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


a5bc50c1f4c67fc1daa9380da5ac478e97d99087
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 /***
3 This file is part of systemd.
4
5 Copyright 2010 Lennart Poettering
6
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
11
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
16
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
19 ***/
20
21 #if HAVE_BLKID
22 #include <blkid.h>
23 #endif
24 #include <errno.h>
25 #include <getopt.h>
26 #include <grp.h>
27 #include <linux/loop.h>
28 #include <pwd.h>
29 #include <sched.h>
30 #if HAVE_SELINUX
31 #include <selinux/selinux.h>
32 #endif
33 #include <signal.h>
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <string.h>
37 #include <sys/file.h>
38 #include <sys/mount.h>
39 #include <sys/personality.h>
40 #include <sys/prctl.h>
41 #include <sys/types.h>
42 #include <sys/wait.h>
43 #include <unistd.h>
44
45 #include "sd-bus.h"
46 #include "sd-daemon.h"
47 #include "sd-id128.h"
48
49 #include "alloc-util.h"
50 #include "barrier.h"
51 #include "base-filesystem.h"
52 #include "blkid-util.h"
53 #include "btrfs-util.h"
54 #include "bus-util.h"
55 #include "cap-list.h"
56 #include "capability-util.h"
57 #include "cgroup-util.h"
58 #include "copy.h"
59 #include "dev-setup.h"
60 #include "dissect-image.h"
61 #include "env-util.h"
62 #include "fd-util.h"
63 #include "fdset.h"
64 #include "fileio.h"
65 #include "format-util.h"
66 #include "fs-util.h"
67 #include "gpt.h"
68 #include "hexdecoct.h"
69 #include "hostname-util.h"
70 #include "id128-util.h"
71 #include "log.h"
72 #include "loop-util.h"
73 #include "loopback-setup.h"
74 #include "machine-image.h"
75 #include "macro.h"
76 #include "missing.h"
77 #include "mkdir.h"
78 #include "mount-util.h"
79 #include "netlink-util.h"
80 #include "nspawn-cgroup.h"
81 #include "nspawn-def.h"
82 #include "nspawn-expose-ports.h"
83 #include "nspawn-mount.h"
84 #include "nspawn-network.h"
85 #include "nspawn-patch-uid.h"
86 #include "nspawn-register.h"
87 #include "nspawn-seccomp.h"
88 #include "nspawn-settings.h"
89 #include "nspawn-setuid.h"
90 #include "nspawn-stub-pid1.h"
91 #include "parse-util.h"
92 #include "path-util.h"
93 #include "process-util.h"
94 #include "ptyfwd.h"
95 #include "random-util.h"
96 #include "raw-clone.h"
97 #include "rm-rf.h"
98 #include "selinux-util.h"
99 #include "signal-util.h"
100 #include "socket-util.h"
101 #include "stat-util.h"
102 #include "stdio-util.h"
103 #include "string-util.h"
104 #include "strv.h"
105 #include "terminal-util.h"
106 #include "udev-util.h"
107 #include "umask-util.h"
108 #include "user-util.h"
109 #include "util.h"
110
111 #if HAVE_SPLIT_USR
112 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
113 #else
114 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
115 #endif
116
117 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
118 * nspawn_notify_socket_path is relative to the container
119 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
120 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
121
122 #define EXIT_FORCE_RESTART 133
123
124 typedef enum ContainerStatus {
125 CONTAINER_TERMINATED,
126 CONTAINER_REBOOTED
127 } ContainerStatus;
128
129 typedef enum LinkJournal {
130 LINK_NO,
131 LINK_AUTO,
132 LINK_HOST,
133 LINK_GUEST
134 } LinkJournal;
135
136 static char *arg_directory = NULL;
137 static char *arg_template = NULL;
138 static char *arg_chdir = NULL;
139 static char *arg_pivot_root_new = NULL;
140 static char *arg_pivot_root_old = NULL;
141 static char *arg_user = NULL;
142 static sd_id128_t arg_uuid = {};
143 static char *arg_machine = NULL;
144 static const char *arg_selinux_context = NULL;
145 static const char *arg_selinux_apifs_context = NULL;
146 static const char *arg_slice = NULL;
147 static bool arg_private_network = false;
148 static bool arg_read_only = false;
149 static StartMode arg_start_mode = START_PID1;
150 static bool arg_ephemeral = false;
151 static LinkJournal arg_link_journal = LINK_AUTO;
152 static bool arg_link_journal_try = false;
153 static uint64_t arg_caps_retain =
154 (1ULL << CAP_AUDIT_CONTROL) |
155 (1ULL << CAP_AUDIT_WRITE) |
156 (1ULL << CAP_CHOWN) |
157 (1ULL << CAP_DAC_OVERRIDE) |
158 (1ULL << CAP_DAC_READ_SEARCH) |
159 (1ULL << CAP_FOWNER) |
160 (1ULL << CAP_FSETID) |
161 (1ULL << CAP_IPC_OWNER) |
162 (1ULL << CAP_KILL) |
163 (1ULL << CAP_LEASE) |
164 (1ULL << CAP_LINUX_IMMUTABLE) |
165 (1ULL << CAP_MKNOD) |
166 (1ULL << CAP_NET_BIND_SERVICE) |
167 (1ULL << CAP_NET_BROADCAST) |
168 (1ULL << CAP_NET_RAW) |
169 (1ULL << CAP_SETFCAP) |
170 (1ULL << CAP_SETGID) |
171 (1ULL << CAP_SETPCAP) |
172 (1ULL << CAP_SETUID) |
173 (1ULL << CAP_SYS_ADMIN) |
174 (1ULL << CAP_SYS_BOOT) |
175 (1ULL << CAP_SYS_CHROOT) |
176 (1ULL << CAP_SYS_NICE) |
177 (1ULL << CAP_SYS_PTRACE) |
178 (1ULL << CAP_SYS_RESOURCE) |
179 (1ULL << CAP_SYS_TTY_CONFIG);
180 static CustomMount *arg_custom_mounts = NULL;
181 static unsigned arg_n_custom_mounts = 0;
182 static char **arg_setenv = NULL;
183 static bool arg_quiet = false;
184 static bool arg_register = true;
185 static bool arg_keep_unit = false;
186 static char **arg_network_interfaces = NULL;
187 static char **arg_network_macvlan = NULL;
188 static char **arg_network_ipvlan = NULL;
189 static bool arg_network_veth = false;
190 static char **arg_network_veth_extra = NULL;
191 static char *arg_network_bridge = NULL;
192 static char *arg_network_zone = NULL;
193 static char *arg_network_namespace_path = NULL;
194 static unsigned long arg_personality = PERSONALITY_INVALID;
195 static char *arg_image = NULL;
196 static VolatileMode arg_volatile_mode = VOLATILE_NO;
197 static ExposePort *arg_expose_ports = NULL;
198 static char **arg_property = NULL;
199 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
200 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
201 static bool arg_userns_chown = false;
202 static int arg_kill_signal = 0;
203 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
204 static SettingsMask arg_settings_mask = 0;
205 static int arg_settings_trusted = -1;
206 static char **arg_parameters = NULL;
207 static const char *arg_container_service_name = "systemd-nspawn";
208 static bool arg_notify_ready = false;
209 static bool arg_use_cgns = true;
210 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
211 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO;
212 static void *arg_root_hash = NULL;
213 static size_t arg_root_hash_size = 0;
214 static char **arg_syscall_whitelist = NULL;
215 static char **arg_syscall_blacklist = NULL;
216
217 static void help(void) {
218 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
219 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
220 " -h --help Show this help\n"
221 " --version Print version string\n"
222 " -q --quiet Do not show status information\n"
223 " -D --directory=PATH Root directory for the container\n"
224 " --template=PATH Initialize root directory from template directory,\n"
225 " if missing\n"
226 " -x --ephemeral Run container with snapshot of root directory, and\n"
227 " remove it after exit\n"
228 " -i --image=PATH File system device or disk image for the container\n"
229 " --root-hash=HASH Specify verity root hash\n"
230 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
231 " -b --boot Boot up full system (i.e. invoke init)\n"
232 " --chdir=PATH Set working directory in the container\n"
233 " --pivot-root=PATH[:PATH]\n"
234 " Pivot root to given directory in the container\n"
235 " -u --user=USER Run the command under specified user or uid\n"
236 " -M --machine=NAME Set the machine name for the container\n"
237 " --uuid=UUID Set a specific machine UUID for the container\n"
238 " -S --slice=SLICE Place the container in the specified slice\n"
239 " --property=NAME=VALUE Set scope unit property\n"
240 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
241 " --private-users[=UIDBASE[:NUIDS]]\n"
242 " Similar, but with user configured UID/GID range\n"
243 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
244 " --private-network Disable network in container\n"
245 " --network-interface=INTERFACE\n"
246 " Assign an existing network interface to the\n"
247 " container\n"
248 " --network-macvlan=INTERFACE\n"
249 " Create a macvlan network interface based on an\n"
250 " existing network interface to the container\n"
251 " --network-ipvlan=INTERFACE\n"
252 " Create a ipvlan network interface based on an\n"
253 " existing network interface to the container\n"
254 " -n --network-veth Add a virtual Ethernet connection between host\n"
255 " and container\n"
256 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
257 " Add an additional virtual Ethernet link between\n"
258 " host and container\n"
259 " --network-bridge=INTERFACE\n"
260 " Add a virtual Ethernet connection to the container\n"
261 " and attach it to an existing bridge on the host\n"
262 " --network-zone=NAME Similar, but attach the new interface to an\n"
263 " an automatically managed bridge interface\n"
264 " --network-namespace-path=PATH\n"
265 " Set network namespace to the one represented by\n"
266 " the specified kernel namespace file node\n"
267 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
268 " Expose a container IP port on the host\n"
269 " -Z --selinux-context=SECLABEL\n"
270 " Set the SELinux security context to be used by\n"
271 " processes in the container\n"
272 " -L --selinux-apifs-context=SECLABEL\n"
273 " Set the SELinux security context to be used by\n"
274 " API/tmpfs file systems in the container\n"
275 " --capability=CAP In addition to the default, retain specified\n"
276 " capability\n"
277 " --drop-capability=CAP Drop the specified capability from the default set\n"
278 " --system-call-filter=LIST|~LIST\n"
279 " Permit/prohibit specific system calls\n"
280 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
281 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
282 " host, try-guest, try-host\n"
283 " -j Equivalent to --link-journal=try-guest\n"
284 " --read-only Mount the root directory read-only\n"
285 " --bind=PATH[:PATH[:OPTIONS]]\n"
286 " Bind mount a file or directory from the host into\n"
287 " the container\n"
288 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
289 " Similar, but creates a read-only bind mount\n"
290 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
291 " --overlay=PATH[:PATH...]:PATH\n"
292 " Create an overlay mount from the host to \n"
293 " the container\n"
294 " --overlay-ro=PATH[:PATH...]:PATH\n"
295 " Similar, but creates a read-only overlay mount\n"
296 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
297 " --register=BOOLEAN Register container as machine\n"
298 " --keep-unit Do not register a scope for the machine, reuse\n"
299 " the service unit nspawn is running in\n"
300 " --volatile[=MODE] Run the system in volatile mode\n"
301 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
302 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
303 , program_invocation_short_name);
304 }
305
306 static int custom_mount_check_all(void) {
307 unsigned i;
308
309 for (i = 0; i < arg_n_custom_mounts; i++) {
310 CustomMount *m = &arg_custom_mounts[i];
311
312 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
313
314 if (arg_userns_chown) {
315 log_error("--private-users-chown may not be combined with custom root mounts.");
316 return -EINVAL;
317 } else if (arg_uid_shift == UID_INVALID) {
318 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
319 return -EINVAL;
320 }
321 }
322 }
323
324 return 0;
325 }
326
327 static int detect_unified_cgroup_hierarchy_from_environment(void) {
328 const char *e;
329 int r;
330
331 /* Allow the user to control whether the unified hierarchy is used */
332 e = getenv("UNIFIED_CGROUP_HIERARCHY");
333 if (e) {
334 r = parse_boolean(e);
335 if (r < 0)
336 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
337 if (r > 0)
338 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
339 else
340 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
341 }
342
343 return 0;
344 }
345
346 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
347 int r;
348
349 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
350 * image actually supports. */
351 r = cg_all_unified();
352 if (r < 0)
353 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
354 if (r > 0) {
355 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
356 * routine only detects 231, so we'll have a false negative here for 230. */
357 r = systemd_installation_has_version(directory, 230);
358 if (r < 0)
359 return log_error_errno(r, "Failed to determine systemd version in container: %m");
360 if (r > 0)
361 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
362 else
363 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
364 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
365 /* Mixed cgroup hierarchy support was added in 233 */
366 r = systemd_installation_has_version(directory, 233);
367 if (r < 0)
368 return log_error_errno(r, "Failed to determine systemd version in container: %m");
369 if (r > 0)
370 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
371 else
372 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
373 } else
374 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
375
376 log_debug("Using %s hierarchy for container.",
377 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
378 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
379
380 return 0;
381 }
382
383 static void parse_share_ns_env(const char *name, unsigned long ns_flag) {
384 int r;
385
386 r = getenv_bool(name);
387 if (r == -ENXIO)
388 return;
389 if (r < 0)
390 log_warning_errno(r, "Failed to parse %s from environment, defaulting to false.", name);
391 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
392 }
393
394 static void parse_mount_settings_env(void) {
395 int r;
396 const char *e;
397
398 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
399 if (!e)
400 return;
401
402 if (streq(e, "network")) {
403 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
404 return;
405 }
406
407 r = parse_boolean(e);
408 if (r < 0) {
409 log_warning_errno(r, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
410 return;
411 }
412
413 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
414 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
415 }
416
417 static int parse_argv(int argc, char *argv[]) {
418
419 enum {
420 ARG_VERSION = 0x100,
421 ARG_PRIVATE_NETWORK,
422 ARG_UUID,
423 ARG_READ_ONLY,
424 ARG_CAPABILITY,
425 ARG_DROP_CAPABILITY,
426 ARG_LINK_JOURNAL,
427 ARG_BIND,
428 ARG_BIND_RO,
429 ARG_TMPFS,
430 ARG_OVERLAY,
431 ARG_OVERLAY_RO,
432 ARG_SHARE_SYSTEM,
433 ARG_REGISTER,
434 ARG_KEEP_UNIT,
435 ARG_NETWORK_INTERFACE,
436 ARG_NETWORK_MACVLAN,
437 ARG_NETWORK_IPVLAN,
438 ARG_NETWORK_BRIDGE,
439 ARG_NETWORK_ZONE,
440 ARG_NETWORK_VETH_EXTRA,
441 ARG_NETWORK_NAMESPACE_PATH,
442 ARG_PERSONALITY,
443 ARG_VOLATILE,
444 ARG_TEMPLATE,
445 ARG_PROPERTY,
446 ARG_PRIVATE_USERS,
447 ARG_KILL_SIGNAL,
448 ARG_SETTINGS,
449 ARG_CHDIR,
450 ARG_PIVOT_ROOT,
451 ARG_PRIVATE_USERS_CHOWN,
452 ARG_NOTIFY_READY,
453 ARG_ROOT_HASH,
454 ARG_SYSTEM_CALL_FILTER,
455 };
456
457 static const struct option options[] = {
458 { "help", no_argument, NULL, 'h' },
459 { "version", no_argument, NULL, ARG_VERSION },
460 { "directory", required_argument, NULL, 'D' },
461 { "template", required_argument, NULL, ARG_TEMPLATE },
462 { "ephemeral", no_argument, NULL, 'x' },
463 { "user", required_argument, NULL, 'u' },
464 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
465 { "as-pid2", no_argument, NULL, 'a' },
466 { "boot", no_argument, NULL, 'b' },
467 { "uuid", required_argument, NULL, ARG_UUID },
468 { "read-only", no_argument, NULL, ARG_READ_ONLY },
469 { "capability", required_argument, NULL, ARG_CAPABILITY },
470 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
471 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
472 { "bind", required_argument, NULL, ARG_BIND },
473 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
474 { "tmpfs", required_argument, NULL, ARG_TMPFS },
475 { "overlay", required_argument, NULL, ARG_OVERLAY },
476 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
477 { "machine", required_argument, NULL, 'M' },
478 { "slice", required_argument, NULL, 'S' },
479 { "setenv", required_argument, NULL, 'E' },
480 { "selinux-context", required_argument, NULL, 'Z' },
481 { "selinux-apifs-context", required_argument, NULL, 'L' },
482 { "quiet", no_argument, NULL, 'q' },
483 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
484 { "register", required_argument, NULL, ARG_REGISTER },
485 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
486 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
487 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
488 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
489 { "network-veth", no_argument, NULL, 'n' },
490 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
491 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
492 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
493 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
494 { "personality", required_argument, NULL, ARG_PERSONALITY },
495 { "image", required_argument, NULL, 'i' },
496 { "volatile", optional_argument, NULL, ARG_VOLATILE },
497 { "port", required_argument, NULL, 'p' },
498 { "property", required_argument, NULL, ARG_PROPERTY },
499 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
500 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
501 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
502 { "settings", required_argument, NULL, ARG_SETTINGS },
503 { "chdir", required_argument, NULL, ARG_CHDIR },
504 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
505 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
506 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
507 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
508 {}
509 };
510
511 int c, r;
512 const char *p, *e;
513 uint64_t plus = 0, minus = 0;
514 bool mask_all_settings = false, mask_no_settings = false;
515
516 assert(argc >= 0);
517 assert(argv);
518
519 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options, NULL)) >= 0)
520
521 switch (c) {
522
523 case 'h':
524 help();
525 return 0;
526
527 case ARG_VERSION:
528 return version();
529
530 case 'D':
531 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
532 if (r < 0)
533 return r;
534 break;
535
536 case ARG_TEMPLATE:
537 r = parse_path_argument_and_warn(optarg, false, &arg_template);
538 if (r < 0)
539 return r;
540 break;
541
542 case 'i':
543 r = parse_path_argument_and_warn(optarg, false, &arg_image);
544 if (r < 0)
545 return r;
546 break;
547
548 case 'x':
549 arg_ephemeral = true;
550 break;
551
552 case 'u':
553 r = free_and_strdup(&arg_user, optarg);
554 if (r < 0)
555 return log_oom();
556
557 arg_settings_mask |= SETTING_USER;
558 break;
559
560 case ARG_NETWORK_ZONE: {
561 char *j;
562
563 j = strappend("vz-", optarg);
564 if (!j)
565 return log_oom();
566
567 if (!ifname_valid(j)) {
568 log_error("Network zone name not valid: %s", j);
569 free(j);
570 return -EINVAL;
571 }
572
573 free(arg_network_zone);
574 arg_network_zone = j;
575
576 arg_network_veth = true;
577 arg_private_network = true;
578 arg_settings_mask |= SETTING_NETWORK;
579 break;
580 }
581
582 case ARG_NETWORK_BRIDGE:
583
584 if (!ifname_valid(optarg)) {
585 log_error("Bridge interface name not valid: %s", optarg);
586 return -EINVAL;
587 }
588
589 r = free_and_strdup(&arg_network_bridge, optarg);
590 if (r < 0)
591 return log_oom();
592
593 _fallthrough_;
594 case 'n':
595 arg_network_veth = true;
596 arg_private_network = true;
597 arg_settings_mask |= SETTING_NETWORK;
598 break;
599
600 case ARG_NETWORK_VETH_EXTRA:
601 r = veth_extra_parse(&arg_network_veth_extra, optarg);
602 if (r < 0)
603 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
604
605 arg_private_network = true;
606 arg_settings_mask |= SETTING_NETWORK;
607 break;
608
609 case ARG_NETWORK_INTERFACE:
610
611 if (!ifname_valid(optarg)) {
612 log_error("Network interface name not valid: %s", optarg);
613 return -EINVAL;
614 }
615
616 if (strv_extend(&arg_network_interfaces, optarg) < 0)
617 return log_oom();
618
619 arg_private_network = true;
620 arg_settings_mask |= SETTING_NETWORK;
621 break;
622
623 case ARG_NETWORK_MACVLAN:
624
625 if (!ifname_valid(optarg)) {
626 log_error("MACVLAN network interface name not valid: %s", optarg);
627 return -EINVAL;
628 }
629
630 if (strv_extend(&arg_network_macvlan, optarg) < 0)
631 return log_oom();
632
633 arg_private_network = true;
634 arg_settings_mask |= SETTING_NETWORK;
635 break;
636
637 case ARG_NETWORK_IPVLAN:
638
639 if (!ifname_valid(optarg)) {
640 log_error("IPVLAN network interface name not valid: %s", optarg);
641 return -EINVAL;
642 }
643
644 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
645 return log_oom();
646
647 _fallthrough_;
648 case ARG_PRIVATE_NETWORK:
649 arg_private_network = true;
650 arg_settings_mask |= SETTING_NETWORK;
651 break;
652
653 case ARG_NETWORK_NAMESPACE_PATH:
654 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
655 if (r < 0)
656 return r;
657
658 break;
659
660 case 'b':
661 if (arg_start_mode == START_PID2) {
662 log_error("--boot and --as-pid2 may not be combined.");
663 return -EINVAL;
664 }
665
666 arg_start_mode = START_BOOT;
667 arg_settings_mask |= SETTING_START_MODE;
668 break;
669
670 case 'a':
671 if (arg_start_mode == START_BOOT) {
672 log_error("--boot and --as-pid2 may not be combined.");
673 return -EINVAL;
674 }
675
676 arg_start_mode = START_PID2;
677 arg_settings_mask |= SETTING_START_MODE;
678 break;
679
680 case ARG_UUID:
681 r = sd_id128_from_string(optarg, &arg_uuid);
682 if (r < 0)
683 return log_error_errno(r, "Invalid UUID: %s", optarg);
684
685 if (sd_id128_is_null(arg_uuid)) {
686 log_error("Machine UUID may not be all zeroes.");
687 return -EINVAL;
688 }
689
690 arg_settings_mask |= SETTING_MACHINE_ID;
691 break;
692
693 case 'S':
694 arg_slice = optarg;
695 break;
696
697 case 'M':
698 if (isempty(optarg))
699 arg_machine = mfree(arg_machine);
700 else {
701 if (!machine_name_is_valid(optarg)) {
702 log_error("Invalid machine name: %s", optarg);
703 return -EINVAL;
704 }
705
706 r = free_and_strdup(&arg_machine, optarg);
707 if (r < 0)
708 return log_oom();
709 }
710 break;
711
712 case 'Z':
713 arg_selinux_context = optarg;
714 break;
715
716 case 'L':
717 arg_selinux_apifs_context = optarg;
718 break;
719
720 case ARG_READ_ONLY:
721 arg_read_only = true;
722 arg_settings_mask |= SETTING_READ_ONLY;
723 break;
724
725 case ARG_CAPABILITY:
726 case ARG_DROP_CAPABILITY: {
727 p = optarg;
728 for (;;) {
729 _cleanup_free_ char *t = NULL;
730
731 r = extract_first_word(&p, &t, ",", 0);
732 if (r < 0)
733 return log_error_errno(r, "Failed to parse capability %s.", t);
734
735 if (r == 0)
736 break;
737
738 if (streq(t, "all")) {
739 if (c == ARG_CAPABILITY)
740 plus = (uint64_t) -1;
741 else
742 minus = (uint64_t) -1;
743 } else {
744 int cap;
745
746 cap = capability_from_name(t);
747 if (cap < 0) {
748 log_error("Failed to parse capability %s.", t);
749 return -EINVAL;
750 }
751
752 if (c == ARG_CAPABILITY)
753 plus |= 1ULL << (uint64_t) cap;
754 else
755 minus |= 1ULL << (uint64_t) cap;
756 }
757 }
758
759 arg_settings_mask |= SETTING_CAPABILITY;
760 break;
761 }
762
763 case 'j':
764 arg_link_journal = LINK_GUEST;
765 arg_link_journal_try = true;
766 break;
767
768 case ARG_LINK_JOURNAL:
769 if (streq(optarg, "auto")) {
770 arg_link_journal = LINK_AUTO;
771 arg_link_journal_try = false;
772 } else if (streq(optarg, "no")) {
773 arg_link_journal = LINK_NO;
774 arg_link_journal_try = false;
775 } else if (streq(optarg, "guest")) {
776 arg_link_journal = LINK_GUEST;
777 arg_link_journal_try = false;
778 } else if (streq(optarg, "host")) {
779 arg_link_journal = LINK_HOST;
780 arg_link_journal_try = false;
781 } else if (streq(optarg, "try-guest")) {
782 arg_link_journal = LINK_GUEST;
783 arg_link_journal_try = true;
784 } else if (streq(optarg, "try-host")) {
785 arg_link_journal = LINK_HOST;
786 arg_link_journal_try = true;
787 } else {
788 log_error("Failed to parse link journal mode %s", optarg);
789 return -EINVAL;
790 }
791
792 break;
793
794 case ARG_BIND:
795 case ARG_BIND_RO:
796 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
797 if (r < 0)
798 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
799
800 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
801 break;
802
803 case ARG_TMPFS:
804 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
805 if (r < 0)
806 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
807
808 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
809 break;
810
811 case ARG_OVERLAY:
812 case ARG_OVERLAY_RO:
813 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
814 if (r == -EADDRNOTAVAIL)
815 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
816 if (r < 0)
817 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
818
819 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
820 break;
821
822 case 'E': {
823 char **n;
824
825 if (!env_assignment_is_valid(optarg)) {
826 log_error("Environment variable assignment '%s' is not valid.", optarg);
827 return -EINVAL;
828 }
829
830 n = strv_env_set(arg_setenv, optarg);
831 if (!n)
832 return log_oom();
833
834 strv_free(arg_setenv);
835 arg_setenv = n;
836
837 arg_settings_mask |= SETTING_ENVIRONMENT;
838 break;
839 }
840
841 case 'q':
842 arg_quiet = true;
843 break;
844
845 case ARG_SHARE_SYSTEM:
846 /* We don't officially support this anymore, except for compat reasons. People should use the
847 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
848 arg_clone_ns_flags = 0;
849 break;
850
851 case ARG_REGISTER:
852 r = parse_boolean(optarg);
853 if (r < 0) {
854 log_error("Failed to parse --register= argument: %s", optarg);
855 return r;
856 }
857
858 arg_register = r;
859 break;
860
861 case ARG_KEEP_UNIT:
862 arg_keep_unit = true;
863 break;
864
865 case ARG_PERSONALITY:
866
867 arg_personality = personality_from_string(optarg);
868 if (arg_personality == PERSONALITY_INVALID) {
869 log_error("Unknown or unsupported personality '%s'.", optarg);
870 return -EINVAL;
871 }
872
873 arg_settings_mask |= SETTING_PERSONALITY;
874 break;
875
876 case ARG_VOLATILE:
877
878 if (!optarg)
879 arg_volatile_mode = VOLATILE_YES;
880 else {
881 VolatileMode m;
882
883 m = volatile_mode_from_string(optarg);
884 if (m < 0) {
885 log_error("Failed to parse --volatile= argument: %s", optarg);
886 return -EINVAL;
887 } else
888 arg_volatile_mode = m;
889 }
890
891 arg_settings_mask |= SETTING_VOLATILE_MODE;
892 break;
893
894 case 'p':
895 r = expose_port_parse(&arg_expose_ports, optarg);
896 if (r == -EEXIST)
897 return log_error_errno(r, "Duplicate port specification: %s", optarg);
898 if (r < 0)
899 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
900
901 arg_settings_mask |= SETTING_EXPOSE_PORTS;
902 break;
903
904 case ARG_PROPERTY:
905 if (strv_extend(&arg_property, optarg) < 0)
906 return log_oom();
907
908 break;
909
910 case ARG_PRIVATE_USERS: {
911 int boolean = -1;
912
913 if (!optarg)
914 boolean = true;
915 else if (!in_charset(optarg, DIGITS))
916 /* do *not* parse numbers as booleans */
917 boolean = parse_boolean(optarg);
918
919 if (boolean == false) {
920 /* no: User namespacing off */
921 arg_userns_mode = USER_NAMESPACE_NO;
922 arg_uid_shift = UID_INVALID;
923 arg_uid_range = UINT32_C(0x10000);
924 } else if (boolean == true) {
925 /* yes: User namespacing on, UID range is read from root dir */
926 arg_userns_mode = USER_NAMESPACE_FIXED;
927 arg_uid_shift = UID_INVALID;
928 arg_uid_range = UINT32_C(0x10000);
929 } else if (streq(optarg, "pick")) {
930 /* pick: User namespacing on, UID range is picked randomly */
931 arg_userns_mode = USER_NAMESPACE_PICK;
932 arg_uid_shift = UID_INVALID;
933 arg_uid_range = UINT32_C(0x10000);
934 } else {
935 _cleanup_free_ char *buffer = NULL;
936 const char *range, *shift;
937
938 /* anything else: User namespacing on, UID range is explicitly configured */
939
940 range = strchr(optarg, ':');
941 if (range) {
942 buffer = strndup(optarg, range - optarg);
943 if (!buffer)
944 return log_oom();
945 shift = buffer;
946
947 range++;
948 r = safe_atou32(range, &arg_uid_range);
949 if (r < 0)
950 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
951 } else
952 shift = optarg;
953
954 r = parse_uid(shift, &arg_uid_shift);
955 if (r < 0)
956 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
957
958 arg_userns_mode = USER_NAMESPACE_FIXED;
959 }
960
961 if (arg_uid_range <= 0) {
962 log_error("UID range cannot be 0.");
963 return -EINVAL;
964 }
965
966 arg_settings_mask |= SETTING_USERNS;
967 break;
968 }
969
970 case 'U':
971 if (userns_supported()) {
972 arg_userns_mode = USER_NAMESPACE_PICK;
973 arg_uid_shift = UID_INVALID;
974 arg_uid_range = UINT32_C(0x10000);
975
976 arg_settings_mask |= SETTING_USERNS;
977 }
978
979 break;
980
981 case ARG_PRIVATE_USERS_CHOWN:
982 arg_userns_chown = true;
983
984 arg_settings_mask |= SETTING_USERNS;
985 break;
986
987 case ARG_KILL_SIGNAL:
988 arg_kill_signal = signal_from_string_try_harder(optarg);
989 if (arg_kill_signal < 0) {
990 log_error("Cannot parse signal: %s", optarg);
991 return -EINVAL;
992 }
993
994 arg_settings_mask |= SETTING_KILL_SIGNAL;
995 break;
996
997 case ARG_SETTINGS:
998
999 /* no → do not read files
1000 * yes → read files, do not override cmdline, trust only subset
1001 * override → read files, override cmdline, trust only subset
1002 * trusted → read files, do not override cmdline, trust all
1003 */
1004
1005 r = parse_boolean(optarg);
1006 if (r < 0) {
1007 if (streq(optarg, "trusted")) {
1008 mask_all_settings = false;
1009 mask_no_settings = false;
1010 arg_settings_trusted = true;
1011
1012 } else if (streq(optarg, "override")) {
1013 mask_all_settings = false;
1014 mask_no_settings = true;
1015 arg_settings_trusted = -1;
1016 } else
1017 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1018 } else if (r > 0) {
1019 /* yes */
1020 mask_all_settings = false;
1021 mask_no_settings = false;
1022 arg_settings_trusted = -1;
1023 } else {
1024 /* no */
1025 mask_all_settings = true;
1026 mask_no_settings = false;
1027 arg_settings_trusted = false;
1028 }
1029
1030 break;
1031
1032 case ARG_CHDIR:
1033 if (!path_is_absolute(optarg)) {
1034 log_error("Working directory %s is not an absolute path.", optarg);
1035 return -EINVAL;
1036 }
1037
1038 r = free_and_strdup(&arg_chdir, optarg);
1039 if (r < 0)
1040 return log_oom();
1041
1042 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1043 break;
1044
1045 case ARG_PIVOT_ROOT:
1046 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1047 if (r < 0)
1048 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1049
1050 arg_settings_mask |= SETTING_PIVOT_ROOT;
1051 break;
1052
1053 case ARG_NOTIFY_READY:
1054 r = parse_boolean(optarg);
1055 if (r < 0) {
1056 log_error("%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1057 return -EINVAL;
1058 }
1059 arg_notify_ready = r;
1060 arg_settings_mask |= SETTING_NOTIFY_READY;
1061 break;
1062
1063 case ARG_ROOT_HASH: {
1064 void *k;
1065 size_t l;
1066
1067 r = unhexmem(optarg, strlen(optarg), &k, &l);
1068 if (r < 0)
1069 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1070 if (l < sizeof(sd_id128_t)) {
1071 log_error("Root hash must be at least 128bit long: %s", optarg);
1072 free(k);
1073 return -EINVAL;
1074 }
1075
1076 free(arg_root_hash);
1077 arg_root_hash = k;
1078 arg_root_hash_size = l;
1079 break;
1080 }
1081
1082 case ARG_SYSTEM_CALL_FILTER: {
1083 bool negative;
1084 const char *items;
1085
1086 negative = optarg[0] == '~';
1087 items = negative ? optarg + 1 : optarg;
1088
1089 for (;;) {
1090 _cleanup_free_ char *word = NULL;
1091
1092 r = extract_first_word(&items, &word, NULL, 0);
1093 if (r == 0)
1094 break;
1095 if (r == -ENOMEM)
1096 return log_oom();
1097 if (r < 0)
1098 return log_error_errno(r, "Failed to parse system call filter: %m");
1099
1100 if (negative)
1101 r = strv_extend(&arg_syscall_blacklist, word);
1102 else
1103 r = strv_extend(&arg_syscall_whitelist, word);
1104 if (r < 0)
1105 return log_oom();
1106 }
1107
1108 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1109 break;
1110 }
1111
1112 case '?':
1113 return -EINVAL;
1114
1115 default:
1116 assert_not_reached("Unhandled option");
1117 }
1118
1119 /* If --network-namespace-path is given with any other network-related option,
1120 * we need to error out, to avoid conflicts between different network options. */
1121 if (arg_network_namespace_path &&
1122 (arg_network_interfaces || arg_network_macvlan ||
1123 arg_network_ipvlan || arg_network_veth_extra ||
1124 arg_network_bridge || arg_network_zone ||
1125 arg_network_veth || arg_private_network)) {
1126 log_error("--network-namespace-path cannot be combined with other network options.");
1127 return -EINVAL;
1128 }
1129
1130 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
1131 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
1132 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
1133 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
1134
1135 if (arg_userns_mode != USER_NAMESPACE_NO)
1136 arg_mount_settings |= MOUNT_USE_USERNS;
1137
1138 if (arg_private_network)
1139 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1140
1141 parse_mount_settings_env();
1142
1143 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1144 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1145 arg_register = false;
1146 if (arg_start_mode != START_PID1) {
1147 log_error("--boot cannot be used without namespacing.");
1148 return -EINVAL;
1149 }
1150 }
1151
1152 if (arg_userns_mode == USER_NAMESPACE_PICK)
1153 arg_userns_chown = true;
1154
1155 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0) {
1156 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1157 * The latter is not technically a user session, but we don't need to labour the point. */
1158 log_error("--keep-unit --register=yes may not be used when invoked from a user session.");
1159 return -EINVAL;
1160 }
1161
1162 if (arg_directory && arg_image) {
1163 log_error("--directory= and --image= may not be combined.");
1164 return -EINVAL;
1165 }
1166
1167 if (arg_template && arg_image) {
1168 log_error("--template= and --image= may not be combined.");
1169 return -EINVAL;
1170 }
1171
1172 if (arg_ephemeral && arg_template && !arg_directory) {
1173 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1174 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1175 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1176 * --directory=". */
1177
1178 arg_directory = TAKE_PTR(arg_template);
1179 }
1180
1181 if (arg_template && !(arg_directory || arg_machine)) {
1182 log_error("--template= needs --directory= or --machine=.");
1183 return -EINVAL;
1184 }
1185
1186 if (arg_ephemeral && arg_template) {
1187 log_error("--ephemeral and --template= may not be combined.");
1188 return -EINVAL;
1189 }
1190
1191 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
1192 log_error("--ephemeral and --link-journal= may not be combined.");
1193 return -EINVAL;
1194 }
1195
1196 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported()) {
1197 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1198 return -EOPNOTSUPP;
1199 }
1200
1201 if (arg_userns_chown && arg_read_only) {
1202 log_error("--read-only and --private-users-chown may not be combined.");
1203 return -EINVAL;
1204 }
1205
1206 if (arg_network_bridge && arg_network_zone) {
1207 log_error("--network-bridge= and --network-zone= may not be combined.");
1208 return -EINVAL;
1209 }
1210
1211 if (argc > optind) {
1212 arg_parameters = strv_copy(argv + optind);
1213 if (!arg_parameters)
1214 return log_oom();
1215
1216 arg_settings_mask |= SETTING_START_MODE;
1217 }
1218
1219 /* Load all settings from .nspawn files */
1220 if (mask_no_settings)
1221 arg_settings_mask = 0;
1222
1223 /* Don't load any settings from .nspawn files */
1224 if (mask_all_settings)
1225 arg_settings_mask = _SETTINGS_MASK_ALL;
1226
1227 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
1228
1229 r = cg_unified_flush();
1230 if (r < 0)
1231 return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
1232
1233 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1234 if (e)
1235 arg_container_service_name = e;
1236
1237 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1238 if (r < 0)
1239 arg_use_cgns = cg_ns_supported();
1240 else
1241 arg_use_cgns = r;
1242
1243 r = custom_mount_check_all();
1244 if (r < 0)
1245 return r;
1246
1247 return 1;
1248 }
1249
1250 static int verify_arguments(void) {
1251 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network) {
1252 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1253 return -EINVAL;
1254 }
1255
1256 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO)) {
1257 log_error("Cannot combine --private-users with read-write mounts.");
1258 return -EINVAL;
1259 }
1260
1261 if (arg_volatile_mode != VOLATILE_NO && arg_read_only) {
1262 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1263 return -EINVAL;
1264 }
1265
1266 if (arg_expose_ports && !arg_private_network) {
1267 log_error("Cannot use --port= without private networking.");
1268 return -EINVAL;
1269 }
1270
1271 #if ! HAVE_LIBIPTC
1272 if (arg_expose_ports) {
1273 log_error("--port= is not supported, compiled without libiptc support.");
1274 return -EOPNOTSUPP;
1275 }
1276 #endif
1277
1278 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1279 arg_kill_signal = SIGRTMIN+3;
1280
1281 return 0;
1282 }
1283
1284 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1285 assert(p);
1286
1287 if (arg_userns_mode == USER_NAMESPACE_NO)
1288 return 0;
1289
1290 if (uid == UID_INVALID && gid == GID_INVALID)
1291 return 0;
1292
1293 if (uid != UID_INVALID) {
1294 uid += arg_uid_shift;
1295
1296 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1297 return -EOVERFLOW;
1298 }
1299
1300 if (gid != GID_INVALID) {
1301 gid += (gid_t) arg_uid_shift;
1302
1303 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1304 return -EOVERFLOW;
1305 }
1306
1307 if (lchown(p, uid, gid) < 0)
1308 return -errno;
1309
1310 return 0;
1311 }
1312
1313 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1314 const char *q;
1315 int r;
1316
1317 q = prefix_roota(root, path);
1318 r = mkdir_errno_wrapper(q, mode);
1319 if (r == -EEXIST)
1320 return 0;
1321 if (r < 0)
1322 return r;
1323
1324 return userns_lchown(q, uid, gid);
1325 }
1326
1327 static int setup_timezone(const char *dest) {
1328 _cleanup_free_ char *p = NULL, *q = NULL;
1329 const char *where, *check, *what;
1330 char *z, *y;
1331 int r;
1332
1333 assert(dest);
1334
1335 /* Fix the timezone, if possible */
1336 r = readlink_malloc("/etc/localtime", &p);
1337 if (r < 0) {
1338 log_warning("host's /etc/localtime is not a symlink, not updating container timezone.");
1339 /* to handle warning, delete /etc/localtime and replace it
1340 * with a symbolic link to a time zone data file.
1341 *
1342 * Example:
1343 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1344 */
1345 return 0;
1346 }
1347
1348 z = path_startswith(p, "../usr/share/zoneinfo/");
1349 if (!z)
1350 z = path_startswith(p, "/usr/share/zoneinfo/");
1351 if (!z) {
1352 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1353 return 0;
1354 }
1355
1356 where = prefix_roota(dest, "/etc/localtime");
1357 r = readlink_malloc(where, &q);
1358 if (r >= 0) {
1359 y = path_startswith(q, "../usr/share/zoneinfo/");
1360 if (!y)
1361 y = path_startswith(q, "/usr/share/zoneinfo/");
1362
1363 /* Already pointing to the right place? Then do nothing .. */
1364 if (y && streq(y, z))
1365 return 0;
1366 }
1367
1368 check = strjoina("/usr/share/zoneinfo/", z);
1369 check = prefix_roota(dest, check);
1370 if (laccess(check, F_OK) < 0) {
1371 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1372 return 0;
1373 }
1374
1375 if (unlink(where) < 0 && errno != ENOENT) {
1376 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1377 errno,
1378 "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1379 return 0;
1380 }
1381
1382 what = strjoina("../usr/share/zoneinfo/", z);
1383 if (symlink(what, where) < 0) {
1384 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1385 errno,
1386 "Failed to correct timezone of container, ignoring: %m");
1387 return 0;
1388 }
1389
1390 r = userns_lchown(where, 0, 0);
1391 if (r < 0)
1392 return log_warning_errno(r, "Failed to chown /etc/localtime: %m");
1393
1394 return 0;
1395 }
1396
1397 static int resolved_listening(void) {
1398 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1399 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1400 int r;
1401
1402 /* Check if resolved is listening */
1403
1404 r = sd_bus_open_system(&bus);
1405 if (r < 0)
1406 return r;
1407
1408 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1409 if (r <= 0)
1410 return r;
1411
1412 r = sd_bus_get_property_string(bus,
1413 "org.freedesktop.resolve1",
1414 "/org/freedesktop/resolve1",
1415 "org.freedesktop.resolve1.Manager",
1416 "DNSStubListener",
1417 NULL,
1418 &dns_stub_listener_mode);
1419 if (r < 0)
1420 return r;
1421
1422 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1423 }
1424
1425 static int setup_resolv_conf(const char *dest) {
1426 _cleanup_free_ char *resolved = NULL, *etc = NULL;
1427 const char *where;
1428 int r, found;
1429
1430 assert(dest);
1431
1432 if (arg_private_network)
1433 return 0;
1434
1435 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1436 if (r < 0) {
1437 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1438 return 0;
1439 }
1440
1441 where = strjoina(etc, "/resolv.conf");
1442 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1443 if (found < 0) {
1444 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1445 return 0;
1446 }
1447
1448 if (access(STATIC_RESOLV_CONF, F_OK) >= 0 &&
1449 resolved_listening() > 0) {
1450
1451 /* resolved is enabled on the host. In this, case bind mount its static resolv.conf file into the
1452 * container, so that the container can use the host's resolver. Given that network namespacing is
1453 * disabled it's only natural of the container also uses the host's resolver. It also has the big
1454 * advantage that the container will be able to follow the host's DNS server configuration changes
1455 * transparently. */
1456
1457 if (found == 0) /* missing? */
1458 (void) touch(resolved);
1459
1460 r = mount_verbose(LOG_DEBUG, STATIC_RESOLV_CONF, resolved, NULL, MS_BIND, NULL);
1461 if (r >= 0)
1462 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1463 }
1464
1465 /* If that didn't work, let's copy the file */
1466 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0, COPY_REFLINK);
1467 if (r < 0) {
1468 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1469 * resolved or something similar runs inside and the symlink points there.
1470 *
1471 * If the disk image is read-only, there's also no point in complaining.
1472 */
1473 log_full_errno(IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1474 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1475 return 0;
1476 }
1477
1478 r = userns_lchown(where, 0, 0);
1479 if (r < 0)
1480 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1481
1482 return 0;
1483 }
1484
1485 static int setup_boot_id(const char *dest) {
1486 sd_id128_t rnd = SD_ID128_NULL;
1487 const char *from, *to;
1488 int r;
1489
1490 /* Generate a new randomized boot ID, so that each boot-up of
1491 * the container gets a new one */
1492
1493 from = prefix_roota(dest, "/run/proc-sys-kernel-random-boot-id");
1494 to = prefix_roota(dest, "/proc/sys/kernel/random/boot_id");
1495
1496 r = sd_id128_randomize(&rnd);
1497 if (r < 0)
1498 return log_error_errno(r, "Failed to generate random boot id: %m");
1499
1500 r = id128_write(from, ID128_UUID, rnd, false);
1501 if (r < 0)
1502 return log_error_errno(r, "Failed to write boot id: %m");
1503
1504 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1505 if (r >= 0)
1506 r = mount_verbose(LOG_ERR, NULL, to, NULL,
1507 MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1508
1509 (void) unlink(from);
1510 return r;
1511 }
1512
1513 static int copy_devnodes(const char *dest) {
1514
1515 static const char devnodes[] =
1516 "null\0"
1517 "zero\0"
1518 "full\0"
1519 "random\0"
1520 "urandom\0"
1521 "tty\0"
1522 "net/tun\0";
1523
1524 const char *d;
1525 int r = 0;
1526 _cleanup_umask_ mode_t u;
1527
1528 assert(dest);
1529
1530 u = umask(0000);
1531
1532 /* Create /dev/net, so that we can create /dev/net/tun in it */
1533 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1534 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1535
1536 NULSTR_FOREACH(d, devnodes) {
1537 _cleanup_free_ char *from = NULL, *to = NULL;
1538 struct stat st;
1539
1540 from = strappend("/dev/", d);
1541 to = prefix_root(dest, from);
1542
1543 if (stat(from, &st) < 0) {
1544
1545 if (errno != ENOENT)
1546 return log_error_errno(errno, "Failed to stat %s: %m", from);
1547
1548 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1549
1550 log_error("%s is not a char or block device, cannot copy.", from);
1551 return -EIO;
1552
1553 } else {
1554 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1555 /* Explicitly warn the user when /dev is already populated. */
1556 if (errno == EEXIST)
1557 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);
1558 if (errno != EPERM)
1559 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1560
1561 /* Some systems abusively restrict mknod but
1562 * allow bind mounts. */
1563 r = touch(to);
1564 if (r < 0)
1565 return log_error_errno(r, "touch (%s) failed: %m", to);
1566 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1567 if (r < 0)
1568 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1569 }
1570
1571 r = userns_lchown(to, 0, 0);
1572 if (r < 0)
1573 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1574 }
1575 }
1576
1577 return r;
1578 }
1579
1580 static int setup_pts(const char *dest) {
1581 _cleanup_free_ char *options = NULL;
1582 const char *p;
1583 int r;
1584
1585 #if HAVE_SELINUX
1586 if (arg_selinux_apifs_context)
1587 (void) asprintf(&options,
1588 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1589 arg_uid_shift + TTY_GID,
1590 arg_selinux_apifs_context);
1591 else
1592 #endif
1593 (void) asprintf(&options,
1594 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1595 arg_uid_shift + TTY_GID);
1596
1597 if (!options)
1598 return log_oom();
1599
1600 /* Mount /dev/pts itself */
1601 p = prefix_roota(dest, "/dev/pts");
1602 r = mkdir_errno_wrapper(p, 0755);
1603 if (r < 0)
1604 return log_error_errno(r, "Failed to create /dev/pts: %m");
1605
1606 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
1607 if (r < 0)
1608 return r;
1609 r = userns_lchown(p, 0, 0);
1610 if (r < 0)
1611 return log_error_errno(r, "Failed to chown /dev/pts: %m");
1612
1613 /* Create /dev/ptmx symlink */
1614 p = prefix_roota(dest, "/dev/ptmx");
1615 if (symlink("pts/ptmx", p) < 0)
1616 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1617 r = userns_lchown(p, 0, 0);
1618 if (r < 0)
1619 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
1620
1621 /* And fix /dev/pts/ptmx ownership */
1622 p = prefix_roota(dest, "/dev/pts/ptmx");
1623 r = userns_lchown(p, 0, 0);
1624 if (r < 0)
1625 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
1626
1627 return 0;
1628 }
1629
1630 static int setup_dev_console(const char *dest, const char *console) {
1631 _cleanup_umask_ mode_t u;
1632 const char *to;
1633 int r;
1634
1635 assert(dest);
1636 assert(console);
1637
1638 u = umask(0000);
1639
1640 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
1641 if (r < 0)
1642 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1643
1644 /* We need to bind mount the right tty to /dev/console since
1645 * ptys can only exist on pts file systems. To have something
1646 * to bind mount things on we create a empty regular file. */
1647
1648 to = prefix_roota(dest, "/dev/console");
1649 r = touch(to);
1650 if (r < 0)
1651 return log_error_errno(r, "touch() for /dev/console failed: %m");
1652
1653 return mount_verbose(LOG_ERR, console, to, NULL, MS_BIND, NULL);
1654 }
1655
1656 static int setup_keyring(void) {
1657 key_serial_t keyring;
1658
1659 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
1660 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
1661 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
1662 * these system calls let's make sure we don't leak anything into the container. */
1663
1664 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
1665 if (keyring == -1) {
1666 if (errno == ENOSYS)
1667 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
1668 else if (IN_SET(errno, EACCES, EPERM))
1669 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
1670 else
1671 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
1672 }
1673
1674 return 0;
1675 }
1676
1677 static int setup_kmsg(const char *dest, int kmsg_socket) {
1678 const char *from, *to;
1679 _cleanup_umask_ mode_t u;
1680 int fd, r;
1681
1682 assert(kmsg_socket >= 0);
1683
1684 u = umask(0000);
1685
1686 /* We create the kmsg FIFO as /run/kmsg, but immediately
1687 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1688 * on the reading side behave very similar to /proc/kmsg,
1689 * their writing side behaves differently from /dev/kmsg in
1690 * that writing blocks when nothing is reading. In order to
1691 * avoid any problems with containers deadlocking due to this
1692 * we simply make /dev/kmsg unavailable to the container. */
1693 from = prefix_roota(dest, "/run/kmsg");
1694 to = prefix_roota(dest, "/proc/kmsg");
1695
1696 if (mkfifo(from, 0600) < 0)
1697 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
1698 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1699 if (r < 0)
1700 return r;
1701
1702 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
1703 if (fd < 0)
1704 return log_error_errno(errno, "Failed to open fifo: %m");
1705
1706 /* Store away the fd in the socket, so that it stays open as
1707 * long as we run the child */
1708 r = send_one_fd(kmsg_socket, fd, 0);
1709 safe_close(fd);
1710
1711 if (r < 0)
1712 return log_error_errno(r, "Failed to send FIFO fd: %m");
1713
1714 /* And now make the FIFO unavailable as /run/kmsg... */
1715 (void) unlink(from);
1716
1717 return 0;
1718 }
1719
1720 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
1721 union in_addr_union *exposed = userdata;
1722
1723 assert(rtnl);
1724 assert(m);
1725 assert(exposed);
1726
1727 expose_port_execute(rtnl, arg_expose_ports, exposed);
1728 return 0;
1729 }
1730
1731 static int setup_hostname(void) {
1732
1733 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
1734 return 0;
1735
1736 if (sethostname_idempotent(arg_machine) < 0)
1737 return -errno;
1738
1739 return 0;
1740 }
1741
1742 static int setup_journal(const char *directory) {
1743 sd_id128_t this_id;
1744 _cleanup_free_ char *d = NULL;
1745 const char *p, *q;
1746 bool try;
1747 char id[33];
1748 int r;
1749
1750 /* Don't link journals in ephemeral mode */
1751 if (arg_ephemeral)
1752 return 0;
1753
1754 if (arg_link_journal == LINK_NO)
1755 return 0;
1756
1757 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
1758
1759 r = sd_id128_get_machine(&this_id);
1760 if (r < 0)
1761 return log_error_errno(r, "Failed to retrieve machine ID: %m");
1762
1763 if (sd_id128_equal(arg_uuid, this_id)) {
1764 log_full(try ? LOG_WARNING : LOG_ERR,
1765 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
1766 if (try)
1767 return 0;
1768 return -EEXIST;
1769 }
1770
1771 r = userns_mkdir(directory, "/var", 0755, 0, 0);
1772 if (r < 0)
1773 return log_error_errno(r, "Failed to create /var: %m");
1774
1775 r = userns_mkdir(directory, "/var/log", 0755, 0, 0);
1776 if (r < 0)
1777 return log_error_errno(r, "Failed to create /var/log: %m");
1778
1779 r = userns_mkdir(directory, "/var/log/journal", 0755, 0, 0);
1780 if (r < 0)
1781 return log_error_errno(r, "Failed to create /var/log/journal: %m");
1782
1783 (void) sd_id128_to_string(arg_uuid, id);
1784
1785 p = strjoina("/var/log/journal/", id);
1786 q = prefix_roota(directory, p);
1787
1788 if (path_is_mount_point(p, NULL, 0) > 0) {
1789 if (try)
1790 return 0;
1791
1792 log_error("%s: already a mount point, refusing to use for journal", p);
1793 return -EEXIST;
1794 }
1795
1796 if (path_is_mount_point(q, NULL, 0) > 0) {
1797 if (try)
1798 return 0;
1799
1800 log_error("%s: already a mount point, refusing to use for journal", q);
1801 return -EEXIST;
1802 }
1803
1804 r = readlink_and_make_absolute(p, &d);
1805 if (r >= 0) {
1806 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
1807 path_equal(d, q)) {
1808
1809 r = userns_mkdir(directory, p, 0755, 0, 0);
1810 if (r < 0)
1811 log_warning_errno(r, "Failed to create directory %s: %m", q);
1812 return 0;
1813 }
1814
1815 if (unlink(p) < 0)
1816 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
1817 } else if (r == -EINVAL) {
1818
1819 if (arg_link_journal == LINK_GUEST &&
1820 rmdir(p) < 0) {
1821
1822 if (errno == ENOTDIR) {
1823 log_error("%s already exists and is neither a symlink nor a directory", p);
1824 return r;
1825 } else
1826 return log_error_errno(errno, "Failed to remove %s: %m", p);
1827 }
1828 } else if (r != -ENOENT)
1829 return log_error_errno(r, "readlink(%s) failed: %m", p);
1830
1831 if (arg_link_journal == LINK_GUEST) {
1832
1833 if (symlink(q, p) < 0) {
1834 if (try) {
1835 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
1836 return 0;
1837 } else
1838 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
1839 }
1840
1841 r = userns_mkdir(directory, p, 0755, 0, 0);
1842 if (r < 0)
1843 log_warning_errno(r, "Failed to create directory %s: %m", q);
1844 return 0;
1845 }
1846
1847 if (arg_link_journal == LINK_HOST) {
1848 /* don't create parents here — if the host doesn't have
1849 * permanent journal set up, don't force it here */
1850
1851 r = mkdir_errno_wrapper(p, 0755);
1852 if (r < 0 && r != -EEXIST) {
1853 if (try) {
1854 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
1855 return 0;
1856 } else
1857 return log_error_errno(r, "Failed to create %s: %m", p);
1858 }
1859
1860 } else if (access(p, F_OK) < 0)
1861 return 0;
1862
1863 if (dir_is_empty(q) == 0)
1864 log_warning("%s is not empty, proceeding anyway.", q);
1865
1866 r = userns_mkdir(directory, p, 0755, 0, 0);
1867 if (r < 0)
1868 return log_error_errno(r, "Failed to create %s: %m", q);
1869
1870 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
1871 if (r < 0)
1872 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
1873
1874 return 0;
1875 }
1876
1877 static int drop_capabilities(void) {
1878 return capability_bounding_set_drop(arg_caps_retain, false);
1879 }
1880
1881 static int reset_audit_loginuid(void) {
1882 _cleanup_free_ char *p = NULL;
1883 int r;
1884
1885 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
1886 return 0;
1887
1888 r = read_one_line_file("/proc/self/loginuid", &p);
1889 if (r == -ENOENT)
1890 return 0;
1891 if (r < 0)
1892 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
1893
1894 /* Already reset? */
1895 if (streq(p, "4294967295"))
1896 return 0;
1897
1898 r = write_string_file("/proc/self/loginuid", "4294967295", 0);
1899 if (r < 0) {
1900 log_error_errno(r,
1901 "Failed to reset audit login UID. This probably means that your kernel is too\n"
1902 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1903 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1904 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1905 "using systemd-nspawn. Sleeping for 5s... (%m)");
1906
1907 sleep(5);
1908 }
1909
1910 return 0;
1911 }
1912
1913
1914 static int setup_propagate(const char *root) {
1915 const char *p, *q;
1916 int r;
1917
1918 (void) mkdir_p("/run/systemd/nspawn/", 0755);
1919 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
1920 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
1921 (void) mkdir_p(p, 0600);
1922
1923 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
1924 if (r < 0)
1925 return log_error_errno(r, "Failed to create /run/systemd: %m");
1926
1927 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
1928 if (r < 0)
1929 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
1930
1931 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
1932 if (r < 0)
1933 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
1934
1935 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
1936 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
1937 if (r < 0)
1938 return r;
1939
1940 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
1941 if (r < 0)
1942 return r;
1943
1944 /* machined will MS_MOVE into that directory, and that's only
1945 * supported for non-shared mounts. */
1946 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
1947 }
1948
1949 static int setup_machine_id(const char *directory) {
1950 const char *etc_machine_id;
1951 sd_id128_t id;
1952 int r;
1953
1954 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
1955 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
1956 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
1957 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
1958 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
1959 * container behaves nicely). */
1960
1961 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
1962
1963 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
1964 if (r < 0) {
1965 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
1966 return log_error_errno(r, "Failed to read machine ID from container image: %m");
1967
1968 if (sd_id128_is_null(arg_uuid)) {
1969 r = sd_id128_randomize(&arg_uuid);
1970 if (r < 0)
1971 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
1972 }
1973 } else {
1974 if (sd_id128_is_null(id)) {
1975 log_error("Machine ID in container image is zero, refusing.");
1976 return -EINVAL;
1977 }
1978
1979 arg_uuid = id;
1980 }
1981
1982 return 0;
1983 }
1984
1985 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
1986 int r;
1987
1988 assert(directory);
1989
1990 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
1991 return 0;
1992
1993 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
1994 if (r == -EOPNOTSUPP)
1995 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.");
1996 if (r == -EBADE)
1997 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
1998 if (r < 0)
1999 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2000 if (r == 0)
2001 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2002 else
2003 log_debug("Patched directory tree to match UID/GID range.");
2004
2005 return r;
2006 }
2007
2008 /*
2009 * Return values:
2010 * < 0 : wait_for_terminate() failed to get the state of the
2011 * container, the container was terminated by a signal, or
2012 * failed for an unknown reason. No change is made to the
2013 * container argument.
2014 * > 0 : The program executed in the container terminated with an
2015 * error. The exit code of the program executed in the
2016 * container is returned. The container argument has been set
2017 * to CONTAINER_TERMINATED.
2018 * 0 : The container is being rebooted, has been shut down or exited
2019 * successfully. The container argument has been set to either
2020 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2021 *
2022 * That is, success is indicated by a return value of zero, and an
2023 * error is indicated by a non-zero value.
2024 */
2025 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2026 siginfo_t status;
2027 int r;
2028
2029 r = wait_for_terminate(pid, &status);
2030 if (r < 0)
2031 return log_warning_errno(r, "Failed to wait for container: %m");
2032
2033 switch (status.si_code) {
2034
2035 case CLD_EXITED:
2036 if (status.si_status == 0)
2037 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2038 else
2039 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2040
2041 *container = CONTAINER_TERMINATED;
2042 return status.si_status;
2043
2044 case CLD_KILLED:
2045 if (status.si_status == SIGINT) {
2046 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2047 *container = CONTAINER_TERMINATED;
2048 return 0;
2049
2050 } else if (status.si_status == SIGHUP) {
2051 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2052 *container = CONTAINER_REBOOTED;
2053 return 0;
2054 }
2055
2056 _fallthrough_;
2057 case CLD_DUMPED:
2058 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2059 return -EIO;
2060
2061 default:
2062 log_error("Container %s failed due to unknown reason.", arg_machine);
2063 return -EIO;
2064 }
2065 }
2066
2067 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2068 pid_t pid;
2069
2070 pid = PTR_TO_PID(userdata);
2071 if (pid > 0) {
2072 if (kill(pid, arg_kill_signal) >= 0) {
2073 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2074 sd_event_source_set_userdata(s, NULL);
2075 return 0;
2076 }
2077 }
2078
2079 sd_event_exit(sd_event_source_get_event(s), 0);
2080 return 0;
2081 }
2082
2083 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2084 pid_t pid;
2085
2086 assert(s);
2087 assert(ssi);
2088
2089 pid = PTR_TO_PID(userdata);
2090
2091 for (;;) {
2092 siginfo_t si = {};
2093
2094 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2095 return log_error_errno(errno, "Failed to waitid(): %m");
2096 if (si.si_pid == 0) /* No pending children. */
2097 break;
2098 if (si.si_pid == pid) {
2099 /* The main process we care for has exited. Return from
2100 * signal handler but leave the zombie. */
2101 sd_event_exit(sd_event_source_get_event(s), 0);
2102 break;
2103 }
2104
2105 /* Reap all other children. */
2106 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2107 }
2108
2109 return 0;
2110 }
2111
2112 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2113 pid_t pid;
2114
2115 assert(m);
2116
2117 pid = PTR_TO_PID(userdata);
2118
2119 if (arg_kill_signal > 0) {
2120 log_info("Container termination requested. Attempting to halt container.");
2121 (void) kill(pid, arg_kill_signal);
2122 } else {
2123 log_info("Container termination requested. Exiting.");
2124 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2125 }
2126
2127 return 0;
2128 }
2129
2130 static int determine_names(void) {
2131 int r;
2132
2133 if (arg_template && !arg_directory && arg_machine) {
2134
2135 /* If --template= was specified then we should not
2136 * search for a machine, but instead create a new one
2137 * in /var/lib/machine. */
2138
2139 arg_directory = strjoin("/var/lib/machines/", arg_machine);
2140 if (!arg_directory)
2141 return log_oom();
2142 }
2143
2144 if (!arg_image && !arg_directory) {
2145 if (arg_machine) {
2146 _cleanup_(image_unrefp) Image *i = NULL;
2147
2148 r = image_find(arg_machine, &i);
2149 if (r < 0)
2150 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2151 if (r == 0) {
2152 log_error("No image for machine '%s'.", arg_machine);
2153 return -ENOENT;
2154 }
2155
2156 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2157 r = free_and_strdup(&arg_image, i->path);
2158 else
2159 r = free_and_strdup(&arg_directory, i->path);
2160 if (r < 0)
2161 return log_oom();
2162
2163 if (!arg_ephemeral)
2164 arg_read_only = arg_read_only || i->read_only;
2165 } else {
2166 r = safe_getcwd(&arg_directory);
2167 if (r < 0)
2168 return log_error_errno(r, "Failed to determine current directory: %m");
2169 }
2170
2171 if (!arg_directory && !arg_image) {
2172 log_error("Failed to determine path, please use -D or -i.");
2173 return -EINVAL;
2174 }
2175 }
2176
2177 if (!arg_machine) {
2178
2179 if (arg_directory && path_equal(arg_directory, "/"))
2180 arg_machine = gethostname_malloc();
2181 else {
2182 if (arg_image) {
2183 char *e;
2184
2185 arg_machine = strdup(basename(arg_image));
2186
2187 /* Truncate suffix if there is one */
2188 e = endswith(arg_machine, ".raw");
2189 if (e)
2190 *e = 0;
2191 } else
2192 arg_machine = strdup(basename(arg_directory));
2193 }
2194 if (!arg_machine)
2195 return log_oom();
2196
2197 hostname_cleanup(arg_machine);
2198 if (!machine_name_is_valid(arg_machine)) {
2199 log_error("Failed to determine machine name automatically, please use -M.");
2200 return -EINVAL;
2201 }
2202
2203 if (arg_ephemeral) {
2204 char *b;
2205
2206 /* Add a random suffix when this is an
2207 * ephemeral machine, so that we can run many
2208 * instances at once without manually having
2209 * to specify -M each time. */
2210
2211 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2212 return log_oom();
2213
2214 free(arg_machine);
2215 arg_machine = b;
2216 }
2217 }
2218
2219 return 0;
2220 }
2221
2222 static int chase_symlinks_and_update(char **p, unsigned flags) {
2223 char *chased;
2224 int r;
2225
2226 assert(p);
2227
2228 if (!*p)
2229 return 0;
2230
2231 r = chase_symlinks(*p, NULL, flags, &chased);
2232 if (r < 0)
2233 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2234
2235 free_and_replace(*p, chased);
2236 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2237 }
2238
2239 static int determine_uid_shift(const char *directory) {
2240 int r;
2241
2242 if (arg_userns_mode == USER_NAMESPACE_NO) {
2243 arg_uid_shift = 0;
2244 return 0;
2245 }
2246
2247 if (arg_uid_shift == UID_INVALID) {
2248 struct stat st;
2249
2250 r = stat(directory, &st);
2251 if (r < 0)
2252 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2253
2254 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2255
2256 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
2257 log_error("UID and GID base of %s don't match.", directory);
2258 return -EINVAL;
2259 }
2260
2261 arg_uid_range = UINT32_C(0x10000);
2262 }
2263
2264 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
2265 log_error("UID base too high for UID range.");
2266 return -EINVAL;
2267 }
2268
2269 return 0;
2270 }
2271
2272 static int inner_child(
2273 Barrier *barrier,
2274 const char *directory,
2275 bool secondary,
2276 int kmsg_socket,
2277 int rtnl_socket,
2278 FDSet *fds) {
2279
2280 _cleanup_free_ char *home = NULL;
2281 char as_uuid[37];
2282 unsigned n_env = 1;
2283 const char *envp[] = {
2284 "PATH=" DEFAULT_PATH_SPLIT_USR,
2285 NULL, /* container */
2286 NULL, /* TERM */
2287 NULL, /* HOME */
2288 NULL, /* USER */
2289 NULL, /* LOGNAME */
2290 NULL, /* container_uuid */
2291 NULL, /* LISTEN_FDS */
2292 NULL, /* LISTEN_PID */
2293 NULL, /* NOTIFY_SOCKET */
2294 NULL
2295 };
2296 const char *exec_target;
2297
2298 _cleanup_strv_free_ char **env_use = NULL;
2299 int r;
2300
2301 assert(barrier);
2302 assert(directory);
2303 assert(kmsg_socket >= 0);
2304
2305 if (arg_userns_mode != USER_NAMESPACE_NO) {
2306 /* Tell the parent, that it now can write the UID map. */
2307 (void) barrier_place(barrier); /* #1 */
2308
2309 /* Wait until the parent wrote the UID map */
2310 if (!barrier_place_and_sync(barrier)) { /* #2 */
2311 log_error("Parent died too early");
2312 return -ESRCH;
2313 }
2314 }
2315
2316 r = reset_uid_gid();
2317 if (r < 0)
2318 return log_error_errno(r, "Couldn't become new root: %m");
2319
2320 r = mount_all(NULL,
2321 arg_mount_settings | MOUNT_IN_USERNS,
2322 arg_uid_shift,
2323 arg_uid_range,
2324 arg_selinux_apifs_context);
2325 if (r < 0)
2326 return r;
2327
2328 if (!arg_network_namespace_path && arg_private_network) {
2329 r = unshare(CLONE_NEWNET);
2330 if (r < 0)
2331 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2332
2333 /* Tell the parent that it can setup network interfaces. */
2334 (void) barrier_place(barrier); /* #3 */
2335 }
2336
2337 r = mount_sysfs(NULL, arg_mount_settings);
2338 if (r < 0)
2339 return r;
2340
2341 /* Wait until we are cgroup-ified, so that we
2342 * can mount the right cgroup path writable */
2343 if (!barrier_place_and_sync(barrier)) { /* #4 */
2344 log_error("Parent died too early");
2345 return -ESRCH;
2346 }
2347
2348 if (arg_use_cgns && cg_ns_supported()) {
2349 r = unshare(CLONE_NEWCGROUP);
2350 if (r < 0)
2351 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2352 r = mount_cgroups(
2353 "",
2354 arg_unified_cgroup_hierarchy,
2355 arg_userns_mode != USER_NAMESPACE_NO,
2356 arg_uid_shift,
2357 arg_uid_range,
2358 arg_selinux_apifs_context,
2359 true);
2360 if (r < 0)
2361 return r;
2362 } else {
2363 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2364 if (r < 0)
2365 return r;
2366 }
2367
2368 r = setup_boot_id(NULL);
2369 if (r < 0)
2370 return r;
2371
2372 r = setup_kmsg(NULL, kmsg_socket);
2373 if (r < 0)
2374 return r;
2375 kmsg_socket = safe_close(kmsg_socket);
2376
2377 umask(0022);
2378
2379 if (setsid() < 0)
2380 return log_error_errno(errno, "setsid() failed: %m");
2381
2382 if (arg_private_network)
2383 loopback_setup();
2384
2385 if (arg_expose_ports) {
2386 r = expose_port_send_rtnl(rtnl_socket);
2387 if (r < 0)
2388 return r;
2389 rtnl_socket = safe_close(rtnl_socket);
2390 }
2391
2392 r = drop_capabilities();
2393 if (r < 0)
2394 return log_error_errno(r, "drop_capabilities() failed: %m");
2395
2396 setup_hostname();
2397
2398 if (arg_personality != PERSONALITY_INVALID) {
2399 r = safe_personality(arg_personality);
2400 if (r < 0)
2401 return log_error_errno(r, "personality() failed: %m");
2402 } else if (secondary) {
2403 r = safe_personality(PER_LINUX32);
2404 if (r < 0)
2405 return log_error_errno(r, "personality() failed: %m");
2406 }
2407
2408 #if HAVE_SELINUX
2409 if (arg_selinux_context)
2410 if (setexeccon(arg_selinux_context) < 0)
2411 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
2412 #endif
2413
2414 r = change_uid_gid(arg_user, &home);
2415 if (r < 0)
2416 return r;
2417
2418 /* LXC sets container=lxc, so follow the scheme here */
2419 envp[n_env++] = strjoina("container=", arg_container_service_name);
2420
2421 envp[n_env] = strv_find_prefix(environ, "TERM=");
2422 if (envp[n_env])
2423 n_env++;
2424
2425 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
2426 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
2427 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
2428 return log_oom();
2429
2430 assert(!sd_id128_is_null(arg_uuid));
2431
2432 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
2433 return log_oom();
2434
2435 if (fdset_size(fds) > 0) {
2436 r = fdset_cloexec(fds, false);
2437 if (r < 0)
2438 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
2439
2440 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
2441 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
2442 return log_oom();
2443 }
2444 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
2445 return log_oom();
2446
2447 env_use = strv_env_merge(2, envp, arg_setenv);
2448 if (!env_use)
2449 return log_oom();
2450
2451 /* Let the parent know that we are ready and
2452 * wait until the parent is ready with the
2453 * setup, too... */
2454 if (!barrier_place_and_sync(barrier)) { /* #5 */
2455 log_error("Parent died too early");
2456 return -ESRCH;
2457 }
2458
2459 if (arg_chdir)
2460 if (chdir(arg_chdir) < 0)
2461 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
2462
2463 if (arg_start_mode == START_PID2) {
2464 r = stub_pid1(arg_uuid);
2465 if (r < 0)
2466 return r;
2467 }
2468
2469 /* Now, explicitly close the log, so that we
2470 * then can close all remaining fds. Closing
2471 * the log explicitly first has the benefit
2472 * that the logging subsystem knows about it,
2473 * and is thus ready to be reopened should we
2474 * need it again. Note that the other fds
2475 * closed here are at least the locking and
2476 * barrier fds. */
2477 log_close();
2478 (void) fdset_close_others(fds);
2479
2480 if (arg_start_mode == START_BOOT) {
2481 char **a;
2482 size_t m;
2483
2484 /* Automatically search for the init system */
2485
2486 m = strv_length(arg_parameters);
2487 a = newa(char*, m + 2);
2488 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
2489 a[1 + m] = NULL;
2490
2491 a[0] = (char*) "/usr/lib/systemd/systemd";
2492 execve(a[0], a, env_use);
2493
2494 a[0] = (char*) "/lib/systemd/systemd";
2495 execve(a[0], a, env_use);
2496
2497 a[0] = (char*) "/sbin/init";
2498 execve(a[0], a, env_use);
2499
2500 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2501 } else if (!strv_isempty(arg_parameters)) {
2502 exec_target = arg_parameters[0];
2503 execvpe(arg_parameters[0], arg_parameters, env_use);
2504 } else {
2505 if (!arg_chdir)
2506 /* If we cannot change the directory, we'll end up in /, that is expected. */
2507 (void) chdir(home ?: "/root");
2508
2509 execle("/bin/bash", "-bash", NULL, env_use);
2510 execle("/bin/sh", "-sh", NULL, env_use);
2511
2512 exec_target = "/bin/bash, /bin/sh";
2513 }
2514
2515 r = -errno;
2516 (void) log_open();
2517 return log_error_errno(r, "execv(%s) failed: %m", exec_target);
2518 }
2519
2520 static int setup_sd_notify_child(void) {
2521 static const int one = 1;
2522 int fd = -1;
2523 union sockaddr_union sa = {
2524 .sa.sa_family = AF_UNIX,
2525 };
2526 int r;
2527
2528 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2529 if (fd < 0)
2530 return log_error_errno(errno, "Failed to allocate notification socket: %m");
2531
2532 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
2533 (void) unlink(NSPAWN_NOTIFY_SOCKET_PATH);
2534
2535 strncpy(sa.un.sun_path, NSPAWN_NOTIFY_SOCKET_PATH, sizeof(sa.un.sun_path)-1);
2536 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
2537 if (r < 0) {
2538 safe_close(fd);
2539 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
2540 }
2541
2542 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
2543 if (r < 0) {
2544 safe_close(fd);
2545 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
2546 }
2547
2548 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
2549 if (r < 0) {
2550 safe_close(fd);
2551 return log_error_errno(errno, "SO_PASSCRED failed: %m");
2552 }
2553
2554 return fd;
2555 }
2556
2557 static int outer_child(
2558 Barrier *barrier,
2559 const char *directory,
2560 const char *console,
2561 DissectedImage *dissected_image,
2562 bool interactive,
2563 bool secondary,
2564 int pid_socket,
2565 int uuid_socket,
2566 int notify_socket,
2567 int kmsg_socket,
2568 int rtnl_socket,
2569 int uid_shift_socket,
2570 int unified_cgroup_hierarchy_socket,
2571 FDSet *fds,
2572 int netns_fd) {
2573
2574 pid_t pid;
2575 ssize_t l;
2576 int r;
2577 _cleanup_close_ int fd = -1;
2578
2579 assert(barrier);
2580 assert(directory);
2581 assert(console);
2582 assert(pid_socket >= 0);
2583 assert(uuid_socket >= 0);
2584 assert(notify_socket >= 0);
2585 assert(kmsg_socket >= 0);
2586
2587 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
2588 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
2589
2590 if (interactive) {
2591 int terminal;
2592
2593 terminal = open_terminal(console, O_RDWR);
2594 if (terminal < 0)
2595 return log_error_errno(terminal, "Failed to open console: %m");
2596
2597 r = rearrange_stdio(terminal, terminal, terminal); /* invalidates 'terminal' on success and failure */
2598 if (r < 0)
2599 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2600 }
2601
2602 r = reset_audit_loginuid();
2603 if (r < 0)
2604 return r;
2605
2606 /* Mark everything as slave, so that we still
2607 * receive mounts from the real root, but don't
2608 * propagate mounts to the real root. */
2609 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
2610 if (r < 0)
2611 return r;
2612
2613 if (dissected_image) {
2614 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
2615 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
2616 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
2617 * makes sure ESP partitions and userns are compatible. */
2618
2619 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2620 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2621 if (r < 0)
2622 return r;
2623 }
2624
2625 r = determine_uid_shift(directory);
2626 if (r < 0)
2627 return r;
2628
2629 if (arg_userns_mode != USER_NAMESPACE_NO) {
2630 /* Let the parent know which UID shift we read from the image */
2631 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
2632 if (l < 0)
2633 return log_error_errno(errno, "Failed to send UID shift: %m");
2634 if (l != sizeof(arg_uid_shift)) {
2635 log_error("Short write while sending UID shift.");
2636 return -EIO;
2637 }
2638
2639 if (arg_userns_mode == USER_NAMESPACE_PICK) {
2640 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2641 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2642 * not it will pick a different one, and send it back to us. */
2643
2644 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
2645 if (l < 0)
2646 return log_error_errno(errno, "Failed to recv UID shift: %m");
2647 if (l != sizeof(arg_uid_shift)) {
2648 log_error("Short read while receiving UID shift.");
2649 return -EIO;
2650 }
2651 }
2652
2653 log_info("Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
2654 }
2655
2656 if (dissected_image) {
2657 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
2658 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2659 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2660 if (r < 0)
2661 return r;
2662 }
2663
2664 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
2665 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
2666
2667 r = detect_unified_cgroup_hierarchy_from_image(directory);
2668 if (r < 0)
2669 return r;
2670
2671 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
2672 if (l < 0)
2673 return log_error_errno(errno, "Failed to send cgroup mode: %m");
2674 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
2675 log_error("Short write while sending cgroup mode: %m");
2676 return -EIO;
2677 }
2678
2679 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
2680 }
2681
2682 /* Turn directory into bind mount */
2683 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
2684 if (r < 0)
2685 return r;
2686
2687 r = setup_pivot_root(
2688 directory,
2689 arg_pivot_root_new,
2690 arg_pivot_root_old);
2691 if (r < 0)
2692 return r;
2693
2694 r = setup_volatile(
2695 directory,
2696 arg_volatile_mode,
2697 arg_userns_mode != USER_NAMESPACE_NO,
2698 arg_uid_shift,
2699 arg_uid_range,
2700 arg_selinux_context);
2701 if (r < 0)
2702 return r;
2703
2704 r = setup_volatile_state(
2705 directory,
2706 arg_volatile_mode,
2707 arg_userns_mode != USER_NAMESPACE_NO,
2708 arg_uid_shift,
2709 arg_uid_range,
2710 arg_selinux_context);
2711 if (r < 0)
2712 return r;
2713
2714 /* Mark everything as shared so our mounts get propagated down. This is
2715 * required to make new bind mounts available in systemd services
2716 * inside the containter that create a new mount namespace.
2717 * See https://github.com/systemd/systemd/issues/3860
2718 * Further submounts (such as /dev) done after this will inherit the
2719 * shared propagation mode. */
2720 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
2721 if (r < 0)
2722 return r;
2723
2724 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
2725 if (r < 0)
2726 return r;
2727
2728 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
2729 if (r < 0)
2730 return r;
2731
2732 if (arg_read_only) {
2733 r = bind_remount_recursive(directory, true, NULL);
2734 if (r < 0)
2735 return log_error_errno(r, "Failed to make tree read-only: %m");
2736 }
2737
2738 r = mount_all(directory,
2739 arg_mount_settings,
2740 arg_uid_shift,
2741 arg_uid_range,
2742 arg_selinux_apifs_context);
2743 if (r < 0)
2744 return r;
2745
2746 r = copy_devnodes(directory);
2747 if (r < 0)
2748 return r;
2749
2750 dev_setup(directory, arg_uid_shift, arg_uid_shift);
2751
2752 r = setup_pts(directory);
2753 if (r < 0)
2754 return r;
2755
2756 r = setup_propagate(directory);
2757 if (r < 0)
2758 return r;
2759
2760 r = setup_dev_console(directory, console);
2761 if (r < 0)
2762 return r;
2763
2764 r = setup_keyring();
2765 if (r < 0)
2766 return r;
2767
2768 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
2769 if (r < 0)
2770 return r;
2771
2772 r = setup_timezone(directory);
2773 if (r < 0)
2774 return r;
2775
2776 r = setup_resolv_conf(directory);
2777 if (r < 0)
2778 return r;
2779
2780 r = setup_machine_id(directory);
2781 if (r < 0)
2782 return r;
2783
2784 r = setup_journal(directory);
2785 if (r < 0)
2786 return r;
2787
2788 r = mount_custom(
2789 directory,
2790 arg_custom_mounts,
2791 arg_n_custom_mounts,
2792 arg_userns_mode != USER_NAMESPACE_NO,
2793 arg_uid_shift,
2794 arg_uid_range,
2795 arg_selinux_apifs_context);
2796 if (r < 0)
2797 return r;
2798
2799 if (!arg_use_cgns || !cg_ns_supported()) {
2800 r = mount_cgroups(
2801 directory,
2802 arg_unified_cgroup_hierarchy,
2803 arg_userns_mode != USER_NAMESPACE_NO,
2804 arg_uid_shift,
2805 arg_uid_range,
2806 arg_selinux_apifs_context,
2807 false);
2808 if (r < 0)
2809 return r;
2810 }
2811
2812 r = mount_move_root(directory);
2813 if (r < 0)
2814 return log_error_errno(r, "Failed to move root directory: %m");
2815
2816 fd = setup_sd_notify_child();
2817 if (fd < 0)
2818 return fd;
2819
2820 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
2821 arg_clone_ns_flags |
2822 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
2823 if (pid < 0)
2824 return log_error_errno(errno, "Failed to fork inner child: %m");
2825 if (pid == 0) {
2826 pid_socket = safe_close(pid_socket);
2827 uuid_socket = safe_close(uuid_socket);
2828 notify_socket = safe_close(notify_socket);
2829 uid_shift_socket = safe_close(uid_shift_socket);
2830
2831 /* The inner child has all namespaces that are
2832 * requested, so that we all are owned by the user if
2833 * user namespaces are turned on. */
2834
2835 if (arg_network_namespace_path) {
2836 r = namespace_enter(-1, -1, netns_fd, -1, -1);
2837 if (r < 0)
2838 return r;
2839 }
2840
2841 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
2842 if (r < 0)
2843 _exit(EXIT_FAILURE);
2844
2845 _exit(EXIT_SUCCESS);
2846 }
2847
2848 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
2849 if (l < 0)
2850 return log_error_errno(errno, "Failed to send PID: %m");
2851 if (l != sizeof(pid)) {
2852 log_error("Short write while sending PID.");
2853 return -EIO;
2854 }
2855
2856 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
2857 if (l < 0)
2858 return log_error_errno(errno, "Failed to send machine ID: %m");
2859 if (l != sizeof(arg_uuid)) {
2860 log_error("Short write while sending machine ID.");
2861 return -EIO;
2862 }
2863
2864 l = send_one_fd(notify_socket, fd, 0);
2865 if (l < 0)
2866 return log_error_errno(errno, "Failed to send notify fd: %m");
2867
2868 pid_socket = safe_close(pid_socket);
2869 uuid_socket = safe_close(uuid_socket);
2870 notify_socket = safe_close(notify_socket);
2871 kmsg_socket = safe_close(kmsg_socket);
2872 rtnl_socket = safe_close(rtnl_socket);
2873 netns_fd = safe_close(netns_fd);
2874
2875 return 0;
2876 }
2877
2878 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
2879 bool tried_hashed = false;
2880 unsigned n_tries = 100;
2881 uid_t candidate;
2882 int r;
2883
2884 assert(shift);
2885 assert(ret_lock_file);
2886 assert(arg_userns_mode == USER_NAMESPACE_PICK);
2887 assert(arg_uid_range == 0x10000U);
2888
2889 candidate = *shift;
2890
2891 (void) mkdir("/run/systemd/nspawn-uid", 0755);
2892
2893 for (;;) {
2894 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
2895 _cleanup_release_lock_file_ LockFile lf = LOCK_FILE_INIT;
2896
2897 if (--n_tries <= 0)
2898 return -EBUSY;
2899
2900 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
2901 goto next;
2902 if ((candidate & UINT32_C(0xFFFF)) != 0)
2903 goto next;
2904
2905 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
2906 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
2907 if (r == -EBUSY) /* Range already taken by another nspawn instance */
2908 goto next;
2909 if (r < 0)
2910 return r;
2911
2912 /* Make some superficial checks whether the range is currently known in the user database */
2913 if (getpwuid(candidate))
2914 goto next;
2915 if (getpwuid(candidate + UINT32_C(0xFFFE)))
2916 goto next;
2917 if (getgrgid(candidate))
2918 goto next;
2919 if (getgrgid(candidate + UINT32_C(0xFFFE)))
2920 goto next;
2921
2922 *ret_lock_file = lf;
2923 lf = (struct LockFile) LOCK_FILE_INIT;
2924 *shift = candidate;
2925 return 0;
2926
2927 next:
2928 if (arg_machine && !tried_hashed) {
2929 /* Try to hash the base from the container name */
2930
2931 static const uint8_t hash_key[] = {
2932 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
2933 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
2934 };
2935
2936 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
2937
2938 tried_hashed = true;
2939 } else
2940 random_bytes(&candidate, sizeof(candidate));
2941
2942 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
2943 candidate &= (uid_t) UINT32_C(0xFFFF0000);
2944 }
2945 }
2946
2947 static int setup_uid_map(pid_t pid) {
2948 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
2949 int r;
2950
2951 assert(pid > 1);
2952
2953 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
2954 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
2955 r = write_string_file(uid_map, line, 0);
2956 if (r < 0)
2957 return log_error_errno(r, "Failed to write UID map: %m");
2958
2959 /* We always assign the same UID and GID ranges */
2960 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
2961 r = write_string_file(uid_map, line, 0);
2962 if (r < 0)
2963 return log_error_errno(r, "Failed to write GID map: %m");
2964
2965 return 0;
2966 }
2967
2968 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2969 char buf[NOTIFY_BUFFER_MAX+1];
2970 char *p = NULL;
2971 struct iovec iovec = {
2972 .iov_base = buf,
2973 .iov_len = sizeof(buf)-1,
2974 };
2975 union {
2976 struct cmsghdr cmsghdr;
2977 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
2978 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
2979 } control = {};
2980 struct msghdr msghdr = {
2981 .msg_iov = &iovec,
2982 .msg_iovlen = 1,
2983 .msg_control = &control,
2984 .msg_controllen = sizeof(control),
2985 };
2986 struct cmsghdr *cmsg;
2987 struct ucred *ucred = NULL;
2988 ssize_t n;
2989 pid_t inner_child_pid;
2990 _cleanup_strv_free_ char **tags = NULL;
2991
2992 assert(userdata);
2993
2994 inner_child_pid = PTR_TO_PID(userdata);
2995
2996 if (revents != EPOLLIN) {
2997 log_warning("Got unexpected poll event for notify fd.");
2998 return 0;
2999 }
3000
3001 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3002 if (n < 0) {
3003 if (IN_SET(errno, EAGAIN, EINTR))
3004 return 0;
3005
3006 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3007 }
3008 cmsg_close_all(&msghdr);
3009
3010 CMSG_FOREACH(cmsg, &msghdr) {
3011 if (cmsg->cmsg_level == SOL_SOCKET &&
3012 cmsg->cmsg_type == SCM_CREDENTIALS &&
3013 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3014
3015 ucred = (struct ucred*) CMSG_DATA(cmsg);
3016 }
3017 }
3018
3019 if (!ucred || ucred->pid != inner_child_pid) {
3020 log_debug("Received notify message without valid credentials. Ignoring.");
3021 return 0;
3022 }
3023
3024 if ((size_t) n >= sizeof(buf)) {
3025 log_warning("Received notify message exceeded maximum size. Ignoring.");
3026 return 0;
3027 }
3028
3029 buf[n] = 0;
3030 tags = strv_split(buf, "\n\r");
3031 if (!tags)
3032 return log_oom();
3033
3034 if (strv_find(tags, "READY=1"))
3035 sd_notifyf(false, "READY=1\n");
3036
3037 p = strv_find_startswith(tags, "STATUS=");
3038 if (p)
3039 sd_notifyf(false, "STATUS=Container running: %s", p);
3040
3041 return 0;
3042 }
3043
3044 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3045 int r;
3046
3047 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3048 if (r < 0)
3049 return log_error_errno(r, "Failed to allocate notify event source: %m");
3050
3051 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3052
3053 return 0;
3054 }
3055
3056 static int load_settings(void) {
3057 _cleanup_(settings_freep) Settings *settings = NULL;
3058 _cleanup_fclose_ FILE *f = NULL;
3059 _cleanup_free_ char *p = NULL;
3060 const char *fn, *i;
3061 int r;
3062
3063 /* If all settings are masked, there's no point in looking for
3064 * the settings file */
3065 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3066 return 0;
3067
3068 fn = strjoina(arg_machine, ".nspawn");
3069
3070 /* We first look in the admin's directories in /etc and /run */
3071 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3072 _cleanup_free_ char *j = NULL;
3073
3074 j = strjoin(i, "/", fn);
3075 if (!j)
3076 return log_oom();
3077
3078 f = fopen(j, "re");
3079 if (f) {
3080 p = TAKE_PTR(j);
3081
3082 /* By default, we trust configuration from /etc and /run */
3083 if (arg_settings_trusted < 0)
3084 arg_settings_trusted = true;
3085
3086 break;
3087 }
3088
3089 if (errno != ENOENT)
3090 return log_error_errno(errno, "Failed to open %s: %m", j);
3091 }
3092
3093 if (!f) {
3094 /* After that, let's look for a file next to the
3095 * actual image we shall boot. */
3096
3097 if (arg_image) {
3098 p = file_in_same_dir(arg_image, fn);
3099 if (!p)
3100 return log_oom();
3101 } else if (arg_directory) {
3102 p = file_in_same_dir(arg_directory, fn);
3103 if (!p)
3104 return log_oom();
3105 }
3106
3107 if (p) {
3108 f = fopen(p, "re");
3109 if (!f && errno != ENOENT)
3110 return log_error_errno(errno, "Failed to open %s: %m", p);
3111
3112 /* By default, we do not trust configuration from /var/lib/machines */
3113 if (arg_settings_trusted < 0)
3114 arg_settings_trusted = false;
3115 }
3116 }
3117
3118 if (!f)
3119 return 0;
3120
3121 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
3122
3123 r = settings_load(f, p, &settings);
3124 if (r < 0)
3125 return r;
3126
3127 /* Copy over bits from the settings, unless they have been
3128 * explicitly masked by command line switches. */
3129
3130 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3131 settings->start_mode >= 0) {
3132 arg_start_mode = settings->start_mode;
3133
3134 strv_free(arg_parameters);
3135 arg_parameters = TAKE_PTR(settings->parameters);
3136 }
3137
3138 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3139 settings->pivot_root_new) {
3140 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3141 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3142 }
3143
3144 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3145 settings->working_directory)
3146 free_and_replace(arg_chdir, settings->working_directory);
3147
3148 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3149 settings->environment) {
3150 strv_free(arg_setenv);
3151 arg_setenv = TAKE_PTR(settings->environment);
3152 }
3153
3154 if ((arg_settings_mask & SETTING_USER) == 0 &&
3155 settings->user)
3156 free_and_replace(arg_user, settings->user);
3157
3158 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3159 uint64_t plus;
3160
3161 plus = settings->capability;
3162 if (settings_private_network(settings))
3163 plus |= (1ULL << CAP_NET_ADMIN);
3164
3165 if (!arg_settings_trusted && plus != 0) {
3166 if (settings->capability != 0)
3167 log_warning("Ignoring Capability= setting, file %s is not trusted.", p);
3168 } else
3169 arg_caps_retain |= plus;
3170
3171 arg_caps_retain &= ~settings->drop_capability;
3172 }
3173
3174 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3175 settings->kill_signal > 0)
3176 arg_kill_signal = settings->kill_signal;
3177
3178 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3179 settings->personality != PERSONALITY_INVALID)
3180 arg_personality = settings->personality;
3181
3182 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3183 !sd_id128_is_null(settings->machine_id)) {
3184
3185 if (!arg_settings_trusted)
3186 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p);
3187 else
3188 arg_uuid = settings->machine_id;
3189 }
3190
3191 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3192 settings->read_only >= 0)
3193 arg_read_only = settings->read_only;
3194
3195 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3196 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3197 arg_volatile_mode = settings->volatile_mode;
3198
3199 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3200 settings->n_custom_mounts > 0) {
3201
3202 if (!arg_settings_trusted)
3203 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p);
3204 else {
3205 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3206 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3207 arg_n_custom_mounts = settings->n_custom_mounts;
3208 settings->n_custom_mounts = 0;
3209 }
3210 }
3211
3212 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3213 (settings->private_network >= 0 ||
3214 settings->network_veth >= 0 ||
3215 settings->network_bridge ||
3216 settings->network_zone ||
3217 settings->network_interfaces ||
3218 settings->network_macvlan ||
3219 settings->network_ipvlan ||
3220 settings->network_veth_extra)) {
3221
3222 if (!arg_settings_trusted)
3223 log_warning("Ignoring network settings, file %s is not trusted.", p);
3224 else {
3225 arg_network_veth = settings_network_veth(settings);
3226 arg_private_network = settings_private_network(settings);
3227
3228 strv_free(arg_network_interfaces);
3229 arg_network_interfaces = TAKE_PTR(settings->network_interfaces);
3230
3231 strv_free(arg_network_macvlan);
3232 arg_network_macvlan = TAKE_PTR(settings->network_macvlan);
3233
3234 strv_free(arg_network_ipvlan);
3235 arg_network_ipvlan = TAKE_PTR(settings->network_ipvlan);
3236
3237 strv_free(arg_network_veth_extra);
3238 arg_network_veth_extra = TAKE_PTR(settings->network_veth_extra);
3239
3240 free_and_replace(arg_network_bridge, settings->network_bridge);
3241 free_and_replace(arg_network_zone, settings->network_zone);
3242 }
3243 }
3244
3245 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3246 settings->expose_ports) {
3247
3248 if (!arg_settings_trusted)
3249 log_warning("Ignoring Port= setting, file %s is not trusted.", p);
3250 else {
3251 expose_port_free_all(arg_expose_ports);
3252 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3253 }
3254 }
3255
3256 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3257 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3258
3259 if (!arg_settings_trusted)
3260 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", p);
3261 else {
3262 arg_userns_mode = settings->userns_mode;
3263 arg_uid_shift = settings->uid_shift;
3264 arg_uid_range = settings->uid_range;
3265 arg_userns_chown = settings->userns_chown;
3266 }
3267 }
3268
3269 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3270 arg_notify_ready = settings->notify_ready;
3271
3272 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3273
3274 if (!arg_settings_trusted && !strv_isempty(arg_syscall_whitelist))
3275 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", p);
3276 else {
3277 strv_free(arg_syscall_whitelist);
3278 strv_free(arg_syscall_blacklist);
3279
3280 arg_syscall_whitelist = TAKE_PTR(settings->syscall_whitelist);
3281 arg_syscall_blacklist = TAKE_PTR(settings->syscall_blacklist);
3282 }
3283 }
3284
3285 return 0;
3286 }
3287
3288 static int run(int master,
3289 const char* console,
3290 DissectedImage *dissected_image,
3291 bool interactive,
3292 bool secondary,
3293 FDSet *fds,
3294 char veth_name[IFNAMSIZ], bool *veth_created,
3295 union in_addr_union *exposed,
3296 pid_t *pid, int *ret) {
3297
3298 static const struct sigaction sa = {
3299 .sa_handler = nop_signal_handler,
3300 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
3301 };
3302
3303 _cleanup_release_lock_file_ LockFile uid_shift_lock = LOCK_FILE_INIT;
3304 _cleanup_close_ int etc_passwd_lock = -1;
3305 _cleanup_close_pair_ int
3306 kmsg_socket_pair[2] = { -1, -1 },
3307 rtnl_socket_pair[2] = { -1, -1 },
3308 pid_socket_pair[2] = { -1, -1 },
3309 uuid_socket_pair[2] = { -1, -1 },
3310 notify_socket_pair[2] = { -1, -1 },
3311 uid_shift_socket_pair[2] = { -1, -1 },
3312 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
3313
3314 _cleanup_close_ int notify_socket= -1;
3315 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
3316 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
3317 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
3318 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
3319 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
3320 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
3321 ContainerStatus container_status = 0;
3322 char last_char = 0;
3323 int ifi = 0, r;
3324 ssize_t l;
3325 sigset_t mask_chld;
3326 _cleanup_close_ int netns_fd = -1;
3327
3328 assert_se(sigemptyset(&mask_chld) == 0);
3329 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
3330
3331 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3332 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3333 * check with getpwuid() if the specific user already exists. Note that /etc might be
3334 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3335 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3336 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3337 * really ours. */
3338
3339 etc_passwd_lock = take_etc_passwd_lock(NULL);
3340 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
3341 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
3342 }
3343
3344 r = barrier_create(&barrier);
3345 if (r < 0)
3346 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
3347
3348 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
3349 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
3350
3351 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
3352 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
3353
3354 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
3355 return log_error_errno(errno, "Failed to create pid socket pair: %m");
3356
3357 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
3358 return log_error_errno(errno, "Failed to create id socket pair: %m");
3359
3360 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
3361 return log_error_errno(errno, "Failed to create notify socket pair: %m");
3362
3363 if (arg_userns_mode != USER_NAMESPACE_NO)
3364 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
3365 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
3366
3367 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
3368 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
3369 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
3370
3371 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3372 * parent's blocking calls and give it a chance to call wait() and terminate. */
3373 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
3374 if (r < 0)
3375 return log_error_errno(errno, "Failed to change the signal mask: %m");
3376
3377 r = sigaction(SIGCHLD, &sa, NULL);
3378 if (r < 0)
3379 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
3380
3381 if (arg_network_namespace_path) {
3382 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
3383 if (netns_fd < 0)
3384 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
3385
3386 r = fd_is_network_ns(netns_fd);
3387 if (r < 0 && r != -ENOTTY)
3388 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
3389 if (r == 0) {
3390 log_error("Path %s doesn't refer to a network namespace", arg_network_namespace_path);
3391 return -EINVAL;
3392 }
3393 }
3394
3395 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
3396 if (*pid < 0)
3397 return log_error_errno(errno, "clone() failed%s: %m",
3398 errno == EINVAL ?
3399 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3400
3401 if (*pid == 0) {
3402 /* The outer child only has a file system namespace. */
3403 barrier_set_role(&barrier, BARRIER_CHILD);
3404
3405 master = safe_close(master);
3406
3407 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
3408 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3409 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
3410 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
3411 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
3412 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
3413 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
3414
3415 (void) reset_all_signal_handlers();
3416 (void) reset_signal_mask();
3417
3418 r = outer_child(&barrier,
3419 arg_directory,
3420 console,
3421 dissected_image,
3422 interactive,
3423 secondary,
3424 pid_socket_pair[1],
3425 uuid_socket_pair[1],
3426 notify_socket_pair[1],
3427 kmsg_socket_pair[1],
3428 rtnl_socket_pair[1],
3429 uid_shift_socket_pair[1],
3430 unified_cgroup_hierarchy_socket_pair[1],
3431 fds,
3432 netns_fd);
3433 if (r < 0)
3434 _exit(EXIT_FAILURE);
3435
3436 _exit(EXIT_SUCCESS);
3437 }
3438
3439 barrier_set_role(&barrier, BARRIER_PARENT);
3440
3441 fds = fdset_free(fds);
3442
3443 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
3444 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
3445 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
3446 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
3447 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
3448 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
3449 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
3450
3451 if (arg_userns_mode != USER_NAMESPACE_NO) {
3452 /* The child just let us know the UID shift it might have read from the image. */
3453 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
3454 if (l < 0)
3455 return log_error_errno(errno, "Failed to read UID shift: %m");
3456 if (l != sizeof arg_uid_shift) {
3457 log_error("Short read while reading UID shift.");
3458 return -EIO;
3459 }
3460
3461 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3462 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3463 * image, but if that's already in use, pick a new one, and report back to the child,
3464 * which one we now picked. */
3465
3466 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
3467 if (r < 0)
3468 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
3469
3470 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
3471 if (l < 0)
3472 return log_error_errno(errno, "Failed to send UID shift: %m");
3473 if (l != sizeof arg_uid_shift) {
3474 log_error("Short write while writing UID shift.");
3475 return -EIO;
3476 }
3477 }
3478 }
3479
3480 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3481 /* The child let us know the support cgroup mode it might have read from the image. */
3482 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
3483 if (l < 0)
3484 return log_error_errno(errno, "Failed to read cgroup mode: %m");
3485 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
3486 log_error("Short read while reading cgroup mode.");
3487 return -EIO;
3488 }
3489 }
3490
3491 /* Wait for the outer child. */
3492 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
3493 if (r < 0)
3494 return r;
3495 if (r != EXIT_SUCCESS)
3496 return -EIO;
3497
3498 /* And now retrieve the PID of the inner child. */
3499 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
3500 if (l < 0)
3501 return log_error_errno(errno, "Failed to read inner child PID: %m");
3502 if (l != sizeof *pid) {
3503 log_error("Short read while reading inner child PID.");
3504 return -EIO;
3505 }
3506
3507 /* We also retrieve container UUID in case it was generated by outer child */
3508 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
3509 if (l < 0)
3510 return log_error_errno(errno, "Failed to read container machine ID: %m");
3511 if (l != sizeof(arg_uuid)) {
3512 log_error("Short read while reading container machined ID.");
3513 return -EIO;
3514 }
3515
3516 /* We also retrieve the socket used for notifications generated by outer child */
3517 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
3518 if (notify_socket < 0)
3519 return log_error_errno(notify_socket,
3520 "Failed to receive notification socket from the outer child: %m");
3521
3522 log_debug("Init process invoked as PID "PID_FMT, *pid);
3523
3524 if (arg_userns_mode != USER_NAMESPACE_NO) {
3525 if (!barrier_place_and_sync(&barrier)) { /* #1 */
3526 log_error("Child died too early.");
3527 return -ESRCH;
3528 }
3529
3530 r = setup_uid_map(*pid);
3531 if (r < 0)
3532 return r;
3533
3534 (void) barrier_place(&barrier); /* #2 */
3535 }
3536
3537 if (arg_private_network) {
3538
3539 if (!arg_network_namespace_path) {
3540 /* Wait until the child has unshared its network namespace. */
3541 if (!barrier_place_and_sync(&barrier)) { /* #3 */
3542 log_error("Child died too early");
3543 return -ESRCH;
3544 }
3545 }
3546
3547 r = move_network_interfaces(*pid, arg_network_interfaces);
3548 if (r < 0)
3549 return r;
3550
3551 if (arg_network_veth) {
3552 r = setup_veth(arg_machine, *pid, veth_name,
3553 arg_network_bridge || arg_network_zone);
3554 if (r < 0)
3555 return r;
3556 else if (r > 0)
3557 ifi = r;
3558
3559 if (arg_network_bridge) {
3560 /* Add the interface to a bridge */
3561 r = setup_bridge(veth_name, arg_network_bridge, false);
3562 if (r < 0)
3563 return r;
3564 if (r > 0)
3565 ifi = r;
3566 } else if (arg_network_zone) {
3567 /* Add the interface to a bridge, possibly creating it */
3568 r = setup_bridge(veth_name, arg_network_zone, true);
3569 if (r < 0)
3570 return r;
3571 if (r > 0)
3572 ifi = r;
3573 }
3574 }
3575
3576 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
3577 if (r < 0)
3578 return r;
3579
3580 /* We created the primary and extra veth links now; let's remember this, so that we know to
3581 remove them later on. Note that we don't bother with removing veth links that were created
3582 here when their setup failed half-way, because in that case the kernel should be able to
3583 remove them on its own, since they cannot be referenced by anything yet. */
3584 *veth_created = true;
3585
3586 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
3587 if (r < 0)
3588 return r;
3589
3590 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
3591 if (r < 0)
3592 return r;
3593 }
3594
3595 if (arg_register || !arg_keep_unit) {
3596 r = sd_bus_default_system(&bus);
3597 if (r < 0)
3598 return log_error_errno(r, "Failed to open system bus: %m");
3599 }
3600
3601 if (!arg_keep_unit) {
3602 /* When a new scope is created for this container, then we'll be registered as its controller, in which
3603 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
3604 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
3605
3606 r = sd_bus_match_signal_async(
3607 bus,
3608 NULL,
3609 "org.freedesktop.systemd1",
3610 NULL,
3611 "org.freedesktop.systemd1.Scope",
3612 "RequestStop",
3613 on_request_stop, NULL, PID_TO_PTR(*pid));
3614 if (r < 0)
3615 return log_error_errno(r, "Failed to request RequestStop match: %m");
3616 }
3617
3618 if (arg_register) {
3619
3620 r = register_machine(
3621 bus,
3622 arg_machine,
3623 *pid,
3624 arg_directory,
3625 arg_uuid,
3626 ifi,
3627 arg_slice,
3628 arg_custom_mounts, arg_n_custom_mounts,
3629 arg_kill_signal,
3630 arg_property,
3631 arg_keep_unit,
3632 arg_container_service_name);
3633 if (r < 0)
3634 return r;
3635
3636 } else if (!arg_keep_unit) {
3637
3638 r = allocate_scope(
3639 bus,
3640 arg_machine,
3641 *pid,
3642 arg_slice,
3643 arg_custom_mounts, arg_n_custom_mounts,
3644 arg_kill_signal,
3645 arg_property);
3646 if (r < 0)
3647 return r;
3648
3649 } else if (arg_slice || arg_property)
3650 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
3651
3652 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3653 if (r < 0)
3654 return r;
3655
3656 if (arg_keep_unit) {
3657 r = create_subcgroup(*pid, arg_unified_cgroup_hierarchy);
3658 if (r < 0)
3659 return r;
3660 }
3661
3662 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3663 if (r < 0)
3664 return r;
3665
3666 /* Notify the child that the parent is ready with all
3667 * its setup (including cgroup-ification), and that
3668 * the child can now hand over control to the code to
3669 * run inside the container. */
3670 (void) barrier_place(&barrier); /* #4 */
3671
3672 /* Block SIGCHLD here, before notifying child.
3673 * process_pty() will handle it with the other signals. */
3674 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
3675
3676 /* Reset signal to default */
3677 r = default_signals(SIGCHLD, -1);
3678 if (r < 0)
3679 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
3680
3681 r = sd_event_new(&event);
3682 if (r < 0)
3683 return log_error_errno(r, "Failed to get default event source: %m");
3684
3685 (void) sd_event_set_watchdog(event, true);
3686
3687 if (bus) {
3688 r = sd_bus_attach_event(bus, event, 0);
3689 if (r < 0)
3690 return log_error_errno(r, "Failed to attach bus to event loop: %m");
3691 }
3692
3693 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
3694 if (r < 0)
3695 return r;
3696
3697 /* Let the child know that we are ready and wait that the child is completely ready now. */
3698 if (!barrier_place_and_sync(&barrier)) { /* #5 */
3699 log_error("Child died too early.");
3700 return -ESRCH;
3701 }
3702
3703 /* At this point we have made use of the UID we picked, and thus nss-mymachines
3704 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
3705 etc_passwd_lock = safe_close(etc_passwd_lock);
3706
3707 sd_notifyf(false,
3708 "STATUS=Container running.\n"
3709 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
3710 if (!arg_notify_ready)
3711 sd_notify(false, "READY=1\n");
3712
3713 if (arg_kill_signal > 0) {
3714 /* Try to kill the init system on SIGINT or SIGTERM */
3715 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
3716 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
3717 } else {
3718 /* Immediately exit */
3719 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
3720 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
3721 }
3722
3723 /* Exit when the child exits */
3724 sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
3725
3726 if (arg_expose_ports) {
3727 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
3728 if (r < 0)
3729 return r;
3730
3731 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
3732 }
3733
3734 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3735
3736 r = pty_forward_new(event, master,
3737 PTY_FORWARD_IGNORE_VHANGUP | (interactive ? 0 : PTY_FORWARD_READ_ONLY),
3738 &forward);
3739 if (r < 0)
3740 return log_error_errno(r, "Failed to create PTY forwarder: %m");
3741
3742 r = sd_event_loop(event);
3743 if (r < 0)
3744 return log_error_errno(r, "Failed to run event loop: %m");
3745
3746 pty_forward_get_last_char(forward, &last_char);
3747
3748 forward = pty_forward_free(forward);
3749
3750 if (!arg_quiet && last_char != '\n')
3751 putc('\n', stdout);
3752
3753 /* Kill if it is not dead yet anyway */
3754 if (arg_register && !arg_keep_unit && bus)
3755 terminate_machine(bus, *pid);
3756
3757 /* Normally redundant, but better safe than sorry */
3758 (void) kill(*pid, SIGKILL);
3759
3760 r = wait_for_container(*pid, &container_status);
3761 *pid = 0;
3762
3763 if (r < 0)
3764 /* We failed to wait for the container, or the container exited abnormally. */
3765 return r;
3766 if (r > 0 || container_status == CONTAINER_TERMINATED) {
3767 /* r > 0 → The container exited with a non-zero status.
3768 * As a special case, we need to replace 133 with a different value,
3769 * because 133 is special-cased in the service file to reboot the container.
3770 * otherwise → The container exited with zero status and a reboot was not requested.
3771 */
3772 if (r == EXIT_FORCE_RESTART)
3773 r = EXIT_FAILURE; /* replace 133 with the general failure code */
3774 *ret = r;
3775 return 0; /* finito */
3776 }
3777
3778 /* CONTAINER_REBOOTED, loop again */
3779
3780 if (arg_keep_unit) {
3781 /* Special handling if we are running as a service: instead of simply
3782 * restarting the machine we want to restart the entire service, so let's
3783 * inform systemd about this with the special exit code 133. The service
3784 * file uses RestartForceExitStatus=133 so that this results in a full
3785 * nspawn restart. This is necessary since we might have cgroup parameters
3786 * set we want to have flushed out. */
3787 *ret = EXIT_FORCE_RESTART;
3788 return 0; /* finito */
3789 }
3790
3791 expose_port_flush(arg_expose_ports, exposed);
3792
3793 (void) remove_veth_links(veth_name, arg_network_veth_extra);
3794 *veth_created = false;
3795 return 1; /* loop again */
3796 }
3797
3798 int main(int argc, char *argv[]) {
3799
3800 _cleanup_free_ char *console = NULL;
3801 _cleanup_close_ int master = -1;
3802 _cleanup_fdset_free_ FDSet *fds = NULL;
3803 int r, n_fd_passed, ret = EXIT_SUCCESS;
3804 char veth_name[IFNAMSIZ] = "";
3805 bool secondary = false, remove_directory = false, remove_image = false;
3806 pid_t pid = 0;
3807 union in_addr_union exposed = {};
3808 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
3809 bool interactive, veth_created = false, remove_tmprootdir = false;
3810 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
3811 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
3812 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3813 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3814
3815 log_parse_environment();
3816 log_open();
3817
3818 /* Make sure rename_process() in the stub init process can work */
3819 saved_argv = argv;
3820 saved_argc = argc;
3821
3822 r = parse_argv(argc, argv);
3823 if (r <= 0)
3824 goto finish;
3825
3826 r = must_be_root();
3827 if (r < 0)
3828 goto finish;
3829
3830 r = determine_names();
3831 if (r < 0)
3832 goto finish;
3833
3834 r = load_settings();
3835 if (r < 0)
3836 goto finish;
3837
3838 r = verify_arguments();
3839 if (r < 0)
3840 goto finish;
3841
3842 r = detect_unified_cgroup_hierarchy_from_environment();
3843 if (r < 0)
3844 goto finish;
3845
3846 n_fd_passed = sd_listen_fds(false);
3847 if (n_fd_passed > 0) {
3848 r = fdset_new_listen_fds(&fds, false);
3849 if (r < 0) {
3850 log_error_errno(r, "Failed to collect file descriptors: %m");
3851 goto finish;
3852 }
3853 }
3854
3855 if (arg_directory) {
3856 assert(!arg_image);
3857
3858 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
3859 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3860 r = -EINVAL;
3861 goto finish;
3862 }
3863
3864 if (arg_ephemeral) {
3865 _cleanup_free_ char *np = NULL;
3866
3867 r = chase_symlinks_and_update(&arg_directory, 0);
3868 if (r < 0)
3869 goto finish;
3870
3871 /* If the specified path is a mount point we
3872 * generate the new snapshot immediately
3873 * inside it under a random name. However if
3874 * the specified is not a mount point we
3875 * create the new snapshot in the parent
3876 * directory, just next to it. */
3877 r = path_is_mount_point(arg_directory, NULL, 0);
3878 if (r < 0) {
3879 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
3880 goto finish;
3881 }
3882 if (r > 0)
3883 r = tempfn_random_child(arg_directory, "machine.", &np);
3884 else
3885 r = tempfn_random(arg_directory, "machine.", &np);
3886 if (r < 0) {
3887 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
3888 goto finish;
3889 }
3890
3891 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3892 if (r < 0) {
3893 log_error_errno(r, "Failed to lock %s: %m", np);
3894 goto finish;
3895 }
3896
3897 r = btrfs_subvol_snapshot(arg_directory, np,
3898 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3899 BTRFS_SNAPSHOT_FALLBACK_COPY |
3900 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3901 BTRFS_SNAPSHOT_RECURSIVE |
3902 BTRFS_SNAPSHOT_QUOTA);
3903 if (r < 0) {
3904 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
3905 goto finish;
3906 }
3907
3908 free_and_replace(arg_directory, np);
3909
3910 remove_directory = true;
3911
3912 } else {
3913 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
3914 if (r < 0)
3915 goto finish;
3916
3917 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3918 if (r == -EBUSY) {
3919 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
3920 goto finish;
3921 }
3922 if (r < 0) {
3923 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
3924 goto finish;
3925 }
3926
3927 if (arg_template) {
3928 r = chase_symlinks_and_update(&arg_template, 0);
3929 if (r < 0)
3930 goto finish;
3931
3932 r = btrfs_subvol_snapshot(arg_template, arg_directory,
3933 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3934 BTRFS_SNAPSHOT_FALLBACK_COPY |
3935 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3936 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
3937 BTRFS_SNAPSHOT_RECURSIVE |
3938 BTRFS_SNAPSHOT_QUOTA);
3939 if (r == -EEXIST) {
3940 if (!arg_quiet)
3941 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
3942 } else if (r < 0) {
3943 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
3944 goto finish;
3945 } else {
3946 if (!arg_quiet)
3947 log_info("Populated %s from template %s.", arg_directory, arg_template);
3948 }
3949 }
3950 }
3951
3952 if (arg_start_mode == START_BOOT) {
3953 if (path_is_os_tree(arg_directory) <= 0) {
3954 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
3955 r = -EINVAL;
3956 goto finish;
3957 }
3958 } else {
3959 const char *p;
3960
3961 p = strjoina(arg_directory, "/usr/");
3962 if (laccess(p, F_OK) < 0) {
3963 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", arg_directory);
3964 r = -EINVAL;
3965 goto finish;
3966 }
3967 }
3968
3969 } else {
3970 assert(arg_image);
3971 assert(!arg_template);
3972
3973 r = chase_symlinks_and_update(&arg_image, 0);
3974 if (r < 0)
3975 goto finish;
3976
3977 if (arg_ephemeral) {
3978 _cleanup_free_ char *np = NULL;
3979
3980 r = tempfn_random(arg_image, "machine.", &np);
3981 if (r < 0) {
3982 log_error_errno(r, "Failed to generate name for image snapshot: %m");
3983 goto finish;
3984 }
3985
3986 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3987 if (r < 0) {
3988 r = log_error_errno(r, "Failed to create image lock: %m");
3989 goto finish;
3990 }
3991
3992 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, COPY_REFLINK);
3993 if (r < 0) {
3994 r = log_error_errno(r, "Failed to copy image file: %m");
3995 goto finish;
3996 }
3997
3998 free_and_replace(arg_image, np);
3999
4000 remove_image = true;
4001 } else {
4002 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4003 if (r == -EBUSY) {
4004 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4005 goto finish;
4006 }
4007 if (r < 0) {
4008 r = log_error_errno(r, "Failed to create image lock: %m");
4009 goto finish;
4010 }
4011
4012 if (!arg_root_hash) {
4013 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4014 if (r < 0) {
4015 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4016 goto finish;
4017 }
4018 }
4019 }
4020
4021 if (!mkdtemp(tmprootdir)) {
4022 r = log_error_errno(errno, "Failed to create temporary directory: %m");
4023 goto finish;
4024 }
4025
4026 remove_tmprootdir = true;
4027
4028 arg_directory = strdup(tmprootdir);
4029 if (!arg_directory) {
4030 r = log_oom();
4031 goto finish;
4032 }
4033
4034 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4035 if (r < 0) {
4036 log_error_errno(r, "Failed to set up loopback block device: %m");
4037 goto finish;
4038 }
4039
4040 r = dissect_image_and_warn(
4041 loop->fd,
4042 arg_image,
4043 arg_root_hash, arg_root_hash_size,
4044 DISSECT_IMAGE_REQUIRE_ROOT,
4045 &dissected_image);
4046 if (r == -ENOPKG) {
4047 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4048 log_notice("Note that the disk image needs to\n"
4049 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4050 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4051 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4052 " d) or contain a file system without a partition table\n"
4053 "in order to be bootable with systemd-nspawn.");
4054 goto finish;
4055 }
4056 if (r < 0)
4057 goto finish;
4058
4059 if (!arg_root_hash && dissected_image->can_verity)
4060 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4061
4062 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
4063 if (r < 0)
4064 goto finish;
4065
4066 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4067 if (remove_image && unlink(arg_image) >= 0)
4068 remove_image = false;
4069 }
4070
4071 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
4072 if (r < 0)
4073 goto finish;
4074
4075 interactive =
4076 isatty(STDIN_FILENO) > 0 &&
4077 isatty(STDOUT_FILENO) > 0;
4078
4079 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
4080 if (master < 0) {
4081 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4082 goto finish;
4083 }
4084
4085 r = ptsname_malloc(master, &console);
4086 if (r < 0) {
4087 r = log_error_errno(r, "Failed to determine tty name: %m");
4088 goto finish;
4089 }
4090
4091 if (arg_selinux_apifs_context) {
4092 r = mac_selinux_apply(console, arg_selinux_apifs_context);
4093 if (r < 0)
4094 goto finish;
4095 }
4096
4097 if (unlockpt(master) < 0) {
4098 r = log_error_errno(errno, "Failed to unlock tty: %m");
4099 goto finish;
4100 }
4101
4102 if (!arg_quiet)
4103 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4104 arg_machine, arg_image ?: arg_directory);
4105
4106 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
4107
4108 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
4109 r = log_error_errno(errno, "Failed to become subreaper: %m");
4110 goto finish;
4111 }
4112
4113 for (;;) {
4114 r = run(master,
4115 console,
4116 dissected_image,
4117 interactive, secondary,
4118 fds,
4119 veth_name, &veth_created,
4120 &exposed,
4121 &pid, &ret);
4122 if (r <= 0)
4123 break;
4124 }
4125
4126 finish:
4127 sd_notify(false,
4128 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
4129 "STOPPING=1\nSTATUS=Terminating...");
4130
4131 if (pid > 0)
4132 (void) kill(pid, SIGKILL);
4133
4134 /* Try to flush whatever is still queued in the pty */
4135 if (master >= 0) {
4136 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
4137 master = safe_close(master);
4138 }
4139
4140 if (pid > 0)
4141 (void) wait_for_terminate(pid, NULL);
4142
4143 if (remove_directory && arg_directory) {
4144 int k;
4145
4146 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
4147 if (k < 0)
4148 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
4149 }
4150
4151 if (remove_image && arg_image) {
4152 if (unlink(arg_image) < 0)
4153 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
4154 }
4155
4156 if (remove_tmprootdir) {
4157 if (rmdir(tmprootdir) < 0)
4158 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
4159 }
4160
4161 if (arg_machine) {
4162 const char *p;
4163
4164 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
4165 (void) rm_rf(p, REMOVE_ROOT);
4166 }
4167
4168 expose_port_flush(arg_expose_ports, &exposed);
4169
4170 if (veth_created)
4171 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4172 (void) remove_bridge(arg_network_zone);
4173
4174 free(arg_directory);
4175 free(arg_template);
4176 free(arg_image);
4177 free(arg_machine);
4178 free(arg_user);
4179 free(arg_pivot_root_new);
4180 free(arg_pivot_root_old);
4181 free(arg_chdir);
4182 strv_free(arg_setenv);
4183 free(arg_network_bridge);
4184 strv_free(arg_network_interfaces);
4185 strv_free(arg_network_macvlan);
4186 strv_free(arg_network_ipvlan);
4187 strv_free(arg_network_veth_extra);
4188 strv_free(arg_parameters);
4189 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4190 expose_port_free_all(arg_expose_ports);
4191 free(arg_root_hash);
4192
4193 return r < 0 ? EXIT_FAILURE : ret;
4194 }
Unnamed repository; edit this file 'description' to name the repository.