3881
3882 if (!arg_settings_trusted)
3883 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
3884 else {
3885 arg_oom_score_adjust = settings->oom_score_adjust;
3886 arg_oom_score_adjust_set = true;
3887 }
3888 }
3889
3890 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
3891 settings->cpu_set.set) {
3892
3893 if (!arg_settings_trusted)
3894 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
3895 else {
3896 cpu_set_reset(&arg_cpu_set);
3897 arg_cpu_set = settings->cpu_set;
3898 settings->cpu_set = (CPUSet) {};
3899 }
3900 }
3901
3902 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
3903 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
3904 arg_resolv_conf = settings->resolv_conf;
3905
3906 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
3907 settings->link_journal != _LINK_JOURNAL_INVALID) {
3908
3909 if (!arg_settings_trusted)
3910 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
3911 else {
3912 arg_link_journal = settings->link_journal;
3913 arg_link_journal_try = settings->link_journal_try;
3914 }
3915 }
3916
3917 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
3918 settings->timezone != _TIMEZONE_MODE_INVALID)
3919 arg_timezone = settings->timezone;
3920
3921 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
3922 settings->slice) {
3923
3924 if (!arg_settings_trusted)
3925 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
3926 else
3927 free_and_replace(arg_slice, settings->slice);
3928 }
3929
3930 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
3931 settings->use_cgns >= 0) {
3932
3933 if (!arg_settings_trusted)
3934 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
3935 else
3936 arg_use_cgns = settings->use_cgns;
3937 }
3938
3939 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
3940 settings->clone_ns_flags != (unsigned long) -1) {
3941
3942 if (!arg_settings_trusted)
3943 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
3944 else
3945 arg_clone_ns_flags = settings->clone_ns_flags;
3946 }
3947
3948 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
3949 settings->console_mode >= 0) {
3950
3951 if (!arg_settings_trusted)
3952 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
3953 else
3954 arg_console_mode = settings->console_mode;
3955 }
3956
3957 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
3958 * don't consult arg_settings_mask for them. */
3959
3960 sd_bus_message_unref(arg_property_message);
3961 arg_property_message = TAKE_PTR(settings->properties);
3962
3963 arg_console_width = settings->console_width;
3964 arg_console_height = settings->console_height;
3965
3966 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
3967 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
3968 arg_n_extra_nodes = settings->n_extra_nodes;
3969
3970 return 0;
3971 }
3972
3973 static int load_settings(void) {
3974 _cleanup_(settings_freep) Settings *settings = NULL;
3975 _cleanup_fclose_ FILE *f = NULL;
3976 _cleanup_free_ char *p = NULL;
3977 const char *fn, *i;
3978 int r;
3979
3980 if (arg_oci_bundle)
3981 return 0;
3982
3983 /* If all settings are masked, there's no point in looking for
3984 * the settings file */
3985 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3986 return 0;
3987
3988 fn = strjoina(arg_machine, ".nspawn");
3989
3990 /* We first look in the admin's directories in /etc and /run */
3991 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3992 _cleanup_free_ char *j = NULL;
3993
3994 j = path_join(i, fn);
3995 if (!j)
3996 return log_oom();
3997
3998 f = fopen(j, "re");
3999 if (f) {
4000 p = TAKE_PTR(j);
4001
4002 /* By default, we trust configuration from /etc and /run */
4003 if (arg_settings_trusted < 0)
4004 arg_settings_trusted = true;
4005
4006 break;
4007 }
4008
4009 if (errno != ENOENT)
4010 return log_error_errno(errno, "Failed to open %s: %m", j);
4011 }
4012
4013 if (!f) {
4014 /* After that, let's look for a file next to the
4015 * actual image we shall boot. */
4016
4017 if (arg_image) {
4018 p = file_in_same_dir(arg_image, fn);
4019 if (!p)
4020 return log_oom();
4021 } else if (arg_directory) {
4022 p = file_in_same_dir(arg_directory, fn);
4023 if (!p)
4024 return log_oom();
4025 }
4026
4027 if (p) {
4028 f = fopen(p, "re");
4029 if (!f && errno != ENOENT)
4030 return log_error_errno(errno, "Failed to open %s: %m", p);
4031
4032 /* By default, we do not trust configuration from /var/lib/machines */
4033 if (arg_settings_trusted < 0)
4034 arg_settings_trusted = false;
4035 }
4036 }
4037
4038 if (!f)
4039 return 0;
4040
4041 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4042
4043 r = settings_load(f, p, &settings);
4044 if (r < 0)
4045 return r;
4046
4047 return merge_settings(settings, p);
4048 }
4049
4050 static int load_oci_bundle(void) {
4051 _cleanup_(settings_freep) Settings *settings = NULL;
4052 int r;
4053
4054 if (!arg_oci_bundle)
4055 return 0;
4056
4057 /* By default let's trust OCI bundles */
4058 if (arg_settings_trusted < 0)
4059 arg_settings_trusted = true;
4060
4061 r = oci_load(NULL, arg_oci_bundle, &settings);
4062 if (r < 0)
4063 return r;
4064
4065 return merge_settings(settings, arg_oci_bundle);
4066 }
4067
4068 static int run_container(
4069 DissectedImage *dissected_image,
4070 bool secondary,
4071 FDSet *fds,
4072 char veth_name[IFNAMSIZ], bool *veth_created,
4073 union in_addr_union *exposed,
4074 int *master, pid_t *pid, int *ret) {
4075
4076 static const struct sigaction sa = {
4077 .sa_handler = nop_signal_handler,
4078 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4079 };
4080
4081 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4082 _cleanup_close_ int etc_passwd_lock = -1;
4083 _cleanup_close_pair_ int
4084 kmsg_socket_pair[2] = { -1, -1 },
4085 rtnl_socket_pair[2] = { -1, -1 },
4086 pid_socket_pair[2] = { -1, -1 },
4087 uuid_socket_pair[2] = { -1, -1 },
4088 notify_socket_pair[2] = { -1, -1 },
4089 uid_shift_socket_pair[2] = { -1, -1 },
4090 master_pty_socket_pair[2] = { -1, -1 },
4091 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4092
4093 _cleanup_close_ int notify_socket = -1;
4094 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4095 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4096 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4097 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4098 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4099 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4100 ContainerStatus container_status = 0;
4101 int ifi = 0, r;
4102 ssize_t l;
4103 sigset_t mask_chld;
4104 _cleanup_close_ int netns_fd = -1;
4105
4106 assert_se(sigemptyset(&mask_chld) == 0);
4107 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4108
4109 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4110 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4111 * check with getpwuid() if the specific user already exists. Note that /etc might be
4112 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4113 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4114 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4115 * really ours. */
4116
4117 etc_passwd_lock = take_etc_passwd_lock(NULL);
4118 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4119 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4120 }
4121
4122 r = barrier_create(&barrier);
4123 if (r < 0)
4124 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4125
4126 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4127 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4128
4129 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4130 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4131
4132 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4133 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4134
4135 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4136 return log_error_errno(errno, "Failed to create id socket pair: %m");
4137
4138 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4139 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4140
4141 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
4142 return log_error_errno(errno, "Failed to create console socket pair: %m");
4143
4144 if (arg_userns_mode != USER_NAMESPACE_NO)
4145 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4146 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4147
4148 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4149 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4150 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4151
4152 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4153 * parent's blocking calls and give it a chance to call wait() and terminate. */
4154 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4155 if (r < 0)
4156 return log_error_errno(errno, "Failed to change the signal mask: %m");
4157
4158 r = sigaction(SIGCHLD, &sa, NULL);
4159 if (r < 0)
4160 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4161
4162 if (arg_network_namespace_path) {
4163 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4164 if (netns_fd < 0)
4165 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4166
4167 r = fd_is_network_ns(netns_fd);
4168 if (r == -EUCLEAN)
4169 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);
4170 else if (r < 0)
4171 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4172 else if (r == 0)
4173 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4174 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4175 }
4176
4177 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4178 if (*pid < 0)
4179 return log_error_errno(errno, "clone() failed%s: %m",
4180 errno == EINVAL ?
4181 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4182
4183 if (*pid == 0) {
4184 /* The outer child only has a file system namespace. */
4185 barrier_set_role(&barrier, BARRIER_CHILD);
4186
4187 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4188 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4189 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4190 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4191 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
4192 master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
4193 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4194 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
4195
4196 (void) reset_all_signal_handlers();
4197 (void) reset_signal_mask();
4198
4199 r = outer_child(&barrier,
4200 arg_directory,
4201 dissected_image,
4202 secondary,
4203 pid_socket_pair[1],
4204 uuid_socket_pair[1],
4205 notify_socket_pair[1],
4206 kmsg_socket_pair[1],
4207 rtnl_socket_pair[1],
4208 uid_shift_socket_pair[1],
4209 master_pty_socket_pair[1],
4210 unified_cgroup_hierarchy_socket_pair[1],
4211 fds,
4212 netns_fd);
4213 if (r < 0)
4214 _exit(EXIT_FAILURE);
4215
4216 _exit(EXIT_SUCCESS);
4217 }
4218
4219 barrier_set_role(&barrier, BARRIER_PARENT);
4220
4221 fdset_close(fds);
4222
4223 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4224 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4225 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4226 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4227 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
4228 master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
4229 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
4230 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
4231
4232 if (arg_userns_mode != USER_NAMESPACE_NO) {
4233 /* The child just let us know the UID shift it might have read from the image. */
4234 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4235 if (l < 0)
4236 return log_error_errno(errno, "Failed to read UID shift: %m");
4237 if (l != sizeof arg_uid_shift)
4238 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
4239
4240 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4241 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4242 * image, but if that's already in use, pick a new one, and report back to the child,
4243 * which one we now picked. */
4244
4245 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4246 if (r < 0)
4247 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4248
4249 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4250 if (l < 0)
4251 return log_error_errno(errno, "Failed to send UID shift: %m");
4252 if (l != sizeof arg_uid_shift)
4253 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
4254 }
4255 }
4256
4257 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4258 /* The child let us know the support cgroup mode it might have read from the image. */
4259 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4260 if (l < 0)
4261 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4262 if (l != sizeof(arg_unified_cgroup_hierarchy))
4263 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zu bytes).%s",
4264 l, l == 0 ? " The child is most likely dead." : "");
4265 }
4266
4267 /* Wait for the outer child. */
4268 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4269 if (r < 0)
4270 return r;
4271 if (r != EXIT_SUCCESS)
4272 return -EIO;
4273
4274 /* And now retrieve the PID of the inner child. */
4275 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4276 if (l < 0)
4277 return log_error_errno(errno, "Failed to read inner child PID: %m");
4278 if (l != sizeof *pid)
4279 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
4280
4281 /* We also retrieve container UUID in case it was generated by outer child */
4282 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4283 if (l < 0)
4284 return log_error_errno(errno, "Failed to read container machine ID: %m");
4285 if (l != sizeof(arg_uuid))
4286 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
4287
4288 /* We also retrieve the socket used for notifications generated by outer child */
4289 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4290 if (notify_socket < 0)
4291 return log_error_errno(notify_socket,
4292 "Failed to receive notification socket from the outer child: %m");
4293
4294 log_debug("Init process invoked as PID "PID_FMT, *pid);
4295
4296 if (arg_userns_mode != USER_NAMESPACE_NO) {
4297 if (!barrier_place_and_sync(&barrier)) /* #1 */
4298 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4299
4300 r = setup_uid_map(*pid);
4301 if (r < 0)
4302 return r;
4303
4304 (void) barrier_place(&barrier); /* #2 */
4305 }
4306
4307 if (arg_private_network) {
4308 if (!arg_network_namespace_path) {
4309 /* Wait until the child has unshared its network namespace. */
4310 if (!barrier_place_and_sync(&barrier)) /* #3 */
4311 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
4312 }
4313
4314 r = move_network_interfaces(*pid, arg_network_interfaces);
4315 if (r < 0)
4316 return r;
4317
4318 if (arg_network_veth) {
4319 r = setup_veth(arg_machine, *pid, veth_name,
4320 arg_network_bridge || arg_network_zone);
4321 if (r < 0)
4322 return r;
4323 else if (r > 0)
4324 ifi = r;
4325
4326 if (arg_network_bridge) {
4327 /* Add the interface to a bridge */
4328 r = setup_bridge(veth_name, arg_network_bridge, false);
4329 if (r < 0)
4330 return r;
4331 if (r > 0)
4332 ifi = r;
4333 } else if (arg_network_zone) {
4334 /* Add the interface to a bridge, possibly creating it */
4335 r = setup_bridge(veth_name, arg_network_zone, true);
4336 if (r < 0)
4337 return r;
4338 if (r > 0)
4339 ifi = r;
4340 }
4341 }
4342
4343 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
4344 if (r < 0)
4345 return r;
4346
4347 /* We created the primary and extra veth links now; let's remember this, so that we know to
4348 remove them later on. Note that we don't bother with removing veth links that were created
4349 here when their setup failed half-way, because in that case the kernel should be able to
4350 remove them on its own, since they cannot be referenced by anything yet. */
4351 *veth_created = true;
4352
4353 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
4354 if (r < 0)
4355 return r;
4356
4357 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
4358 if (r < 0)
4359 return r;
4360 }
4361
4362 if (arg_register || !arg_keep_unit) {
4363 r = sd_bus_default_system(&bus);
4364 if (r < 0)
4365 return log_error_errno(r, "Failed to open system bus: %m");
4366
4367 r = sd_bus_set_close_on_exit(bus, false);
4368 if (r < 0)
4369 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
4370 }
4371
4372 if (!arg_keep_unit) {
4373 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4374 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4375 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4376
4377 r = sd_bus_match_signal_async(
4378 bus,
4379 NULL,
4380 "org.freedesktop.systemd1",
4381 NULL,
4382 "org.freedesktop.systemd1.Scope",
4383 "RequestStop",
4384 on_request_stop, NULL, PID_TO_PTR(*pid));
4385 if (r < 0)
4386 return log_error_errno(r, "Failed to request RequestStop match: %m");
4387 }
4388
4389 if (arg_register) {
4390 r = register_machine(
4391 bus,
4392 arg_machine,
4393 *pid,
4394 arg_directory,
4395 arg_uuid,
4396 ifi,
4397 arg_slice,
4398 arg_custom_mounts, arg_n_custom_mounts,
4399 arg_kill_signal,
4400 arg_property,
4401 arg_property_message,
4402 arg_keep_unit,
4403 arg_container_service_name);
4404 if (r < 0)
4405 return r;
4406
4407 } else if (!arg_keep_unit) {
4408 r = allocate_scope(
4409 bus,
4410 arg_machine,
4411 *pid,
4412 arg_slice,
4413 arg_custom_mounts, arg_n_custom_mounts,
4414 arg_kill_signal,
4415 arg_property,
4416 arg_property_message);
4417 if (r < 0)
4418 return r;
4419
4420 } else if (arg_slice || arg_property)
4421 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4422
4423 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
4424 if (r < 0)
4425 return r;
4426
4427 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4428 if (r < 0)
4429 return r;
4430
4431 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4432 if (r < 0)
4433 return r;
4434
4435 /* Notify the child that the parent is ready with all
4436 * its setup (including cgroup-ification), and that
4437 * the child can now hand over control to the code to
4438 * run inside the container. */
4439 (void) barrier_place(&barrier); /* #4 */
4440
4441 /* Block SIGCHLD here, before notifying child.
4442 * process_pty() will handle it with the other signals. */
4443 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4444
4445 /* Reset signal to default */
4446 r = default_signals(SIGCHLD, -1);
4447 if (r < 0)
4448 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4449
4450 r = sd_event_new(&event);
4451 if (r < 0)
4452 return log_error_errno(r, "Failed to get default event source: %m");
4453
4454 (void) sd_event_set_watchdog(event, true);
4455
4456 if (bus) {
4457 r = sd_bus_attach_event(bus, event, 0);
4458 if (r < 0)
4459 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4460 }
4461
4462 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
4463 if (r < 0)
4464 return r;
4465
4466 /* Let the child know that we are ready and wait that the child is completely ready now. */
4467 if (!barrier_place_and_sync(&barrier)) /* #5 */
4468 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4469
4470 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4471 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4472 etc_passwd_lock = safe_close(etc_passwd_lock);
4473
4474 (void) sd_notifyf(false,
4475 "STATUS=Container running.\n"
4476 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
4477 if (!arg_notify_ready)
4478 (void) sd_notify(false, "READY=1\n");
4479
4480 if (arg_kill_signal > 0) {
4481 /* Try to kill the init system on SIGINT or SIGTERM */
4482 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4483 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
4484 } else {
4485 /* Immediately exit */
4486 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4487 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4488 }
4489
4490 /* Exit when the child exits */
4491 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
4492
4493 if (arg_expose_ports) {
4494 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
4495 if (r < 0)
4496 return r;
4497
4498 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
4499 }
4500
4501 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4502
4503 if (arg_console_mode != CONSOLE_PIPE) {
4504 _cleanup_close_ int fd = -1;
4505 PTYForwardFlags flags = 0;
4506
4507 /* Retrieve the master pty allocated by inner child */
4508 fd = receive_one_fd(master_pty_socket_pair[0], 0);
4509 if (fd < 0)
4510 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
4511
4512 switch (arg_console_mode) {
4513
4514 case CONSOLE_READ_ONLY:
4515 flags |= PTY_FORWARD_READ_ONLY;
4516
4517 _fallthrough_;
4518
4519 case CONSOLE_INTERACTIVE:
4520 flags |= PTY_FORWARD_IGNORE_VHANGUP;
4521
4522 r = pty_forward_new(event, fd, flags, &forward);
4523 if (r < 0)
4524 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4525
4526 if (arg_console_width != (unsigned) -1 || arg_console_height != (unsigned) -1)
4527 (void) pty_forward_set_width_height(forward,
4528 arg_console_width,
4529 arg_console_height);
4530 break;
4531
4532 default:
4533 assert(arg_console_mode == CONSOLE_PASSIVE);
4534 }
4535
4536 *master = TAKE_FD(fd);
4537 }
4538
4539 r = sd_event_loop(event);
4540 if (r < 0)
4541 return log_error_errno(r, "Failed to run event loop: %m");
4542
4543 if (forward) {
4544 char last_char = 0;
4545
4546 (void) pty_forward_get_last_char(forward, &last_char);
4547 forward = pty_forward_free(forward);
4548
4549 if (!arg_quiet && last_char != '\n')
4550 putc('\n', stdout);
4551 }
4552
4553 /* Kill if it is not dead yet anyway */
4554 if (bus) {
4555 if (arg_register)
4556 terminate_machine(bus, arg_machine);
4557 else if (!arg_keep_unit)
4558 terminate_scope(bus, arg_machine);
4559 }
4560
4561 /* Normally redundant, but better safe than sorry */
4562 (void) kill(*pid, SIGKILL);
4563
4564 r = wait_for_container(*pid, &container_status);
4565 *pid = 0;
4566
4567 if (r < 0)
4568 /* We failed to wait for the container, or the container exited abnormally. */
4569 return r;
4570 if (r > 0 || container_status == CONTAINER_TERMINATED) {
4571 /* r > 0 → The container exited with a non-zero status.
4572 * As a special case, we need to replace 133 with a different value,
4573 * because 133 is special-cased in the service file to reboot the container.
4574 * otherwise → The container exited with zero status and a reboot was not requested.
4575 */
4576 if (r == EXIT_FORCE_RESTART)
4577 r = EXIT_FAILURE; /* replace 133 with the general failure code */
4578 *ret = r;
4579 return 0; /* finito */
4580 }
4581
4582 /* CONTAINER_REBOOTED, loop again */
4583
4584 if (arg_keep_unit) {
4585 /* Special handling if we are running as a service: instead of simply
4586 * restarting the machine we want to restart the entire service, so let's
4587 * inform systemd about this with the special exit code 133. The service
4588 * file uses RestartForceExitStatus=133 so that this results in a full
4589 * nspawn restart. This is necessary since we might have cgroup parameters
4590 * set we want to have flushed out. */
4591 *ret = EXIT_FORCE_RESTART;
4592 return 0; /* finito */
4593 }
4594
4595 expose_port_flush(arg_expose_ports, exposed);
4596
4597 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4598 *veth_created = false;
4599 return 1; /* loop again */
4600 }
4601
4602 static int initialize_rlimits(void) {
4603 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4604 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4605 * container execution environments. */
4606
4607 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
4608 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
4609 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
4610 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
4611 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
4612 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
4613 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
4614 [RLIMIT_MEMLOCK] = { 65536, 65536 },
4615 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
4616 [RLIMIT_NICE] = { 0, 0 },
4617 [RLIMIT_NOFILE] = { 1024, 4096 },
4618 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
4619 [RLIMIT_RTPRIO] = { 0, 0 },
4620 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
4621 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
4622
4623 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4624 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4625 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4626 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4627 * that PID 1 changes a number of other resource limits during early initialization which is why we
4628 * don't read the other limits from PID 1 but prefer the static table above. */
4629 };
4630
4631 int rl;
4632
4633 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
4634 /* Let's only fill in what the user hasn't explicitly configured anyway */
4635 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
4636 const struct rlimit *v;
4637 struct rlimit buffer;
4638
4639 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
4640 /* For these two let's read the limits off PID 1. See above for an explanation. */
4641
4642 if (prlimit(1, rl, NULL, &buffer) < 0)
4643 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
4644
4645 v = &buffer;
4646 } else
4647 v = kernel_defaults + rl;
4648
4649 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
4650 if (!arg_rlimit[rl])
4651 return log_oom();
4652 }
4653
4654 if (DEBUG_LOGGING) {
4655 _cleanup_free_ char *k = NULL;
4656
4657 (void) rlimit_format(arg_rlimit[rl], &k);
4658 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
4659 }
4660 }
4661
4662 return 0;
4663 }
4664
4665 static int run(int argc, char *argv[]) {
4666 bool secondary = false, remove_directory = false, remove_image = false,
4667 veth_created = false, remove_tmprootdir = false;
4668 _cleanup_close_ int master = -1;
4669 _cleanup_fdset_free_ FDSet *fds = NULL;
4670 int r, n_fd_passed, ret = EXIT_SUCCESS;
4671 char veth_name[IFNAMSIZ] = "";
4672 union in_addr_union exposed = {};
4673 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4674 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
4675 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
4676 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4677 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
4678 pid_t pid = 0;
4679
4680 log_parse_environment();
4681 log_open();
4682
4683 r = parse_argv(argc, argv);
4684 if (r <= 0)
4685 goto finish;
4686
4687 r = must_be_root();
4688 if (r < 0)
4689 goto finish;
4690
4691 r = initialize_rlimits();
4692 if (r < 0)
4693 goto finish;
4694
4695 r = load_oci_bundle();
4696 if (r < 0)
4697 goto finish;
4698
4699 r = determine_names();
4700 if (r < 0)
4701 goto finish;
4702
4703 r = load_settings();
4704 if (r < 0)
4705 goto finish;
4706
4707 r = cg_unified_flush();
4708 if (r < 0) {
4709 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4710 goto finish;
4711 }
4712
4713 r = verify_arguments();
4714 if (r < 0)
4715 goto finish;
4716
4717 r = detect_unified_cgroup_hierarchy_from_environment();
4718 if (r < 0)
4719 goto finish;
4720
4721 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4722 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4723 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4724 (void) ignore_signals(SIGPIPE, -1);
4725
4726 n_fd_passed = sd_listen_fds(false);
4727 if (n_fd_passed > 0) {
4728 r = fdset_new_listen_fds(&fds, false);
4729 if (r < 0) {
4730 log_error_errno(r, "Failed to collect file descriptors: %m");
4731 goto finish;
4732 }
4733 }
4734
4735 /* The "default" umask. This is appropriate for most file and directory
4736 * operations performed by nspawn, and is the umask that will be used for
4737 * the child. Functions like copy_devnodes() change the umask temporarily. */
4738 umask(0022);
4739
4740 if (arg_directory) {
4741 assert(!arg_image);
4742
4743 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
4744 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4745 r = -EINVAL;
4746 goto finish;
4747 }
4748
4749 if (arg_ephemeral) {
4750 _cleanup_free_ char *np = NULL;
4751
4752 r = chase_symlinks_and_update(&arg_directory, 0);
4753 if (r < 0)
4754 goto finish;
4755
4756 /* If the specified path is a mount point we generate the new snapshot immediately
4757 * inside it under a random name. However if the specified is not a mount point we
4758 * create the new snapshot in the parent directory, just next to it. */
4759 r = path_is_mount_point(arg_directory, NULL, 0);
4760 if (r < 0) {
4761 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4762 goto finish;
4763 }
4764 if (r > 0)
4765 r = tempfn_random_child(arg_directory, "machine.", &np);
4766 else
4767 r = tempfn_random(arg_directory, "machine.", &np);
4768 if (r < 0) {
4769 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
4770 goto finish;
4771 }
4772
4773 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4774 if (r < 0) {
4775 log_error_errno(r, "Failed to lock %s: %m", np);
4776 goto finish;
4777 }
4778
4779 {
4780 BLOCK_SIGNALS(SIGINT);
4781 r = btrfs_subvol_snapshot(arg_directory, np,
4782 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4783 BTRFS_SNAPSHOT_FALLBACK_COPY |
4784 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4785 BTRFS_SNAPSHOT_RECURSIVE |
4786 BTRFS_SNAPSHOT_QUOTA |
4787 BTRFS_SNAPSHOT_SIGINT);
4788 }
4789 if (r == -EINTR) {
4790 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
4791 goto finish;
4792 }
4793 if (r < 0) {
4794 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4795 goto finish;
4796 }
4797
4798 free_and_replace(arg_directory, np);
4799 remove_directory = true;
4800 } else {
4801 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
4802 if (r < 0)
4803 goto finish;
4804
4805 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4806 if (r == -EBUSY) {
4807 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4808 goto finish;
4809 }
4810 if (r < 0) {
4811 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4812 goto finish;
4813 }
4814
4815 if (arg_template) {
4816 r = chase_symlinks_and_update(&arg_template, 0);
4817 if (r < 0)
4818 goto finish;
4819
4820 {
4821 BLOCK_SIGNALS(SIGINT);
4822 r = btrfs_subvol_snapshot(arg_template, arg_directory,
4823 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4824 BTRFS_SNAPSHOT_FALLBACK_COPY |
4825 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4826 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
4827 BTRFS_SNAPSHOT_RECURSIVE |
4828 BTRFS_SNAPSHOT_QUOTA |
4829 BTRFS_SNAPSHOT_SIGINT);
4830 }
4831 if (r == -EEXIST)
4832 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4833 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4834 else if (r == -EINTR) {
4835 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
4836 goto finish;
4837 } else if (r < 0) {
4838 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4839 goto finish;
4840 } else
4841 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4842 "Populated %s from template %s.", arg_directory, arg_template);
4843 }
4844 }
4845
4846 if (arg_start_mode == START_BOOT) {
4847 const char *p;
4848
4849 if (arg_pivot_root_new)
4850 p = prefix_roota(arg_directory, arg_pivot_root_new);
4851 else
4852 p = arg_directory;
4853
4854 if (path_is_os_tree(p) <= 0) {
4855 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
4856 r = -EINVAL;
4857 goto finish;
4858 }
4859 } else {
4860 const char *p, *q;
4861
4862 if (arg_pivot_root_new)
4863 p = prefix_roota(arg_directory, arg_pivot_root_new);
4864 else
4865 p = arg_directory;
4866
4867 q = strjoina(p, "/usr/");
4868
4869 if (laccess(q, F_OK) < 0) {
4870 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
4871 r = -EINVAL;
4872 goto finish;
4873 }
4874 }
4875
4876 } else {
4877 assert(arg_image);
4878 assert(!arg_template);
4879
4880 r = chase_symlinks_and_update(&arg_image, 0);
4881 if (r < 0)
4882 goto finish;
4883
4884 if (arg_ephemeral) {
4885 _cleanup_free_ char *np = NULL;
4886
4887 r = tempfn_random(arg_image, "machine.", &np);
4888 if (r < 0) {
4889 log_error_errno(r, "Failed to generate name for image snapshot: %m");
4890 goto finish;
4891 }
4892
4893 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4894 if (r < 0) {
4895 r = log_error_errno(r, "Failed to create image lock: %m");
4896 goto finish;
4897 }
4898
4899 {
4900 BLOCK_SIGNALS(SIGINT);
4901 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME|COPY_SIGINT);
4902 }
4903 if (r == -EINTR) {
4904 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
4905 goto finish;
4906 }
4907 if (r < 0) {
4908 r = log_error_errno(r, "Failed to copy image file: %m");
4909 goto finish;
4910 }
4911
4912 free_and_replace(arg_image, np);
4913 remove_image = true;
4914 } else {
4915 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4916 if (r == -EBUSY) {
4917 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4918 goto finish;
4919 }
4920 if (r < 0) {
4921 r = log_error_errno(r, "Failed to create image lock: %m");
4922 goto finish;
4923 }
4924
4925 if (!arg_root_hash) {
4926 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4927 if (r < 0) {
4928 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4929 goto finish;
4930 }
4931 }
4932 }
4933
4934 if (!mkdtemp(tmprootdir)) {
4935 r = log_error_errno(errno, "Failed to create temporary directory: %m");
4936 goto finish;
4937 }
4938
4939 remove_tmprootdir = true;
4940
4941 arg_directory = strdup(tmprootdir);
4942 if (!arg_directory) {
4943 r = log_oom();
4944 goto finish;
4945 }
4946
4947 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4948 if (r < 0) {
4949 log_error_errno(r, "Failed to set up loopback block device: %m");
4950 goto finish;
4951 }
4952
4953 r = dissect_image_and_warn(
4954 loop->fd,
4955 arg_image,
4956 arg_root_hash, arg_root_hash_size,
4957 DISSECT_IMAGE_REQUIRE_ROOT,
4958 &dissected_image);
4959 if (r == -ENOPKG) {
4960 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4961 log_notice("Note that the disk image needs to\n"
4962 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4963 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4964 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4965 " d) or contain a file system without a partition table\n"
4966 "in order to be bootable with systemd-nspawn.");
4967 goto finish;
4968 }
4969 if (r < 0)
4970 goto finish;
4971
4972 if (!arg_root_hash && dissected_image->can_verity)
4973 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4974
4975 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
4976 if (r < 0)
4977 goto finish;
4978
4979 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4980 if (remove_image && unlink(arg_image) >= 0)
4981 remove_image = false;
4982 }
4983
4984 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
4985 if (r < 0)
4986 goto finish;
4987
4988 if (arg_console_mode < 0)
4989 arg_console_mode =
4990 isatty(STDIN_FILENO) > 0 &&
4991 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
4992
4993 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
4994 arg_quiet = true;
4995
4996 if (!arg_quiet)
4997 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4998 arg_machine, arg_image ?: arg_directory);
4999
5000 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
5001
5002 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
5003 r = log_error_errno(errno, "Failed to become subreaper: %m");
5004 goto finish;
5005 }
5006
5007 for (;;) {
5008 r = run_container(dissected_image,
5009 secondary,
5010 fds,
5011 veth_name, &veth_created,
5012 &exposed, &master,
5013 &pid, &ret);
5014 if (r <= 0)
5015 break;
5016 }
5017
5018 finish:
5019 (void) sd_notify(false,
5020 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5021 "STOPPING=1\nSTATUS=Terminating...");
5022
5023 if (pid > 0)
5024 (void) kill(pid, SIGKILL);
5025
5026 /* Try to flush whatever is still queued in the pty */
5027 if (master >= 0) {
5028 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
5029 master = safe_close(master);
5030 }
5031
5032 if (pid > 0)
5033 (void) wait_for_terminate(pid, NULL);
5034
5035 pager_close();
5036
5037 if (remove_directory && arg_directory) {
5038 int k;
5039
5040 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5041 if (k < 0)
5042 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5043 }
5044
5045 if (remove_image && arg_image) {
5046 if (unlink(arg_image) < 0)
5047 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5048 }
5049
5050 if (remove_tmprootdir) {
5051 if (rmdir(tmprootdir) < 0)
5052 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5053 }
5054
5055 if (arg_machine) {
5056 const char *p;
5057
5058 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5059 (void) rm_rf(p, REMOVE_ROOT);
5060 }
5061
5062 expose_port_flush(arg_expose_ports, &exposed);
5063
5064 if (veth_created)
5065 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5066 (void) remove_bridge(arg_network_zone);
5067
5068 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5069 expose_port_free_all(arg_expose_ports);
5070 rlimit_free_all(arg_rlimit);
5071 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
5072
5073 if (r < 0)
5074 return r;
5075
5076 return ret;
5077 }
5078
5079 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.