]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
copy: change the various copy_xyz() calls to take a unified flags parameter
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #ifdef HAVE_BLKID
21 #include <blkid/blkid.h>
22 #endif
23 #include <errno.h>
24 #include <getopt.h>
25 #include <grp.h>
26 #include <linux/loop.h>
27 #include <pwd.h>
28 #include <sched.h>
29 #ifdef HAVE_SELINUX
30 #include <selinux/selinux.h>
31 #endif
32 #include <signal.h>
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <string.h>
36 #include <sys/file.h>
37 #include <sys/mount.h>
38 #include <sys/personality.h>
39 #include <sys/prctl.h>
40 #include <sys/types.h>
41 #include <sys/wait.h>
42 #include <unistd.h>
43
44 #include "sd-daemon.h"
45 #include "sd-id128.h"
46
47 #include "alloc-util.h"
48 #include "barrier.h"
49 #include "base-filesystem.h"
50 #include "blkid-util.h"
51 #include "btrfs-util.h"
52 #include "cap-list.h"
53 #include "capability-util.h"
54 #include "cgroup-util.h"
55 #include "copy.h"
56 #include "dev-setup.h"
57 #include "dissect-image.h"
58 #include "env-util.h"
59 #include "fd-util.h"
60 #include "fdset.h"
61 #include "fileio.h"
62 #include "format-util.h"
63 #include "fs-util.h"
64 #include "gpt.h"
65 #include "hexdecoct.h"
66 #include "hostname-util.h"
67 #include "id128-util.h"
68 #include "log.h"
69 #include "loop-util.h"
70 #include "loopback-setup.h"
71 #include "machine-image.h"
72 #include "macro.h"
73 #include "missing.h"
74 #include "mkdir.h"
75 #include "mount-util.h"
76 #include "netlink-util.h"
77 #include "nspawn-cgroup.h"
78 #include "nspawn-expose-ports.h"
79 #include "nspawn-mount.h"
80 #include "nspawn-network.h"
81 #include "nspawn-patch-uid.h"
82 #include "nspawn-register.h"
83 #include "nspawn-seccomp.h"
84 #include "nspawn-settings.h"
85 #include "nspawn-setuid.h"
86 #include "nspawn-stub-pid1.h"
87 #include "parse-util.h"
88 #include "path-util.h"
89 #include "process-util.h"
90 #include "ptyfwd.h"
91 #include "random-util.h"
92 #include "raw-clone.h"
93 #include "rm-rf.h"
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-util.h"
100 #include "strv.h"
101 #include "terminal-util.h"
102 #include "udev-util.h"
103 #include "umask-util.h"
104 #include "user-util.h"
105 #include "util.h"
106
107 /* Note that devpts's gid= parameter parses GIDs as signed values, hence we stay away from the upper half of the 32bit
108 * UID range here. We leave a bit of room at the lower end and a lot of room at the upper end, so that other subsystems
109 * may have their own allocation ranges too. */
110 #define UID_SHIFT_PICK_MIN ((uid_t) UINT32_C(0x00080000))
111 #define UID_SHIFT_PICK_MAX ((uid_t) UINT32_C(0x6FFF0000))
112
113 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
114 * nspawn_notify_socket_path is relative to the container
115 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
116 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
117
118 #define EXIT_FORCE_RESTART 133
119
120 typedef enum ContainerStatus {
121 CONTAINER_TERMINATED,
122 CONTAINER_REBOOTED
123 } ContainerStatus;
124
125 typedef enum LinkJournal {
126 LINK_NO,
127 LINK_AUTO,
128 LINK_HOST,
129 LINK_GUEST
130 } LinkJournal;
131
132 static char *arg_directory = NULL;
133 static char *arg_template = NULL;
134 static char *arg_chdir = NULL;
135 static char *arg_pivot_root_new = NULL;
136 static char *arg_pivot_root_old = NULL;
137 static char *arg_user = NULL;
138 static sd_id128_t arg_uuid = {};
139 static char *arg_machine = NULL;
140 static const char *arg_selinux_context = NULL;
141 static const char *arg_selinux_apifs_context = NULL;
142 static const 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 CustomMount *arg_custom_mounts = NULL;
177 static unsigned arg_n_custom_mounts = 0;
178 static char **arg_setenv = NULL;
179 static bool arg_quiet = false;
180 static bool arg_register = true;
181 static bool arg_keep_unit = false;
182 static char **arg_network_interfaces = NULL;
183 static char **arg_network_macvlan = NULL;
184 static char **arg_network_ipvlan = NULL;
185 static bool arg_network_veth = false;
186 static char **arg_network_veth_extra = NULL;
187 static char *arg_network_bridge = NULL;
188 static char *arg_network_zone = NULL;
189 static unsigned long arg_personality = PERSONALITY_INVALID;
190 static char *arg_image = NULL;
191 static VolatileMode arg_volatile_mode = VOLATILE_NO;
192 static ExposePort *arg_expose_ports = NULL;
193 static char **arg_property = NULL;
194 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
195 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
196 static bool arg_userns_chown = false;
197 static int arg_kill_signal = 0;
198 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
199 static SettingsMask arg_settings_mask = 0;
200 static int arg_settings_trusted = -1;
201 static char **arg_parameters = NULL;
202 static const char *arg_container_service_name = "systemd-nspawn";
203 static bool arg_notify_ready = false;
204 static bool arg_use_cgns = true;
205 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
206 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO;
207 static void *arg_root_hash = NULL;
208 static size_t arg_root_hash_size = 0;
209
210 static void help(void) {
211 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
212 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
213 " -h --help Show this help\n"
214 " --version Print version string\n"
215 " -q --quiet Do not show status information\n"
216 " -D --directory=PATH Root directory for the container\n"
217 " --template=PATH Initialize root directory from template directory,\n"
218 " if missing\n"
219 " -x --ephemeral Run container with snapshot of root directory, and\n"
220 " remove it after exit\n"
221 " -i --image=PATH File system device or disk image for the container\n"
222 " --root-hash=HASH Specify verity root hash\n"
223 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
224 " -b --boot Boot up full system (i.e. invoke init)\n"
225 " --chdir=PATH Set working directory in the container\n"
226 " --pivot-root=PATH[:PATH]\n"
227 " Pivot root to given directory in the container\n"
228 " -u --user=USER Run the command under specified user or uid\n"
229 " -M --machine=NAME Set the machine name for the container\n"
230 " --uuid=UUID Set a specific machine UUID for the container\n"
231 " -S --slice=SLICE Place the container in the specified slice\n"
232 " --property=NAME=VALUE Set scope unit property\n"
233 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
234 " --private-users[=UIDBASE[:NUIDS]]\n"
235 " Similar, but with user configured UID/GID range\n"
236 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
237 " --private-network Disable network in container\n"
238 " --network-interface=INTERFACE\n"
239 " Assign an existing network interface to the\n"
240 " container\n"
241 " --network-macvlan=INTERFACE\n"
242 " Create a macvlan network interface based on an\n"
243 " existing network interface to the container\n"
244 " --network-ipvlan=INTERFACE\n"
245 " Create a ipvlan network interface based on an\n"
246 " existing network interface to the container\n"
247 " -n --network-veth Add a virtual Ethernet connection between host\n"
248 " and container\n"
249 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
250 " Add an additional virtual Ethernet link between\n"
251 " host and container\n"
252 " --network-bridge=INTERFACE\n"
253 " Add a virtual Ethernet connection to the container\n"
254 " and attach it to an existing bridge on the host\n"
255 " --network-zone=NAME Similar, but attach the new interface to an\n"
256 " an automatically managed bridge interface\n"
257 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
258 " Expose a container IP port on the host\n"
259 " -Z --selinux-context=SECLABEL\n"
260 " Set the SELinux security context to be used by\n"
261 " processes in the container\n"
262 " -L --selinux-apifs-context=SECLABEL\n"
263 " Set the SELinux security context to be used by\n"
264 " API/tmpfs file systems in the container\n"
265 " --capability=CAP In addition to the default, retain specified\n"
266 " capability\n"
267 " --drop-capability=CAP Drop the specified capability from the default set\n"
268 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
269 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
270 " host, try-guest, try-host\n"
271 " -j Equivalent to --link-journal=try-guest\n"
272 " --read-only Mount the root directory read-only\n"
273 " --bind=PATH[:PATH[:OPTIONS]]\n"
274 " Bind mount a file or directory from the host into\n"
275 " the container\n"
276 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
277 " Similar, but creates a read-only bind mount\n"
278 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
279 " --overlay=PATH[:PATH...]:PATH\n"
280 " Create an overlay mount from the host to \n"
281 " the container\n"
282 " --overlay-ro=PATH[:PATH...]:PATH\n"
283 " Similar, but creates a read-only overlay mount\n"
284 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
285 " --register=BOOLEAN Register container as machine\n"
286 " --keep-unit Do not register a scope for the machine, reuse\n"
287 " the service unit nspawn is running in\n"
288 " --volatile[=MODE] Run the system in volatile mode\n"
289 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
290 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
291 , program_invocation_short_name);
292 }
293
294 static int custom_mount_check_all(void) {
295 unsigned i;
296
297 for (i = 0; i < arg_n_custom_mounts; i++) {
298 CustomMount *m = &arg_custom_mounts[i];
299
300 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
301
302 if (arg_userns_chown) {
303 log_error("--private-users-chown may not be combined with custom root mounts.");
304 return -EINVAL;
305 } else if (arg_uid_shift == UID_INVALID) {
306 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
307 return -EINVAL;
308 }
309 }
310 }
311
312 return 0;
313 }
314
315 static int detect_unified_cgroup_hierarchy(const char *directory) {
316 const char *e;
317 int r, all_unified, systemd_unified;
318
319 /* Allow the user to control whether the unified hierarchy is used */
320 e = getenv("UNIFIED_CGROUP_HIERARCHY");
321 if (e) {
322 r = parse_boolean(e);
323 if (r < 0)
324 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
325 if (r > 0)
326 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
327 else
328 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
329
330 return 0;
331 }
332
333 all_unified = cg_all_unified();
334 systemd_unified = cg_unified(SYSTEMD_CGROUP_CONTROLLER);
335
336 if (all_unified < 0 || systemd_unified < 0)
337 return log_error_errno(all_unified < 0 ? all_unified : systemd_unified,
338 "Failed to determine whether the unified cgroups hierarchy is used: %m");
339
340 /* Otherwise inherit the default from the host system */
341 if (all_unified > 0) {
342 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
343 * routine only detects 231, so we'll have a false negative here for 230. */
344 r = systemd_installation_has_version(directory, 230);
345 if (r < 0)
346 return log_error_errno(r, "Failed to determine systemd version in container: %m");
347 if (r > 0)
348 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
349 else
350 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
351 } else if (systemd_unified > 0) {
352 /* Mixed cgroup hierarchy support was added in 232 */
353 r = systemd_installation_has_version(directory, 232);
354 if (r < 0)
355 return log_error_errno(r, "Failed to determine systemd version in container: %m");
356 if (r > 0)
357 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
358 else
359 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
360 } else
361 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
362
363 return 0;
364 }
365
366 static void parse_share_ns_env(const char *name, unsigned long ns_flag) {
367 int r;
368
369 r = getenv_bool(name);
370 if (r == -ENXIO)
371 return;
372 if (r < 0)
373 log_warning_errno(r, "Failed to parse %s from environment, defaulting to false.", name);
374 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
375 }
376
377 static void parse_mount_settings_env(void) {
378 int r;
379 const char *e;
380
381 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
382 if (!e)
383 return;
384
385 if (streq(e, "network")) {
386 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
387 return;
388 }
389
390 r = parse_boolean(e);
391 if (r < 0) {
392 log_warning_errno(r, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
393 return;
394 } else if (r > 0)
395 arg_mount_settings &= ~MOUNT_APPLY_APIVFS_RO;
396 else
397 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO;
398
399 arg_mount_settings &= ~MOUNT_APPLY_APIVFS_NETNS;
400 }
401
402 static int parse_argv(int argc, char *argv[]) {
403
404 enum {
405 ARG_VERSION = 0x100,
406 ARG_PRIVATE_NETWORK,
407 ARG_UUID,
408 ARG_READ_ONLY,
409 ARG_CAPABILITY,
410 ARG_DROP_CAPABILITY,
411 ARG_LINK_JOURNAL,
412 ARG_BIND,
413 ARG_BIND_RO,
414 ARG_TMPFS,
415 ARG_OVERLAY,
416 ARG_OVERLAY_RO,
417 ARG_SHARE_SYSTEM,
418 ARG_REGISTER,
419 ARG_KEEP_UNIT,
420 ARG_NETWORK_INTERFACE,
421 ARG_NETWORK_MACVLAN,
422 ARG_NETWORK_IPVLAN,
423 ARG_NETWORK_BRIDGE,
424 ARG_NETWORK_ZONE,
425 ARG_NETWORK_VETH_EXTRA,
426 ARG_PERSONALITY,
427 ARG_VOLATILE,
428 ARG_TEMPLATE,
429 ARG_PROPERTY,
430 ARG_PRIVATE_USERS,
431 ARG_KILL_SIGNAL,
432 ARG_SETTINGS,
433 ARG_CHDIR,
434 ARG_PIVOT_ROOT,
435 ARG_PRIVATE_USERS_CHOWN,
436 ARG_NOTIFY_READY,
437 ARG_ROOT_HASH,
438 };
439
440 static const struct option options[] = {
441 { "help", no_argument, NULL, 'h' },
442 { "version", no_argument, NULL, ARG_VERSION },
443 { "directory", required_argument, NULL, 'D' },
444 { "template", required_argument, NULL, ARG_TEMPLATE },
445 { "ephemeral", no_argument, NULL, 'x' },
446 { "user", required_argument, NULL, 'u' },
447 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
448 { "as-pid2", no_argument, NULL, 'a' },
449 { "boot", no_argument, NULL, 'b' },
450 { "uuid", required_argument, NULL, ARG_UUID },
451 { "read-only", no_argument, NULL, ARG_READ_ONLY },
452 { "capability", required_argument, NULL, ARG_CAPABILITY },
453 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
454 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
455 { "bind", required_argument, NULL, ARG_BIND },
456 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
457 { "tmpfs", required_argument, NULL, ARG_TMPFS },
458 { "overlay", required_argument, NULL, ARG_OVERLAY },
459 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
460 { "machine", required_argument, NULL, 'M' },
461 { "slice", required_argument, NULL, 'S' },
462 { "setenv", required_argument, NULL, 'E' },
463 { "selinux-context", required_argument, NULL, 'Z' },
464 { "selinux-apifs-context", required_argument, NULL, 'L' },
465 { "quiet", no_argument, NULL, 'q' },
466 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
467 { "register", required_argument, NULL, ARG_REGISTER },
468 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
469 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
470 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
471 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
472 { "network-veth", no_argument, NULL, 'n' },
473 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
474 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
475 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
476 { "personality", required_argument, NULL, ARG_PERSONALITY },
477 { "image", required_argument, NULL, 'i' },
478 { "volatile", optional_argument, NULL, ARG_VOLATILE },
479 { "port", required_argument, NULL, 'p' },
480 { "property", required_argument, NULL, ARG_PROPERTY },
481 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
482 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
483 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
484 { "settings", required_argument, NULL, ARG_SETTINGS },
485 { "chdir", required_argument, NULL, ARG_CHDIR },
486 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
487 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
488 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
489 {}
490 };
491
492 int c, r;
493 const char *p, *e;
494 uint64_t plus = 0, minus = 0;
495 bool mask_all_settings = false, mask_no_settings = false;
496
497 assert(argc >= 0);
498 assert(argv);
499
500 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options, NULL)) >= 0)
501
502 switch (c) {
503
504 case 'h':
505 help();
506 return 0;
507
508 case ARG_VERSION:
509 return version();
510
511 case 'D':
512 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
513 if (r < 0)
514 return r;
515 break;
516
517 case ARG_TEMPLATE:
518 r = parse_path_argument_and_warn(optarg, false, &arg_template);
519 if (r < 0)
520 return r;
521 break;
522
523 case 'i':
524 r = parse_path_argument_and_warn(optarg, false, &arg_image);
525 if (r < 0)
526 return r;
527 break;
528
529 case 'x':
530 arg_ephemeral = true;
531 break;
532
533 case 'u':
534 r = free_and_strdup(&arg_user, optarg);
535 if (r < 0)
536 return log_oom();
537
538 arg_settings_mask |= SETTING_USER;
539 break;
540
541 case ARG_NETWORK_ZONE: {
542 char *j;
543
544 j = strappend("vz-", optarg);
545 if (!j)
546 return log_oom();
547
548 if (!ifname_valid(j)) {
549 log_error("Network zone name not valid: %s", j);
550 free(j);
551 return -EINVAL;
552 }
553
554 free(arg_network_zone);
555 arg_network_zone = j;
556
557 arg_network_veth = true;
558 arg_private_network = true;
559 arg_settings_mask |= SETTING_NETWORK;
560 break;
561 }
562
563 case ARG_NETWORK_BRIDGE:
564
565 if (!ifname_valid(optarg)) {
566 log_error("Bridge interface name not valid: %s", optarg);
567 return -EINVAL;
568 }
569
570 r = free_and_strdup(&arg_network_bridge, optarg);
571 if (r < 0)
572 return log_oom();
573
574 /* fall through */
575
576 case 'n':
577 arg_network_veth = true;
578 arg_private_network = true;
579 arg_settings_mask |= SETTING_NETWORK;
580 break;
581
582 case ARG_NETWORK_VETH_EXTRA:
583 r = veth_extra_parse(&arg_network_veth_extra, optarg);
584 if (r < 0)
585 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
586
587 arg_private_network = true;
588 arg_settings_mask |= SETTING_NETWORK;
589 break;
590
591 case ARG_NETWORK_INTERFACE:
592
593 if (!ifname_valid(optarg)) {
594 log_error("Network interface name not valid: %s", optarg);
595 return -EINVAL;
596 }
597
598 if (strv_extend(&arg_network_interfaces, optarg) < 0)
599 return log_oom();
600
601 arg_private_network = true;
602 arg_settings_mask |= SETTING_NETWORK;
603 break;
604
605 case ARG_NETWORK_MACVLAN:
606
607 if (!ifname_valid(optarg)) {
608 log_error("MACVLAN network interface name not valid: %s", optarg);
609 return -EINVAL;
610 }
611
612 if (strv_extend(&arg_network_macvlan, optarg) < 0)
613 return log_oom();
614
615 arg_private_network = true;
616 arg_settings_mask |= SETTING_NETWORK;
617 break;
618
619 case ARG_NETWORK_IPVLAN:
620
621 if (!ifname_valid(optarg)) {
622 log_error("IPVLAN network interface name not valid: %s", optarg);
623 return -EINVAL;
624 }
625
626 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
627 return log_oom();
628
629 /* fall through */
630
631 case ARG_PRIVATE_NETWORK:
632 arg_private_network = true;
633 arg_settings_mask |= SETTING_NETWORK;
634 break;
635
636 case 'b':
637 if (arg_start_mode == START_PID2) {
638 log_error("--boot and --as-pid2 may not be combined.");
639 return -EINVAL;
640 }
641
642 arg_start_mode = START_BOOT;
643 arg_settings_mask |= SETTING_START_MODE;
644 break;
645
646 case 'a':
647 if (arg_start_mode == START_BOOT) {
648 log_error("--boot and --as-pid2 may not be combined.");
649 return -EINVAL;
650 }
651
652 arg_start_mode = START_PID2;
653 arg_settings_mask |= SETTING_START_MODE;
654 break;
655
656 case ARG_UUID:
657 r = sd_id128_from_string(optarg, &arg_uuid);
658 if (r < 0)
659 return log_error_errno(r, "Invalid UUID: %s", optarg);
660
661 if (sd_id128_is_null(arg_uuid)) {
662 log_error("Machine UUID may not be all zeroes.");
663 return -EINVAL;
664 }
665
666 arg_settings_mask |= SETTING_MACHINE_ID;
667 break;
668
669 case 'S':
670 arg_slice = optarg;
671 break;
672
673 case 'M':
674 if (isempty(optarg))
675 arg_machine = mfree(arg_machine);
676 else {
677 if (!machine_name_is_valid(optarg)) {
678 log_error("Invalid machine name: %s", optarg);
679 return -EINVAL;
680 }
681
682 r = free_and_strdup(&arg_machine, optarg);
683 if (r < 0)
684 return log_oom();
685 }
686 break;
687
688 case 'Z':
689 arg_selinux_context = optarg;
690 break;
691
692 case 'L':
693 arg_selinux_apifs_context = optarg;
694 break;
695
696 case ARG_READ_ONLY:
697 arg_read_only = true;
698 arg_settings_mask |= SETTING_READ_ONLY;
699 break;
700
701 case ARG_CAPABILITY:
702 case ARG_DROP_CAPABILITY: {
703 p = optarg;
704 for (;;) {
705 _cleanup_free_ char *t = NULL;
706
707 r = extract_first_word(&p, &t, ",", 0);
708 if (r < 0)
709 return log_error_errno(r, "Failed to parse capability %s.", t);
710
711 if (r == 0)
712 break;
713
714 if (streq(t, "all")) {
715 if (c == ARG_CAPABILITY)
716 plus = (uint64_t) -1;
717 else
718 minus = (uint64_t) -1;
719 } else {
720 int cap;
721
722 cap = capability_from_name(t);
723 if (cap < 0) {
724 log_error("Failed to parse capability %s.", t);
725 return -EINVAL;
726 }
727
728 if (c == ARG_CAPABILITY)
729 plus |= 1ULL << (uint64_t) cap;
730 else
731 minus |= 1ULL << (uint64_t) cap;
732 }
733 }
734
735 arg_settings_mask |= SETTING_CAPABILITY;
736 break;
737 }
738
739 case 'j':
740 arg_link_journal = LINK_GUEST;
741 arg_link_journal_try = true;
742 break;
743
744 case ARG_LINK_JOURNAL:
745 if (streq(optarg, "auto")) {
746 arg_link_journal = LINK_AUTO;
747 arg_link_journal_try = false;
748 } else if (streq(optarg, "no")) {
749 arg_link_journal = LINK_NO;
750 arg_link_journal_try = false;
751 } else if (streq(optarg, "guest")) {
752 arg_link_journal = LINK_GUEST;
753 arg_link_journal_try = false;
754 } else if (streq(optarg, "host")) {
755 arg_link_journal = LINK_HOST;
756 arg_link_journal_try = false;
757 } else if (streq(optarg, "try-guest")) {
758 arg_link_journal = LINK_GUEST;
759 arg_link_journal_try = true;
760 } else if (streq(optarg, "try-host")) {
761 arg_link_journal = LINK_HOST;
762 arg_link_journal_try = true;
763 } else {
764 log_error("Failed to parse link journal mode %s", optarg);
765 return -EINVAL;
766 }
767
768 break;
769
770 case ARG_BIND:
771 case ARG_BIND_RO:
772 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
773 if (r < 0)
774 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
775
776 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
777 break;
778
779 case ARG_TMPFS:
780 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
781 if (r < 0)
782 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
783
784 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
785 break;
786
787 case ARG_OVERLAY:
788 case ARG_OVERLAY_RO:
789 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
790 if (r == -EADDRNOTAVAIL)
791 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
792 if (r < 0)
793 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
794
795 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
796 break;
797
798 case 'E': {
799 char **n;
800
801 if (!env_assignment_is_valid(optarg)) {
802 log_error("Environment variable assignment '%s' is not valid.", optarg);
803 return -EINVAL;
804 }
805
806 n = strv_env_set(arg_setenv, optarg);
807 if (!n)
808 return log_oom();
809
810 strv_free(arg_setenv);
811 arg_setenv = n;
812
813 arg_settings_mask |= SETTING_ENVIRONMENT;
814 break;
815 }
816
817 case 'q':
818 arg_quiet = true;
819 break;
820
821 case ARG_SHARE_SYSTEM:
822 /* We don't officially support this anymore, except for compat reasons. People should use the
823 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
824 arg_clone_ns_flags = 0;
825 break;
826
827 case ARG_REGISTER:
828 r = parse_boolean(optarg);
829 if (r < 0) {
830 log_error("Failed to parse --register= argument: %s", optarg);
831 return r;
832 }
833
834 arg_register = r;
835 break;
836
837 case ARG_KEEP_UNIT:
838 arg_keep_unit = true;
839 break;
840
841 case ARG_PERSONALITY:
842
843 arg_personality = personality_from_string(optarg);
844 if (arg_personality == PERSONALITY_INVALID) {
845 log_error("Unknown or unsupported personality '%s'.", optarg);
846 return -EINVAL;
847 }
848
849 arg_settings_mask |= SETTING_PERSONALITY;
850 break;
851
852 case ARG_VOLATILE:
853
854 if (!optarg)
855 arg_volatile_mode = VOLATILE_YES;
856 else {
857 VolatileMode m;
858
859 m = volatile_mode_from_string(optarg);
860 if (m < 0) {
861 log_error("Failed to parse --volatile= argument: %s", optarg);
862 return -EINVAL;
863 } else
864 arg_volatile_mode = m;
865 }
866
867 arg_settings_mask |= SETTING_VOLATILE_MODE;
868 break;
869
870 case 'p':
871 r = expose_port_parse(&arg_expose_ports, optarg);
872 if (r == -EEXIST)
873 return log_error_errno(r, "Duplicate port specification: %s", optarg);
874 if (r < 0)
875 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
876
877 arg_settings_mask |= SETTING_EXPOSE_PORTS;
878 break;
879
880 case ARG_PROPERTY:
881 if (strv_extend(&arg_property, optarg) < 0)
882 return log_oom();
883
884 break;
885
886 case ARG_PRIVATE_USERS: {
887 int boolean = -1;
888
889 if (!optarg)
890 boolean = true;
891 else if (!in_charset(optarg, DIGITS))
892 /* do *not* parse numbers as booleans */
893 boolean = parse_boolean(optarg);
894
895 if (boolean == false) {
896 /* no: User namespacing off */
897 arg_userns_mode = USER_NAMESPACE_NO;
898 arg_uid_shift = UID_INVALID;
899 arg_uid_range = UINT32_C(0x10000);
900 } else if (boolean == true) {
901 /* yes: User namespacing on, UID range is read from root dir */
902 arg_userns_mode = USER_NAMESPACE_FIXED;
903 arg_uid_shift = UID_INVALID;
904 arg_uid_range = UINT32_C(0x10000);
905 } else if (streq(optarg, "pick")) {
906 /* pick: User namespacing on, UID range is picked randomly */
907 arg_userns_mode = USER_NAMESPACE_PICK;
908 arg_uid_shift = UID_INVALID;
909 arg_uid_range = UINT32_C(0x10000);
910 } else {
911 _cleanup_free_ char *buffer = NULL;
912 const char *range, *shift;
913
914 /* anything else: User namespacing on, UID range is explicitly configured */
915
916 range = strchr(optarg, ':');
917 if (range) {
918 buffer = strndup(optarg, range - optarg);
919 if (!buffer)
920 return log_oom();
921 shift = buffer;
922
923 range++;
924 r = safe_atou32(range, &arg_uid_range);
925 if (r < 0)
926 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
927 } else
928 shift = optarg;
929
930 r = parse_uid(shift, &arg_uid_shift);
931 if (r < 0)
932 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
933
934 arg_userns_mode = USER_NAMESPACE_FIXED;
935 }
936
937 if (arg_uid_range <= 0) {
938 log_error("UID range cannot be 0.");
939 return -EINVAL;
940 }
941
942 arg_settings_mask |= SETTING_USERNS;
943 break;
944 }
945
946 case 'U':
947 if (userns_supported()) {
948 arg_userns_mode = USER_NAMESPACE_PICK;
949 arg_uid_shift = UID_INVALID;
950 arg_uid_range = UINT32_C(0x10000);
951
952 arg_settings_mask |= SETTING_USERNS;
953 }
954
955 break;
956
957 case ARG_PRIVATE_USERS_CHOWN:
958 arg_userns_chown = true;
959
960 arg_settings_mask |= SETTING_USERNS;
961 break;
962
963 case ARG_KILL_SIGNAL:
964 arg_kill_signal = signal_from_string_try_harder(optarg);
965 if (arg_kill_signal < 0) {
966 log_error("Cannot parse signal: %s", optarg);
967 return -EINVAL;
968 }
969
970 arg_settings_mask |= SETTING_KILL_SIGNAL;
971 break;
972
973 case ARG_SETTINGS:
974
975 /* no → do not read files
976 * yes → read files, do not override cmdline, trust only subset
977 * override → read files, override cmdline, trust only subset
978 * trusted → read files, do not override cmdline, trust all
979 */
980
981 r = parse_boolean(optarg);
982 if (r < 0) {
983 if (streq(optarg, "trusted")) {
984 mask_all_settings = false;
985 mask_no_settings = false;
986 arg_settings_trusted = true;
987
988 } else if (streq(optarg, "override")) {
989 mask_all_settings = false;
990 mask_no_settings = true;
991 arg_settings_trusted = -1;
992 } else
993 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
994 } else if (r > 0) {
995 /* yes */
996 mask_all_settings = false;
997 mask_no_settings = false;
998 arg_settings_trusted = -1;
999 } else {
1000 /* no */
1001 mask_all_settings = true;
1002 mask_no_settings = false;
1003 arg_settings_trusted = false;
1004 }
1005
1006 break;
1007
1008 case ARG_CHDIR:
1009 if (!path_is_absolute(optarg)) {
1010 log_error("Working directory %s is not an absolute path.", optarg);
1011 return -EINVAL;
1012 }
1013
1014 r = free_and_strdup(&arg_chdir, optarg);
1015 if (r < 0)
1016 return log_oom();
1017
1018 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1019 break;
1020
1021 case ARG_PIVOT_ROOT:
1022 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1023 if (r < 0)
1024 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1025
1026 arg_settings_mask |= SETTING_PIVOT_ROOT;
1027 break;
1028
1029 case ARG_NOTIFY_READY:
1030 r = parse_boolean(optarg);
1031 if (r < 0) {
1032 log_error("%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1033 return -EINVAL;
1034 }
1035 arg_notify_ready = r;
1036 arg_settings_mask |= SETTING_NOTIFY_READY;
1037 break;
1038
1039 case ARG_ROOT_HASH: {
1040 void *k;
1041 size_t l;
1042
1043 r = unhexmem(optarg, strlen(optarg), &k, &l);
1044 if (r < 0)
1045 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1046 if (l < sizeof(sd_id128_t)) {
1047 log_error("Root hash must be at least 128bit long: %s", optarg);
1048 free(k);
1049 return -EINVAL;
1050 }
1051
1052 free(arg_root_hash);
1053 arg_root_hash = k;
1054 arg_root_hash_size = l;
1055 break;
1056 }
1057
1058 case '?':
1059 return -EINVAL;
1060
1061 default:
1062 assert_not_reached("Unhandled option");
1063 }
1064
1065 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
1066 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
1067 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
1068 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
1069
1070 if (arg_userns_mode != USER_NAMESPACE_NO)
1071 arg_mount_settings |= MOUNT_USE_USERNS;
1072
1073 if (arg_private_network)
1074 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1075
1076 parse_mount_settings_env();
1077
1078 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1079 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1080 arg_register = false;
1081 if (arg_start_mode != START_PID1) {
1082 log_error("--boot cannot be used without namespacing.");
1083 return -EINVAL;
1084 }
1085 }
1086
1087 if (arg_userns_mode == USER_NAMESPACE_PICK)
1088 arg_userns_chown = true;
1089
1090 if (arg_keep_unit && cg_pid_get_owner_uid(0, NULL) >= 0) {
1091 log_error("--keep-unit may not be used when invoked from a user session.");
1092 return -EINVAL;
1093 }
1094
1095 if (arg_directory && arg_image) {
1096 log_error("--directory= and --image= may not be combined.");
1097 return -EINVAL;
1098 }
1099
1100 if (arg_template && arg_image) {
1101 log_error("--template= and --image= may not be combined.");
1102 return -EINVAL;
1103 }
1104
1105 if (arg_ephemeral && arg_template && !arg_directory) {
1106 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1107 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1108 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1109 * --directory=". */
1110
1111 arg_directory = arg_template;
1112 arg_template = NULL;
1113 }
1114
1115 if (arg_template && !(arg_directory || arg_machine)) {
1116 log_error("--template= needs --directory= or --machine=.");
1117 return -EINVAL;
1118 }
1119
1120 if (arg_ephemeral && arg_template) {
1121 log_error("--ephemeral and --template= may not be combined.");
1122 return -EINVAL;
1123 }
1124
1125 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
1126 log_error("--ephemeral and --link-journal= may not be combined.");
1127 return -EINVAL;
1128 }
1129
1130 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported()) {
1131 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1132 return -EOPNOTSUPP;
1133 }
1134
1135 if (arg_userns_chown && arg_read_only) {
1136 log_error("--read-only and --private-users-chown may not be combined.");
1137 return -EINVAL;
1138 }
1139
1140 if (arg_network_bridge && arg_network_zone) {
1141 log_error("--network-bridge= and --network-zone= may not be combined.");
1142 return -EINVAL;
1143 }
1144
1145 if (argc > optind) {
1146 arg_parameters = strv_copy(argv + optind);
1147 if (!arg_parameters)
1148 return log_oom();
1149
1150 arg_settings_mask |= SETTING_START_MODE;
1151 }
1152
1153 /* Load all settings from .nspawn files */
1154 if (mask_no_settings)
1155 arg_settings_mask = 0;
1156
1157 /* Don't load any settings from .nspawn files */
1158 if (mask_all_settings)
1159 arg_settings_mask = _SETTINGS_MASK_ALL;
1160
1161 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
1162
1163 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1164 if (e)
1165 arg_container_service_name = e;
1166
1167 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1168 if (r < 0)
1169 arg_use_cgns = cg_ns_supported();
1170 else
1171 arg_use_cgns = r;
1172
1173 r = custom_mount_check_all();
1174 if (r < 0)
1175 return r;
1176
1177 return 1;
1178 }
1179
1180 static int verify_arguments(void) {
1181 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network) {
1182 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1183 return -EINVAL;
1184 }
1185
1186 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO)) {
1187 log_error("Cannot combine --private-users with read-write mounts.");
1188 return -EINVAL;
1189 }
1190
1191 if (arg_volatile_mode != VOLATILE_NO && arg_read_only) {
1192 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1193 return -EINVAL;
1194 }
1195
1196 if (arg_expose_ports && !arg_private_network) {
1197 log_error("Cannot use --port= without private networking.");
1198 return -EINVAL;
1199 }
1200
1201 #ifndef HAVE_LIBIPTC
1202 if (arg_expose_ports) {
1203 log_error("--port= is not supported, compiled without libiptc support.");
1204 return -EOPNOTSUPP;
1205 }
1206 #endif
1207
1208 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1209 arg_kill_signal = SIGRTMIN+3;
1210
1211 return 0;
1212 }
1213
1214 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1215 assert(p);
1216
1217 if (arg_userns_mode == USER_NAMESPACE_NO)
1218 return 0;
1219
1220 if (uid == UID_INVALID && gid == GID_INVALID)
1221 return 0;
1222
1223 if (uid != UID_INVALID) {
1224 uid += arg_uid_shift;
1225
1226 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1227 return -EOVERFLOW;
1228 }
1229
1230 if (gid != GID_INVALID) {
1231 gid += (gid_t) arg_uid_shift;
1232
1233 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1234 return -EOVERFLOW;
1235 }
1236
1237 if (lchown(p, uid, gid) < 0)
1238 return -errno;
1239
1240 return 0;
1241 }
1242
1243 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1244 const char *q;
1245
1246 q = prefix_roota(root, path);
1247 if (mkdir(q, mode) < 0) {
1248 if (errno == EEXIST)
1249 return 0;
1250 return -errno;
1251 }
1252
1253 return userns_lchown(q, uid, gid);
1254 }
1255
1256 static int setup_timezone(const char *dest) {
1257 _cleanup_free_ char *p = NULL, *q = NULL;
1258 const char *where, *check, *what;
1259 char *z, *y;
1260 int r;
1261
1262 assert(dest);
1263
1264 /* Fix the timezone, if possible */
1265 r = readlink_malloc("/etc/localtime", &p);
1266 if (r < 0) {
1267 log_warning("host's /etc/localtime is not a symlink, not updating container timezone.");
1268 /* to handle warning, delete /etc/localtime and replace it
1269 * with a symbolic link to a time zone data file.
1270 *
1271 * Example:
1272 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1273 */
1274 return 0;
1275 }
1276
1277 z = path_startswith(p, "../usr/share/zoneinfo/");
1278 if (!z)
1279 z = path_startswith(p, "/usr/share/zoneinfo/");
1280 if (!z) {
1281 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1282 return 0;
1283 }
1284
1285 where = prefix_roota(dest, "/etc/localtime");
1286 r = readlink_malloc(where, &q);
1287 if (r >= 0) {
1288 y = path_startswith(q, "../usr/share/zoneinfo/");
1289 if (!y)
1290 y = path_startswith(q, "/usr/share/zoneinfo/");
1291
1292 /* Already pointing to the right place? Then do nothing .. */
1293 if (y && streq(y, z))
1294 return 0;
1295 }
1296
1297 check = strjoina("/usr/share/zoneinfo/", z);
1298 check = prefix_roota(dest, check);
1299 if (laccess(check, F_OK) < 0) {
1300 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1301 return 0;
1302 }
1303
1304 if (unlink(where) < 0 && errno != ENOENT) {
1305 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1306 errno,
1307 "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1308 return 0;
1309 }
1310
1311 what = strjoina("../usr/share/zoneinfo/", z);
1312 if (symlink(what, where) < 0) {
1313 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1314 errno,
1315 "Failed to correct timezone of container, ignoring: %m");
1316 return 0;
1317 }
1318
1319 r = userns_lchown(where, 0, 0);
1320 if (r < 0)
1321 return log_warning_errno(r, "Failed to chown /etc/localtime: %m");
1322
1323 return 0;
1324 }
1325
1326 static int setup_resolv_conf(const char *dest) {
1327 _cleanup_free_ char *resolved = NULL, *etc = NULL;
1328 const char *where;
1329 int r, found;
1330
1331 assert(dest);
1332
1333 if (arg_private_network)
1334 return 0;
1335
1336 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1337 if (r < 0) {
1338 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1339 return 0;
1340 }
1341
1342 where = strjoina(etc, "/resolv.conf");
1343 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1344 if (found < 0) {
1345 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1346 return 0;
1347 }
1348
1349 if (access("/run/systemd/resolve/resolv.conf", F_OK) >= 0 &&
1350 access("/usr/lib/systemd/resolv.conf", F_OK) >= 0) {
1351
1352 /* resolved is enabled on the host. In this, case bind mount its static resolv.conf file into the
1353 * container, so that the container can use the host's resolver. Given that network namespacing is
1354 * disabled it's only natural of the container also uses the host's resolver. It also has the big
1355 * advantage that the container will be able to follow the host's DNS server configuration changes
1356 * transparently. */
1357
1358 if (found == 0) /* missing? */
1359 (void) touch(resolved);
1360
1361 r = mount_verbose(LOG_DEBUG, "/usr/lib/systemd/resolv.conf", resolved, NULL, MS_BIND, NULL);
1362 if (r >= 0)
1363 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1364 }
1365
1366 /* If that didn't work, let's copy the file */
1367 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0, COPY_REFLINK);
1368 if (r < 0) {
1369 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1370 * resolved or something similar runs inside and the symlink points there.
1371 *
1372 * If the disk image is read-only, there's also no point in complaining.
1373 */
1374 log_full_errno(IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1375 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1376 return 0;
1377 }
1378
1379 r = userns_lchown(where, 0, 0);
1380 if (r < 0)
1381 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1382
1383 return 0;
1384 }
1385
1386 static int setup_boot_id(const char *dest) {
1387 sd_id128_t rnd = SD_ID128_NULL;
1388 const char *from, *to;
1389 int r;
1390
1391 /* Generate a new randomized boot ID, so that each boot-up of
1392 * the container gets a new one */
1393
1394 from = prefix_roota(dest, "/run/proc-sys-kernel-random-boot-id");
1395 to = prefix_roota(dest, "/proc/sys/kernel/random/boot_id");
1396
1397 r = sd_id128_randomize(&rnd);
1398 if (r < 0)
1399 return log_error_errno(r, "Failed to generate random boot id: %m");
1400
1401 r = id128_write(from, ID128_UUID, rnd, false);
1402 if (r < 0)
1403 return log_error_errno(r, "Failed to write boot id: %m");
1404
1405 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1406 if (r >= 0)
1407 r = mount_verbose(LOG_ERR, NULL, to, NULL,
1408 MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1409
1410 (void) unlink(from);
1411 return r;
1412 }
1413
1414 static int copy_devnodes(const char *dest) {
1415
1416 static const char devnodes[] =
1417 "null\0"
1418 "zero\0"
1419 "full\0"
1420 "random\0"
1421 "urandom\0"
1422 "tty\0"
1423 "net/tun\0";
1424
1425 const char *d;
1426 int r = 0;
1427 _cleanup_umask_ mode_t u;
1428
1429 assert(dest);
1430
1431 u = umask(0000);
1432
1433 /* Create /dev/net, so that we can create /dev/net/tun in it */
1434 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1435 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1436
1437 NULSTR_FOREACH(d, devnodes) {
1438 _cleanup_free_ char *from = NULL, *to = NULL;
1439 struct stat st;
1440
1441 from = strappend("/dev/", d);
1442 to = prefix_root(dest, from);
1443
1444 if (stat(from, &st) < 0) {
1445
1446 if (errno != ENOENT)
1447 return log_error_errno(errno, "Failed to stat %s: %m", from);
1448
1449 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1450
1451 log_error("%s is not a char or block device, cannot copy.", from);
1452 return -EIO;
1453
1454 } else {
1455 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1456 /* Explicitly warn the user when /dev is already populated. */
1457 if (errno == EEXIST)
1458 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);
1459 if (errno != EPERM)
1460 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1461
1462 /* Some systems abusively restrict mknod but
1463 * allow bind mounts. */
1464 r = touch(to);
1465 if (r < 0)
1466 return log_error_errno(r, "touch (%s) failed: %m", to);
1467 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1468 if (r < 0)
1469 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1470 }
1471
1472 r = userns_lchown(to, 0, 0);
1473 if (r < 0)
1474 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1475 }
1476 }
1477
1478 return r;
1479 }
1480
1481 static int setup_pts(const char *dest) {
1482 _cleanup_free_ char *options = NULL;
1483 const char *p;
1484 int r;
1485
1486 #ifdef HAVE_SELINUX
1487 if (arg_selinux_apifs_context)
1488 (void) asprintf(&options,
1489 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1490 arg_uid_shift + TTY_GID,
1491 arg_selinux_apifs_context);
1492 else
1493 #endif
1494 (void) asprintf(&options,
1495 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1496 arg_uid_shift + TTY_GID);
1497
1498 if (!options)
1499 return log_oom();
1500
1501 /* Mount /dev/pts itself */
1502 p = prefix_roota(dest, "/dev/pts");
1503 if (mkdir(p, 0755) < 0)
1504 return log_error_errno(errno, "Failed to create /dev/pts: %m");
1505 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
1506 if (r < 0)
1507 return r;
1508 r = userns_lchown(p, 0, 0);
1509 if (r < 0)
1510 return log_error_errno(r, "Failed to chown /dev/pts: %m");
1511
1512 /* Create /dev/ptmx symlink */
1513 p = prefix_roota(dest, "/dev/ptmx");
1514 if (symlink("pts/ptmx", p) < 0)
1515 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1516 r = userns_lchown(p, 0, 0);
1517 if (r < 0)
1518 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
1519
1520 /* And fix /dev/pts/ptmx ownership */
1521 p = prefix_roota(dest, "/dev/pts/ptmx");
1522 r = userns_lchown(p, 0, 0);
1523 if (r < 0)
1524 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
1525
1526 return 0;
1527 }
1528
1529 static int setup_dev_console(const char *dest, const char *console) {
1530 _cleanup_umask_ mode_t u;
1531 const char *to;
1532 int r;
1533
1534 assert(dest);
1535 assert(console);
1536
1537 u = umask(0000);
1538
1539 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
1540 if (r < 0)
1541 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1542
1543 /* We need to bind mount the right tty to /dev/console since
1544 * ptys can only exist on pts file systems. To have something
1545 * to bind mount things on we create a empty regular file. */
1546
1547 to = prefix_roota(dest, "/dev/console");
1548 r = touch(to);
1549 if (r < 0)
1550 return log_error_errno(r, "touch() for /dev/console failed: %m");
1551
1552 return mount_verbose(LOG_ERR, console, to, NULL, MS_BIND, NULL);
1553 }
1554
1555 static int setup_kmsg(const char *dest, int kmsg_socket) {
1556 const char *from, *to;
1557 _cleanup_umask_ mode_t u;
1558 int fd, r;
1559
1560 assert(kmsg_socket >= 0);
1561
1562 u = umask(0000);
1563
1564 /* We create the kmsg FIFO as /run/kmsg, but immediately
1565 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1566 * on the reading side behave very similar to /proc/kmsg,
1567 * their writing side behaves differently from /dev/kmsg in
1568 * that writing blocks when nothing is reading. In order to
1569 * avoid any problems with containers deadlocking due to this
1570 * we simply make /dev/kmsg unavailable to the container. */
1571 from = prefix_roota(dest, "/run/kmsg");
1572 to = prefix_roota(dest, "/proc/kmsg");
1573
1574 if (mkfifo(from, 0600) < 0)
1575 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
1576 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1577 if (r < 0)
1578 return r;
1579
1580 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
1581 if (fd < 0)
1582 return log_error_errno(errno, "Failed to open fifo: %m");
1583
1584 /* Store away the fd in the socket, so that it stays open as
1585 * long as we run the child */
1586 r = send_one_fd(kmsg_socket, fd, 0);
1587 safe_close(fd);
1588
1589 if (r < 0)
1590 return log_error_errno(r, "Failed to send FIFO fd: %m");
1591
1592 /* And now make the FIFO unavailable as /run/kmsg... */
1593 (void) unlink(from);
1594
1595 return 0;
1596 }
1597
1598 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
1599 union in_addr_union *exposed = userdata;
1600
1601 assert(rtnl);
1602 assert(m);
1603 assert(exposed);
1604
1605 expose_port_execute(rtnl, arg_expose_ports, exposed);
1606 return 0;
1607 }
1608
1609 static int setup_hostname(void) {
1610
1611 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
1612 return 0;
1613
1614 if (sethostname_idempotent(arg_machine) < 0)
1615 return -errno;
1616
1617 return 0;
1618 }
1619
1620 static int setup_journal(const char *directory) {
1621 sd_id128_t this_id;
1622 _cleanup_free_ char *d = NULL;
1623 const char *p, *q;
1624 bool try;
1625 char id[33];
1626 int r;
1627
1628 /* Don't link journals in ephemeral mode */
1629 if (arg_ephemeral)
1630 return 0;
1631
1632 if (arg_link_journal == LINK_NO)
1633 return 0;
1634
1635 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
1636
1637 r = sd_id128_get_machine(&this_id);
1638 if (r < 0)
1639 return log_error_errno(r, "Failed to retrieve machine ID: %m");
1640
1641 if (sd_id128_equal(arg_uuid, this_id)) {
1642 log_full(try ? LOG_WARNING : LOG_ERR,
1643 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
1644 if (try)
1645 return 0;
1646 return -EEXIST;
1647 }
1648
1649 r = userns_mkdir(directory, "/var", 0755, 0, 0);
1650 if (r < 0)
1651 return log_error_errno(r, "Failed to create /var: %m");
1652
1653 r = userns_mkdir(directory, "/var/log", 0755, 0, 0);
1654 if (r < 0)
1655 return log_error_errno(r, "Failed to create /var/log: %m");
1656
1657 r = userns_mkdir(directory, "/var/log/journal", 0755, 0, 0);
1658 if (r < 0)
1659 return log_error_errno(r, "Failed to create /var/log/journal: %m");
1660
1661 (void) sd_id128_to_string(arg_uuid, id);
1662
1663 p = strjoina("/var/log/journal/", id);
1664 q = prefix_roota(directory, p);
1665
1666 if (path_is_mount_point(p, NULL, 0) > 0) {
1667 if (try)
1668 return 0;
1669
1670 log_error("%s: already a mount point, refusing to use for journal", p);
1671 return -EEXIST;
1672 }
1673
1674 if (path_is_mount_point(q, NULL, 0) > 0) {
1675 if (try)
1676 return 0;
1677
1678 log_error("%s: already a mount point, refusing to use for journal", q);
1679 return -EEXIST;
1680 }
1681
1682 r = readlink_and_make_absolute(p, &d);
1683 if (r >= 0) {
1684 if ((arg_link_journal == LINK_GUEST ||
1685 arg_link_journal == LINK_AUTO) &&
1686 path_equal(d, q)) {
1687
1688 r = userns_mkdir(directory, p, 0755, 0, 0);
1689 if (r < 0)
1690 log_warning_errno(r, "Failed to create directory %s: %m", q);
1691 return 0;
1692 }
1693
1694 if (unlink(p) < 0)
1695 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
1696 } else if (r == -EINVAL) {
1697
1698 if (arg_link_journal == LINK_GUEST &&
1699 rmdir(p) < 0) {
1700
1701 if (errno == ENOTDIR) {
1702 log_error("%s already exists and is neither a symlink nor a directory", p);
1703 return r;
1704 } else
1705 return log_error_errno(errno, "Failed to remove %s: %m", p);
1706 }
1707 } else if (r != -ENOENT)
1708 return log_error_errno(r, "readlink(%s) failed: %m", p);
1709
1710 if (arg_link_journal == LINK_GUEST) {
1711
1712 if (symlink(q, p) < 0) {
1713 if (try) {
1714 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
1715 return 0;
1716 } else
1717 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
1718 }
1719
1720 r = userns_mkdir(directory, p, 0755, 0, 0);
1721 if (r < 0)
1722 log_warning_errno(r, "Failed to create directory %s: %m", q);
1723 return 0;
1724 }
1725
1726 if (arg_link_journal == LINK_HOST) {
1727 /* don't create parents here — if the host doesn't have
1728 * permanent journal set up, don't force it here */
1729
1730 if (mkdir(p, 0755) < 0 && errno != EEXIST) {
1731 if (try) {
1732 log_debug_errno(errno, "Failed to create %s, skipping journal setup: %m", p);
1733 return 0;
1734 } else
1735 return log_error_errno(errno, "Failed to create %s: %m", p);
1736 }
1737
1738 } else if (access(p, F_OK) < 0)
1739 return 0;
1740
1741 if (dir_is_empty(q) == 0)
1742 log_warning("%s is not empty, proceeding anyway.", q);
1743
1744 r = userns_mkdir(directory, p, 0755, 0, 0);
1745 if (r < 0)
1746 return log_error_errno(r, "Failed to create %s: %m", q);
1747
1748 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
1749 if (r < 0)
1750 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
1751
1752 return 0;
1753 }
1754
1755 static int drop_capabilities(void) {
1756 return capability_bounding_set_drop(arg_caps_retain, false);
1757 }
1758
1759 static int reset_audit_loginuid(void) {
1760 _cleanup_free_ char *p = NULL;
1761 int r;
1762
1763 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
1764 return 0;
1765
1766 r = read_one_line_file("/proc/self/loginuid", &p);
1767 if (r == -ENOENT)
1768 return 0;
1769 if (r < 0)
1770 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
1771
1772 /* Already reset? */
1773 if (streq(p, "4294967295"))
1774 return 0;
1775
1776 r = write_string_file("/proc/self/loginuid", "4294967295", 0);
1777 if (r < 0) {
1778 log_error_errno(r,
1779 "Failed to reset audit login UID. This probably means that your kernel is too\n"
1780 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1781 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1782 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1783 "using systemd-nspawn. Sleeping for 5s... (%m)");
1784
1785 sleep(5);
1786 }
1787
1788 return 0;
1789 }
1790
1791
1792 static int setup_propagate(const char *root) {
1793 const char *p, *q;
1794 int r;
1795
1796 (void) mkdir_p("/run/systemd/nspawn/", 0755);
1797 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
1798 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
1799 (void) mkdir_p(p, 0600);
1800
1801 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
1802 if (r < 0)
1803 return log_error_errno(r, "Failed to create /run/systemd: %m");
1804
1805 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
1806 if (r < 0)
1807 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
1808
1809 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
1810 if (r < 0)
1811 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
1812
1813 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
1814 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
1815 if (r < 0)
1816 return r;
1817
1818 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
1819 if (r < 0)
1820 return r;
1821
1822 /* machined will MS_MOVE into that directory, and that's only
1823 * supported for non-shared mounts. */
1824 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
1825 }
1826
1827 static int setup_machine_id(const char *directory) {
1828 const char *etc_machine_id;
1829 sd_id128_t id;
1830 int r;
1831
1832 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
1833 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
1834 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
1835 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
1836 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
1837 * container behaves nicely). */
1838
1839 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
1840
1841 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
1842 if (r < 0) {
1843 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
1844 return log_error_errno(r, "Failed to read machine ID from container image: %m");
1845
1846 if (sd_id128_is_null(arg_uuid)) {
1847 r = sd_id128_randomize(&arg_uuid);
1848 if (r < 0)
1849 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
1850 }
1851 } else {
1852 if (sd_id128_is_null(id)) {
1853 log_error("Machine ID in container image is zero, refusing.");
1854 return -EINVAL;
1855 }
1856
1857 arg_uuid = id;
1858 }
1859
1860 return 0;
1861 }
1862
1863 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
1864 int r;
1865
1866 assert(directory);
1867
1868 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
1869 return 0;
1870
1871 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
1872 if (r == -EOPNOTSUPP)
1873 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.");
1874 if (r == -EBADE)
1875 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
1876 if (r < 0)
1877 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
1878 if (r == 0)
1879 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
1880 else
1881 log_debug("Patched directory tree to match UID/GID range.");
1882
1883 return r;
1884 }
1885
1886 /*
1887 * Return values:
1888 * < 0 : wait_for_terminate() failed to get the state of the
1889 * container, the container was terminated by a signal, or
1890 * failed for an unknown reason. No change is made to the
1891 * container argument.
1892 * > 0 : The program executed in the container terminated with an
1893 * error. The exit code of the program executed in the
1894 * container is returned. The container argument has been set
1895 * to CONTAINER_TERMINATED.
1896 * 0 : The container is being rebooted, has been shut down or exited
1897 * successfully. The container argument has been set to either
1898 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
1899 *
1900 * That is, success is indicated by a return value of zero, and an
1901 * error is indicated by a non-zero value.
1902 */
1903 static int wait_for_container(pid_t pid, ContainerStatus *container) {
1904 siginfo_t status;
1905 int r;
1906
1907 r = wait_for_terminate(pid, &status);
1908 if (r < 0)
1909 return log_warning_errno(r, "Failed to wait for container: %m");
1910
1911 switch (status.si_code) {
1912
1913 case CLD_EXITED:
1914 if (status.si_status == 0)
1915 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
1916 else
1917 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
1918
1919 *container = CONTAINER_TERMINATED;
1920 return status.si_status;
1921
1922 case CLD_KILLED:
1923 if (status.si_status == SIGINT) {
1924 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
1925 *container = CONTAINER_TERMINATED;
1926 return 0;
1927
1928 } else if (status.si_status == SIGHUP) {
1929 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
1930 *container = CONTAINER_REBOOTED;
1931 return 0;
1932 }
1933
1934 /* fall through */
1935
1936 case CLD_DUMPED:
1937 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
1938 return -EIO;
1939
1940 default:
1941 log_error("Container %s failed due to unknown reason.", arg_machine);
1942 return -EIO;
1943 }
1944 }
1945
1946 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1947 pid_t pid;
1948
1949 pid = PTR_TO_PID(userdata);
1950 if (pid > 0) {
1951 if (kill(pid, arg_kill_signal) >= 0) {
1952 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
1953 sd_event_source_set_userdata(s, NULL);
1954 return 0;
1955 }
1956 }
1957
1958 sd_event_exit(sd_event_source_get_event(s), 0);
1959 return 0;
1960 }
1961
1962 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
1963 for (;;) {
1964 siginfo_t si = {};
1965 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
1966 return log_error_errno(errno, "Failed to waitid(): %m");
1967 if (si.si_pid == 0) /* No pending children. */
1968 break;
1969 if (si.si_pid == PTR_TO_PID(userdata)) {
1970 /* The main process we care for has exited. Return from
1971 * signal handler but leave the zombie. */
1972 sd_event_exit(sd_event_source_get_event(s), 0);
1973 break;
1974 }
1975 /* Reap all other children. */
1976 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
1977 }
1978
1979 return 0;
1980 }
1981
1982 static int determine_names(void) {
1983 int r;
1984
1985 if (arg_template && !arg_directory && arg_machine) {
1986
1987 /* If --template= was specified then we should not
1988 * search for a machine, but instead create a new one
1989 * in /var/lib/machine. */
1990
1991 arg_directory = strjoin("/var/lib/machines/", arg_machine);
1992 if (!arg_directory)
1993 return log_oom();
1994 }
1995
1996 if (!arg_image && !arg_directory) {
1997 if (arg_machine) {
1998 _cleanup_(image_unrefp) Image *i = NULL;
1999
2000 r = image_find(arg_machine, &i);
2001 if (r < 0)
2002 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2003 if (r == 0) {
2004 log_error("No image for machine '%s': %m", arg_machine);
2005 return -ENOENT;
2006 }
2007
2008 if (i->type == IMAGE_RAW)
2009 r = free_and_strdup(&arg_image, i->path);
2010 else
2011 r = free_and_strdup(&arg_directory, i->path);
2012 if (r < 0)
2013 return log_oom();
2014
2015 if (!arg_ephemeral)
2016 arg_read_only = arg_read_only || i->read_only;
2017 } else
2018 arg_directory = get_current_dir_name();
2019
2020 if (!arg_directory && !arg_image) {
2021 log_error("Failed to determine path, please use -D or -i.");
2022 return -EINVAL;
2023 }
2024 }
2025
2026 if (!arg_machine) {
2027
2028 if (arg_directory && path_equal(arg_directory, "/"))
2029 arg_machine = gethostname_malloc();
2030 else {
2031 if (arg_image) {
2032 char *e;
2033
2034 arg_machine = strdup(basename(arg_image));
2035
2036 /* Truncate suffix if there is one */
2037 e = endswith(arg_machine, ".raw");
2038 if (e)
2039 *e = 0;
2040 } else
2041 arg_machine = strdup(basename(arg_directory));
2042 }
2043 if (!arg_machine)
2044 return log_oom();
2045
2046 hostname_cleanup(arg_machine);
2047 if (!machine_name_is_valid(arg_machine)) {
2048 log_error("Failed to determine machine name automatically, please use -M.");
2049 return -EINVAL;
2050 }
2051
2052 if (arg_ephemeral) {
2053 char *b;
2054
2055 /* Add a random suffix when this is an
2056 * ephemeral machine, so that we can run many
2057 * instances at once without manually having
2058 * to specify -M each time. */
2059
2060 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2061 return log_oom();
2062
2063 free(arg_machine);
2064 arg_machine = b;
2065 }
2066 }
2067
2068 return 0;
2069 }
2070
2071 static int chase_symlinks_and_update(char **p, unsigned flags) {
2072 char *chased;
2073 int r;
2074
2075 assert(p);
2076
2077 if (!*p)
2078 return 0;
2079
2080 r = chase_symlinks(*p, NULL, flags, &chased);
2081 if (r < 0)
2082 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2083
2084 free(*p);
2085 *p = chased;
2086
2087 return 0;
2088 }
2089
2090 static int determine_uid_shift(const char *directory) {
2091 int r;
2092
2093 if (arg_userns_mode == USER_NAMESPACE_NO) {
2094 arg_uid_shift = 0;
2095 return 0;
2096 }
2097
2098 if (arg_uid_shift == UID_INVALID) {
2099 struct stat st;
2100
2101 r = stat(directory, &st);
2102 if (r < 0)
2103 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2104
2105 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2106
2107 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
2108 log_error("UID and GID base of %s don't match.", directory);
2109 return -EINVAL;
2110 }
2111
2112 arg_uid_range = UINT32_C(0x10000);
2113 }
2114
2115 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
2116 log_error("UID base too high for UID range.");
2117 return -EINVAL;
2118 }
2119
2120 return 0;
2121 }
2122
2123 static int inner_child(
2124 Barrier *barrier,
2125 const char *directory,
2126 bool secondary,
2127 int kmsg_socket,
2128 int rtnl_socket,
2129 FDSet *fds) {
2130
2131 _cleanup_free_ char *home = NULL;
2132 char as_uuid[37];
2133 unsigned n_env = 1;
2134 const char *envp[] = {
2135 "PATH=" DEFAULT_PATH_SPLIT_USR,
2136 NULL, /* container */
2137 NULL, /* TERM */
2138 NULL, /* HOME */
2139 NULL, /* USER */
2140 NULL, /* LOGNAME */
2141 NULL, /* container_uuid */
2142 NULL, /* LISTEN_FDS */
2143 NULL, /* LISTEN_PID */
2144 NULL, /* NOTIFY_SOCKET */
2145 NULL
2146 };
2147 const char *exec_target;
2148
2149 _cleanup_strv_free_ char **env_use = NULL;
2150 int r;
2151
2152 assert(barrier);
2153 assert(directory);
2154 assert(kmsg_socket >= 0);
2155
2156 cg_unified_flush();
2157
2158 if (arg_userns_mode != USER_NAMESPACE_NO) {
2159 /* Tell the parent, that it now can write the UID map. */
2160 (void) barrier_place(barrier); /* #1 */
2161
2162 /* Wait until the parent wrote the UID map */
2163 if (!barrier_place_and_sync(barrier)) { /* #2 */
2164 log_error("Parent died too early");
2165 return -ESRCH;
2166 }
2167 }
2168
2169 r = reset_uid_gid();
2170 if (r < 0)
2171 return log_error_errno(r, "Couldn't become new root: %m");
2172
2173 r = mount_all(NULL,
2174 arg_mount_settings | MOUNT_IN_USERNS,
2175 arg_uid_shift,
2176 arg_uid_range,
2177 arg_selinux_apifs_context);
2178
2179 if (r < 0)
2180 return r;
2181
2182 r = mount_sysfs(NULL, arg_mount_settings);
2183 if (r < 0)
2184 return r;
2185
2186 /* Wait until we are cgroup-ified, so that we
2187 * can mount the right cgroup path writable */
2188 if (!barrier_place_and_sync(barrier)) { /* #3 */
2189 log_error("Parent died too early");
2190 return -ESRCH;
2191 }
2192
2193 if (arg_use_cgns && cg_ns_supported()) {
2194 r = unshare(CLONE_NEWCGROUP);
2195 if (r < 0)
2196 return log_error_errno(errno, "Failed to unshare cgroup namespace");
2197 r = mount_cgroups(
2198 "",
2199 arg_unified_cgroup_hierarchy,
2200 arg_userns_mode != USER_NAMESPACE_NO,
2201 arg_uid_shift,
2202 arg_uid_range,
2203 arg_selinux_apifs_context,
2204 true);
2205 if (r < 0)
2206 return r;
2207 } else {
2208 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2209 if (r < 0)
2210 return r;
2211 }
2212
2213 r = setup_boot_id(NULL);
2214 if (r < 0)
2215 return r;
2216
2217 r = setup_kmsg(NULL, kmsg_socket);
2218 if (r < 0)
2219 return r;
2220 kmsg_socket = safe_close(kmsg_socket);
2221
2222 umask(0022);
2223
2224 if (setsid() < 0)
2225 return log_error_errno(errno, "setsid() failed: %m");
2226
2227 if (arg_private_network)
2228 loopback_setup();
2229
2230 if (arg_expose_ports) {
2231 r = expose_port_send_rtnl(rtnl_socket);
2232 if (r < 0)
2233 return r;
2234 rtnl_socket = safe_close(rtnl_socket);
2235 }
2236
2237 r = drop_capabilities();
2238 if (r < 0)
2239 return log_error_errno(r, "drop_capabilities() failed: %m");
2240
2241 setup_hostname();
2242
2243 if (arg_personality != PERSONALITY_INVALID) {
2244 if (personality(arg_personality) < 0)
2245 return log_error_errno(errno, "personality() failed: %m");
2246 } else if (secondary) {
2247 if (personality(PER_LINUX32) < 0)
2248 return log_error_errno(errno, "personality() failed: %m");
2249 }
2250
2251 #ifdef HAVE_SELINUX
2252 if (arg_selinux_context)
2253 if (setexeccon(arg_selinux_context) < 0)
2254 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
2255 #endif
2256
2257 r = change_uid_gid(arg_user, &home);
2258 if (r < 0)
2259 return r;
2260
2261 /* LXC sets container=lxc, so follow the scheme here */
2262 envp[n_env++] = strjoina("container=", arg_container_service_name);
2263
2264 envp[n_env] = strv_find_prefix(environ, "TERM=");
2265 if (envp[n_env])
2266 n_env++;
2267
2268 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
2269 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
2270 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
2271 return log_oom();
2272
2273 assert(!sd_id128_is_null(arg_uuid));
2274
2275 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
2276 return log_oom();
2277
2278 if (fdset_size(fds) > 0) {
2279 r = fdset_cloexec(fds, false);
2280 if (r < 0)
2281 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
2282
2283 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
2284 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
2285 return log_oom();
2286 }
2287 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
2288 return log_oom();
2289
2290 env_use = strv_env_merge(2, envp, arg_setenv);
2291 if (!env_use)
2292 return log_oom();
2293
2294 /* Let the parent know that we are ready and
2295 * wait until the parent is ready with the
2296 * setup, too... */
2297 if (!barrier_place_and_sync(barrier)) { /* #4 */
2298 log_error("Parent died too early");
2299 return -ESRCH;
2300 }
2301
2302 if (arg_chdir)
2303 if (chdir(arg_chdir) < 0)
2304 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
2305
2306 if (arg_start_mode == START_PID2) {
2307 r = stub_pid1(arg_uuid);
2308 if (r < 0)
2309 return r;
2310 }
2311
2312 /* Now, explicitly close the log, so that we
2313 * then can close all remaining fds. Closing
2314 * the log explicitly first has the benefit
2315 * that the logging subsystem knows about it,
2316 * and is thus ready to be reopened should we
2317 * need it again. Note that the other fds
2318 * closed here are at least the locking and
2319 * barrier fds. */
2320 log_close();
2321 (void) fdset_close_others(fds);
2322
2323 if (arg_start_mode == START_BOOT) {
2324 char **a;
2325 size_t m;
2326
2327 /* Automatically search for the init system */
2328
2329 m = strv_length(arg_parameters);
2330 a = newa(char*, m + 2);
2331 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
2332 a[1 + m] = NULL;
2333
2334 a[0] = (char*) "/usr/lib/systemd/systemd";
2335 execve(a[0], a, env_use);
2336
2337 a[0] = (char*) "/lib/systemd/systemd";
2338 execve(a[0], a, env_use);
2339
2340 a[0] = (char*) "/sbin/init";
2341 execve(a[0], a, env_use);
2342
2343 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2344 } else if (!strv_isempty(arg_parameters)) {
2345 exec_target = arg_parameters[0];
2346 execvpe(arg_parameters[0], arg_parameters, env_use);
2347 } else {
2348 if (!arg_chdir)
2349 /* If we cannot change the directory, we'll end up in /, that is expected. */
2350 (void) chdir(home ?: "/root");
2351
2352 execle("/bin/bash", "-bash", NULL, env_use);
2353 execle("/bin/sh", "-sh", NULL, env_use);
2354
2355 exec_target = "/bin/bash, /bin/sh";
2356 }
2357
2358 r = -errno;
2359 (void) log_open();
2360 return log_error_errno(r, "execv(%s) failed: %m", exec_target);
2361 }
2362
2363 static int setup_sd_notify_child(void) {
2364 static const int one = 1;
2365 int fd = -1;
2366 union sockaddr_union sa = {
2367 .sa.sa_family = AF_UNIX,
2368 };
2369 int r;
2370
2371 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2372 if (fd < 0)
2373 return log_error_errno(errno, "Failed to allocate notification socket: %m");
2374
2375 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
2376 (void) unlink(NSPAWN_NOTIFY_SOCKET_PATH);
2377
2378 strncpy(sa.un.sun_path, NSPAWN_NOTIFY_SOCKET_PATH, sizeof(sa.un.sun_path)-1);
2379 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
2380 if (r < 0) {
2381 safe_close(fd);
2382 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
2383 }
2384
2385 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
2386 if (r < 0) {
2387 safe_close(fd);
2388 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
2389 }
2390
2391 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
2392 if (r < 0) {
2393 safe_close(fd);
2394 return log_error_errno(errno, "SO_PASSCRED failed: %m");
2395 }
2396
2397 return fd;
2398 }
2399
2400 static int outer_child(
2401 Barrier *barrier,
2402 const char *directory,
2403 const char *console,
2404 DissectedImage *dissected_image,
2405 bool interactive,
2406 bool secondary,
2407 int pid_socket,
2408 int uuid_socket,
2409 int notify_socket,
2410 int kmsg_socket,
2411 int rtnl_socket,
2412 int uid_shift_socket,
2413 FDSet *fds) {
2414
2415 pid_t pid;
2416 ssize_t l;
2417 int r;
2418 _cleanup_close_ int fd = -1;
2419
2420 assert(barrier);
2421 assert(directory);
2422 assert(console);
2423 assert(pid_socket >= 0);
2424 assert(uuid_socket >= 0);
2425 assert(notify_socket >= 0);
2426 assert(kmsg_socket >= 0);
2427
2428 cg_unified_flush();
2429
2430 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
2431 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
2432
2433 if (interactive) {
2434 close_nointr(STDIN_FILENO);
2435 close_nointr(STDOUT_FILENO);
2436 close_nointr(STDERR_FILENO);
2437
2438 r = open_terminal(console, O_RDWR);
2439 if (r != STDIN_FILENO) {
2440 if (r >= 0) {
2441 safe_close(r);
2442 r = -EINVAL;
2443 }
2444
2445 return log_error_errno(r, "Failed to open console: %m");
2446 }
2447
2448 if (dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
2449 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
2450 return log_error_errno(errno, "Failed to duplicate console: %m");
2451 }
2452
2453 r = reset_audit_loginuid();
2454 if (r < 0)
2455 return r;
2456
2457 /* Mark everything as slave, so that we still
2458 * receive mounts from the real root, but don't
2459 * propagate mounts to the real root. */
2460 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
2461 if (r < 0)
2462 return r;
2463
2464 if (dissected_image) {
2465 r = dissected_image_mount(dissected_image, directory, DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2466 if (r < 0)
2467 return r;
2468 }
2469
2470 r = determine_uid_shift(directory);
2471 if (r < 0)
2472 return r;
2473
2474 r = detect_unified_cgroup_hierarchy(directory);
2475 if (r < 0)
2476 return r;
2477
2478 if (arg_userns_mode != USER_NAMESPACE_NO) {
2479 /* Let the parent know which UID shift we read from the image */
2480 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
2481 if (l < 0)
2482 return log_error_errno(errno, "Failed to send UID shift: %m");
2483 if (l != sizeof(arg_uid_shift)) {
2484 log_error("Short write while sending UID shift.");
2485 return -EIO;
2486 }
2487
2488 if (arg_userns_mode == USER_NAMESPACE_PICK) {
2489 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2490 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2491 * not it will pick a different one, and send it back to us. */
2492
2493 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
2494 if (l < 0)
2495 return log_error_errno(errno, "Failed to recv UID shift: %m");
2496 if (l != sizeof(arg_uid_shift)) {
2497 log_error("Short read while receiving UID shift.");
2498 return -EIO;
2499 }
2500 }
2501
2502 log_info("Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
2503 }
2504
2505 /* Turn directory into bind mount */
2506 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
2507 if (r < 0)
2508 return r;
2509
2510 r = setup_pivot_root(
2511 directory,
2512 arg_pivot_root_new,
2513 arg_pivot_root_old);
2514 if (r < 0)
2515 return r;
2516
2517 r = setup_volatile(
2518 directory,
2519 arg_volatile_mode,
2520 arg_userns_mode != USER_NAMESPACE_NO,
2521 arg_uid_shift,
2522 arg_uid_range,
2523 arg_selinux_context);
2524 if (r < 0)
2525 return r;
2526
2527 r = setup_volatile_state(
2528 directory,
2529 arg_volatile_mode,
2530 arg_userns_mode != USER_NAMESPACE_NO,
2531 arg_uid_shift,
2532 arg_uid_range,
2533 arg_selinux_context);
2534 if (r < 0)
2535 return r;
2536
2537 /* Mark everything as shared so our mounts get propagated down. This is
2538 * required to make new bind mounts available in systemd services
2539 * inside the containter that create a new mount namespace.
2540 * See https://github.com/systemd/systemd/issues/3860
2541 * Further submounts (such as /dev) done after this will inherit the
2542 * shared propagation mode.*/
2543 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
2544 if (r < 0)
2545 return r;
2546
2547 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
2548 if (r < 0)
2549 return r;
2550
2551 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
2552 if (r < 0)
2553 return r;
2554
2555 if (arg_read_only) {
2556 r = bind_remount_recursive(directory, true, NULL);
2557 if (r < 0)
2558 return log_error_errno(r, "Failed to make tree read-only: %m");
2559 }
2560
2561 r = mount_all(directory,
2562 arg_mount_settings,
2563 arg_uid_shift,
2564 arg_uid_range,
2565 arg_selinux_apifs_context);
2566 if (r < 0)
2567 return r;
2568
2569 r = copy_devnodes(directory);
2570 if (r < 0)
2571 return r;
2572
2573 dev_setup(directory, arg_uid_shift, arg_uid_shift);
2574
2575 r = setup_pts(directory);
2576 if (r < 0)
2577 return r;
2578
2579 r = setup_propagate(directory);
2580 if (r < 0)
2581 return r;
2582
2583 r = setup_dev_console(directory, console);
2584 if (r < 0)
2585 return r;
2586
2587 r = setup_seccomp(arg_caps_retain);
2588 if (r < 0)
2589 return r;
2590
2591 r = setup_timezone(directory);
2592 if (r < 0)
2593 return r;
2594
2595 r = setup_resolv_conf(directory);
2596 if (r < 0)
2597 return r;
2598
2599 r = setup_machine_id(directory);
2600 if (r < 0)
2601 return r;
2602
2603 r = setup_journal(directory);
2604 if (r < 0)
2605 return r;
2606
2607 r = mount_custom(
2608 directory,
2609 arg_custom_mounts,
2610 arg_n_custom_mounts,
2611 arg_userns_mode != USER_NAMESPACE_NO,
2612 arg_uid_shift,
2613 arg_uid_range,
2614 arg_selinux_apifs_context);
2615 if (r < 0)
2616 return r;
2617
2618 if (!arg_use_cgns || !cg_ns_supported()) {
2619 r = mount_cgroups(
2620 directory,
2621 arg_unified_cgroup_hierarchy,
2622 arg_userns_mode != USER_NAMESPACE_NO,
2623 arg_uid_shift,
2624 arg_uid_range,
2625 arg_selinux_apifs_context,
2626 false);
2627 if (r < 0)
2628 return r;
2629 }
2630
2631 r = mount_move_root(directory);
2632 if (r < 0)
2633 return log_error_errno(r, "Failed to move root directory: %m");
2634
2635 fd = setup_sd_notify_child();
2636 if (fd < 0)
2637 return fd;
2638
2639 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
2640 arg_clone_ns_flags |
2641 (arg_private_network ? CLONE_NEWNET : 0) |
2642 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
2643 if (pid < 0)
2644 return log_error_errno(errno, "Failed to fork inner child: %m");
2645 if (pid == 0) {
2646 pid_socket = safe_close(pid_socket);
2647 uuid_socket = safe_close(uuid_socket);
2648 notify_socket = safe_close(notify_socket);
2649 uid_shift_socket = safe_close(uid_shift_socket);
2650
2651 /* The inner child has all namespaces that are
2652 * requested, so that we all are owned by the user if
2653 * user namespaces are turned on. */
2654
2655 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
2656 if (r < 0)
2657 _exit(EXIT_FAILURE);
2658
2659 _exit(EXIT_SUCCESS);
2660 }
2661
2662 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
2663 if (l < 0)
2664 return log_error_errno(errno, "Failed to send PID: %m");
2665 if (l != sizeof(pid)) {
2666 log_error("Short write while sending PID.");
2667 return -EIO;
2668 }
2669
2670 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
2671 if (l < 0)
2672 return log_error_errno(errno, "Failed to send machine ID: %m");
2673 if (l != sizeof(arg_uuid)) {
2674 log_error("Short write while sending machine ID.");
2675 return -EIO;
2676 }
2677
2678 l = send_one_fd(notify_socket, fd, 0);
2679 if (l < 0)
2680 return log_error_errno(errno, "Failed to send notify fd: %m");
2681
2682 pid_socket = safe_close(pid_socket);
2683 uuid_socket = safe_close(uuid_socket);
2684 notify_socket = safe_close(notify_socket);
2685 kmsg_socket = safe_close(kmsg_socket);
2686 rtnl_socket = safe_close(rtnl_socket);
2687
2688 return 0;
2689 }
2690
2691 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
2692 unsigned n_tries = 100;
2693 uid_t candidate;
2694 int r;
2695
2696 assert(shift);
2697 assert(ret_lock_file);
2698 assert(arg_userns_mode == USER_NAMESPACE_PICK);
2699 assert(arg_uid_range == 0x10000U);
2700
2701 candidate = *shift;
2702
2703 (void) mkdir("/run/systemd/nspawn-uid", 0755);
2704
2705 for (;;) {
2706 char lock_path[strlen("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
2707 _cleanup_release_lock_file_ LockFile lf = LOCK_FILE_INIT;
2708
2709 if (--n_tries <= 0)
2710 return -EBUSY;
2711
2712 if (candidate < UID_SHIFT_PICK_MIN || candidate > UID_SHIFT_PICK_MAX)
2713 goto next;
2714 if ((candidate & UINT32_C(0xFFFF)) != 0)
2715 goto next;
2716
2717 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
2718 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
2719 if (r == -EBUSY) /* Range already taken by another nspawn instance */
2720 goto next;
2721 if (r < 0)
2722 return r;
2723
2724 /* Make some superficial checks whether the range is currently known in the user database */
2725 if (getpwuid(candidate))
2726 goto next;
2727 if (getpwuid(candidate + UINT32_C(0xFFFE)))
2728 goto next;
2729 if (getgrgid(candidate))
2730 goto next;
2731 if (getgrgid(candidate + UINT32_C(0xFFFE)))
2732 goto next;
2733
2734 *ret_lock_file = lf;
2735 lf = (struct LockFile) LOCK_FILE_INIT;
2736 *shift = candidate;
2737 return 0;
2738
2739 next:
2740 random_bytes(&candidate, sizeof(candidate));
2741 candidate = (candidate % (UID_SHIFT_PICK_MAX - UID_SHIFT_PICK_MIN)) + UID_SHIFT_PICK_MIN;
2742 candidate &= (uid_t) UINT32_C(0xFFFF0000);
2743 }
2744 }
2745
2746 static int setup_uid_map(pid_t pid) {
2747 char uid_map[strlen("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
2748 int r;
2749
2750 assert(pid > 1);
2751
2752 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
2753 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
2754 r = write_string_file(uid_map, line, 0);
2755 if (r < 0)
2756 return log_error_errno(r, "Failed to write UID map: %m");
2757
2758 /* We always assign the same UID and GID ranges */
2759 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
2760 r = write_string_file(uid_map, line, 0);
2761 if (r < 0)
2762 return log_error_errno(r, "Failed to write GID map: %m");
2763
2764 return 0;
2765 }
2766
2767 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2768 char buf[NOTIFY_BUFFER_MAX+1];
2769 char *p = NULL;
2770 struct iovec iovec = {
2771 .iov_base = buf,
2772 .iov_len = sizeof(buf)-1,
2773 };
2774 union {
2775 struct cmsghdr cmsghdr;
2776 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
2777 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
2778 } control = {};
2779 struct msghdr msghdr = {
2780 .msg_iov = &iovec,
2781 .msg_iovlen = 1,
2782 .msg_control = &control,
2783 .msg_controllen = sizeof(control),
2784 };
2785 struct cmsghdr *cmsg;
2786 struct ucred *ucred = NULL;
2787 ssize_t n;
2788 pid_t inner_child_pid;
2789 _cleanup_strv_free_ char **tags = NULL;
2790
2791 assert(userdata);
2792
2793 inner_child_pid = PTR_TO_PID(userdata);
2794
2795 if (revents != EPOLLIN) {
2796 log_warning("Got unexpected poll event for notify fd.");
2797 return 0;
2798 }
2799
2800 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
2801 if (n < 0) {
2802 if (errno == EAGAIN || errno == EINTR)
2803 return 0;
2804
2805 return log_warning_errno(errno, "Couldn't read notification socket: %m");
2806 }
2807 cmsg_close_all(&msghdr);
2808
2809 CMSG_FOREACH(cmsg, &msghdr) {
2810 if (cmsg->cmsg_level == SOL_SOCKET &&
2811 cmsg->cmsg_type == SCM_CREDENTIALS &&
2812 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2813
2814 ucred = (struct ucred*) CMSG_DATA(cmsg);
2815 }
2816 }
2817
2818 if (!ucred || ucred->pid != inner_child_pid) {
2819 log_warning("Received notify message without valid credentials. Ignoring.");
2820 return 0;
2821 }
2822
2823 if ((size_t) n >= sizeof(buf)) {
2824 log_warning("Received notify message exceeded maximum size. Ignoring.");
2825 return 0;
2826 }
2827
2828 buf[n] = 0;
2829 tags = strv_split(buf, "\n\r");
2830 if (!tags)
2831 return log_oom();
2832
2833 if (strv_find(tags, "READY=1"))
2834 sd_notifyf(false, "READY=1\n");
2835
2836 p = strv_find_startswith(tags, "STATUS=");
2837 if (p)
2838 sd_notifyf(false, "STATUS=Container running: %s", p);
2839
2840 return 0;
2841 }
2842
2843 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
2844 int r;
2845
2846 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
2847 if (r < 0)
2848 return log_error_errno(r, "Failed to allocate notify event source: %m");
2849
2850 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
2851
2852 return 0;
2853 }
2854
2855 static int load_settings(void) {
2856 _cleanup_(settings_freep) Settings *settings = NULL;
2857 _cleanup_fclose_ FILE *f = NULL;
2858 _cleanup_free_ char *p = NULL;
2859 const char *fn, *i;
2860 int r;
2861
2862 /* If all settings are masked, there's no point in looking for
2863 * the settings file */
2864 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
2865 return 0;
2866
2867 fn = strjoina(arg_machine, ".nspawn");
2868
2869 /* We first look in the admin's directories in /etc and /run */
2870 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
2871 _cleanup_free_ char *j = NULL;
2872
2873 j = strjoin(i, "/", fn);
2874 if (!j)
2875 return log_oom();
2876
2877 f = fopen(j, "re");
2878 if (f) {
2879 p = j;
2880 j = NULL;
2881
2882 /* By default, we trust configuration from /etc and /run */
2883 if (arg_settings_trusted < 0)
2884 arg_settings_trusted = true;
2885
2886 break;
2887 }
2888
2889 if (errno != ENOENT)
2890 return log_error_errno(errno, "Failed to open %s: %m", j);
2891 }
2892
2893 if (!f) {
2894 /* After that, let's look for a file next to the
2895 * actual image we shall boot. */
2896
2897 if (arg_image) {
2898 p = file_in_same_dir(arg_image, fn);
2899 if (!p)
2900 return log_oom();
2901 } else if (arg_directory) {
2902 p = file_in_same_dir(arg_directory, fn);
2903 if (!p)
2904 return log_oom();
2905 }
2906
2907 if (p) {
2908 f = fopen(p, "re");
2909 if (!f && errno != ENOENT)
2910 return log_error_errno(errno, "Failed to open %s: %m", p);
2911
2912 /* By default, we do not trust configuration from /var/lib/machines */
2913 if (arg_settings_trusted < 0)
2914 arg_settings_trusted = false;
2915 }
2916 }
2917
2918 if (!f)
2919 return 0;
2920
2921 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
2922
2923 r = settings_load(f, p, &settings);
2924 if (r < 0)
2925 return r;
2926
2927 /* Copy over bits from the settings, unless they have been
2928 * explicitly masked by command line switches. */
2929
2930 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
2931 settings->start_mode >= 0) {
2932 arg_start_mode = settings->start_mode;
2933
2934 strv_free(arg_parameters);
2935 arg_parameters = settings->parameters;
2936 settings->parameters = NULL;
2937 }
2938
2939 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
2940 settings->pivot_root_new) {
2941 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
2942 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
2943 }
2944
2945 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
2946 settings->working_directory) {
2947 free(arg_chdir);
2948 arg_chdir = settings->working_directory;
2949 settings->working_directory = NULL;
2950 }
2951
2952 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
2953 settings->environment) {
2954 strv_free(arg_setenv);
2955 arg_setenv = settings->environment;
2956 settings->environment = NULL;
2957 }
2958
2959 if ((arg_settings_mask & SETTING_USER) == 0 &&
2960 settings->user) {
2961 free(arg_user);
2962 arg_user = settings->user;
2963 settings->user = NULL;
2964 }
2965
2966 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
2967 uint64_t plus;
2968
2969 plus = settings->capability;
2970 if (settings_private_network(settings))
2971 plus |= (1ULL << CAP_NET_ADMIN);
2972
2973 if (!arg_settings_trusted && plus != 0) {
2974 if (settings->capability != 0)
2975 log_warning("Ignoring Capability= setting, file %s is not trusted.", p);
2976 } else
2977 arg_caps_retain |= plus;
2978
2979 arg_caps_retain &= ~settings->drop_capability;
2980 }
2981
2982 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
2983 settings->kill_signal > 0)
2984 arg_kill_signal = settings->kill_signal;
2985
2986 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
2987 settings->personality != PERSONALITY_INVALID)
2988 arg_personality = settings->personality;
2989
2990 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
2991 !sd_id128_is_null(settings->machine_id)) {
2992
2993 if (!arg_settings_trusted)
2994 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p);
2995 else
2996 arg_uuid = settings->machine_id;
2997 }
2998
2999 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3000 settings->read_only >= 0)
3001 arg_read_only = settings->read_only;
3002
3003 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3004 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3005 arg_volatile_mode = settings->volatile_mode;
3006
3007 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3008 settings->n_custom_mounts > 0) {
3009
3010 if (!arg_settings_trusted)
3011 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p);
3012 else {
3013 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3014 arg_custom_mounts = settings->custom_mounts;
3015 arg_n_custom_mounts = settings->n_custom_mounts;
3016
3017 settings->custom_mounts = NULL;
3018 settings->n_custom_mounts = 0;
3019 }
3020 }
3021
3022 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3023 (settings->private_network >= 0 ||
3024 settings->network_veth >= 0 ||
3025 settings->network_bridge ||
3026 settings->network_zone ||
3027 settings->network_interfaces ||
3028 settings->network_macvlan ||
3029 settings->network_ipvlan ||
3030 settings->network_veth_extra)) {
3031
3032 if (!arg_settings_trusted)
3033 log_warning("Ignoring network settings, file %s is not trusted.", p);
3034 else {
3035 arg_network_veth = settings_network_veth(settings);
3036 arg_private_network = settings_private_network(settings);
3037
3038 strv_free(arg_network_interfaces);
3039 arg_network_interfaces = settings->network_interfaces;
3040 settings->network_interfaces = NULL;
3041
3042 strv_free(arg_network_macvlan);
3043 arg_network_macvlan = settings->network_macvlan;
3044 settings->network_macvlan = NULL;
3045
3046 strv_free(arg_network_ipvlan);
3047 arg_network_ipvlan = settings->network_ipvlan;
3048 settings->network_ipvlan = NULL;
3049
3050 strv_free(arg_network_veth_extra);
3051 arg_network_veth_extra = settings->network_veth_extra;
3052 settings->network_veth_extra = NULL;
3053
3054 free(arg_network_bridge);
3055 arg_network_bridge = settings->network_bridge;
3056 settings->network_bridge = NULL;
3057
3058 free(arg_network_zone);
3059 arg_network_zone = settings->network_zone;
3060 settings->network_zone = NULL;
3061 }
3062 }
3063
3064 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3065 settings->expose_ports) {
3066
3067 if (!arg_settings_trusted)
3068 log_warning("Ignoring Port= setting, file %s is not trusted.", p);
3069 else {
3070 expose_port_free_all(arg_expose_ports);
3071 arg_expose_ports = settings->expose_ports;
3072 settings->expose_ports = NULL;
3073 }
3074 }
3075
3076 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3077 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3078
3079 if (!arg_settings_trusted)
3080 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", p);
3081 else {
3082 arg_userns_mode = settings->userns_mode;
3083 arg_uid_shift = settings->uid_shift;
3084 arg_uid_range = settings->uid_range;
3085 arg_userns_chown = settings->userns_chown;
3086 }
3087 }
3088
3089 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3090 arg_notify_ready = settings->notify_ready;
3091
3092 return 0;
3093 }
3094
3095 static int run(int master,
3096 const char* console,
3097 DissectedImage *dissected_image,
3098 bool interactive,
3099 bool secondary,
3100 FDSet *fds,
3101 char veth_name[IFNAMSIZ], bool *veth_created,
3102 union in_addr_union *exposed,
3103 pid_t *pid, int *ret) {
3104
3105 static const struct sigaction sa = {
3106 .sa_handler = nop_signal_handler,
3107 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
3108 };
3109
3110 _cleanup_release_lock_file_ LockFile uid_shift_lock = LOCK_FILE_INIT;
3111 _cleanup_close_ int etc_passwd_lock = -1;
3112 _cleanup_close_pair_ int
3113 kmsg_socket_pair[2] = { -1, -1 },
3114 rtnl_socket_pair[2] = { -1, -1 },
3115 pid_socket_pair[2] = { -1, -1 },
3116 uuid_socket_pair[2] = { -1, -1 },
3117 notify_socket_pair[2] = { -1, -1 },
3118 uid_shift_socket_pair[2] = { -1, -1 };
3119 _cleanup_close_ int notify_socket= -1;
3120 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
3121 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
3122 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
3123 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
3124 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
3125 ContainerStatus container_status = 0;
3126 char last_char = 0;
3127 int ifi = 0, r;
3128 ssize_t l;
3129 sigset_t mask_chld;
3130
3131 assert_se(sigemptyset(&mask_chld) == 0);
3132 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
3133
3134 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3135 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3136 * check with getpwuid() if the specific user already exists. Note that /etc might be
3137 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3138 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3139 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3140 * really ours. */
3141
3142 etc_passwd_lock = take_etc_passwd_lock(NULL);
3143 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
3144 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
3145 }
3146
3147 r = barrier_create(&barrier);
3148 if (r < 0)
3149 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
3150
3151 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
3152 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
3153
3154 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
3155 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
3156
3157 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
3158 return log_error_errno(errno, "Failed to create pid socket pair: %m");
3159
3160 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
3161 return log_error_errno(errno, "Failed to create id socket pair: %m");
3162
3163 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
3164 return log_error_errno(errno, "Failed to create notify socket pair: %m");
3165
3166 if (arg_userns_mode != USER_NAMESPACE_NO)
3167 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
3168 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
3169
3170 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3171 * parent's blocking calls and give it a chance to call wait() and terminate. */
3172 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
3173 if (r < 0)
3174 return log_error_errno(errno, "Failed to change the signal mask: %m");
3175
3176 r = sigaction(SIGCHLD, &sa, NULL);
3177 if (r < 0)
3178 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
3179
3180 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
3181 if (*pid < 0)
3182 return log_error_errno(errno, "clone() failed%s: %m",
3183 errno == EINVAL ?
3184 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3185
3186 if (*pid == 0) {
3187 /* The outer child only has a file system namespace. */
3188 barrier_set_role(&barrier, BARRIER_CHILD);
3189
3190 master = safe_close(master);
3191
3192 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
3193 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3194 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
3195 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
3196 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
3197 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
3198
3199 (void) reset_all_signal_handlers();
3200 (void) reset_signal_mask();
3201
3202 r = outer_child(&barrier,
3203 arg_directory,
3204 console,
3205 dissected_image,
3206 interactive,
3207 secondary,
3208 pid_socket_pair[1],
3209 uuid_socket_pair[1],
3210 notify_socket_pair[1],
3211 kmsg_socket_pair[1],
3212 rtnl_socket_pair[1],
3213 uid_shift_socket_pair[1],
3214 fds);
3215 if (r < 0)
3216 _exit(EXIT_FAILURE);
3217
3218 _exit(EXIT_SUCCESS);
3219 }
3220
3221 barrier_set_role(&barrier, BARRIER_PARENT);
3222
3223 fds = fdset_free(fds);
3224
3225 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
3226 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
3227 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
3228 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
3229 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
3230 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
3231
3232 if (arg_userns_mode != USER_NAMESPACE_NO) {
3233 /* The child just let us know the UID shift it might have read from the image. */
3234 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
3235 if (l < 0)
3236 return log_error_errno(errno, "Failed to read UID shift: %m");
3237 if (l != sizeof arg_uid_shift) {
3238 log_error("Short read while reading UID shift.");
3239 return -EIO;
3240 }
3241
3242 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3243 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3244 * image, but if that's already in use, pick a new one, and report back to the child,
3245 * which one we now picked. */
3246
3247 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
3248 if (r < 0)
3249 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
3250
3251 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
3252 if (l < 0)
3253 return log_error_errno(errno, "Failed to send UID shift: %m");
3254 if (l != sizeof arg_uid_shift) {
3255 log_error("Short write while writing UID shift.");
3256 return -EIO;
3257 }
3258 }
3259 }
3260
3261 /* Wait for the outer child. */
3262 r = wait_for_terminate_and_warn("namespace helper", *pid, NULL);
3263 if (r != 0)
3264 return r < 0 ? r : -EIO;
3265
3266 /* And now retrieve the PID of the inner child. */
3267 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
3268 if (l < 0)
3269 return log_error_errno(errno, "Failed to read inner child PID: %m");
3270 if (l != sizeof *pid) {
3271 log_error("Short read while reading inner child PID.");
3272 return -EIO;
3273 }
3274
3275 /* We also retrieve container UUID in case it was generated by outer child */
3276 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
3277 if (l < 0)
3278 return log_error_errno(errno, "Failed to read container machine ID: %m");
3279 if (l != sizeof(arg_uuid)) {
3280 log_error("Short read while reading container machined ID.");
3281 return -EIO;
3282 }
3283
3284 /* We also retrieve the socket used for notifications generated by outer child */
3285 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
3286 if (notify_socket < 0)
3287 return log_error_errno(notify_socket,
3288 "Failed to receive notification socket from the outer child: %m");
3289
3290 log_debug("Init process invoked as PID "PID_FMT, *pid);
3291
3292 if (arg_userns_mode != USER_NAMESPACE_NO) {
3293 if (!barrier_place_and_sync(&barrier)) { /* #1 */
3294 log_error("Child died too early.");
3295 return -ESRCH;
3296 }
3297
3298 r = setup_uid_map(*pid);
3299 if (r < 0)
3300 return r;
3301
3302 (void) barrier_place(&barrier); /* #2 */
3303 }
3304
3305 if (arg_private_network) {
3306
3307 r = move_network_interfaces(*pid, arg_network_interfaces);
3308 if (r < 0)
3309 return r;
3310
3311 if (arg_network_veth) {
3312 r = setup_veth(arg_machine, *pid, veth_name,
3313 arg_network_bridge || arg_network_zone);
3314 if (r < 0)
3315 return r;
3316 else if (r > 0)
3317 ifi = r;
3318
3319 if (arg_network_bridge) {
3320 /* Add the interface to a bridge */
3321 r = setup_bridge(veth_name, arg_network_bridge, false);
3322 if (r < 0)
3323 return r;
3324 if (r > 0)
3325 ifi = r;
3326 } else if (arg_network_zone) {
3327 /* Add the interface to a bridge, possibly creating it */
3328 r = setup_bridge(veth_name, arg_network_zone, true);
3329 if (r < 0)
3330 return r;
3331 if (r > 0)
3332 ifi = r;
3333 }
3334 }
3335
3336 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
3337 if (r < 0)
3338 return r;
3339
3340 /* We created the primary and extra veth links now; let's remember this, so that we know to
3341 remove them later on. Note that we don't bother with removing veth links that were created
3342 here when their setup failed half-way, because in that case the kernel should be able to
3343 remove them on its own, since they cannot be referenced by anything yet. */
3344 *veth_created = true;
3345
3346 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
3347 if (r < 0)
3348 return r;
3349
3350 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
3351 if (r < 0)
3352 return r;
3353 }
3354
3355 if (arg_register) {
3356 r = register_machine(
3357 arg_machine,
3358 *pid,
3359 arg_directory,
3360 arg_uuid,
3361 ifi,
3362 arg_slice,
3363 arg_custom_mounts, arg_n_custom_mounts,
3364 arg_kill_signal,
3365 arg_property,
3366 arg_keep_unit,
3367 arg_container_service_name);
3368 if (r < 0)
3369 return r;
3370 }
3371
3372 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3373 if (r < 0)
3374 return r;
3375
3376 if (arg_keep_unit) {
3377 r = create_subcgroup(*pid, arg_unified_cgroup_hierarchy);
3378 if (r < 0)
3379 return r;
3380 }
3381
3382 r = chown_cgroup(*pid, arg_uid_shift);
3383 if (r < 0)
3384 return r;
3385
3386 /* Notify the child that the parent is ready with all
3387 * its setup (including cgroup-ification), and that
3388 * the child can now hand over control to the code to
3389 * run inside the container. */
3390 (void) barrier_place(&barrier); /* #3 */
3391
3392 /* Block SIGCHLD here, before notifying child.
3393 * process_pty() will handle it with the other signals. */
3394 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
3395
3396 /* Reset signal to default */
3397 r = default_signals(SIGCHLD, -1);
3398 if (r < 0)
3399 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
3400
3401 r = sd_event_new(&event);
3402 if (r < 0)
3403 return log_error_errno(r, "Failed to get default event source: %m");
3404
3405 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
3406 if (r < 0)
3407 return r;
3408
3409 /* Let the child know that we are ready and wait that the child is completely ready now. */
3410 if (!barrier_place_and_sync(&barrier)) { /* #4 */
3411 log_error("Child died too early.");
3412 return -ESRCH;
3413 }
3414
3415 /* At this point we have made use of the UID we picked, and thus nss-mymachines
3416 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
3417 etc_passwd_lock = safe_close(etc_passwd_lock);
3418
3419 sd_notifyf(false,
3420 "STATUS=Container running.\n"
3421 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
3422 if (!arg_notify_ready)
3423 sd_notify(false, "READY=1\n");
3424
3425 if (arg_kill_signal > 0) {
3426 /* Try to kill the init system on SIGINT or SIGTERM */
3427 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
3428 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
3429 } else {
3430 /* Immediately exit */
3431 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
3432 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
3433 }
3434
3435 /* Exit when the child exits */
3436 sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
3437
3438 if (arg_expose_ports) {
3439 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
3440 if (r < 0)
3441 return r;
3442
3443 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
3444 }
3445
3446 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3447
3448 r = pty_forward_new(event, master,
3449 PTY_FORWARD_IGNORE_VHANGUP | (interactive ? 0 : PTY_FORWARD_READ_ONLY),
3450 &forward);
3451 if (r < 0)
3452 return log_error_errno(r, "Failed to create PTY forwarder: %m");
3453
3454 r = sd_event_loop(event);
3455 if (r < 0)
3456 return log_error_errno(r, "Failed to run event loop: %m");
3457
3458 pty_forward_get_last_char(forward, &last_char);
3459
3460 forward = pty_forward_free(forward);
3461
3462 if (!arg_quiet && last_char != '\n')
3463 putc('\n', stdout);
3464
3465 /* Kill if it is not dead yet anyway */
3466 if (arg_register && !arg_keep_unit)
3467 terminate_machine(*pid);
3468
3469 /* Normally redundant, but better safe than sorry */
3470 (void) kill(*pid, SIGKILL);
3471
3472 r = wait_for_container(*pid, &container_status);
3473 *pid = 0;
3474
3475 if (r < 0)
3476 /* We failed to wait for the container, or the container exited abnormally. */
3477 return r;
3478 if (r > 0 || container_status == CONTAINER_TERMINATED) {
3479 /* r > 0 → The container exited with a non-zero status.
3480 * As a special case, we need to replace 133 with a different value,
3481 * because 133 is special-cased in the service file to reboot the container.
3482 * otherwise → The container exited with zero status and a reboot was not requested.
3483 */
3484 if (r == EXIT_FORCE_RESTART)
3485 r = EXIT_FAILURE; /* replace 133 with the general failure code */
3486 *ret = r;
3487 return 0; /* finito */
3488 }
3489
3490 /* CONTAINER_REBOOTED, loop again */
3491
3492 if (arg_keep_unit) {
3493 /* Special handling if we are running as a service: instead of simply
3494 * restarting the machine we want to restart the entire service, so let's
3495 * inform systemd about this with the special exit code 133. The service
3496 * file uses RestartForceExitStatus=133 so that this results in a full
3497 * nspawn restart. This is necessary since we might have cgroup parameters
3498 * set we want to have flushed out. */
3499 *ret = EXIT_FORCE_RESTART;
3500 return 0; /* finito */
3501 }
3502
3503 expose_port_flush(arg_expose_ports, exposed);
3504
3505 (void) remove_veth_links(veth_name, arg_network_veth_extra);
3506 *veth_created = false;
3507 return 1; /* loop again */
3508 }
3509
3510 int main(int argc, char *argv[]) {
3511
3512 _cleanup_free_ char *console = NULL;
3513 _cleanup_close_ int master = -1;
3514 _cleanup_fdset_free_ FDSet *fds = NULL;
3515 int r, n_fd_passed, ret = EXIT_SUCCESS;
3516 char veth_name[IFNAMSIZ] = "";
3517 bool secondary = false, remove_directory = false, remove_image = false;
3518 pid_t pid = 0;
3519 union in_addr_union exposed = {};
3520 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
3521 bool interactive, veth_created = false, remove_tmprootdir = false;
3522 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
3523 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
3524 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3525 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3526
3527 log_parse_environment();
3528 log_open();
3529
3530 /* Make sure rename_process() in the stub init process can work */
3531 saved_argv = argv;
3532 saved_argc = argc;
3533
3534 r = parse_argv(argc, argv);
3535 if (r <= 0)
3536 goto finish;
3537
3538 if (geteuid() != 0) {
3539 log_error("Need to be root.");
3540 r = -EPERM;
3541 goto finish;
3542 }
3543 r = determine_names();
3544 if (r < 0)
3545 goto finish;
3546
3547 r = load_settings();
3548 if (r < 0)
3549 goto finish;
3550
3551 r = verify_arguments();
3552 if (r < 0)
3553 goto finish;
3554
3555 n_fd_passed = sd_listen_fds(false);
3556 if (n_fd_passed > 0) {
3557 r = fdset_new_listen_fds(&fds, false);
3558 if (r < 0) {
3559 log_error_errno(r, "Failed to collect file descriptors: %m");
3560 goto finish;
3561 }
3562 }
3563
3564 if (arg_directory) {
3565 assert(!arg_image);
3566
3567 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
3568 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3569 r = -EINVAL;
3570 goto finish;
3571 }
3572
3573 if (arg_ephemeral) {
3574 _cleanup_free_ char *np = NULL;
3575
3576 r = chase_symlinks_and_update(&arg_directory, 0);
3577 if (r < 0)
3578 goto finish;
3579
3580 /* If the specified path is a mount point we
3581 * generate the new snapshot immediately
3582 * inside it under a random name. However if
3583 * the specified is not a mount point we
3584 * create the new snapshot in the parent
3585 * directory, just next to it. */
3586 r = path_is_mount_point(arg_directory, NULL, 0);
3587 if (r < 0) {
3588 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
3589 goto finish;
3590 }
3591 if (r > 0)
3592 r = tempfn_random_child(arg_directory, "machine.", &np);
3593 else
3594 r = tempfn_random(arg_directory, "machine.", &np);
3595 if (r < 0) {
3596 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
3597 goto finish;
3598 }
3599
3600 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3601 if (r < 0) {
3602 log_error_errno(r, "Failed to lock %s: %m", np);
3603 goto finish;
3604 }
3605
3606 r = btrfs_subvol_snapshot(arg_directory, np,
3607 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3608 BTRFS_SNAPSHOT_FALLBACK_COPY |
3609 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3610 BTRFS_SNAPSHOT_RECURSIVE |
3611 BTRFS_SNAPSHOT_QUOTA);
3612 if (r < 0) {
3613 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
3614 goto finish;
3615 }
3616
3617 free(arg_directory);
3618 arg_directory = np;
3619 np = NULL;
3620
3621 remove_directory = true;
3622
3623 } else {
3624 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
3625 if (r < 0)
3626 goto finish;
3627
3628 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3629 if (r == -EBUSY) {
3630 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
3631 goto finish;
3632 }
3633 if (r < 0) {
3634 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
3635 goto finish;
3636 }
3637
3638 if (arg_template) {
3639 r = chase_symlinks_and_update(&arg_template, 0);
3640 if (r < 0)
3641 goto finish;
3642
3643 r = btrfs_subvol_snapshot(arg_template, arg_directory,
3644 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3645 BTRFS_SNAPSHOT_FALLBACK_COPY |
3646 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3647 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
3648 BTRFS_SNAPSHOT_RECURSIVE |
3649 BTRFS_SNAPSHOT_QUOTA);
3650 if (r == -EEXIST) {
3651 if (!arg_quiet)
3652 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
3653 } else if (r < 0) {
3654 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
3655 goto finish;
3656 } else {
3657 if (!arg_quiet)
3658 log_info("Populated %s from template %s.", arg_directory, arg_template);
3659 }
3660 }
3661 }
3662
3663 if (arg_start_mode == START_BOOT) {
3664 if (path_is_os_tree(arg_directory) <= 0) {
3665 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
3666 r = -EINVAL;
3667 goto finish;
3668 }
3669 } else {
3670 const char *p;
3671
3672 p = strjoina(arg_directory, "/usr/");
3673 if (laccess(p, F_OK) < 0) {
3674 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", arg_directory);
3675 r = -EINVAL;
3676 goto finish;
3677 }
3678 }
3679
3680 } else {
3681 assert(arg_image);
3682 assert(!arg_template);
3683
3684 r = chase_symlinks_and_update(&arg_image, 0);
3685 if (r < 0)
3686 goto finish;
3687
3688 if (arg_ephemeral) {
3689 _cleanup_free_ char *np = NULL;
3690
3691 r = tempfn_random(arg_image, "machine.", &np);
3692 if (r < 0) {
3693 log_error_errno(r, "Failed to generate name for image snapshot: %m");
3694 goto finish;
3695 }
3696
3697 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3698 if (r < 0) {
3699 r = log_error_errno(r, "Failed to create image lock: %m");
3700 goto finish;
3701 }
3702
3703 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, COPY_REFLINK);
3704 if (r < 0) {
3705 r = log_error_errno(r, "Failed to copy image file: %m");
3706 goto finish;
3707 }
3708
3709 free(arg_image);
3710 arg_image = np;
3711 np = NULL;
3712
3713 remove_image = true;
3714 } else {
3715 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3716 if (r == -EBUSY) {
3717 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
3718 goto finish;
3719 }
3720 if (r < 0) {
3721 r = log_error_errno(r, "Failed to create image lock: %m");
3722 goto finish;
3723 }
3724
3725 if (!arg_root_hash) {
3726 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
3727 if (r < 0) {
3728 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
3729 goto finish;
3730 }
3731 }
3732 }
3733
3734 if (!mkdtemp(tmprootdir)) {
3735 r = log_error_errno(errno, "Failed to create temporary directory: %m");
3736 goto finish;
3737 }
3738
3739 remove_tmprootdir = true;
3740
3741 arg_directory = strdup(tmprootdir);
3742 if (!arg_directory) {
3743 r = log_oom();
3744 goto finish;
3745 }
3746
3747 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
3748 if (r < 0) {
3749 log_error_errno(r, "Failed to set up loopback block device: %m");
3750 goto finish;
3751 }
3752
3753 r = dissect_image(
3754 loop->fd,
3755 arg_root_hash, arg_root_hash_size,
3756 DISSECT_IMAGE_REQUIRE_ROOT,
3757 &dissected_image);
3758 if (r == -ENOPKG) {
3759 log_error_errno(r, "Could not find a suitable file system or partition table in image: %s", arg_image);
3760
3761 log_notice("Note that the disk image needs to\n"
3762 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
3763 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
3764 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
3765 " d) or contain a file system without a partition table\n"
3766 "in order to be bootable with systemd-nspawn.");
3767 goto finish;
3768 }
3769 if (r == -EADDRNOTAVAIL) {
3770 log_error_errno(r, "No root partition for specified root hash found.");
3771 goto finish;
3772 }
3773 if (r == -EOPNOTSUPP) {
3774 log_error_errno(r, "--image= is not supported, compiled without blkid support.");
3775 goto finish;
3776 }
3777 if (r < 0) {
3778 log_error_errno(r, "Failed to dissect image: %m");
3779 goto finish;
3780 }
3781
3782 if (!arg_root_hash && dissected_image->can_verity)
3783 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
3784
3785 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
3786 if (r < 0)
3787 goto finish;
3788
3789 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
3790 if (remove_image && unlink(arg_image) >= 0)
3791 remove_image = false;
3792 }
3793
3794 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
3795 if (r < 0)
3796 goto finish;
3797
3798 interactive =
3799 isatty(STDIN_FILENO) > 0 &&
3800 isatty(STDOUT_FILENO) > 0;
3801
3802 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
3803 if (master < 0) {
3804 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
3805 goto finish;
3806 }
3807
3808 r = ptsname_malloc(master, &console);
3809 if (r < 0) {
3810 r = log_error_errno(r, "Failed to determine tty name: %m");
3811 goto finish;
3812 }
3813
3814 if (arg_selinux_apifs_context) {
3815 r = mac_selinux_apply(console, arg_selinux_apifs_context);
3816 if (r < 0)
3817 goto finish;
3818 }
3819
3820 if (unlockpt(master) < 0) {
3821 r = log_error_errno(errno, "Failed to unlock tty: %m");
3822 goto finish;
3823 }
3824
3825 if (!arg_quiet)
3826 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
3827 arg_machine, arg_image ?: arg_directory);
3828
3829 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
3830
3831 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
3832 r = log_error_errno(errno, "Failed to become subreaper: %m");
3833 goto finish;
3834 }
3835
3836 for (;;) {
3837 r = run(master,
3838 console,
3839 dissected_image,
3840 interactive, secondary,
3841 fds,
3842 veth_name, &veth_created,
3843 &exposed,
3844 &pid, &ret);
3845 if (r <= 0)
3846 break;
3847 }
3848
3849 finish:
3850 sd_notify(false,
3851 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
3852 "STOPPING=1\nSTATUS=Terminating...");
3853
3854 if (pid > 0)
3855 (void) kill(pid, SIGKILL);
3856
3857 /* Try to flush whatever is still queued in the pty */
3858 if (master >= 0) {
3859 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
3860 master = safe_close(master);
3861 }
3862
3863 if (pid > 0)
3864 (void) wait_for_terminate(pid, NULL);
3865
3866 if (remove_directory && arg_directory) {
3867 int k;
3868
3869 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
3870 if (k < 0)
3871 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
3872 }
3873
3874 if (remove_image && arg_image) {
3875 if (unlink(arg_image) < 0)
3876 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
3877 }
3878
3879 if (remove_tmprootdir) {
3880 if (rmdir(tmprootdir) < 0)
3881 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
3882 }
3883
3884 if (arg_machine) {
3885 const char *p;
3886
3887 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
3888 (void) rm_rf(p, REMOVE_ROOT);
3889 }
3890
3891 expose_port_flush(arg_expose_ports, &exposed);
3892
3893 if (veth_created)
3894 (void) remove_veth_links(veth_name, arg_network_veth_extra);
3895 (void) remove_bridge(arg_network_zone);
3896
3897 free(arg_directory);
3898 free(arg_template);
3899 free(arg_image);
3900 free(arg_machine);
3901 free(arg_user);
3902 free(arg_pivot_root_new);
3903 free(arg_pivot_root_old);
3904 free(arg_chdir);
3905 strv_free(arg_setenv);
3906 free(arg_network_bridge);
3907 strv_free(arg_network_interfaces);
3908 strv_free(arg_network_macvlan);
3909 strv_free(arg_network_ipvlan);
3910 strv_free(arg_network_veth_extra);
3911 strv_free(arg_parameters);
3912 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3913 expose_port_free_all(arg_expose_ports);
3914 free(arg_root_hash);
3915
3916 return r < 0 ? EXIT_FAILURE : ret;
3917 }
Unnamed repository; edit this file 'description' to name the repository.