]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #8487 from keszybz/oss-fuzz-fixes
[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 = arg_template;
1179 arg_template = NULL;
1180 }
1181
1182 if (arg_template && !(arg_directory || arg_machine)) {
1183 log_error("--template= needs --directory= or --machine=.");
1184 return -EINVAL;
1185 }
1186
1187 if (arg_ephemeral && arg_template) {
1188 log_error("--ephemeral and --template= may not be combined.");
1189 return -EINVAL;
1190 }
1191
1192 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
1193 log_error("--ephemeral and --link-journal= may not be combined.");
1194 return -EINVAL;
1195 }
1196
1197 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported()) {
1198 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1199 return -EOPNOTSUPP;
1200 }
1201
1202 if (arg_userns_chown && arg_read_only) {
1203 log_error("--read-only and --private-users-chown may not be combined.");
1204 return -EINVAL;
1205 }
1206
1207 if (arg_network_bridge && arg_network_zone) {
1208 log_error("--network-bridge= and --network-zone= may not be combined.");
1209 return -EINVAL;
1210 }
1211
1212 if (argc > optind) {
1213 arg_parameters = strv_copy(argv + optind);
1214 if (!arg_parameters)
1215 return log_oom();
1216
1217 arg_settings_mask |= SETTING_START_MODE;
1218 }
1219
1220 /* Load all settings from .nspawn files */
1221 if (mask_no_settings)
1222 arg_settings_mask = 0;
1223
1224 /* Don't load any settings from .nspawn files */
1225 if (mask_all_settings)
1226 arg_settings_mask = _SETTINGS_MASK_ALL;
1227
1228 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
1229
1230 r = cg_unified_flush();
1231 if (r < 0)
1232 return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
1233
1234 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1235 if (e)
1236 arg_container_service_name = e;
1237
1238 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1239 if (r < 0)
1240 arg_use_cgns = cg_ns_supported();
1241 else
1242 arg_use_cgns = r;
1243
1244 r = custom_mount_check_all();
1245 if (r < 0)
1246 return r;
1247
1248 return 1;
1249 }
1250
1251 static int verify_arguments(void) {
1252 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network) {
1253 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1254 return -EINVAL;
1255 }
1256
1257 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO)) {
1258 log_error("Cannot combine --private-users with read-write mounts.");
1259 return -EINVAL;
1260 }
1261
1262 if (arg_volatile_mode != VOLATILE_NO && arg_read_only) {
1263 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1264 return -EINVAL;
1265 }
1266
1267 if (arg_expose_ports && !arg_private_network) {
1268 log_error("Cannot use --port= without private networking.");
1269 return -EINVAL;
1270 }
1271
1272 #if ! HAVE_LIBIPTC
1273 if (arg_expose_ports) {
1274 log_error("--port= is not supported, compiled without libiptc support.");
1275 return -EOPNOTSUPP;
1276 }
1277 #endif
1278
1279 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1280 arg_kill_signal = SIGRTMIN+3;
1281
1282 return 0;
1283 }
1284
1285 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1286 assert(p);
1287
1288 if (arg_userns_mode == USER_NAMESPACE_NO)
1289 return 0;
1290
1291 if (uid == UID_INVALID && gid == GID_INVALID)
1292 return 0;
1293
1294 if (uid != UID_INVALID) {
1295 uid += arg_uid_shift;
1296
1297 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1298 return -EOVERFLOW;
1299 }
1300
1301 if (gid != GID_INVALID) {
1302 gid += (gid_t) arg_uid_shift;
1303
1304 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1305 return -EOVERFLOW;
1306 }
1307
1308 if (lchown(p, uid, gid) < 0)
1309 return -errno;
1310
1311 return 0;
1312 }
1313
1314 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1315 const char *q;
1316 int r;
1317
1318 q = prefix_roota(root, path);
1319 r = mkdir_errno_wrapper(q, mode);
1320 if (r == -EEXIST)
1321 return 0;
1322 if (r < 0)
1323 return r;
1324
1325 return userns_lchown(q, uid, gid);
1326 }
1327
1328 static int setup_timezone(const char *dest) {
1329 _cleanup_free_ char *p = NULL, *q = NULL;
1330 const char *where, *check, *what;
1331 char *z, *y;
1332 int r;
1333
1334 assert(dest);
1335
1336 /* Fix the timezone, if possible */
1337 r = readlink_malloc("/etc/localtime", &p);
1338 if (r < 0) {
1339 log_warning("host's /etc/localtime is not a symlink, not updating container timezone.");
1340 /* to handle warning, delete /etc/localtime and replace it
1341 * with a symbolic link to a time zone data file.
1342 *
1343 * Example:
1344 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1345 */
1346 return 0;
1347 }
1348
1349 z = path_startswith(p, "../usr/share/zoneinfo/");
1350 if (!z)
1351 z = path_startswith(p, "/usr/share/zoneinfo/");
1352 if (!z) {
1353 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1354 return 0;
1355 }
1356
1357 where = prefix_roota(dest, "/etc/localtime");
1358 r = readlink_malloc(where, &q);
1359 if (r >= 0) {
1360 y = path_startswith(q, "../usr/share/zoneinfo/");
1361 if (!y)
1362 y = path_startswith(q, "/usr/share/zoneinfo/");
1363
1364 /* Already pointing to the right place? Then do nothing .. */
1365 if (y && streq(y, z))
1366 return 0;
1367 }
1368
1369 check = strjoina("/usr/share/zoneinfo/", z);
1370 check = prefix_roota(dest, check);
1371 if (laccess(check, F_OK) < 0) {
1372 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1373 return 0;
1374 }
1375
1376 if (unlink(where) < 0 && errno != ENOENT) {
1377 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1378 errno,
1379 "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1380 return 0;
1381 }
1382
1383 what = strjoina("../usr/share/zoneinfo/", z);
1384 if (symlink(what, where) < 0) {
1385 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1386 errno,
1387 "Failed to correct timezone of container, ignoring: %m");
1388 return 0;
1389 }
1390
1391 r = userns_lchown(where, 0, 0);
1392 if (r < 0)
1393 return log_warning_errno(r, "Failed to chown /etc/localtime: %m");
1394
1395 return 0;
1396 }
1397
1398 static int resolved_listening(void) {
1399 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1400 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1401 int r;
1402
1403 /* Check if resolved is listening */
1404
1405 r = sd_bus_open_system(&bus);
1406 if (r < 0)
1407 return r;
1408
1409 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1410 if (r <= 0)
1411 return r;
1412
1413 r = sd_bus_get_property_string(bus,
1414 "org.freedesktop.resolve1",
1415 "/org/freedesktop/resolve1",
1416 "org.freedesktop.resolve1.Manager",
1417 "DNSStubListener",
1418 NULL,
1419 &dns_stub_listener_mode);
1420 if (r < 0)
1421 return r;
1422
1423 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1424 }
1425
1426 static int setup_resolv_conf(const char *dest) {
1427 _cleanup_free_ char *resolved = NULL, *etc = NULL;
1428 const char *where;
1429 int r, found;
1430
1431 assert(dest);
1432
1433 if (arg_private_network)
1434 return 0;
1435
1436 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1437 if (r < 0) {
1438 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1439 return 0;
1440 }
1441
1442 where = strjoina(etc, "/resolv.conf");
1443 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1444 if (found < 0) {
1445 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1446 return 0;
1447 }
1448
1449 if (access(STATIC_RESOLV_CONF, F_OK) >= 0 &&
1450 resolved_listening() > 0) {
1451
1452 /* resolved is enabled on the host. In this, case bind mount its static resolv.conf file into the
1453 * container, so that the container can use the host's resolver. Given that network namespacing is
1454 * disabled it's only natural of the container also uses the host's resolver. It also has the big
1455 * advantage that the container will be able to follow the host's DNS server configuration changes
1456 * transparently. */
1457
1458 if (found == 0) /* missing? */
1459 (void) touch(resolved);
1460
1461 r = mount_verbose(LOG_DEBUG, STATIC_RESOLV_CONF, resolved, NULL, MS_BIND, NULL);
1462 if (r >= 0)
1463 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1464 }
1465
1466 /* If that didn't work, let's copy the file */
1467 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0, COPY_REFLINK);
1468 if (r < 0) {
1469 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1470 * resolved or something similar runs inside and the symlink points there.
1471 *
1472 * If the disk image is read-only, there's also no point in complaining.
1473 */
1474 log_full_errno(IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1475 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1476 return 0;
1477 }
1478
1479 r = userns_lchown(where, 0, 0);
1480 if (r < 0)
1481 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1482
1483 return 0;
1484 }
1485
1486 static int setup_boot_id(const char *dest) {
1487 sd_id128_t rnd = SD_ID128_NULL;
1488 const char *from, *to;
1489 int r;
1490
1491 /* Generate a new randomized boot ID, so that each boot-up of
1492 * the container gets a new one */
1493
1494 from = prefix_roota(dest, "/run/proc-sys-kernel-random-boot-id");
1495 to = prefix_roota(dest, "/proc/sys/kernel/random/boot_id");
1496
1497 r = sd_id128_randomize(&rnd);
1498 if (r < 0)
1499 return log_error_errno(r, "Failed to generate random boot id: %m");
1500
1501 r = id128_write(from, ID128_UUID, rnd, false);
1502 if (r < 0)
1503 return log_error_errno(r, "Failed to write boot id: %m");
1504
1505 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1506 if (r >= 0)
1507 r = mount_verbose(LOG_ERR, NULL, to, NULL,
1508 MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1509
1510 (void) unlink(from);
1511 return r;
1512 }
1513
1514 static int copy_devnodes(const char *dest) {
1515
1516 static const char devnodes[] =
1517 "null\0"
1518 "zero\0"
1519 "full\0"
1520 "random\0"
1521 "urandom\0"
1522 "tty\0"
1523 "net/tun\0";
1524
1525 const char *d;
1526 int r = 0;
1527 _cleanup_umask_ mode_t u;
1528
1529 assert(dest);
1530
1531 u = umask(0000);
1532
1533 /* Create /dev/net, so that we can create /dev/net/tun in it */
1534 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1535 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1536
1537 NULSTR_FOREACH(d, devnodes) {
1538 _cleanup_free_ char *from = NULL, *to = NULL;
1539 struct stat st;
1540
1541 from = strappend("/dev/", d);
1542 to = prefix_root(dest, from);
1543
1544 if (stat(from, &st) < 0) {
1545
1546 if (errno != ENOENT)
1547 return log_error_errno(errno, "Failed to stat %s: %m", from);
1548
1549 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1550
1551 log_error("%s is not a char or block device, cannot copy.", from);
1552 return -EIO;
1553
1554 } else {
1555 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1556 /* Explicitly warn the user when /dev is already populated. */
1557 if (errno == EEXIST)
1558 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);
1559 if (errno != EPERM)
1560 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1561
1562 /* Some systems abusively restrict mknod but
1563 * allow bind mounts. */
1564 r = touch(to);
1565 if (r < 0)
1566 return log_error_errno(r, "touch (%s) failed: %m", to);
1567 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1568 if (r < 0)
1569 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1570 }
1571
1572 r = userns_lchown(to, 0, 0);
1573 if (r < 0)
1574 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1575 }
1576 }
1577
1578 return r;
1579 }
1580
1581 static int setup_pts(const char *dest) {
1582 _cleanup_free_ char *options = NULL;
1583 const char *p;
1584 int r;
1585
1586 #if HAVE_SELINUX
1587 if (arg_selinux_apifs_context)
1588 (void) asprintf(&options,
1589 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1590 arg_uid_shift + TTY_GID,
1591 arg_selinux_apifs_context);
1592 else
1593 #endif
1594 (void) asprintf(&options,
1595 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1596 arg_uid_shift + TTY_GID);
1597
1598 if (!options)
1599 return log_oom();
1600
1601 /* Mount /dev/pts itself */
1602 p = prefix_roota(dest, "/dev/pts");
1603 r = mkdir_errno_wrapper(p, 0755);
1604 if (r < 0)
1605 return log_error_errno(r, "Failed to create /dev/pts: %m");
1606
1607 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
1608 if (r < 0)
1609 return r;
1610 r = userns_lchown(p, 0, 0);
1611 if (r < 0)
1612 return log_error_errno(r, "Failed to chown /dev/pts: %m");
1613
1614 /* Create /dev/ptmx symlink */
1615 p = prefix_roota(dest, "/dev/ptmx");
1616 if (symlink("pts/ptmx", p) < 0)
1617 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1618 r = userns_lchown(p, 0, 0);
1619 if (r < 0)
1620 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
1621
1622 /* And fix /dev/pts/ptmx ownership */
1623 p = prefix_roota(dest, "/dev/pts/ptmx");
1624 r = userns_lchown(p, 0, 0);
1625 if (r < 0)
1626 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
1627
1628 return 0;
1629 }
1630
1631 static int setup_dev_console(const char *dest, const char *console) {
1632 _cleanup_umask_ mode_t u;
1633 const char *to;
1634 int r;
1635
1636 assert(dest);
1637 assert(console);
1638
1639 u = umask(0000);
1640
1641 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
1642 if (r < 0)
1643 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1644
1645 /* We need to bind mount the right tty to /dev/console since
1646 * ptys can only exist on pts file systems. To have something
1647 * to bind mount things on we create a empty regular file. */
1648
1649 to = prefix_roota(dest, "/dev/console");
1650 r = touch(to);
1651 if (r < 0)
1652 return log_error_errno(r, "touch() for /dev/console failed: %m");
1653
1654 return mount_verbose(LOG_ERR, console, to, NULL, MS_BIND, NULL);
1655 }
1656
1657 static int setup_keyring(void) {
1658 key_serial_t keyring;
1659
1660 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
1661 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
1662 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
1663 * these system calls let's make sure we don't leak anything into the container. */
1664
1665 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
1666 if (keyring == -1) {
1667 if (errno == ENOSYS)
1668 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
1669 else if (IN_SET(errno, EACCES, EPERM))
1670 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
1671 else
1672 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
1673 }
1674
1675 return 0;
1676 }
1677
1678 static int setup_kmsg(const char *dest, int kmsg_socket) {
1679 const char *from, *to;
1680 _cleanup_umask_ mode_t u;
1681 int fd, r;
1682
1683 assert(kmsg_socket >= 0);
1684
1685 u = umask(0000);
1686
1687 /* We create the kmsg FIFO as /run/kmsg, but immediately
1688 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1689 * on the reading side behave very similar to /proc/kmsg,
1690 * their writing side behaves differently from /dev/kmsg in
1691 * that writing blocks when nothing is reading. In order to
1692 * avoid any problems with containers deadlocking due to this
1693 * we simply make /dev/kmsg unavailable to the container. */
1694 from = prefix_roota(dest, "/run/kmsg");
1695 to = prefix_roota(dest, "/proc/kmsg");
1696
1697 if (mkfifo(from, 0600) < 0)
1698 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
1699 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1700 if (r < 0)
1701 return r;
1702
1703 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
1704 if (fd < 0)
1705 return log_error_errno(errno, "Failed to open fifo: %m");
1706
1707 /* Store away the fd in the socket, so that it stays open as
1708 * long as we run the child */
1709 r = send_one_fd(kmsg_socket, fd, 0);
1710 safe_close(fd);
1711
1712 if (r < 0)
1713 return log_error_errno(r, "Failed to send FIFO fd: %m");
1714
1715 /* And now make the FIFO unavailable as /run/kmsg... */
1716 (void) unlink(from);
1717
1718 return 0;
1719 }
1720
1721 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
1722 union in_addr_union *exposed = userdata;
1723
1724 assert(rtnl);
1725 assert(m);
1726 assert(exposed);
1727
1728 expose_port_execute(rtnl, arg_expose_ports, exposed);
1729 return 0;
1730 }
1731
1732 static int setup_hostname(void) {
1733
1734 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
1735 return 0;
1736
1737 if (sethostname_idempotent(arg_machine) < 0)
1738 return -errno;
1739
1740 return 0;
1741 }
1742
1743 static int setup_journal(const char *directory) {
1744 sd_id128_t this_id;
1745 _cleanup_free_ char *d = NULL;
1746 const char *p, *q;
1747 bool try;
1748 char id[33];
1749 int r;
1750
1751 /* Don't link journals in ephemeral mode */
1752 if (arg_ephemeral)
1753 return 0;
1754
1755 if (arg_link_journal == LINK_NO)
1756 return 0;
1757
1758 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
1759
1760 r = sd_id128_get_machine(&this_id);
1761 if (r < 0)
1762 return log_error_errno(r, "Failed to retrieve machine ID: %m");
1763
1764 if (sd_id128_equal(arg_uuid, this_id)) {
1765 log_full(try ? LOG_WARNING : LOG_ERR,
1766 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
1767 if (try)
1768 return 0;
1769 return -EEXIST;
1770 }
1771
1772 r = userns_mkdir(directory, "/var", 0755, 0, 0);
1773 if (r < 0)
1774 return log_error_errno(r, "Failed to create /var: %m");
1775
1776 r = userns_mkdir(directory, "/var/log", 0755, 0, 0);
1777 if (r < 0)
1778 return log_error_errno(r, "Failed to create /var/log: %m");
1779
1780 r = userns_mkdir(directory, "/var/log/journal", 0755, 0, 0);
1781 if (r < 0)
1782 return log_error_errno(r, "Failed to create /var/log/journal: %m");
1783
1784 (void) sd_id128_to_string(arg_uuid, id);
1785
1786 p = strjoina("/var/log/journal/", id);
1787 q = prefix_roota(directory, p);
1788
1789 if (path_is_mount_point(p, NULL, 0) > 0) {
1790 if (try)
1791 return 0;
1792
1793 log_error("%s: already a mount point, refusing to use for journal", p);
1794 return -EEXIST;
1795 }
1796
1797 if (path_is_mount_point(q, NULL, 0) > 0) {
1798 if (try)
1799 return 0;
1800
1801 log_error("%s: already a mount point, refusing to use for journal", q);
1802 return -EEXIST;
1803 }
1804
1805 r = readlink_and_make_absolute(p, &d);
1806 if (r >= 0) {
1807 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
1808 path_equal(d, q)) {
1809
1810 r = userns_mkdir(directory, p, 0755, 0, 0);
1811 if (r < 0)
1812 log_warning_errno(r, "Failed to create directory %s: %m", q);
1813 return 0;
1814 }
1815
1816 if (unlink(p) < 0)
1817 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
1818 } else if (r == -EINVAL) {
1819
1820 if (arg_link_journal == LINK_GUEST &&
1821 rmdir(p) < 0) {
1822
1823 if (errno == ENOTDIR) {
1824 log_error("%s already exists and is neither a symlink nor a directory", p);
1825 return r;
1826 } else
1827 return log_error_errno(errno, "Failed to remove %s: %m", p);
1828 }
1829 } else if (r != -ENOENT)
1830 return log_error_errno(r, "readlink(%s) failed: %m", p);
1831
1832 if (arg_link_journal == LINK_GUEST) {
1833
1834 if (symlink(q, p) < 0) {
1835 if (try) {
1836 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
1837 return 0;
1838 } else
1839 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
1840 }
1841
1842 r = userns_mkdir(directory, p, 0755, 0, 0);
1843 if (r < 0)
1844 log_warning_errno(r, "Failed to create directory %s: %m", q);
1845 return 0;
1846 }
1847
1848 if (arg_link_journal == LINK_HOST) {
1849 /* don't create parents here — if the host doesn't have
1850 * permanent journal set up, don't force it here */
1851
1852 r = mkdir_errno_wrapper(p, 0755);
1853 if (r < 0 && r != -EEXIST) {
1854 if (try) {
1855 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
1856 return 0;
1857 } else
1858 return log_error_errno(r, "Failed to create %s: %m", p);
1859 }
1860
1861 } else if (access(p, F_OK) < 0)
1862 return 0;
1863
1864 if (dir_is_empty(q) == 0)
1865 log_warning("%s is not empty, proceeding anyway.", q);
1866
1867 r = userns_mkdir(directory, p, 0755, 0, 0);
1868 if (r < 0)
1869 return log_error_errno(r, "Failed to create %s: %m", q);
1870
1871 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
1872 if (r < 0)
1873 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
1874
1875 return 0;
1876 }
1877
1878 static int drop_capabilities(void) {
1879 return capability_bounding_set_drop(arg_caps_retain, false);
1880 }
1881
1882 static int reset_audit_loginuid(void) {
1883 _cleanup_free_ char *p = NULL;
1884 int r;
1885
1886 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
1887 return 0;
1888
1889 r = read_one_line_file("/proc/self/loginuid", &p);
1890 if (r == -ENOENT)
1891 return 0;
1892 if (r < 0)
1893 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
1894
1895 /* Already reset? */
1896 if (streq(p, "4294967295"))
1897 return 0;
1898
1899 r = write_string_file("/proc/self/loginuid", "4294967295", 0);
1900 if (r < 0) {
1901 log_error_errno(r,
1902 "Failed to reset audit login UID. This probably means that your kernel is too\n"
1903 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1904 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1905 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1906 "using systemd-nspawn. Sleeping for 5s... (%m)");
1907
1908 sleep(5);
1909 }
1910
1911 return 0;
1912 }
1913
1914
1915 static int setup_propagate(const char *root) {
1916 const char *p, *q;
1917 int r;
1918
1919 (void) mkdir_p("/run/systemd/nspawn/", 0755);
1920 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
1921 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
1922 (void) mkdir_p(p, 0600);
1923
1924 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
1925 if (r < 0)
1926 return log_error_errno(r, "Failed to create /run/systemd: %m");
1927
1928 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
1929 if (r < 0)
1930 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
1931
1932 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
1933 if (r < 0)
1934 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
1935
1936 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
1937 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
1938 if (r < 0)
1939 return r;
1940
1941 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
1942 if (r < 0)
1943 return r;
1944
1945 /* machined will MS_MOVE into that directory, and that's only
1946 * supported for non-shared mounts. */
1947 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
1948 }
1949
1950 static int setup_machine_id(const char *directory) {
1951 const char *etc_machine_id;
1952 sd_id128_t id;
1953 int r;
1954
1955 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
1956 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
1957 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
1958 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
1959 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
1960 * container behaves nicely). */
1961
1962 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
1963
1964 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
1965 if (r < 0) {
1966 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
1967 return log_error_errno(r, "Failed to read machine ID from container image: %m");
1968
1969 if (sd_id128_is_null(arg_uuid)) {
1970 r = sd_id128_randomize(&arg_uuid);
1971 if (r < 0)
1972 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
1973 }
1974 } else {
1975 if (sd_id128_is_null(id)) {
1976 log_error("Machine ID in container image is zero, refusing.");
1977 return -EINVAL;
1978 }
1979
1980 arg_uuid = id;
1981 }
1982
1983 return 0;
1984 }
1985
1986 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
1987 int r;
1988
1989 assert(directory);
1990
1991 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
1992 return 0;
1993
1994 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
1995 if (r == -EOPNOTSUPP)
1996 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.");
1997 if (r == -EBADE)
1998 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
1999 if (r < 0)
2000 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2001 if (r == 0)
2002 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2003 else
2004 log_debug("Patched directory tree to match UID/GID range.");
2005
2006 return r;
2007 }
2008
2009 /*
2010 * Return values:
2011 * < 0 : wait_for_terminate() failed to get the state of the
2012 * container, the container was terminated by a signal, or
2013 * failed for an unknown reason. No change is made to the
2014 * container argument.
2015 * > 0 : The program executed in the container terminated with an
2016 * error. The exit code of the program executed in the
2017 * container is returned. The container argument has been set
2018 * to CONTAINER_TERMINATED.
2019 * 0 : The container is being rebooted, has been shut down or exited
2020 * successfully. The container argument has been set to either
2021 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2022 *
2023 * That is, success is indicated by a return value of zero, and an
2024 * error is indicated by a non-zero value.
2025 */
2026 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2027 siginfo_t status;
2028 int r;
2029
2030 r = wait_for_terminate(pid, &status);
2031 if (r < 0)
2032 return log_warning_errno(r, "Failed to wait for container: %m");
2033
2034 switch (status.si_code) {
2035
2036 case CLD_EXITED:
2037 if (status.si_status == 0)
2038 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2039 else
2040 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2041
2042 *container = CONTAINER_TERMINATED;
2043 return status.si_status;
2044
2045 case CLD_KILLED:
2046 if (status.si_status == SIGINT) {
2047 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2048 *container = CONTAINER_TERMINATED;
2049 return 0;
2050
2051 } else if (status.si_status == SIGHUP) {
2052 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2053 *container = CONTAINER_REBOOTED;
2054 return 0;
2055 }
2056
2057 _fallthrough_;
2058 case CLD_DUMPED:
2059 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2060 return -EIO;
2061
2062 default:
2063 log_error("Container %s failed due to unknown reason.", arg_machine);
2064 return -EIO;
2065 }
2066 }
2067
2068 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2069 pid_t pid;
2070
2071 pid = PTR_TO_PID(userdata);
2072 if (pid > 0) {
2073 if (kill(pid, arg_kill_signal) >= 0) {
2074 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2075 sd_event_source_set_userdata(s, NULL);
2076 return 0;
2077 }
2078 }
2079
2080 sd_event_exit(sd_event_source_get_event(s), 0);
2081 return 0;
2082 }
2083
2084 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2085 pid_t pid;
2086
2087 assert(s);
2088 assert(ssi);
2089
2090 pid = PTR_TO_PID(userdata);
2091
2092 for (;;) {
2093 siginfo_t si = {};
2094
2095 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2096 return log_error_errno(errno, "Failed to waitid(): %m");
2097 if (si.si_pid == 0) /* No pending children. */
2098 break;
2099 if (si.si_pid == pid) {
2100 /* The main process we care for has exited. Return from
2101 * signal handler but leave the zombie. */
2102 sd_event_exit(sd_event_source_get_event(s), 0);
2103 break;
2104 }
2105
2106 /* Reap all other children. */
2107 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2108 }
2109
2110 return 0;
2111 }
2112
2113 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2114 pid_t pid;
2115
2116 assert(m);
2117
2118 pid = PTR_TO_PID(userdata);
2119
2120 if (arg_kill_signal > 0) {
2121 log_info("Container termination requested. Attempting to halt container.");
2122 (void) kill(pid, arg_kill_signal);
2123 } else {
2124 log_info("Container termination requested. Exiting.");
2125 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2126 }
2127
2128 return 0;
2129 }
2130
2131 static int determine_names(void) {
2132 int r;
2133
2134 if (arg_template && !arg_directory && arg_machine) {
2135
2136 /* If --template= was specified then we should not
2137 * search for a machine, but instead create a new one
2138 * in /var/lib/machine. */
2139
2140 arg_directory = strjoin("/var/lib/machines/", arg_machine);
2141 if (!arg_directory)
2142 return log_oom();
2143 }
2144
2145 if (!arg_image && !arg_directory) {
2146 if (arg_machine) {
2147 _cleanup_(image_unrefp) Image *i = NULL;
2148
2149 r = image_find(arg_machine, &i);
2150 if (r < 0)
2151 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2152 if (r == 0) {
2153 log_error("No image for machine '%s'.", arg_machine);
2154 return -ENOENT;
2155 }
2156
2157 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2158 r = free_and_strdup(&arg_image, i->path);
2159 else
2160 r = free_and_strdup(&arg_directory, i->path);
2161 if (r < 0)
2162 return log_oom();
2163
2164 if (!arg_ephemeral)
2165 arg_read_only = arg_read_only || i->read_only;
2166 } else {
2167 r = safe_getcwd(&arg_directory);
2168 if (r < 0)
2169 return log_error_errno(r, "Failed to determine current directory: %m");
2170 }
2171
2172 if (!arg_directory && !arg_image) {
2173 log_error("Failed to determine path, please use -D or -i.");
2174 return -EINVAL;
2175 }
2176 }
2177
2178 if (!arg_machine) {
2179
2180 if (arg_directory && path_equal(arg_directory, "/"))
2181 arg_machine = gethostname_malloc();
2182 else {
2183 if (arg_image) {
2184 char *e;
2185
2186 arg_machine = strdup(basename(arg_image));
2187
2188 /* Truncate suffix if there is one */
2189 e = endswith(arg_machine, ".raw");
2190 if (e)
2191 *e = 0;
2192 } else
2193 arg_machine = strdup(basename(arg_directory));
2194 }
2195 if (!arg_machine)
2196 return log_oom();
2197
2198 hostname_cleanup(arg_machine);
2199 if (!machine_name_is_valid(arg_machine)) {
2200 log_error("Failed to determine machine name automatically, please use -M.");
2201 return -EINVAL;
2202 }
2203
2204 if (arg_ephemeral) {
2205 char *b;
2206
2207 /* Add a random suffix when this is an
2208 * ephemeral machine, so that we can run many
2209 * instances at once without manually having
2210 * to specify -M each time. */
2211
2212 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2213 return log_oom();
2214
2215 free(arg_machine);
2216 arg_machine = b;
2217 }
2218 }
2219
2220 return 0;
2221 }
2222
2223 static int chase_symlinks_and_update(char **p, unsigned flags) {
2224 char *chased;
2225 int r;
2226
2227 assert(p);
2228
2229 if (!*p)
2230 return 0;
2231
2232 r = chase_symlinks(*p, NULL, flags, &chased);
2233 if (r < 0)
2234 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2235
2236 free_and_replace(*p, chased);
2237 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2238 }
2239
2240 static int determine_uid_shift(const char *directory) {
2241 int r;
2242
2243 if (arg_userns_mode == USER_NAMESPACE_NO) {
2244 arg_uid_shift = 0;
2245 return 0;
2246 }
2247
2248 if (arg_uid_shift == UID_INVALID) {
2249 struct stat st;
2250
2251 r = stat(directory, &st);
2252 if (r < 0)
2253 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2254
2255 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2256
2257 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
2258 log_error("UID and GID base of %s don't match.", directory);
2259 return -EINVAL;
2260 }
2261
2262 arg_uid_range = UINT32_C(0x10000);
2263 }
2264
2265 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
2266 log_error("UID base too high for UID range.");
2267 return -EINVAL;
2268 }
2269
2270 return 0;
2271 }
2272
2273 static int inner_child(
2274 Barrier *barrier,
2275 const char *directory,
2276 bool secondary,
2277 int kmsg_socket,
2278 int rtnl_socket,
2279 FDSet *fds) {
2280
2281 _cleanup_free_ char *home = NULL;
2282 char as_uuid[37];
2283 unsigned n_env = 1;
2284 const char *envp[] = {
2285 "PATH=" DEFAULT_PATH_SPLIT_USR,
2286 NULL, /* container */
2287 NULL, /* TERM */
2288 NULL, /* HOME */
2289 NULL, /* USER */
2290 NULL, /* LOGNAME */
2291 NULL, /* container_uuid */
2292 NULL, /* LISTEN_FDS */
2293 NULL, /* LISTEN_PID */
2294 NULL, /* NOTIFY_SOCKET */
2295 NULL
2296 };
2297 const char *exec_target;
2298
2299 _cleanup_strv_free_ char **env_use = NULL;
2300 int r;
2301
2302 assert(barrier);
2303 assert(directory);
2304 assert(kmsg_socket >= 0);
2305
2306 if (arg_userns_mode != USER_NAMESPACE_NO) {
2307 /* Tell the parent, that it now can write the UID map. */
2308 (void) barrier_place(barrier); /* #1 */
2309
2310 /* Wait until the parent wrote the UID map */
2311 if (!barrier_place_and_sync(barrier)) { /* #2 */
2312 log_error("Parent died too early");
2313 return -ESRCH;
2314 }
2315 }
2316
2317 r = reset_uid_gid();
2318 if (r < 0)
2319 return log_error_errno(r, "Couldn't become new root: %m");
2320
2321 r = mount_all(NULL,
2322 arg_mount_settings | MOUNT_IN_USERNS,
2323 arg_uid_shift,
2324 arg_uid_range,
2325 arg_selinux_apifs_context);
2326 if (r < 0)
2327 return r;
2328
2329 if (!arg_network_namespace_path && arg_private_network) {
2330 r = unshare(CLONE_NEWNET);
2331 if (r < 0)
2332 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2333 }
2334
2335 r = mount_sysfs(NULL, arg_mount_settings);
2336 if (r < 0)
2337 return r;
2338
2339 /* Wait until we are cgroup-ified, so that we
2340 * can mount the right cgroup path writable */
2341 if (!barrier_place_and_sync(barrier)) { /* #3 */
2342 log_error("Parent died too early");
2343 return -ESRCH;
2344 }
2345
2346 if (arg_use_cgns && cg_ns_supported()) {
2347 r = unshare(CLONE_NEWCGROUP);
2348 if (r < 0)
2349 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2350 r = mount_cgroups(
2351 "",
2352 arg_unified_cgroup_hierarchy,
2353 arg_userns_mode != USER_NAMESPACE_NO,
2354 arg_uid_shift,
2355 arg_uid_range,
2356 arg_selinux_apifs_context,
2357 true);
2358 if (r < 0)
2359 return r;
2360 } else {
2361 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2362 if (r < 0)
2363 return r;
2364 }
2365
2366 r = setup_boot_id(NULL);
2367 if (r < 0)
2368 return r;
2369
2370 r = setup_kmsg(NULL, kmsg_socket);
2371 if (r < 0)
2372 return r;
2373 kmsg_socket = safe_close(kmsg_socket);
2374
2375 umask(0022);
2376
2377 if (setsid() < 0)
2378 return log_error_errno(errno, "setsid() failed: %m");
2379
2380 if (arg_private_network)
2381 loopback_setup();
2382
2383 if (arg_expose_ports) {
2384 r = expose_port_send_rtnl(rtnl_socket);
2385 if (r < 0)
2386 return r;
2387 rtnl_socket = safe_close(rtnl_socket);
2388 }
2389
2390 r = drop_capabilities();
2391 if (r < 0)
2392 return log_error_errno(r, "drop_capabilities() failed: %m");
2393
2394 setup_hostname();
2395
2396 if (arg_personality != PERSONALITY_INVALID) {
2397 r = safe_personality(arg_personality);
2398 if (r < 0)
2399 return log_error_errno(r, "personality() failed: %m");
2400 } else if (secondary) {
2401 r = safe_personality(PER_LINUX32);
2402 if (r < 0)
2403 return log_error_errno(r, "personality() failed: %m");
2404 }
2405
2406 #if HAVE_SELINUX
2407 if (arg_selinux_context)
2408 if (setexeccon(arg_selinux_context) < 0)
2409 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
2410 #endif
2411
2412 r = change_uid_gid(arg_user, &home);
2413 if (r < 0)
2414 return r;
2415
2416 /* LXC sets container=lxc, so follow the scheme here */
2417 envp[n_env++] = strjoina("container=", arg_container_service_name);
2418
2419 envp[n_env] = strv_find_prefix(environ, "TERM=");
2420 if (envp[n_env])
2421 n_env++;
2422
2423 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
2424 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
2425 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
2426 return log_oom();
2427
2428 assert(!sd_id128_is_null(arg_uuid));
2429
2430 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
2431 return log_oom();
2432
2433 if (fdset_size(fds) > 0) {
2434 r = fdset_cloexec(fds, false);
2435 if (r < 0)
2436 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
2437
2438 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
2439 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
2440 return log_oom();
2441 }
2442 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
2443 return log_oom();
2444
2445 env_use = strv_env_merge(2, envp, arg_setenv);
2446 if (!env_use)
2447 return log_oom();
2448
2449 /* Let the parent know that we are ready and
2450 * wait until the parent is ready with the
2451 * setup, too... */
2452 if (!barrier_place_and_sync(barrier)) { /* #4 */
2453 log_error("Parent died too early");
2454 return -ESRCH;
2455 }
2456
2457 if (arg_chdir)
2458 if (chdir(arg_chdir) < 0)
2459 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
2460
2461 if (arg_start_mode == START_PID2) {
2462 r = stub_pid1(arg_uuid);
2463 if (r < 0)
2464 return r;
2465 }
2466
2467 /* Now, explicitly close the log, so that we
2468 * then can close all remaining fds. Closing
2469 * the log explicitly first has the benefit
2470 * that the logging subsystem knows about it,
2471 * and is thus ready to be reopened should we
2472 * need it again. Note that the other fds
2473 * closed here are at least the locking and
2474 * barrier fds. */
2475 log_close();
2476 (void) fdset_close_others(fds);
2477
2478 if (arg_start_mode == START_BOOT) {
2479 char **a;
2480 size_t m;
2481
2482 /* Automatically search for the init system */
2483
2484 m = strv_length(arg_parameters);
2485 a = newa(char*, m + 2);
2486 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
2487 a[1 + m] = NULL;
2488
2489 a[0] = (char*) "/usr/lib/systemd/systemd";
2490 execve(a[0], a, env_use);
2491
2492 a[0] = (char*) "/lib/systemd/systemd";
2493 execve(a[0], a, env_use);
2494
2495 a[0] = (char*) "/sbin/init";
2496 execve(a[0], a, env_use);
2497
2498 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2499 } else if (!strv_isempty(arg_parameters)) {
2500 exec_target = arg_parameters[0];
2501 execvpe(arg_parameters[0], arg_parameters, env_use);
2502 } else {
2503 if (!arg_chdir)
2504 /* If we cannot change the directory, we'll end up in /, that is expected. */
2505 (void) chdir(home ?: "/root");
2506
2507 execle("/bin/bash", "-bash", NULL, env_use);
2508 execle("/bin/sh", "-sh", NULL, env_use);
2509
2510 exec_target = "/bin/bash, /bin/sh";
2511 }
2512
2513 r = -errno;
2514 (void) log_open();
2515 return log_error_errno(r, "execv(%s) failed: %m", exec_target);
2516 }
2517
2518 static int setup_sd_notify_child(void) {
2519 static const int one = 1;
2520 int fd = -1;
2521 union sockaddr_union sa = {
2522 .sa.sa_family = AF_UNIX,
2523 };
2524 int r;
2525
2526 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2527 if (fd < 0)
2528 return log_error_errno(errno, "Failed to allocate notification socket: %m");
2529
2530 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
2531 (void) unlink(NSPAWN_NOTIFY_SOCKET_PATH);
2532
2533 strncpy(sa.un.sun_path, NSPAWN_NOTIFY_SOCKET_PATH, sizeof(sa.un.sun_path)-1);
2534 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
2535 if (r < 0) {
2536 safe_close(fd);
2537 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
2538 }
2539
2540 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
2541 if (r < 0) {
2542 safe_close(fd);
2543 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
2544 }
2545
2546 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
2547 if (r < 0) {
2548 safe_close(fd);
2549 return log_error_errno(errno, "SO_PASSCRED failed: %m");
2550 }
2551
2552 return fd;
2553 }
2554
2555 static int outer_child(
2556 Barrier *barrier,
2557 const char *directory,
2558 const char *console,
2559 DissectedImage *dissected_image,
2560 bool interactive,
2561 bool secondary,
2562 int pid_socket,
2563 int uuid_socket,
2564 int notify_socket,
2565 int kmsg_socket,
2566 int rtnl_socket,
2567 int uid_shift_socket,
2568 int unified_cgroup_hierarchy_socket,
2569 FDSet *fds,
2570 int netns_fd) {
2571
2572 pid_t pid;
2573 ssize_t l;
2574 int r;
2575 _cleanup_close_ int fd = -1;
2576
2577 assert(barrier);
2578 assert(directory);
2579 assert(console);
2580 assert(pid_socket >= 0);
2581 assert(uuid_socket >= 0);
2582 assert(notify_socket >= 0);
2583 assert(kmsg_socket >= 0);
2584
2585 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
2586 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
2587
2588 if (interactive) {
2589 int terminal;
2590
2591 terminal = open_terminal(console, O_RDWR);
2592 if (terminal < 0)
2593 return log_error_errno(terminal, "Failed to open console: %m");
2594
2595 r = rearrange_stdio(terminal, terminal, terminal); /* invalidates 'terminal' on success and failure */
2596 if (r < 0)
2597 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2598 }
2599
2600 r = reset_audit_loginuid();
2601 if (r < 0)
2602 return r;
2603
2604 /* Mark everything as slave, so that we still
2605 * receive mounts from the real root, but don't
2606 * propagate mounts to the real root. */
2607 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
2608 if (r < 0)
2609 return r;
2610
2611 if (dissected_image) {
2612 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
2613 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
2614 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
2615 * makes sure ESP partitions and userns are compatible. */
2616
2617 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2618 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2619 if (r < 0)
2620 return r;
2621 }
2622
2623 r = determine_uid_shift(directory);
2624 if (r < 0)
2625 return r;
2626
2627 if (arg_userns_mode != USER_NAMESPACE_NO) {
2628 /* Let the parent know which UID shift we read from the image */
2629 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
2630 if (l < 0)
2631 return log_error_errno(errno, "Failed to send UID shift: %m");
2632 if (l != sizeof(arg_uid_shift)) {
2633 log_error("Short write while sending UID shift.");
2634 return -EIO;
2635 }
2636
2637 if (arg_userns_mode == USER_NAMESPACE_PICK) {
2638 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2639 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2640 * not it will pick a different one, and send it back to us. */
2641
2642 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
2643 if (l < 0)
2644 return log_error_errno(errno, "Failed to recv UID shift: %m");
2645 if (l != sizeof(arg_uid_shift)) {
2646 log_error("Short read while receiving UID shift.");
2647 return -EIO;
2648 }
2649 }
2650
2651 log_info("Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
2652 }
2653
2654 if (dissected_image) {
2655 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
2656 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2657 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2658 if (r < 0)
2659 return r;
2660 }
2661
2662 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
2663 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
2664
2665 r = detect_unified_cgroup_hierarchy_from_image(directory);
2666 if (r < 0)
2667 return r;
2668
2669 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
2670 if (l < 0)
2671 return log_error_errno(errno, "Failed to send cgroup mode: %m");
2672 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
2673 log_error("Short write while sending cgroup mode: %m");
2674 return -EIO;
2675 }
2676
2677 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
2678 }
2679
2680 /* Turn directory into bind mount */
2681 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
2682 if (r < 0)
2683 return r;
2684
2685 r = setup_pivot_root(
2686 directory,
2687 arg_pivot_root_new,
2688 arg_pivot_root_old);
2689 if (r < 0)
2690 return r;
2691
2692 r = setup_volatile(
2693 directory,
2694 arg_volatile_mode,
2695 arg_userns_mode != USER_NAMESPACE_NO,
2696 arg_uid_shift,
2697 arg_uid_range,
2698 arg_selinux_context);
2699 if (r < 0)
2700 return r;
2701
2702 r = setup_volatile_state(
2703 directory,
2704 arg_volatile_mode,
2705 arg_userns_mode != USER_NAMESPACE_NO,
2706 arg_uid_shift,
2707 arg_uid_range,
2708 arg_selinux_context);
2709 if (r < 0)
2710 return r;
2711
2712 /* Mark everything as shared so our mounts get propagated down. This is
2713 * required to make new bind mounts available in systemd services
2714 * inside the containter that create a new mount namespace.
2715 * See https://github.com/systemd/systemd/issues/3860
2716 * Further submounts (such as /dev) done after this will inherit the
2717 * shared propagation mode. */
2718 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
2719 if (r < 0)
2720 return r;
2721
2722 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
2723 if (r < 0)
2724 return r;
2725
2726 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
2727 if (r < 0)
2728 return r;
2729
2730 if (arg_read_only) {
2731 r = bind_remount_recursive(directory, true, NULL);
2732 if (r < 0)
2733 return log_error_errno(r, "Failed to make tree read-only: %m");
2734 }
2735
2736 r = mount_all(directory,
2737 arg_mount_settings,
2738 arg_uid_shift,
2739 arg_uid_range,
2740 arg_selinux_apifs_context);
2741 if (r < 0)
2742 return r;
2743
2744 r = copy_devnodes(directory);
2745 if (r < 0)
2746 return r;
2747
2748 dev_setup(directory, arg_uid_shift, arg_uid_shift);
2749
2750 r = setup_pts(directory);
2751 if (r < 0)
2752 return r;
2753
2754 r = setup_propagate(directory);
2755 if (r < 0)
2756 return r;
2757
2758 r = setup_dev_console(directory, console);
2759 if (r < 0)
2760 return r;
2761
2762 r = setup_keyring();
2763 if (r < 0)
2764 return r;
2765
2766 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
2767 if (r < 0)
2768 return r;
2769
2770 r = setup_timezone(directory);
2771 if (r < 0)
2772 return r;
2773
2774 r = setup_resolv_conf(directory);
2775 if (r < 0)
2776 return r;
2777
2778 r = setup_machine_id(directory);
2779 if (r < 0)
2780 return r;
2781
2782 r = setup_journal(directory);
2783 if (r < 0)
2784 return r;
2785
2786 r = mount_custom(
2787 directory,
2788 arg_custom_mounts,
2789 arg_n_custom_mounts,
2790 arg_userns_mode != USER_NAMESPACE_NO,
2791 arg_uid_shift,
2792 arg_uid_range,
2793 arg_selinux_apifs_context);
2794 if (r < 0)
2795 return r;
2796
2797 if (!arg_use_cgns || !cg_ns_supported()) {
2798 r = mount_cgroups(
2799 directory,
2800 arg_unified_cgroup_hierarchy,
2801 arg_userns_mode != USER_NAMESPACE_NO,
2802 arg_uid_shift,
2803 arg_uid_range,
2804 arg_selinux_apifs_context,
2805 false);
2806 if (r < 0)
2807 return r;
2808 }
2809
2810 r = mount_move_root(directory);
2811 if (r < 0)
2812 return log_error_errno(r, "Failed to move root directory: %m");
2813
2814 fd = setup_sd_notify_child();
2815 if (fd < 0)
2816 return fd;
2817
2818 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
2819 arg_clone_ns_flags |
2820 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
2821 if (pid < 0)
2822 return log_error_errno(errno, "Failed to fork inner child: %m");
2823 if (pid == 0) {
2824 pid_socket = safe_close(pid_socket);
2825 uuid_socket = safe_close(uuid_socket);
2826 notify_socket = safe_close(notify_socket);
2827 uid_shift_socket = safe_close(uid_shift_socket);
2828
2829 /* The inner child has all namespaces that are
2830 * requested, so that we all are owned by the user if
2831 * user namespaces are turned on. */
2832
2833 if (arg_network_namespace_path) {
2834 r = namespace_enter(-1, -1, netns_fd, -1, -1);
2835 if (r < 0)
2836 return r;
2837 }
2838
2839 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
2840 if (r < 0)
2841 _exit(EXIT_FAILURE);
2842
2843 _exit(EXIT_SUCCESS);
2844 }
2845
2846 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
2847 if (l < 0)
2848 return log_error_errno(errno, "Failed to send PID: %m");
2849 if (l != sizeof(pid)) {
2850 log_error("Short write while sending PID.");
2851 return -EIO;
2852 }
2853
2854 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
2855 if (l < 0)
2856 return log_error_errno(errno, "Failed to send machine ID: %m");
2857 if (l != sizeof(arg_uuid)) {
2858 log_error("Short write while sending machine ID.");
2859 return -EIO;
2860 }
2861
2862 l = send_one_fd(notify_socket, fd, 0);
2863 if (l < 0)
2864 return log_error_errno(errno, "Failed to send notify fd: %m");
2865
2866 pid_socket = safe_close(pid_socket);
2867 uuid_socket = safe_close(uuid_socket);
2868 notify_socket = safe_close(notify_socket);
2869 kmsg_socket = safe_close(kmsg_socket);
2870 rtnl_socket = safe_close(rtnl_socket);
2871 netns_fd = safe_close(netns_fd);
2872
2873 return 0;
2874 }
2875
2876 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
2877 bool tried_hashed = false;
2878 unsigned n_tries = 100;
2879 uid_t candidate;
2880 int r;
2881
2882 assert(shift);
2883 assert(ret_lock_file);
2884 assert(arg_userns_mode == USER_NAMESPACE_PICK);
2885 assert(arg_uid_range == 0x10000U);
2886
2887 candidate = *shift;
2888
2889 (void) mkdir("/run/systemd/nspawn-uid", 0755);
2890
2891 for (;;) {
2892 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
2893 _cleanup_release_lock_file_ LockFile lf = LOCK_FILE_INIT;
2894
2895 if (--n_tries <= 0)
2896 return -EBUSY;
2897
2898 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
2899 goto next;
2900 if ((candidate & UINT32_C(0xFFFF)) != 0)
2901 goto next;
2902
2903 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
2904 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
2905 if (r == -EBUSY) /* Range already taken by another nspawn instance */
2906 goto next;
2907 if (r < 0)
2908 return r;
2909
2910 /* Make some superficial checks whether the range is currently known in the user database */
2911 if (getpwuid(candidate))
2912 goto next;
2913 if (getpwuid(candidate + UINT32_C(0xFFFE)))
2914 goto next;
2915 if (getgrgid(candidate))
2916 goto next;
2917 if (getgrgid(candidate + UINT32_C(0xFFFE)))
2918 goto next;
2919
2920 *ret_lock_file = lf;
2921 lf = (struct LockFile) LOCK_FILE_INIT;
2922 *shift = candidate;
2923 return 0;
2924
2925 next:
2926 if (arg_machine && !tried_hashed) {
2927 /* Try to hash the base from the container name */
2928
2929 static const uint8_t hash_key[] = {
2930 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
2931 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
2932 };
2933
2934 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
2935
2936 tried_hashed = true;
2937 } else
2938 random_bytes(&candidate, sizeof(candidate));
2939
2940 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
2941 candidate &= (uid_t) UINT32_C(0xFFFF0000);
2942 }
2943 }
2944
2945 static int setup_uid_map(pid_t pid) {
2946 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
2947 int r;
2948
2949 assert(pid > 1);
2950
2951 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
2952 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
2953 r = write_string_file(uid_map, line, 0);
2954 if (r < 0)
2955 return log_error_errno(r, "Failed to write UID map: %m");
2956
2957 /* We always assign the same UID and GID ranges */
2958 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
2959 r = write_string_file(uid_map, line, 0);
2960 if (r < 0)
2961 return log_error_errno(r, "Failed to write GID map: %m");
2962
2963 return 0;
2964 }
2965
2966 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2967 char buf[NOTIFY_BUFFER_MAX+1];
2968 char *p = NULL;
2969 struct iovec iovec = {
2970 .iov_base = buf,
2971 .iov_len = sizeof(buf)-1,
2972 };
2973 union {
2974 struct cmsghdr cmsghdr;
2975 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
2976 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
2977 } control = {};
2978 struct msghdr msghdr = {
2979 .msg_iov = &iovec,
2980 .msg_iovlen = 1,
2981 .msg_control = &control,
2982 .msg_controllen = sizeof(control),
2983 };
2984 struct cmsghdr *cmsg;
2985 struct ucred *ucred = NULL;
2986 ssize_t n;
2987 pid_t inner_child_pid;
2988 _cleanup_strv_free_ char **tags = NULL;
2989
2990 assert(userdata);
2991
2992 inner_child_pid = PTR_TO_PID(userdata);
2993
2994 if (revents != EPOLLIN) {
2995 log_warning("Got unexpected poll event for notify fd.");
2996 return 0;
2997 }
2998
2999 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3000 if (n < 0) {
3001 if (IN_SET(errno, EAGAIN, EINTR))
3002 return 0;
3003
3004 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3005 }
3006 cmsg_close_all(&msghdr);
3007
3008 CMSG_FOREACH(cmsg, &msghdr) {
3009 if (cmsg->cmsg_level == SOL_SOCKET &&
3010 cmsg->cmsg_type == SCM_CREDENTIALS &&
3011 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3012
3013 ucred = (struct ucred*) CMSG_DATA(cmsg);
3014 }
3015 }
3016
3017 if (!ucred || ucred->pid != inner_child_pid) {
3018 log_debug("Received notify message without valid credentials. Ignoring.");
3019 return 0;
3020 }
3021
3022 if ((size_t) n >= sizeof(buf)) {
3023 log_warning("Received notify message exceeded maximum size. Ignoring.");
3024 return 0;
3025 }
3026
3027 buf[n] = 0;
3028 tags = strv_split(buf, "\n\r");
3029 if (!tags)
3030 return log_oom();
3031
3032 if (strv_find(tags, "READY=1"))
3033 sd_notifyf(false, "READY=1\n");
3034
3035 p = strv_find_startswith(tags, "STATUS=");
3036 if (p)
3037 sd_notifyf(false, "STATUS=Container running: %s", p);
3038
3039 return 0;
3040 }
3041
3042 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3043 int r;
3044
3045 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3046 if (r < 0)
3047 return log_error_errno(r, "Failed to allocate notify event source: %m");
3048
3049 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3050
3051 return 0;
3052 }
3053
3054 static int load_settings(void) {
3055 _cleanup_(settings_freep) Settings *settings = NULL;
3056 _cleanup_fclose_ FILE *f = NULL;
3057 _cleanup_free_ char *p = NULL;
3058 const char *fn, *i;
3059 int r;
3060
3061 /* If all settings are masked, there's no point in looking for
3062 * the settings file */
3063 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3064 return 0;
3065
3066 fn = strjoina(arg_machine, ".nspawn");
3067
3068 /* We first look in the admin's directories in /etc and /run */
3069 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3070 _cleanup_free_ char *j = NULL;
3071
3072 j = strjoin(i, "/", fn);
3073 if (!j)
3074 return log_oom();
3075
3076 f = fopen(j, "re");
3077 if (f) {
3078 p = j;
3079 j = NULL;
3080
3081 /* By default, we trust configuration from /etc and /run */
3082 if (arg_settings_trusted < 0)
3083 arg_settings_trusted = true;
3084
3085 break;
3086 }
3087
3088 if (errno != ENOENT)
3089 return log_error_errno(errno, "Failed to open %s: %m", j);
3090 }
3091
3092 if (!f) {
3093 /* After that, let's look for a file next to the
3094 * actual image we shall boot. */
3095
3096 if (arg_image) {
3097 p = file_in_same_dir(arg_image, fn);
3098 if (!p)
3099 return log_oom();
3100 } else if (arg_directory) {
3101 p = file_in_same_dir(arg_directory, fn);
3102 if (!p)
3103 return log_oom();
3104 }
3105
3106 if (p) {
3107 f = fopen(p, "re");
3108 if (!f && errno != ENOENT)
3109 return log_error_errno(errno, "Failed to open %s: %m", p);
3110
3111 /* By default, we do not trust configuration from /var/lib/machines */
3112 if (arg_settings_trusted < 0)
3113 arg_settings_trusted = false;
3114 }
3115 }
3116
3117 if (!f)
3118 return 0;
3119
3120 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
3121
3122 r = settings_load(f, p, &settings);
3123 if (r < 0)
3124 return r;
3125
3126 /* Copy over bits from the settings, unless they have been
3127 * explicitly masked by command line switches. */
3128
3129 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3130 settings->start_mode >= 0) {
3131 arg_start_mode = settings->start_mode;
3132
3133 strv_free(arg_parameters);
3134 arg_parameters = settings->parameters;
3135 settings->parameters = NULL;
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(arg_chdir);
3147 arg_chdir = settings->working_directory;
3148 settings->working_directory = NULL;
3149 }
3150
3151 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3152 settings->environment) {
3153 strv_free(arg_setenv);
3154 arg_setenv = settings->environment;
3155 settings->environment = NULL;
3156 }
3157
3158 if ((arg_settings_mask & SETTING_USER) == 0 &&
3159 settings->user) {
3160 free(arg_user);
3161 arg_user = settings->user;
3162 settings->user = NULL;
3163 }
3164
3165 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3166 uint64_t plus;
3167
3168 plus = settings->capability;
3169 if (settings_private_network(settings))
3170 plus |= (1ULL << CAP_NET_ADMIN);
3171
3172 if (!arg_settings_trusted && plus != 0) {
3173 if (settings->capability != 0)
3174 log_warning("Ignoring Capability= setting, file %s is not trusted.", p);
3175 } else
3176 arg_caps_retain |= plus;
3177
3178 arg_caps_retain &= ~settings->drop_capability;
3179 }
3180
3181 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3182 settings->kill_signal > 0)
3183 arg_kill_signal = settings->kill_signal;
3184
3185 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3186 settings->personality != PERSONALITY_INVALID)
3187 arg_personality = settings->personality;
3188
3189 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3190 !sd_id128_is_null(settings->machine_id)) {
3191
3192 if (!arg_settings_trusted)
3193 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p);
3194 else
3195 arg_uuid = settings->machine_id;
3196 }
3197
3198 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3199 settings->read_only >= 0)
3200 arg_read_only = settings->read_only;
3201
3202 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3203 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3204 arg_volatile_mode = settings->volatile_mode;
3205
3206 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3207 settings->n_custom_mounts > 0) {
3208
3209 if (!arg_settings_trusted)
3210 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p);
3211 else {
3212 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3213 arg_custom_mounts = settings->custom_mounts;
3214 arg_n_custom_mounts = settings->n_custom_mounts;
3215
3216 settings->custom_mounts = NULL;
3217 settings->n_custom_mounts = 0;
3218 }
3219 }
3220
3221 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3222 (settings->private_network >= 0 ||
3223 settings->network_veth >= 0 ||
3224 settings->network_bridge ||
3225 settings->network_zone ||
3226 settings->network_interfaces ||
3227 settings->network_macvlan ||
3228 settings->network_ipvlan ||
3229 settings->network_veth_extra)) {
3230
3231 if (!arg_settings_trusted)
3232 log_warning("Ignoring network settings, file %s is not trusted.", p);
3233 else {
3234 arg_network_veth = settings_network_veth(settings);
3235 arg_private_network = settings_private_network(settings);
3236
3237 strv_free(arg_network_interfaces);
3238 arg_network_interfaces = settings->network_interfaces;
3239 settings->network_interfaces = NULL;
3240
3241 strv_free(arg_network_macvlan);
3242 arg_network_macvlan = settings->network_macvlan;
3243 settings->network_macvlan = NULL;
3244
3245 strv_free(arg_network_ipvlan);
3246 arg_network_ipvlan = settings->network_ipvlan;
3247 settings->network_ipvlan = NULL;
3248
3249 strv_free(arg_network_veth_extra);
3250 arg_network_veth_extra = settings->network_veth_extra;
3251 settings->network_veth_extra = NULL;
3252
3253 free(arg_network_bridge);
3254 arg_network_bridge = settings->network_bridge;
3255 settings->network_bridge = NULL;
3256
3257 free(arg_network_zone);
3258 arg_network_zone = settings->network_zone;
3259 settings->network_zone = NULL;
3260 }
3261 }
3262
3263 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3264 settings->expose_ports) {
3265
3266 if (!arg_settings_trusted)
3267 log_warning("Ignoring Port= setting, file %s is not trusted.", p);
3268 else {
3269 expose_port_free_all(arg_expose_ports);
3270 arg_expose_ports = settings->expose_ports;
3271 settings->expose_ports = NULL;
3272 }
3273 }
3274
3275 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3276 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3277
3278 if (!arg_settings_trusted)
3279 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", p);
3280 else {
3281 arg_userns_mode = settings->userns_mode;
3282 arg_uid_shift = settings->uid_shift;
3283 arg_uid_range = settings->uid_range;
3284 arg_userns_chown = settings->userns_chown;
3285 }
3286 }
3287
3288 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3289 arg_notify_ready = settings->notify_ready;
3290
3291 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3292
3293 if (!arg_settings_trusted && !strv_isempty(arg_syscall_whitelist))
3294 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", p);
3295 else {
3296 strv_free(arg_syscall_whitelist);
3297 strv_free(arg_syscall_blacklist);
3298
3299 arg_syscall_whitelist = settings->syscall_whitelist;
3300 arg_syscall_blacklist = settings->syscall_blacklist;
3301
3302 settings->syscall_whitelist = settings->syscall_blacklist = NULL;
3303 }
3304 }
3305
3306 return 0;
3307 }
3308
3309 static int run(int master,
3310 const char* console,
3311 DissectedImage *dissected_image,
3312 bool interactive,
3313 bool secondary,
3314 FDSet *fds,
3315 char veth_name[IFNAMSIZ], bool *veth_created,
3316 union in_addr_union *exposed,
3317 pid_t *pid, int *ret) {
3318
3319 static const struct sigaction sa = {
3320 .sa_handler = nop_signal_handler,
3321 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
3322 };
3323
3324 _cleanup_release_lock_file_ LockFile uid_shift_lock = LOCK_FILE_INIT;
3325 _cleanup_close_ int etc_passwd_lock = -1;
3326 _cleanup_close_pair_ int
3327 kmsg_socket_pair[2] = { -1, -1 },
3328 rtnl_socket_pair[2] = { -1, -1 },
3329 pid_socket_pair[2] = { -1, -1 },
3330 uuid_socket_pair[2] = { -1, -1 },
3331 notify_socket_pair[2] = { -1, -1 },
3332 uid_shift_socket_pair[2] = { -1, -1 },
3333 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
3334
3335 _cleanup_close_ int notify_socket= -1;
3336 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
3337 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
3338 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
3339 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
3340 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
3341 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
3342 ContainerStatus container_status = 0;
3343 char last_char = 0;
3344 int ifi = 0, r;
3345 ssize_t l;
3346 sigset_t mask_chld;
3347 _cleanup_close_ int netns_fd = -1;
3348
3349 assert_se(sigemptyset(&mask_chld) == 0);
3350 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
3351
3352 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3353 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3354 * check with getpwuid() if the specific user already exists. Note that /etc might be
3355 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3356 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3357 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3358 * really ours. */
3359
3360 etc_passwd_lock = take_etc_passwd_lock(NULL);
3361 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
3362 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
3363 }
3364
3365 r = barrier_create(&barrier);
3366 if (r < 0)
3367 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
3368
3369 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
3370 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
3371
3372 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
3373 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
3374
3375 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
3376 return log_error_errno(errno, "Failed to create pid socket pair: %m");
3377
3378 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
3379 return log_error_errno(errno, "Failed to create id socket pair: %m");
3380
3381 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
3382 return log_error_errno(errno, "Failed to create notify socket pair: %m");
3383
3384 if (arg_userns_mode != USER_NAMESPACE_NO)
3385 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
3386 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
3387
3388 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
3389 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
3390 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
3391
3392 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3393 * parent's blocking calls and give it a chance to call wait() and terminate. */
3394 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
3395 if (r < 0)
3396 return log_error_errno(errno, "Failed to change the signal mask: %m");
3397
3398 r = sigaction(SIGCHLD, &sa, NULL);
3399 if (r < 0)
3400 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
3401
3402 if (arg_network_namespace_path) {
3403 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
3404 if (netns_fd < 0)
3405 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
3406
3407 r = fd_is_network_ns(netns_fd);
3408 if (r < 0 && r != -ENOTTY)
3409 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
3410 if (r == 0) {
3411 log_error("Path %s doesn't refer to a network namespace", arg_network_namespace_path);
3412 return -EINVAL;
3413 }
3414 }
3415
3416 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
3417 if (*pid < 0)
3418 return log_error_errno(errno, "clone() failed%s: %m",
3419 errno == EINVAL ?
3420 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3421
3422 if (*pid == 0) {
3423 /* The outer child only has a file system namespace. */
3424 barrier_set_role(&barrier, BARRIER_CHILD);
3425
3426 master = safe_close(master);
3427
3428 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
3429 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3430 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
3431 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
3432 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
3433 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
3434 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
3435
3436 (void) reset_all_signal_handlers();
3437 (void) reset_signal_mask();
3438
3439 r = outer_child(&barrier,
3440 arg_directory,
3441 console,
3442 dissected_image,
3443 interactive,
3444 secondary,
3445 pid_socket_pair[1],
3446 uuid_socket_pair[1],
3447 notify_socket_pair[1],
3448 kmsg_socket_pair[1],
3449 rtnl_socket_pair[1],
3450 uid_shift_socket_pair[1],
3451 unified_cgroup_hierarchy_socket_pair[1],
3452 fds,
3453 netns_fd);
3454 if (r < 0)
3455 _exit(EXIT_FAILURE);
3456
3457 _exit(EXIT_SUCCESS);
3458 }
3459
3460 barrier_set_role(&barrier, BARRIER_PARENT);
3461
3462 fds = fdset_free(fds);
3463
3464 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
3465 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
3466 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
3467 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
3468 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
3469 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
3470 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
3471
3472 if (arg_userns_mode != USER_NAMESPACE_NO) {
3473 /* The child just let us know the UID shift it might have read from the image. */
3474 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
3475 if (l < 0)
3476 return log_error_errno(errno, "Failed to read UID shift: %m");
3477 if (l != sizeof arg_uid_shift) {
3478 log_error("Short read while reading UID shift.");
3479 return -EIO;
3480 }
3481
3482 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3483 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3484 * image, but if that's already in use, pick a new one, and report back to the child,
3485 * which one we now picked. */
3486
3487 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
3488 if (r < 0)
3489 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
3490
3491 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
3492 if (l < 0)
3493 return log_error_errno(errno, "Failed to send UID shift: %m");
3494 if (l != sizeof arg_uid_shift) {
3495 log_error("Short write while writing UID shift.");
3496 return -EIO;
3497 }
3498 }
3499 }
3500
3501 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3502 /* The child let us know the support cgroup mode it might have read from the image. */
3503 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
3504 if (l < 0)
3505 return log_error_errno(errno, "Failed to read cgroup mode: %m");
3506 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
3507 log_error("Short read while reading cgroup mode.");
3508 return -EIO;
3509 }
3510 }
3511
3512 /* Wait for the outer child. */
3513 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
3514 if (r < 0)
3515 return r;
3516 if (r != EXIT_SUCCESS)
3517 return -EIO;
3518
3519 /* And now retrieve the PID of the inner child. */
3520 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
3521 if (l < 0)
3522 return log_error_errno(errno, "Failed to read inner child PID: %m");
3523 if (l != sizeof *pid) {
3524 log_error("Short read while reading inner child PID.");
3525 return -EIO;
3526 }
3527
3528 /* We also retrieve container UUID in case it was generated by outer child */
3529 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
3530 if (l < 0)
3531 return log_error_errno(errno, "Failed to read container machine ID: %m");
3532 if (l != sizeof(arg_uuid)) {
3533 log_error("Short read while reading container machined ID.");
3534 return -EIO;
3535 }
3536
3537 /* We also retrieve the socket used for notifications generated by outer child */
3538 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
3539 if (notify_socket < 0)
3540 return log_error_errno(notify_socket,
3541 "Failed to receive notification socket from the outer child: %m");
3542
3543 log_debug("Init process invoked as PID "PID_FMT, *pid);
3544
3545 if (arg_userns_mode != USER_NAMESPACE_NO) {
3546 if (!barrier_place_and_sync(&barrier)) { /* #1 */
3547 log_error("Child died too early.");
3548 return -ESRCH;
3549 }
3550
3551 r = setup_uid_map(*pid);
3552 if (r < 0)
3553 return r;
3554
3555 (void) barrier_place(&barrier); /* #2 */
3556 }
3557
3558 if (arg_private_network) {
3559
3560 r = move_network_interfaces(*pid, arg_network_interfaces);
3561 if (r < 0)
3562 return r;
3563
3564 if (arg_network_veth) {
3565 r = setup_veth(arg_machine, *pid, veth_name,
3566 arg_network_bridge || arg_network_zone);
3567 if (r < 0)
3568 return r;
3569 else if (r > 0)
3570 ifi = r;
3571
3572 if (arg_network_bridge) {
3573 /* Add the interface to a bridge */
3574 r = setup_bridge(veth_name, arg_network_bridge, false);
3575 if (r < 0)
3576 return r;
3577 if (r > 0)
3578 ifi = r;
3579 } else if (arg_network_zone) {
3580 /* Add the interface to a bridge, possibly creating it */
3581 r = setup_bridge(veth_name, arg_network_zone, true);
3582 if (r < 0)
3583 return r;
3584 if (r > 0)
3585 ifi = r;
3586 }
3587 }
3588
3589 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
3590 if (r < 0)
3591 return r;
3592
3593 /* We created the primary and extra veth links now; let's remember this, so that we know to
3594 remove them later on. Note that we don't bother with removing veth links that were created
3595 here when their setup failed half-way, because in that case the kernel should be able to
3596 remove them on its own, since they cannot be referenced by anything yet. */
3597 *veth_created = true;
3598
3599 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
3600 if (r < 0)
3601 return r;
3602
3603 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
3604 if (r < 0)
3605 return r;
3606 }
3607
3608 if (arg_register || !arg_keep_unit) {
3609 r = sd_bus_default_system(&bus);
3610 if (r < 0)
3611 return log_error_errno(r, "Failed to open system bus: %m");
3612 }
3613
3614 if (!arg_keep_unit) {
3615 /* When a new scope is created for this container, then we'll be registered as its controller, in which
3616 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
3617 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
3618
3619 r = sd_bus_match_signal_async(
3620 bus,
3621 NULL,
3622 "org.freedesktop.systemd1",
3623 NULL,
3624 "org.freedesktop.systemd1.Scope",
3625 "RequestStop",
3626 on_request_stop, NULL, PID_TO_PTR(*pid));
3627 if (r < 0)
3628 return log_error_errno(r, "Failed to request RequestStop match: %m");
3629 }
3630
3631 if (arg_register) {
3632
3633 r = register_machine(
3634 bus,
3635 arg_machine,
3636 *pid,
3637 arg_directory,
3638 arg_uuid,
3639 ifi,
3640 arg_slice,
3641 arg_custom_mounts, arg_n_custom_mounts,
3642 arg_kill_signal,
3643 arg_property,
3644 arg_keep_unit,
3645 arg_container_service_name);
3646 if (r < 0)
3647 return r;
3648
3649 } else if (!arg_keep_unit) {
3650
3651 r = allocate_scope(
3652 bus,
3653 arg_machine,
3654 *pid,
3655 arg_slice,
3656 arg_custom_mounts, arg_n_custom_mounts,
3657 arg_kill_signal,
3658 arg_property);
3659 if (r < 0)
3660 return r;
3661
3662 } else if (arg_slice || arg_property)
3663 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
3664
3665 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3666 if (r < 0)
3667 return r;
3668
3669 if (arg_keep_unit) {
3670 r = create_subcgroup(*pid, arg_unified_cgroup_hierarchy);
3671 if (r < 0)
3672 return r;
3673 }
3674
3675 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3676 if (r < 0)
3677 return r;
3678
3679 /* Notify the child that the parent is ready with all
3680 * its setup (including cgroup-ification), and that
3681 * the child can now hand over control to the code to
3682 * run inside the container. */
3683 (void) barrier_place(&barrier); /* #3 */
3684
3685 /* Block SIGCHLD here, before notifying child.
3686 * process_pty() will handle it with the other signals. */
3687 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
3688
3689 /* Reset signal to default */
3690 r = default_signals(SIGCHLD, -1);
3691 if (r < 0)
3692 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
3693
3694 r = sd_event_new(&event);
3695 if (r < 0)
3696 return log_error_errno(r, "Failed to get default event source: %m");
3697
3698 (void) sd_event_set_watchdog(event, true);
3699
3700 if (bus) {
3701 r = sd_bus_attach_event(bus, event, 0);
3702 if (r < 0)
3703 return log_error_errno(r, "Failed to attach bus to event loop: %m");
3704 }
3705
3706 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
3707 if (r < 0)
3708 return r;
3709
3710 /* Let the child know that we are ready and wait that the child is completely ready now. */
3711 if (!barrier_place_and_sync(&barrier)) { /* #4 */
3712 log_error("Child died too early.");
3713 return -ESRCH;
3714 }
3715
3716 /* At this point we have made use of the UID we picked, and thus nss-mymachines
3717 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
3718 etc_passwd_lock = safe_close(etc_passwd_lock);
3719
3720 sd_notifyf(false,
3721 "STATUS=Container running.\n"
3722 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
3723 if (!arg_notify_ready)
3724 sd_notify(false, "READY=1\n");
3725
3726 if (arg_kill_signal > 0) {
3727 /* Try to kill the init system on SIGINT or SIGTERM */
3728 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
3729 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
3730 } else {
3731 /* Immediately exit */
3732 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
3733 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
3734 }
3735
3736 /* Exit when the child exits */
3737 sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
3738
3739 if (arg_expose_ports) {
3740 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
3741 if (r < 0)
3742 return r;
3743
3744 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
3745 }
3746
3747 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3748
3749 r = pty_forward_new(event, master,
3750 PTY_FORWARD_IGNORE_VHANGUP | (interactive ? 0 : PTY_FORWARD_READ_ONLY),
3751 &forward);
3752 if (r < 0)
3753 return log_error_errno(r, "Failed to create PTY forwarder: %m");
3754
3755 r = sd_event_loop(event);
3756 if (r < 0)
3757 return log_error_errno(r, "Failed to run event loop: %m");
3758
3759 pty_forward_get_last_char(forward, &last_char);
3760
3761 forward = pty_forward_free(forward);
3762
3763 if (!arg_quiet && last_char != '\n')
3764 putc('\n', stdout);
3765
3766 /* Kill if it is not dead yet anyway */
3767 if (arg_register && !arg_keep_unit && bus)
3768 terminate_machine(bus, *pid);
3769
3770 /* Normally redundant, but better safe than sorry */
3771 (void) kill(*pid, SIGKILL);
3772
3773 r = wait_for_container(*pid, &container_status);
3774 *pid = 0;
3775
3776 if (r < 0)
3777 /* We failed to wait for the container, or the container exited abnormally. */
3778 return r;
3779 if (r > 0 || container_status == CONTAINER_TERMINATED) {
3780 /* r > 0 → The container exited with a non-zero status.
3781 * As a special case, we need to replace 133 with a different value,
3782 * because 133 is special-cased in the service file to reboot the container.
3783 * otherwise → The container exited with zero status and a reboot was not requested.
3784 */
3785 if (r == EXIT_FORCE_RESTART)
3786 r = EXIT_FAILURE; /* replace 133 with the general failure code */
3787 *ret = r;
3788 return 0; /* finito */
3789 }
3790
3791 /* CONTAINER_REBOOTED, loop again */
3792
3793 if (arg_keep_unit) {
3794 /* Special handling if we are running as a service: instead of simply
3795 * restarting the machine we want to restart the entire service, so let's
3796 * inform systemd about this with the special exit code 133. The service
3797 * file uses RestartForceExitStatus=133 so that this results in a full
3798 * nspawn restart. This is necessary since we might have cgroup parameters
3799 * set we want to have flushed out. */
3800 *ret = EXIT_FORCE_RESTART;
3801 return 0; /* finito */
3802 }
3803
3804 expose_port_flush(arg_expose_ports, exposed);
3805
3806 (void) remove_veth_links(veth_name, arg_network_veth_extra);
3807 *veth_created = false;
3808 return 1; /* loop again */
3809 }
3810
3811 int main(int argc, char *argv[]) {
3812
3813 _cleanup_free_ char *console = NULL;
3814 _cleanup_close_ int master = -1;
3815 _cleanup_fdset_free_ FDSet *fds = NULL;
3816 int r, n_fd_passed, ret = EXIT_SUCCESS;
3817 char veth_name[IFNAMSIZ] = "";
3818 bool secondary = false, remove_directory = false, remove_image = false;
3819 pid_t pid = 0;
3820 union in_addr_union exposed = {};
3821 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
3822 bool interactive, veth_created = false, remove_tmprootdir = false;
3823 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
3824 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
3825 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3826 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3827
3828 log_parse_environment();
3829 log_open();
3830
3831 /* Make sure rename_process() in the stub init process can work */
3832 saved_argv = argv;
3833 saved_argc = argc;
3834
3835 r = parse_argv(argc, argv);
3836 if (r <= 0)
3837 goto finish;
3838
3839 r = must_be_root();
3840 if (r < 0)
3841 goto finish;
3842
3843 r = determine_names();
3844 if (r < 0)
3845 goto finish;
3846
3847 r = load_settings();
3848 if (r < 0)
3849 goto finish;
3850
3851 r = verify_arguments();
3852 if (r < 0)
3853 goto finish;
3854
3855 r = detect_unified_cgroup_hierarchy_from_environment();
3856 if (r < 0)
3857 goto finish;
3858
3859 n_fd_passed = sd_listen_fds(false);
3860 if (n_fd_passed > 0) {
3861 r = fdset_new_listen_fds(&fds, false);
3862 if (r < 0) {
3863 log_error_errno(r, "Failed to collect file descriptors: %m");
3864 goto finish;
3865 }
3866 }
3867
3868 if (arg_directory) {
3869 assert(!arg_image);
3870
3871 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
3872 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3873 r = -EINVAL;
3874 goto finish;
3875 }
3876
3877 if (arg_ephemeral) {
3878 _cleanup_free_ char *np = NULL;
3879
3880 r = chase_symlinks_and_update(&arg_directory, 0);
3881 if (r < 0)
3882 goto finish;
3883
3884 /* If the specified path is a mount point we
3885 * generate the new snapshot immediately
3886 * inside it under a random name. However if
3887 * the specified is not a mount point we
3888 * create the new snapshot in the parent
3889 * directory, just next to it. */
3890 r = path_is_mount_point(arg_directory, NULL, 0);
3891 if (r < 0) {
3892 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
3893 goto finish;
3894 }
3895 if (r > 0)
3896 r = tempfn_random_child(arg_directory, "machine.", &np);
3897 else
3898 r = tempfn_random(arg_directory, "machine.", &np);
3899 if (r < 0) {
3900 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
3901 goto finish;
3902 }
3903
3904 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3905 if (r < 0) {
3906 log_error_errno(r, "Failed to lock %s: %m", np);
3907 goto finish;
3908 }
3909
3910 r = btrfs_subvol_snapshot(arg_directory, np,
3911 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3912 BTRFS_SNAPSHOT_FALLBACK_COPY |
3913 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3914 BTRFS_SNAPSHOT_RECURSIVE |
3915 BTRFS_SNAPSHOT_QUOTA);
3916 if (r < 0) {
3917 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
3918 goto finish;
3919 }
3920
3921 free(arg_directory);
3922 arg_directory = np;
3923 np = NULL;
3924
3925 remove_directory = true;
3926
3927 } else {
3928 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
3929 if (r < 0)
3930 goto finish;
3931
3932 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3933 if (r == -EBUSY) {
3934 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
3935 goto finish;
3936 }
3937 if (r < 0) {
3938 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
3939 goto finish;
3940 }
3941
3942 if (arg_template) {
3943 r = chase_symlinks_and_update(&arg_template, 0);
3944 if (r < 0)
3945 goto finish;
3946
3947 r = btrfs_subvol_snapshot(arg_template, arg_directory,
3948 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3949 BTRFS_SNAPSHOT_FALLBACK_COPY |
3950 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3951 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
3952 BTRFS_SNAPSHOT_RECURSIVE |
3953 BTRFS_SNAPSHOT_QUOTA);
3954 if (r == -EEXIST) {
3955 if (!arg_quiet)
3956 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
3957 } else if (r < 0) {
3958 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
3959 goto finish;
3960 } else {
3961 if (!arg_quiet)
3962 log_info("Populated %s from template %s.", arg_directory, arg_template);
3963 }
3964 }
3965 }
3966
3967 if (arg_start_mode == START_BOOT) {
3968 if (path_is_os_tree(arg_directory) <= 0) {
3969 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
3970 r = -EINVAL;
3971 goto finish;
3972 }
3973 } else {
3974 const char *p;
3975
3976 p = strjoina(arg_directory, "/usr/");
3977 if (laccess(p, F_OK) < 0) {
3978 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", arg_directory);
3979 r = -EINVAL;
3980 goto finish;
3981 }
3982 }
3983
3984 } else {
3985 assert(arg_image);
3986 assert(!arg_template);
3987
3988 r = chase_symlinks_and_update(&arg_image, 0);
3989 if (r < 0)
3990 goto finish;
3991
3992 if (arg_ephemeral) {
3993 _cleanup_free_ char *np = NULL;
3994
3995 r = tempfn_random(arg_image, "machine.", &np);
3996 if (r < 0) {
3997 log_error_errno(r, "Failed to generate name for image snapshot: %m");
3998 goto finish;
3999 }
4000
4001 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4002 if (r < 0) {
4003 r = log_error_errno(r, "Failed to create image lock: %m");
4004 goto finish;
4005 }
4006
4007 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, COPY_REFLINK);
4008 if (r < 0) {
4009 r = log_error_errno(r, "Failed to copy image file: %m");
4010 goto finish;
4011 }
4012
4013 free(arg_image);
4014 arg_image = np;
4015 np = NULL;
4016
4017 remove_image = true;
4018 } else {
4019 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4020 if (r == -EBUSY) {
4021 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4022 goto finish;
4023 }
4024 if (r < 0) {
4025 r = log_error_errno(r, "Failed to create image lock: %m");
4026 goto finish;
4027 }
4028
4029 if (!arg_root_hash) {
4030 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4031 if (r < 0) {
4032 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4033 goto finish;
4034 }
4035 }
4036 }
4037
4038 if (!mkdtemp(tmprootdir)) {
4039 r = log_error_errno(errno, "Failed to create temporary directory: %m");
4040 goto finish;
4041 }
4042
4043 remove_tmprootdir = true;
4044
4045 arg_directory = strdup(tmprootdir);
4046 if (!arg_directory) {
4047 r = log_oom();
4048 goto finish;
4049 }
4050
4051 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4052 if (r < 0) {
4053 log_error_errno(r, "Failed to set up loopback block device: %m");
4054 goto finish;
4055 }
4056
4057 r = dissect_image(
4058 loop->fd,
4059 arg_root_hash, arg_root_hash_size,
4060 DISSECT_IMAGE_REQUIRE_ROOT,
4061 &dissected_image);
4062 if (r == -ENOPKG) {
4063 log_error_errno(r, "Could not find a suitable file system or partition table in image: %s", arg_image);
4064
4065 log_notice("Note that the disk image needs to\n"
4066 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4067 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4068 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4069 " d) or contain a file system without a partition table\n"
4070 "in order to be bootable with systemd-nspawn.");
4071 goto finish;
4072 }
4073 if (r == -EADDRNOTAVAIL) {
4074 log_error_errno(r, "No root partition for specified root hash found.");
4075 goto finish;
4076 }
4077 if (r == -EOPNOTSUPP) {
4078 log_error_errno(r, "--image= is not supported, compiled without blkid support.");
4079 goto finish;
4080 }
4081 if (r == -EPROTONOSUPPORT) {
4082 log_error_errno(r, "Device is loopback block device with partition scanning turned off, please turn it on.");
4083 goto finish;
4084 }
4085 if (r < 0) {
4086 log_error_errno(r, "Failed to dissect image: %m");
4087 goto finish;
4088 }
4089
4090 if (!arg_root_hash && dissected_image->can_verity)
4091 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4092
4093 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
4094 if (r < 0)
4095 goto finish;
4096
4097 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4098 if (remove_image && unlink(arg_image) >= 0)
4099 remove_image = false;
4100 }
4101
4102 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
4103 if (r < 0)
4104 goto finish;
4105
4106 interactive =
4107 isatty(STDIN_FILENO) > 0 &&
4108 isatty(STDOUT_FILENO) > 0;
4109
4110 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
4111 if (master < 0) {
4112 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4113 goto finish;
4114 }
4115
4116 r = ptsname_malloc(master, &console);
4117 if (r < 0) {
4118 r = log_error_errno(r, "Failed to determine tty name: %m");
4119 goto finish;
4120 }
4121
4122 if (arg_selinux_apifs_context) {
4123 r = mac_selinux_apply(console, arg_selinux_apifs_context);
4124 if (r < 0)
4125 goto finish;
4126 }
4127
4128 if (unlockpt(master) < 0) {
4129 r = log_error_errno(errno, "Failed to unlock tty: %m");
4130 goto finish;
4131 }
4132
4133 if (!arg_quiet)
4134 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4135 arg_machine, arg_image ?: arg_directory);
4136
4137 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
4138
4139 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
4140 r = log_error_errno(errno, "Failed to become subreaper: %m");
4141 goto finish;
4142 }
4143
4144 for (;;) {
4145 r = run(master,
4146 console,
4147 dissected_image,
4148 interactive, secondary,
4149 fds,
4150 veth_name, &veth_created,
4151 &exposed,
4152 &pid, &ret);
4153 if (r <= 0)
4154 break;
4155 }
4156
4157 finish:
4158 sd_notify(false,
4159 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
4160 "STOPPING=1\nSTATUS=Terminating...");
4161
4162 if (pid > 0)
4163 (void) kill(pid, SIGKILL);
4164
4165 /* Try to flush whatever is still queued in the pty */
4166 if (master >= 0) {
4167 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
4168 master = safe_close(master);
4169 }
4170
4171 if (pid > 0)
4172 (void) wait_for_terminate(pid, NULL);
4173
4174 if (remove_directory && arg_directory) {
4175 int k;
4176
4177 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
4178 if (k < 0)
4179 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
4180 }
4181
4182 if (remove_image && arg_image) {
4183 if (unlink(arg_image) < 0)
4184 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
4185 }
4186
4187 if (remove_tmprootdir) {
4188 if (rmdir(tmprootdir) < 0)
4189 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
4190 }
4191
4192 if (arg_machine) {
4193 const char *p;
4194
4195 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
4196 (void) rm_rf(p, REMOVE_ROOT);
4197 }
4198
4199 expose_port_flush(arg_expose_ports, &exposed);
4200
4201 if (veth_created)
4202 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4203 (void) remove_bridge(arg_network_zone);
4204
4205 free(arg_directory);
4206 free(arg_template);
4207 free(arg_image);
4208 free(arg_machine);
4209 free(arg_user);
4210 free(arg_pivot_root_new);
4211 free(arg_pivot_root_old);
4212 free(arg_chdir);
4213 strv_free(arg_setenv);
4214 free(arg_network_bridge);
4215 strv_free(arg_network_interfaces);
4216 strv_free(arg_network_macvlan);
4217 strv_free(arg_network_ipvlan);
4218 strv_free(arg_network_veth_extra);
4219 strv_free(arg_parameters);
4220 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4221 expose_port_free_all(arg_expose_ports);
4222 free(arg_root_hash);
4223
4224 return r < 0 ? EXIT_FAILURE : ret;
4225 }
Unnamed repository; edit this file 'description' to name the repository.