]> 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 #8676 from keszybz/drop-license-boilerplate
[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
8 #if HAVE_BLKID
9 #include <blkid.h>
10 #endif
11 #include <errno.h>
12 #include <getopt.h>
13 #include <grp.h>
14 #include <linux/loop.h>
15 #include <pwd.h>
16 #include <sched.h>
17 #if HAVE_SELINUX
18 #include <selinux/selinux.h>
19 #endif
20 #include <signal.h>
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <sys/file.h>
25 #include <sys/mount.h>
26 #include <sys/personality.h>
27 #include <sys/prctl.h>
28 #include <sys/types.h>
29 #include <sys/wait.h>
30 #include <unistd.h>
31
32 #include "sd-bus.h"
33 #include "sd-daemon.h"
34 #include "sd-id128.h"
35
36 #include "alloc-util.h"
37 #include "barrier.h"
38 #include "base-filesystem.h"
39 #include "blkid-util.h"
40 #include "btrfs-util.h"
41 #include "bus-util.h"
42 #include "cap-list.h"
43 #include "capability-util.h"
44 #include "cgroup-util.h"
45 #include "copy.h"
46 #include "dev-setup.h"
47 #include "dissect-image.h"
48 #include "env-util.h"
49 #include "fd-util.h"
50 #include "fdset.h"
51 #include "fileio.h"
52 #include "format-util.h"
53 #include "fs-util.h"
54 #include "gpt.h"
55 #include "hexdecoct.h"
56 #include "hostname-util.h"
57 #include "id128-util.h"
58 #include "log.h"
59 #include "loop-util.h"
60 #include "loopback-setup.h"
61 #include "machine-image.h"
62 #include "macro.h"
63 #include "missing.h"
64 #include "mkdir.h"
65 #include "mount-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-def.h"
69 #include "nspawn-expose-ports.h"
70 #include "nspawn-mount.h"
71 #include "nspawn-network.h"
72 #include "nspawn-patch-uid.h"
73 #include "nspawn-register.h"
74 #include "nspawn-seccomp.h"
75 #include "nspawn-settings.h"
76 #include "nspawn-setuid.h"
77 #include "nspawn-stub-pid1.h"
78 #include "parse-util.h"
79 #include "path-util.h"
80 #include "process-util.h"
81 #include "ptyfwd.h"
82 #include "random-util.h"
83 #include "raw-clone.h"
84 #include "rm-rf.h"
85 #include "selinux-util.h"
86 #include "signal-util.h"
87 #include "socket-util.h"
88 #include "stat-util.h"
89 #include "stdio-util.h"
90 #include "string-util.h"
91 #include "strv.h"
92 #include "terminal-util.h"
93 #include "udev-util.h"
94 #include "umask-util.h"
95 #include "user-util.h"
96 #include "util.h"
97
98 #if HAVE_SPLIT_USR
99 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
100 #else
101 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
102 #endif
103
104 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
105 * nspawn_notify_socket_path is relative to the container
106 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
107 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
108
109 #define EXIT_FORCE_RESTART 133
110
111 typedef enum ContainerStatus {
112 CONTAINER_TERMINATED,
113 CONTAINER_REBOOTED
114 } ContainerStatus;
115
116 typedef enum LinkJournal {
117 LINK_NO,
118 LINK_AUTO,
119 LINK_HOST,
120 LINK_GUEST
121 } LinkJournal;
122
123 static char *arg_directory = NULL;
124 static char *arg_template = NULL;
125 static char *arg_chdir = NULL;
126 static char *arg_pivot_root_new = NULL;
127 static char *arg_pivot_root_old = NULL;
128 static char *arg_user = NULL;
129 static sd_id128_t arg_uuid = {};
130 static char *arg_machine = NULL;
131 static const char *arg_selinux_context = NULL;
132 static const char *arg_selinux_apifs_context = NULL;
133 static const char *arg_slice = NULL;
134 static bool arg_private_network = false;
135 static bool arg_read_only = false;
136 static StartMode arg_start_mode = START_PID1;
137 static bool arg_ephemeral = false;
138 static LinkJournal arg_link_journal = LINK_AUTO;
139 static bool arg_link_journal_try = false;
140 static uint64_t arg_caps_retain =
141 (1ULL << CAP_AUDIT_CONTROL) |
142 (1ULL << CAP_AUDIT_WRITE) |
143 (1ULL << CAP_CHOWN) |
144 (1ULL << CAP_DAC_OVERRIDE) |
145 (1ULL << CAP_DAC_READ_SEARCH) |
146 (1ULL << CAP_FOWNER) |
147 (1ULL << CAP_FSETID) |
148 (1ULL << CAP_IPC_OWNER) |
149 (1ULL << CAP_KILL) |
150 (1ULL << CAP_LEASE) |
151 (1ULL << CAP_LINUX_IMMUTABLE) |
152 (1ULL << CAP_MKNOD) |
153 (1ULL << CAP_NET_BIND_SERVICE) |
154 (1ULL << CAP_NET_BROADCAST) |
155 (1ULL << CAP_NET_RAW) |
156 (1ULL << CAP_SETFCAP) |
157 (1ULL << CAP_SETGID) |
158 (1ULL << CAP_SETPCAP) |
159 (1ULL << CAP_SETUID) |
160 (1ULL << CAP_SYS_ADMIN) |
161 (1ULL << CAP_SYS_BOOT) |
162 (1ULL << CAP_SYS_CHROOT) |
163 (1ULL << CAP_SYS_NICE) |
164 (1ULL << CAP_SYS_PTRACE) |
165 (1ULL << CAP_SYS_RESOURCE) |
166 (1ULL << CAP_SYS_TTY_CONFIG);
167 static CustomMount *arg_custom_mounts = NULL;
168 static unsigned arg_n_custom_mounts = 0;
169 static char **arg_setenv = NULL;
170 static bool arg_quiet = false;
171 static bool arg_register = true;
172 static bool arg_keep_unit = false;
173 static char **arg_network_interfaces = NULL;
174 static char **arg_network_macvlan = NULL;
175 static char **arg_network_ipvlan = NULL;
176 static bool arg_network_veth = false;
177 static char **arg_network_veth_extra = NULL;
178 static char *arg_network_bridge = NULL;
179 static char *arg_network_zone = NULL;
180 static char *arg_network_namespace_path = NULL;
181 static unsigned long arg_personality = PERSONALITY_INVALID;
182 static char *arg_image = NULL;
183 static VolatileMode arg_volatile_mode = VOLATILE_NO;
184 static ExposePort *arg_expose_ports = NULL;
185 static char **arg_property = NULL;
186 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
187 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
188 static bool arg_userns_chown = false;
189 static int arg_kill_signal = 0;
190 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
191 static SettingsMask arg_settings_mask = 0;
192 static int arg_settings_trusted = -1;
193 static char **arg_parameters = NULL;
194 static const char *arg_container_service_name = "systemd-nspawn";
195 static bool arg_notify_ready = false;
196 static bool arg_use_cgns = true;
197 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
198 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO;
199 static void *arg_root_hash = NULL;
200 static size_t arg_root_hash_size = 0;
201 static char **arg_syscall_whitelist = NULL;
202 static char **arg_syscall_blacklist = NULL;
203
204 static void help(void) {
205 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
206 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
207 " -h --help Show this help\n"
208 " --version Print version string\n"
209 " -q --quiet Do not show status information\n"
210 " -D --directory=PATH Root directory for the container\n"
211 " --template=PATH Initialize root directory from template directory,\n"
212 " if missing\n"
213 " -x --ephemeral Run container with snapshot of root directory, and\n"
214 " remove it after exit\n"
215 " -i --image=PATH File system device or disk image for the container\n"
216 " --root-hash=HASH Specify verity root hash\n"
217 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
218 " -b --boot Boot up full system (i.e. invoke init)\n"
219 " --chdir=PATH Set working directory in the container\n"
220 " --pivot-root=PATH[:PATH]\n"
221 " Pivot root to given directory in the container\n"
222 " -u --user=USER Run the command under specified user or uid\n"
223 " -M --machine=NAME Set the machine name for the container\n"
224 " --uuid=UUID Set a specific machine UUID for the container\n"
225 " -S --slice=SLICE Place the container in the specified slice\n"
226 " --property=NAME=VALUE Set scope unit property\n"
227 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
228 " --private-users[=UIDBASE[:NUIDS]]\n"
229 " Similar, but with user configured UID/GID range\n"
230 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n"
231 " --private-network Disable network in container\n"
232 " --network-interface=INTERFACE\n"
233 " Assign an existing network interface to the\n"
234 " container\n"
235 " --network-macvlan=INTERFACE\n"
236 " Create a macvlan network interface based on an\n"
237 " existing network interface to the container\n"
238 " --network-ipvlan=INTERFACE\n"
239 " Create a ipvlan network interface based on an\n"
240 " existing network interface to the container\n"
241 " -n --network-veth Add a virtual Ethernet connection between host\n"
242 " and container\n"
243 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
244 " Add an additional virtual Ethernet link between\n"
245 " host and container\n"
246 " --network-bridge=INTERFACE\n"
247 " Add a virtual Ethernet connection to the container\n"
248 " and attach it to an existing bridge on the host\n"
249 " --network-zone=NAME Similar, but attach the new interface to an\n"
250 " an automatically managed bridge interface\n"
251 " --network-namespace-path=PATH\n"
252 " Set network namespace to the one represented by\n"
253 " the specified kernel namespace file node\n"
254 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
255 " Expose a container IP port on the host\n"
256 " -Z --selinux-context=SECLABEL\n"
257 " Set the SELinux security context to be used by\n"
258 " processes in the container\n"
259 " -L --selinux-apifs-context=SECLABEL\n"
260 " Set the SELinux security context to be used by\n"
261 " API/tmpfs file systems in the container\n"
262 " --capability=CAP In addition to the default, retain specified\n"
263 " capability\n"
264 " --drop-capability=CAP Drop the specified capability from the default set\n"
265 " --system-call-filter=LIST|~LIST\n"
266 " Permit/prohibit specific system calls\n"
267 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
268 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
269 " host, try-guest, try-host\n"
270 " -j Equivalent to --link-journal=try-guest\n"
271 " --read-only Mount the root directory read-only\n"
272 " --bind=PATH[:PATH[:OPTIONS]]\n"
273 " Bind mount a file or directory from the host into\n"
274 " the container\n"
275 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
276 " Similar, but creates a read-only bind mount\n"
277 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
278 " --overlay=PATH[:PATH...]:PATH\n"
279 " Create an overlay mount from the host to \n"
280 " the container\n"
281 " --overlay-ro=PATH[:PATH...]:PATH\n"
282 " Similar, but creates a read-only overlay mount\n"
283 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
284 " --register=BOOLEAN Register container as machine\n"
285 " --keep-unit Do not register a scope for the machine, reuse\n"
286 " the service unit nspawn is running in\n"
287 " --volatile[=MODE] Run the system in volatile mode\n"
288 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
289 " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
290 , program_invocation_short_name);
291 }
292
293 static int custom_mount_check_all(void) {
294 unsigned i;
295
296 for (i = 0; i < arg_n_custom_mounts; i++) {
297 CustomMount *m = &arg_custom_mounts[i];
298
299 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
300
301 if (arg_userns_chown) {
302 log_error("--private-users-chown may not be combined with custom root mounts.");
303 return -EINVAL;
304 } else if (arg_uid_shift == UID_INVALID) {
305 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
306 return -EINVAL;
307 }
308 }
309 }
310
311 return 0;
312 }
313
314 static int detect_unified_cgroup_hierarchy_from_environment(void) {
315 const char *e;
316 int r;
317
318 /* Allow the user to control whether the unified hierarchy is used */
319 e = getenv("UNIFIED_CGROUP_HIERARCHY");
320 if (e) {
321 r = parse_boolean(e);
322 if (r < 0)
323 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
324 if (r > 0)
325 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
326 else
327 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
328 }
329
330 return 0;
331 }
332
333 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
334 int r;
335
336 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd in the
337 * image actually supports. */
338 r = cg_all_unified();
339 if (r < 0)
340 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
341 if (r > 0) {
342 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
343 * routine only detects 231, so we'll have a false negative here for 230. */
344 r = systemd_installation_has_version(directory, 230);
345 if (r < 0)
346 return log_error_errno(r, "Failed to determine systemd version in container: %m");
347 if (r > 0)
348 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
349 else
350 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
351 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
352 /* Mixed cgroup hierarchy support was added in 233 */
353 r = systemd_installation_has_version(directory, 233);
354 if (r < 0)
355 return log_error_errno(r, "Failed to determine systemd version in container: %m");
356 if (r > 0)
357 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
358 else
359 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
360 } else
361 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
362
363 log_debug("Using %s hierarchy for container.",
364 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
365 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
366
367 return 0;
368 }
369
370 static void parse_share_ns_env(const char *name, unsigned long ns_flag) {
371 int r;
372
373 r = getenv_bool(name);
374 if (r == -ENXIO)
375 return;
376 if (r < 0)
377 log_warning_errno(r, "Failed to parse %s from environment, defaulting to false.", name);
378 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
379 }
380
381 static void parse_mount_settings_env(void) {
382 int r;
383 const char *e;
384
385 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
386 if (!e)
387 return;
388
389 if (streq(e, "network")) {
390 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
391 return;
392 }
393
394 r = parse_boolean(e);
395 if (r < 0) {
396 log_warning_errno(r, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
397 return;
398 }
399
400 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
401 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
402 }
403
404 static int parse_argv(int argc, char *argv[]) {
405
406 enum {
407 ARG_VERSION = 0x100,
408 ARG_PRIVATE_NETWORK,
409 ARG_UUID,
410 ARG_READ_ONLY,
411 ARG_CAPABILITY,
412 ARG_DROP_CAPABILITY,
413 ARG_LINK_JOURNAL,
414 ARG_BIND,
415 ARG_BIND_RO,
416 ARG_TMPFS,
417 ARG_OVERLAY,
418 ARG_OVERLAY_RO,
419 ARG_SHARE_SYSTEM,
420 ARG_REGISTER,
421 ARG_KEEP_UNIT,
422 ARG_NETWORK_INTERFACE,
423 ARG_NETWORK_MACVLAN,
424 ARG_NETWORK_IPVLAN,
425 ARG_NETWORK_BRIDGE,
426 ARG_NETWORK_ZONE,
427 ARG_NETWORK_VETH_EXTRA,
428 ARG_NETWORK_NAMESPACE_PATH,
429 ARG_PERSONALITY,
430 ARG_VOLATILE,
431 ARG_TEMPLATE,
432 ARG_PROPERTY,
433 ARG_PRIVATE_USERS,
434 ARG_KILL_SIGNAL,
435 ARG_SETTINGS,
436 ARG_CHDIR,
437 ARG_PIVOT_ROOT,
438 ARG_PRIVATE_USERS_CHOWN,
439 ARG_NOTIFY_READY,
440 ARG_ROOT_HASH,
441 ARG_SYSTEM_CALL_FILTER,
442 };
443
444 static const struct option options[] = {
445 { "help", no_argument, NULL, 'h' },
446 { "version", no_argument, NULL, ARG_VERSION },
447 { "directory", required_argument, NULL, 'D' },
448 { "template", required_argument, NULL, ARG_TEMPLATE },
449 { "ephemeral", no_argument, NULL, 'x' },
450 { "user", required_argument, NULL, 'u' },
451 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
452 { "as-pid2", no_argument, NULL, 'a' },
453 { "boot", no_argument, NULL, 'b' },
454 { "uuid", required_argument, NULL, ARG_UUID },
455 { "read-only", no_argument, NULL, ARG_READ_ONLY },
456 { "capability", required_argument, NULL, ARG_CAPABILITY },
457 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
458 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
459 { "bind", required_argument, NULL, ARG_BIND },
460 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
461 { "tmpfs", required_argument, NULL, ARG_TMPFS },
462 { "overlay", required_argument, NULL, ARG_OVERLAY },
463 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
464 { "machine", required_argument, NULL, 'M' },
465 { "slice", required_argument, NULL, 'S' },
466 { "setenv", required_argument, NULL, 'E' },
467 { "selinux-context", required_argument, NULL, 'Z' },
468 { "selinux-apifs-context", required_argument, NULL, 'L' },
469 { "quiet", no_argument, NULL, 'q' },
470 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
471 { "register", required_argument, NULL, ARG_REGISTER },
472 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
473 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
474 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
475 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
476 { "network-veth", no_argument, NULL, 'n' },
477 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
478 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
479 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
480 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
481 { "personality", required_argument, NULL, ARG_PERSONALITY },
482 { "image", required_argument, NULL, 'i' },
483 { "volatile", optional_argument, NULL, ARG_VOLATILE },
484 { "port", required_argument, NULL, 'p' },
485 { "property", required_argument, NULL, ARG_PROPERTY },
486 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
487 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
488 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
489 { "settings", required_argument, NULL, ARG_SETTINGS },
490 { "chdir", required_argument, NULL, ARG_CHDIR },
491 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
492 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
493 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
494 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
495 {}
496 };
497
498 int c, r;
499 const char *p, *e;
500 uint64_t plus = 0, minus = 0;
501 bool mask_all_settings = false, mask_no_settings = false;
502
503 assert(argc >= 0);
504 assert(argv);
505
506 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:", options, NULL)) >= 0)
507
508 switch (c) {
509
510 case 'h':
511 help();
512 return 0;
513
514 case ARG_VERSION:
515 return version();
516
517 case 'D':
518 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
519 if (r < 0)
520 return r;
521 break;
522
523 case ARG_TEMPLATE:
524 r = parse_path_argument_and_warn(optarg, false, &arg_template);
525 if (r < 0)
526 return r;
527 break;
528
529 case 'i':
530 r = parse_path_argument_and_warn(optarg, false, &arg_image);
531 if (r < 0)
532 return r;
533 break;
534
535 case 'x':
536 arg_ephemeral = true;
537 break;
538
539 case 'u':
540 r = free_and_strdup(&arg_user, optarg);
541 if (r < 0)
542 return log_oom();
543
544 arg_settings_mask |= SETTING_USER;
545 break;
546
547 case ARG_NETWORK_ZONE: {
548 char *j;
549
550 j = strappend("vz-", optarg);
551 if (!j)
552 return log_oom();
553
554 if (!ifname_valid(j)) {
555 log_error("Network zone name not valid: %s", j);
556 free(j);
557 return -EINVAL;
558 }
559
560 free(arg_network_zone);
561 arg_network_zone = j;
562
563 arg_network_veth = true;
564 arg_private_network = true;
565 arg_settings_mask |= SETTING_NETWORK;
566 break;
567 }
568
569 case ARG_NETWORK_BRIDGE:
570
571 if (!ifname_valid(optarg)) {
572 log_error("Bridge interface name not valid: %s", optarg);
573 return -EINVAL;
574 }
575
576 r = free_and_strdup(&arg_network_bridge, optarg);
577 if (r < 0)
578 return log_oom();
579
580 _fallthrough_;
581 case 'n':
582 arg_network_veth = true;
583 arg_private_network = true;
584 arg_settings_mask |= SETTING_NETWORK;
585 break;
586
587 case ARG_NETWORK_VETH_EXTRA:
588 r = veth_extra_parse(&arg_network_veth_extra, optarg);
589 if (r < 0)
590 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
591
592 arg_private_network = true;
593 arg_settings_mask |= SETTING_NETWORK;
594 break;
595
596 case ARG_NETWORK_INTERFACE:
597
598 if (!ifname_valid(optarg)) {
599 log_error("Network interface name not valid: %s", optarg);
600 return -EINVAL;
601 }
602
603 if (strv_extend(&arg_network_interfaces, optarg) < 0)
604 return log_oom();
605
606 arg_private_network = true;
607 arg_settings_mask |= SETTING_NETWORK;
608 break;
609
610 case ARG_NETWORK_MACVLAN:
611
612 if (!ifname_valid(optarg)) {
613 log_error("MACVLAN network interface name not valid: %s", optarg);
614 return -EINVAL;
615 }
616
617 if (strv_extend(&arg_network_macvlan, optarg) < 0)
618 return log_oom();
619
620 arg_private_network = true;
621 arg_settings_mask |= SETTING_NETWORK;
622 break;
623
624 case ARG_NETWORK_IPVLAN:
625
626 if (!ifname_valid(optarg)) {
627 log_error("IPVLAN network interface name not valid: %s", optarg);
628 return -EINVAL;
629 }
630
631 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
632 return log_oom();
633
634 _fallthrough_;
635 case ARG_PRIVATE_NETWORK:
636 arg_private_network = true;
637 arg_settings_mask |= SETTING_NETWORK;
638 break;
639
640 case ARG_NETWORK_NAMESPACE_PATH:
641 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
642 if (r < 0)
643 return r;
644
645 break;
646
647 case 'b':
648 if (arg_start_mode == START_PID2) {
649 log_error("--boot and --as-pid2 may not be combined.");
650 return -EINVAL;
651 }
652
653 arg_start_mode = START_BOOT;
654 arg_settings_mask |= SETTING_START_MODE;
655 break;
656
657 case 'a':
658 if (arg_start_mode == START_BOOT) {
659 log_error("--boot and --as-pid2 may not be combined.");
660 return -EINVAL;
661 }
662
663 arg_start_mode = START_PID2;
664 arg_settings_mask |= SETTING_START_MODE;
665 break;
666
667 case ARG_UUID:
668 r = sd_id128_from_string(optarg, &arg_uuid);
669 if (r < 0)
670 return log_error_errno(r, "Invalid UUID: %s", optarg);
671
672 if (sd_id128_is_null(arg_uuid)) {
673 log_error("Machine UUID may not be all zeroes.");
674 return -EINVAL;
675 }
676
677 arg_settings_mask |= SETTING_MACHINE_ID;
678 break;
679
680 case 'S':
681 arg_slice = optarg;
682 break;
683
684 case 'M':
685 if (isempty(optarg))
686 arg_machine = mfree(arg_machine);
687 else {
688 if (!machine_name_is_valid(optarg)) {
689 log_error("Invalid machine name: %s", optarg);
690 return -EINVAL;
691 }
692
693 r = free_and_strdup(&arg_machine, optarg);
694 if (r < 0)
695 return log_oom();
696 }
697 break;
698
699 case 'Z':
700 arg_selinux_context = optarg;
701 break;
702
703 case 'L':
704 arg_selinux_apifs_context = optarg;
705 break;
706
707 case ARG_READ_ONLY:
708 arg_read_only = true;
709 arg_settings_mask |= SETTING_READ_ONLY;
710 break;
711
712 case ARG_CAPABILITY:
713 case ARG_DROP_CAPABILITY: {
714 p = optarg;
715 for (;;) {
716 _cleanup_free_ char *t = NULL;
717
718 r = extract_first_word(&p, &t, ",", 0);
719 if (r < 0)
720 return log_error_errno(r, "Failed to parse capability %s.", t);
721
722 if (r == 0)
723 break;
724
725 if (streq(t, "all")) {
726 if (c == ARG_CAPABILITY)
727 plus = (uint64_t) -1;
728 else
729 minus = (uint64_t) -1;
730 } else {
731 int cap;
732
733 cap = capability_from_name(t);
734 if (cap < 0) {
735 log_error("Failed to parse capability %s.", t);
736 return -EINVAL;
737 }
738
739 if (c == ARG_CAPABILITY)
740 plus |= 1ULL << (uint64_t) cap;
741 else
742 minus |= 1ULL << (uint64_t) cap;
743 }
744 }
745
746 arg_settings_mask |= SETTING_CAPABILITY;
747 break;
748 }
749
750 case 'j':
751 arg_link_journal = LINK_GUEST;
752 arg_link_journal_try = true;
753 break;
754
755 case ARG_LINK_JOURNAL:
756 if (streq(optarg, "auto")) {
757 arg_link_journal = LINK_AUTO;
758 arg_link_journal_try = false;
759 } else if (streq(optarg, "no")) {
760 arg_link_journal = LINK_NO;
761 arg_link_journal_try = false;
762 } else if (streq(optarg, "guest")) {
763 arg_link_journal = LINK_GUEST;
764 arg_link_journal_try = false;
765 } else if (streq(optarg, "host")) {
766 arg_link_journal = LINK_HOST;
767 arg_link_journal_try = false;
768 } else if (streq(optarg, "try-guest")) {
769 arg_link_journal = LINK_GUEST;
770 arg_link_journal_try = true;
771 } else if (streq(optarg, "try-host")) {
772 arg_link_journal = LINK_HOST;
773 arg_link_journal_try = true;
774 } else {
775 log_error("Failed to parse link journal mode %s", optarg);
776 return -EINVAL;
777 }
778
779 break;
780
781 case ARG_BIND:
782 case ARG_BIND_RO:
783 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
784 if (r < 0)
785 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
786
787 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
788 break;
789
790 case ARG_TMPFS:
791 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
792 if (r < 0)
793 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
794
795 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
796 break;
797
798 case ARG_OVERLAY:
799 case ARG_OVERLAY_RO:
800 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
801 if (r == -EADDRNOTAVAIL)
802 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
803 if (r < 0)
804 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
805
806 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
807 break;
808
809 case 'E': {
810 char **n;
811
812 if (!env_assignment_is_valid(optarg)) {
813 log_error("Environment variable assignment '%s' is not valid.", optarg);
814 return -EINVAL;
815 }
816
817 n = strv_env_set(arg_setenv, optarg);
818 if (!n)
819 return log_oom();
820
821 strv_free(arg_setenv);
822 arg_setenv = n;
823
824 arg_settings_mask |= SETTING_ENVIRONMENT;
825 break;
826 }
827
828 case 'q':
829 arg_quiet = true;
830 break;
831
832 case ARG_SHARE_SYSTEM:
833 /* We don't officially support this anymore, except for compat reasons. People should use the
834 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
835 arg_clone_ns_flags = 0;
836 break;
837
838 case ARG_REGISTER:
839 r = parse_boolean(optarg);
840 if (r < 0) {
841 log_error("Failed to parse --register= argument: %s", optarg);
842 return r;
843 }
844
845 arg_register = r;
846 break;
847
848 case ARG_KEEP_UNIT:
849 arg_keep_unit = true;
850 break;
851
852 case ARG_PERSONALITY:
853
854 arg_personality = personality_from_string(optarg);
855 if (arg_personality == PERSONALITY_INVALID) {
856 log_error("Unknown or unsupported personality '%s'.", optarg);
857 return -EINVAL;
858 }
859
860 arg_settings_mask |= SETTING_PERSONALITY;
861 break;
862
863 case ARG_VOLATILE:
864
865 if (!optarg)
866 arg_volatile_mode = VOLATILE_YES;
867 else {
868 VolatileMode m;
869
870 m = volatile_mode_from_string(optarg);
871 if (m < 0) {
872 log_error("Failed to parse --volatile= argument: %s", optarg);
873 return -EINVAL;
874 } else
875 arg_volatile_mode = m;
876 }
877
878 arg_settings_mask |= SETTING_VOLATILE_MODE;
879 break;
880
881 case 'p':
882 r = expose_port_parse(&arg_expose_ports, optarg);
883 if (r == -EEXIST)
884 return log_error_errno(r, "Duplicate port specification: %s", optarg);
885 if (r < 0)
886 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
887
888 arg_settings_mask |= SETTING_EXPOSE_PORTS;
889 break;
890
891 case ARG_PROPERTY:
892 if (strv_extend(&arg_property, optarg) < 0)
893 return log_oom();
894
895 break;
896
897 case ARG_PRIVATE_USERS: {
898 int boolean = -1;
899
900 if (!optarg)
901 boolean = true;
902 else if (!in_charset(optarg, DIGITS))
903 /* do *not* parse numbers as booleans */
904 boolean = parse_boolean(optarg);
905
906 if (boolean == false) {
907 /* no: User namespacing off */
908 arg_userns_mode = USER_NAMESPACE_NO;
909 arg_uid_shift = UID_INVALID;
910 arg_uid_range = UINT32_C(0x10000);
911 } else if (boolean == true) {
912 /* yes: User namespacing on, UID range is read from root dir */
913 arg_userns_mode = USER_NAMESPACE_FIXED;
914 arg_uid_shift = UID_INVALID;
915 arg_uid_range = UINT32_C(0x10000);
916 } else if (streq(optarg, "pick")) {
917 /* pick: User namespacing on, UID range is picked randomly */
918 arg_userns_mode = USER_NAMESPACE_PICK;
919 arg_uid_shift = UID_INVALID;
920 arg_uid_range = UINT32_C(0x10000);
921 } else {
922 _cleanup_free_ char *buffer = NULL;
923 const char *range, *shift;
924
925 /* anything else: User namespacing on, UID range is explicitly configured */
926
927 range = strchr(optarg, ':');
928 if (range) {
929 buffer = strndup(optarg, range - optarg);
930 if (!buffer)
931 return log_oom();
932 shift = buffer;
933
934 range++;
935 r = safe_atou32(range, &arg_uid_range);
936 if (r < 0)
937 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
938 } else
939 shift = optarg;
940
941 r = parse_uid(shift, &arg_uid_shift);
942 if (r < 0)
943 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
944
945 arg_userns_mode = USER_NAMESPACE_FIXED;
946 }
947
948 if (arg_uid_range <= 0) {
949 log_error("UID range cannot be 0.");
950 return -EINVAL;
951 }
952
953 arg_settings_mask |= SETTING_USERNS;
954 break;
955 }
956
957 case 'U':
958 if (userns_supported()) {
959 arg_userns_mode = USER_NAMESPACE_PICK;
960 arg_uid_shift = UID_INVALID;
961 arg_uid_range = UINT32_C(0x10000);
962
963 arg_settings_mask |= SETTING_USERNS;
964 }
965
966 break;
967
968 case ARG_PRIVATE_USERS_CHOWN:
969 arg_userns_chown = true;
970
971 arg_settings_mask |= SETTING_USERNS;
972 break;
973
974 case ARG_KILL_SIGNAL:
975 arg_kill_signal = signal_from_string_try_harder(optarg);
976 if (arg_kill_signal < 0) {
977 log_error("Cannot parse signal: %s", optarg);
978 return -EINVAL;
979 }
980
981 arg_settings_mask |= SETTING_KILL_SIGNAL;
982 break;
983
984 case ARG_SETTINGS:
985
986 /* no → do not read files
987 * yes → read files, do not override cmdline, trust only subset
988 * override → read files, override cmdline, trust only subset
989 * trusted → read files, do not override cmdline, trust all
990 */
991
992 r = parse_boolean(optarg);
993 if (r < 0) {
994 if (streq(optarg, "trusted")) {
995 mask_all_settings = false;
996 mask_no_settings = false;
997 arg_settings_trusted = true;
998
999 } else if (streq(optarg, "override")) {
1000 mask_all_settings = false;
1001 mask_no_settings = true;
1002 arg_settings_trusted = -1;
1003 } else
1004 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1005 } else if (r > 0) {
1006 /* yes */
1007 mask_all_settings = false;
1008 mask_no_settings = false;
1009 arg_settings_trusted = -1;
1010 } else {
1011 /* no */
1012 mask_all_settings = true;
1013 mask_no_settings = false;
1014 arg_settings_trusted = false;
1015 }
1016
1017 break;
1018
1019 case ARG_CHDIR:
1020 if (!path_is_absolute(optarg)) {
1021 log_error("Working directory %s is not an absolute path.", optarg);
1022 return -EINVAL;
1023 }
1024
1025 r = free_and_strdup(&arg_chdir, optarg);
1026 if (r < 0)
1027 return log_oom();
1028
1029 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1030 break;
1031
1032 case ARG_PIVOT_ROOT:
1033 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1034 if (r < 0)
1035 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1036
1037 arg_settings_mask |= SETTING_PIVOT_ROOT;
1038 break;
1039
1040 case ARG_NOTIFY_READY:
1041 r = parse_boolean(optarg);
1042 if (r < 0) {
1043 log_error("%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1044 return -EINVAL;
1045 }
1046 arg_notify_ready = r;
1047 arg_settings_mask |= SETTING_NOTIFY_READY;
1048 break;
1049
1050 case ARG_ROOT_HASH: {
1051 void *k;
1052 size_t l;
1053
1054 r = unhexmem(optarg, strlen(optarg), &k, &l);
1055 if (r < 0)
1056 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1057 if (l < sizeof(sd_id128_t)) {
1058 log_error("Root hash must be at least 128bit long: %s", optarg);
1059 free(k);
1060 return -EINVAL;
1061 }
1062
1063 free(arg_root_hash);
1064 arg_root_hash = k;
1065 arg_root_hash_size = l;
1066 break;
1067 }
1068
1069 case ARG_SYSTEM_CALL_FILTER: {
1070 bool negative;
1071 const char *items;
1072
1073 negative = optarg[0] == '~';
1074 items = negative ? optarg + 1 : optarg;
1075
1076 for (;;) {
1077 _cleanup_free_ char *word = NULL;
1078
1079 r = extract_first_word(&items, &word, NULL, 0);
1080 if (r == 0)
1081 break;
1082 if (r == -ENOMEM)
1083 return log_oom();
1084 if (r < 0)
1085 return log_error_errno(r, "Failed to parse system call filter: %m");
1086
1087 if (negative)
1088 r = strv_extend(&arg_syscall_blacklist, word);
1089 else
1090 r = strv_extend(&arg_syscall_whitelist, word);
1091 if (r < 0)
1092 return log_oom();
1093 }
1094
1095 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1096 break;
1097 }
1098
1099 case '?':
1100 return -EINVAL;
1101
1102 default:
1103 assert_not_reached("Unhandled option");
1104 }
1105
1106 /* If --network-namespace-path is given with any other network-related option,
1107 * we need to error out, to avoid conflicts between different network options. */
1108 if (arg_network_namespace_path &&
1109 (arg_network_interfaces || arg_network_macvlan ||
1110 arg_network_ipvlan || arg_network_veth_extra ||
1111 arg_network_bridge || arg_network_zone ||
1112 arg_network_veth || arg_private_network)) {
1113 log_error("--network-namespace-path cannot be combined with other network options.");
1114 return -EINVAL;
1115 }
1116
1117 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
1118 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
1119 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
1120 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
1121
1122 if (arg_userns_mode != USER_NAMESPACE_NO)
1123 arg_mount_settings |= MOUNT_USE_USERNS;
1124
1125 if (arg_private_network)
1126 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1127
1128 parse_mount_settings_env();
1129
1130 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1131 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1132 arg_register = false;
1133 if (arg_start_mode != START_PID1) {
1134 log_error("--boot cannot be used without namespacing.");
1135 return -EINVAL;
1136 }
1137 }
1138
1139 if (arg_userns_mode == USER_NAMESPACE_PICK)
1140 arg_userns_chown = true;
1141
1142 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0) {
1143 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1144 * The latter is not technically a user session, but we don't need to labour the point. */
1145 log_error("--keep-unit --register=yes may not be used when invoked from a user session.");
1146 return -EINVAL;
1147 }
1148
1149 if (arg_directory && arg_image) {
1150 log_error("--directory= and --image= may not be combined.");
1151 return -EINVAL;
1152 }
1153
1154 if (arg_template && arg_image) {
1155 log_error("--template= and --image= may not be combined.");
1156 return -EINVAL;
1157 }
1158
1159 if (arg_ephemeral && arg_template && !arg_directory) {
1160 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1161 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1162 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1163 * --directory=". */
1164
1165 arg_directory = TAKE_PTR(arg_template);
1166 }
1167
1168 if (arg_template && !(arg_directory || arg_machine)) {
1169 log_error("--template= needs --directory= or --machine=.");
1170 return -EINVAL;
1171 }
1172
1173 if (arg_ephemeral && arg_template) {
1174 log_error("--ephemeral and --template= may not be combined.");
1175 return -EINVAL;
1176 }
1177
1178 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
1179 log_error("--ephemeral and --link-journal= may not be combined.");
1180 return -EINVAL;
1181 }
1182
1183 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported()) {
1184 log_error("--private-users= is not supported, kernel compiled without user namespace support.");
1185 return -EOPNOTSUPP;
1186 }
1187
1188 if (arg_userns_chown && arg_read_only) {
1189 log_error("--read-only and --private-users-chown may not be combined.");
1190 return -EINVAL;
1191 }
1192
1193 if (arg_network_bridge && arg_network_zone) {
1194 log_error("--network-bridge= and --network-zone= may not be combined.");
1195 return -EINVAL;
1196 }
1197
1198 if (argc > optind) {
1199 arg_parameters = strv_copy(argv + optind);
1200 if (!arg_parameters)
1201 return log_oom();
1202
1203 arg_settings_mask |= SETTING_START_MODE;
1204 }
1205
1206 /* Load all settings from .nspawn files */
1207 if (mask_no_settings)
1208 arg_settings_mask = 0;
1209
1210 /* Don't load any settings from .nspawn files */
1211 if (mask_all_settings)
1212 arg_settings_mask = _SETTINGS_MASK_ALL;
1213
1214 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
1215
1216 r = cg_unified_flush();
1217 if (r < 0)
1218 return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
1219
1220 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
1221 if (e)
1222 arg_container_service_name = e;
1223
1224 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
1225 if (r < 0)
1226 arg_use_cgns = cg_ns_supported();
1227 else
1228 arg_use_cgns = r;
1229
1230 r = custom_mount_check_all();
1231 if (r < 0)
1232 return r;
1233
1234 return 1;
1235 }
1236
1237 static int verify_arguments(void) {
1238 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network) {
1239 log_error("Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1240 return -EINVAL;
1241 }
1242
1243 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO)) {
1244 log_error("Cannot combine --private-users with read-write mounts.");
1245 return -EINVAL;
1246 }
1247
1248 if (arg_volatile_mode != VOLATILE_NO && arg_read_only) {
1249 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1250 return -EINVAL;
1251 }
1252
1253 if (arg_expose_ports && !arg_private_network) {
1254 log_error("Cannot use --port= without private networking.");
1255 return -EINVAL;
1256 }
1257
1258 #if ! HAVE_LIBIPTC
1259 if (arg_expose_ports) {
1260 log_error("--port= is not supported, compiled without libiptc support.");
1261 return -EOPNOTSUPP;
1262 }
1263 #endif
1264
1265 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1266 arg_kill_signal = SIGRTMIN+3;
1267
1268 return 0;
1269 }
1270
1271 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1272 assert(p);
1273
1274 if (arg_userns_mode == USER_NAMESPACE_NO)
1275 return 0;
1276
1277 if (uid == UID_INVALID && gid == GID_INVALID)
1278 return 0;
1279
1280 if (uid != UID_INVALID) {
1281 uid += arg_uid_shift;
1282
1283 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1284 return -EOVERFLOW;
1285 }
1286
1287 if (gid != GID_INVALID) {
1288 gid += (gid_t) arg_uid_shift;
1289
1290 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1291 return -EOVERFLOW;
1292 }
1293
1294 if (lchown(p, uid, gid) < 0)
1295 return -errno;
1296
1297 return 0;
1298 }
1299
1300 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1301 const char *q;
1302 int r;
1303
1304 q = prefix_roota(root, path);
1305 r = mkdir_errno_wrapper(q, mode);
1306 if (r == -EEXIST)
1307 return 0;
1308 if (r < 0)
1309 return r;
1310
1311 return userns_lchown(q, uid, gid);
1312 }
1313
1314 static int setup_timezone(const char *dest) {
1315 _cleanup_free_ char *p = NULL, *q = NULL;
1316 const char *where, *check, *what;
1317 char *z, *y;
1318 int r;
1319
1320 assert(dest);
1321
1322 /* Fix the timezone, if possible */
1323 r = readlink_malloc("/etc/localtime", &p);
1324 if (r < 0) {
1325 log_warning("host's /etc/localtime is not a symlink, not updating container timezone.");
1326 /* to handle warning, delete /etc/localtime and replace it
1327 * with a symbolic link to a time zone data file.
1328 *
1329 * Example:
1330 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1331 */
1332 return 0;
1333 }
1334
1335 z = path_startswith(p, "../usr/share/zoneinfo/");
1336 if (!z)
1337 z = path_startswith(p, "/usr/share/zoneinfo/");
1338 if (!z) {
1339 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1340 return 0;
1341 }
1342
1343 where = prefix_roota(dest, "/etc/localtime");
1344 r = readlink_malloc(where, &q);
1345 if (r >= 0) {
1346 y = path_startswith(q, "../usr/share/zoneinfo/");
1347 if (!y)
1348 y = path_startswith(q, "/usr/share/zoneinfo/");
1349
1350 /* Already pointing to the right place? Then do nothing .. */
1351 if (y && streq(y, z))
1352 return 0;
1353 }
1354
1355 check = strjoina("/usr/share/zoneinfo/", z);
1356 check = prefix_roota(dest, check);
1357 if (laccess(check, F_OK) < 0) {
1358 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1359 return 0;
1360 }
1361
1362 if (unlink(where) < 0 && errno != ENOENT) {
1363 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1364 errno,
1365 "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1366 return 0;
1367 }
1368
1369 what = strjoina("../usr/share/zoneinfo/", z);
1370 if (symlink(what, where) < 0) {
1371 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1372 errno,
1373 "Failed to correct timezone of container, ignoring: %m");
1374 return 0;
1375 }
1376
1377 r = userns_lchown(where, 0, 0);
1378 if (r < 0)
1379 return log_warning_errno(r, "Failed to chown /etc/localtime: %m");
1380
1381 return 0;
1382 }
1383
1384 static int resolved_listening(void) {
1385 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1386 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1387 int r;
1388
1389 /* Check if resolved is listening */
1390
1391 r = sd_bus_open_system(&bus);
1392 if (r < 0)
1393 return r;
1394
1395 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1396 if (r <= 0)
1397 return r;
1398
1399 r = sd_bus_get_property_string(bus,
1400 "org.freedesktop.resolve1",
1401 "/org/freedesktop/resolve1",
1402 "org.freedesktop.resolve1.Manager",
1403 "DNSStubListener",
1404 NULL,
1405 &dns_stub_listener_mode);
1406 if (r < 0)
1407 return r;
1408
1409 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1410 }
1411
1412 static int setup_resolv_conf(const char *dest) {
1413 _cleanup_free_ char *resolved = NULL, *etc = NULL;
1414 const char *where;
1415 int r, found;
1416
1417 assert(dest);
1418
1419 if (arg_private_network)
1420 return 0;
1421
1422 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1423 if (r < 0) {
1424 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1425 return 0;
1426 }
1427
1428 where = strjoina(etc, "/resolv.conf");
1429 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1430 if (found < 0) {
1431 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1432 return 0;
1433 }
1434
1435 if (access(STATIC_RESOLV_CONF, F_OK) >= 0 &&
1436 resolved_listening() > 0) {
1437
1438 /* resolved is enabled on the host. In this, case bind mount its static resolv.conf file into the
1439 * container, so that the container can use the host's resolver. Given that network namespacing is
1440 * disabled it's only natural of the container also uses the host's resolver. It also has the big
1441 * advantage that the container will be able to follow the host's DNS server configuration changes
1442 * transparently. */
1443
1444 if (found == 0) /* missing? */
1445 (void) touch(resolved);
1446
1447 r = mount_verbose(LOG_DEBUG, STATIC_RESOLV_CONF, resolved, NULL, MS_BIND, NULL);
1448 if (r >= 0)
1449 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1450 }
1451
1452 /* If that didn't work, let's copy the file */
1453 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0, COPY_REFLINK);
1454 if (r < 0) {
1455 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1456 * resolved or something similar runs inside and the symlink points there.
1457 *
1458 * If the disk image is read-only, there's also no point in complaining.
1459 */
1460 log_full_errno(IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1461 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1462 return 0;
1463 }
1464
1465 r = userns_lchown(where, 0, 0);
1466 if (r < 0)
1467 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1468
1469 return 0;
1470 }
1471
1472 static int setup_boot_id(const char *dest) {
1473 sd_id128_t rnd = SD_ID128_NULL;
1474 const char *from, *to;
1475 int r;
1476
1477 /* Generate a new randomized boot ID, so that each boot-up of
1478 * the container gets a new one */
1479
1480 from = prefix_roota(dest, "/run/proc-sys-kernel-random-boot-id");
1481 to = prefix_roota(dest, "/proc/sys/kernel/random/boot_id");
1482
1483 r = sd_id128_randomize(&rnd);
1484 if (r < 0)
1485 return log_error_errno(r, "Failed to generate random boot id: %m");
1486
1487 r = id128_write(from, ID128_UUID, rnd, false);
1488 if (r < 0)
1489 return log_error_errno(r, "Failed to write boot id: %m");
1490
1491 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1492 if (r >= 0)
1493 r = mount_verbose(LOG_ERR, NULL, to, NULL,
1494 MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1495
1496 (void) unlink(from);
1497 return r;
1498 }
1499
1500 static int copy_devnodes(const char *dest) {
1501
1502 static const char devnodes[] =
1503 "null\0"
1504 "zero\0"
1505 "full\0"
1506 "random\0"
1507 "urandom\0"
1508 "tty\0"
1509 "net/tun\0";
1510
1511 const char *d;
1512 int r = 0;
1513 _cleanup_umask_ mode_t u;
1514
1515 assert(dest);
1516
1517 u = umask(0000);
1518
1519 /* Create /dev/net, so that we can create /dev/net/tun in it */
1520 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1521 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1522
1523 NULSTR_FOREACH(d, devnodes) {
1524 _cleanup_free_ char *from = NULL, *to = NULL;
1525 struct stat st;
1526
1527 from = strappend("/dev/", d);
1528 to = prefix_root(dest, from);
1529
1530 if (stat(from, &st) < 0) {
1531
1532 if (errno != ENOENT)
1533 return log_error_errno(errno, "Failed to stat %s: %m", from);
1534
1535 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1536
1537 log_error("%s is not a char or block device, cannot copy.", from);
1538 return -EIO;
1539
1540 } else {
1541 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1542 /* Explicitly warn the user when /dev is already populated. */
1543 if (errno == EEXIST)
1544 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);
1545 if (errno != EPERM)
1546 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1547
1548 /* Some systems abusively restrict mknod but
1549 * allow bind mounts. */
1550 r = touch(to);
1551 if (r < 0)
1552 return log_error_errno(r, "touch (%s) failed: %m", to);
1553 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1554 if (r < 0)
1555 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1556 }
1557
1558 r = userns_lchown(to, 0, 0);
1559 if (r < 0)
1560 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1561 }
1562 }
1563
1564 return r;
1565 }
1566
1567 static int setup_pts(const char *dest) {
1568 _cleanup_free_ char *options = NULL;
1569 const char *p;
1570 int r;
1571
1572 #if HAVE_SELINUX
1573 if (arg_selinux_apifs_context)
1574 (void) asprintf(&options,
1575 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1576 arg_uid_shift + TTY_GID,
1577 arg_selinux_apifs_context);
1578 else
1579 #endif
1580 (void) asprintf(&options,
1581 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1582 arg_uid_shift + TTY_GID);
1583
1584 if (!options)
1585 return log_oom();
1586
1587 /* Mount /dev/pts itself */
1588 p = prefix_roota(dest, "/dev/pts");
1589 r = mkdir_errno_wrapper(p, 0755);
1590 if (r < 0)
1591 return log_error_errno(r, "Failed to create /dev/pts: %m");
1592
1593 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
1594 if (r < 0)
1595 return r;
1596 r = userns_lchown(p, 0, 0);
1597 if (r < 0)
1598 return log_error_errno(r, "Failed to chown /dev/pts: %m");
1599
1600 /* Create /dev/ptmx symlink */
1601 p = prefix_roota(dest, "/dev/ptmx");
1602 if (symlink("pts/ptmx", p) < 0)
1603 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1604 r = userns_lchown(p, 0, 0);
1605 if (r < 0)
1606 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
1607
1608 /* And fix /dev/pts/ptmx ownership */
1609 p = prefix_roota(dest, "/dev/pts/ptmx");
1610 r = userns_lchown(p, 0, 0);
1611 if (r < 0)
1612 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
1613
1614 return 0;
1615 }
1616
1617 static int setup_dev_console(const char *dest, const char *console) {
1618 _cleanup_umask_ mode_t u;
1619 const char *to;
1620 int r;
1621
1622 assert(dest);
1623 assert(console);
1624
1625 u = umask(0000);
1626
1627 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
1628 if (r < 0)
1629 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1630
1631 /* We need to bind mount the right tty to /dev/console since
1632 * ptys can only exist on pts file systems. To have something
1633 * to bind mount things on we create a empty regular file. */
1634
1635 to = prefix_roota(dest, "/dev/console");
1636 r = touch(to);
1637 if (r < 0)
1638 return log_error_errno(r, "touch() for /dev/console failed: %m");
1639
1640 return mount_verbose(LOG_ERR, console, to, NULL, MS_BIND, NULL);
1641 }
1642
1643 static int setup_keyring(void) {
1644 key_serial_t keyring;
1645
1646 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
1647 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
1648 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
1649 * these system calls let's make sure we don't leak anything into the container. */
1650
1651 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
1652 if (keyring == -1) {
1653 if (errno == ENOSYS)
1654 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
1655 else if (IN_SET(errno, EACCES, EPERM))
1656 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
1657 else
1658 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
1659 }
1660
1661 return 0;
1662 }
1663
1664 static int setup_kmsg(const char *dest, int kmsg_socket) {
1665 const char *from, *to;
1666 _cleanup_umask_ mode_t u;
1667 int fd, r;
1668
1669 assert(kmsg_socket >= 0);
1670
1671 u = umask(0000);
1672
1673 /* We create the kmsg FIFO as /run/kmsg, but immediately
1674 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1675 * on the reading side behave very similar to /proc/kmsg,
1676 * their writing side behaves differently from /dev/kmsg in
1677 * that writing blocks when nothing is reading. In order to
1678 * avoid any problems with containers deadlocking due to this
1679 * we simply make /dev/kmsg unavailable to the container. */
1680 from = prefix_roota(dest, "/run/kmsg");
1681 to = prefix_roota(dest, "/proc/kmsg");
1682
1683 if (mkfifo(from, 0600) < 0)
1684 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
1685 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1686 if (r < 0)
1687 return r;
1688
1689 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
1690 if (fd < 0)
1691 return log_error_errno(errno, "Failed to open fifo: %m");
1692
1693 /* Store away the fd in the socket, so that it stays open as
1694 * long as we run the child */
1695 r = send_one_fd(kmsg_socket, fd, 0);
1696 safe_close(fd);
1697
1698 if (r < 0)
1699 return log_error_errno(r, "Failed to send FIFO fd: %m");
1700
1701 /* And now make the FIFO unavailable as /run/kmsg... */
1702 (void) unlink(from);
1703
1704 return 0;
1705 }
1706
1707 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
1708 union in_addr_union *exposed = userdata;
1709
1710 assert(rtnl);
1711 assert(m);
1712 assert(exposed);
1713
1714 expose_port_execute(rtnl, arg_expose_ports, exposed);
1715 return 0;
1716 }
1717
1718 static int setup_hostname(void) {
1719
1720 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
1721 return 0;
1722
1723 if (sethostname_idempotent(arg_machine) < 0)
1724 return -errno;
1725
1726 return 0;
1727 }
1728
1729 static int setup_journal(const char *directory) {
1730 sd_id128_t this_id;
1731 _cleanup_free_ char *d = NULL;
1732 const char *p, *q;
1733 bool try;
1734 char id[33];
1735 int r;
1736
1737 /* Don't link journals in ephemeral mode */
1738 if (arg_ephemeral)
1739 return 0;
1740
1741 if (arg_link_journal == LINK_NO)
1742 return 0;
1743
1744 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
1745
1746 r = sd_id128_get_machine(&this_id);
1747 if (r < 0)
1748 return log_error_errno(r, "Failed to retrieve machine ID: %m");
1749
1750 if (sd_id128_equal(arg_uuid, this_id)) {
1751 log_full(try ? LOG_WARNING : LOG_ERR,
1752 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
1753 if (try)
1754 return 0;
1755 return -EEXIST;
1756 }
1757
1758 r = userns_mkdir(directory, "/var", 0755, 0, 0);
1759 if (r < 0)
1760 return log_error_errno(r, "Failed to create /var: %m");
1761
1762 r = userns_mkdir(directory, "/var/log", 0755, 0, 0);
1763 if (r < 0)
1764 return log_error_errno(r, "Failed to create /var/log: %m");
1765
1766 r = userns_mkdir(directory, "/var/log/journal", 0755, 0, 0);
1767 if (r < 0)
1768 return log_error_errno(r, "Failed to create /var/log/journal: %m");
1769
1770 (void) sd_id128_to_string(arg_uuid, id);
1771
1772 p = strjoina("/var/log/journal/", id);
1773 q = prefix_roota(directory, p);
1774
1775 if (path_is_mount_point(p, NULL, 0) > 0) {
1776 if (try)
1777 return 0;
1778
1779 log_error("%s: already a mount point, refusing to use for journal", p);
1780 return -EEXIST;
1781 }
1782
1783 if (path_is_mount_point(q, NULL, 0) > 0) {
1784 if (try)
1785 return 0;
1786
1787 log_error("%s: already a mount point, refusing to use for journal", q);
1788 return -EEXIST;
1789 }
1790
1791 r = readlink_and_make_absolute(p, &d);
1792 if (r >= 0) {
1793 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
1794 path_equal(d, q)) {
1795
1796 r = userns_mkdir(directory, p, 0755, 0, 0);
1797 if (r < 0)
1798 log_warning_errno(r, "Failed to create directory %s: %m", q);
1799 return 0;
1800 }
1801
1802 if (unlink(p) < 0)
1803 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
1804 } else if (r == -EINVAL) {
1805
1806 if (arg_link_journal == LINK_GUEST &&
1807 rmdir(p) < 0) {
1808
1809 if (errno == ENOTDIR) {
1810 log_error("%s already exists and is neither a symlink nor a directory", p);
1811 return r;
1812 } else
1813 return log_error_errno(errno, "Failed to remove %s: %m", p);
1814 }
1815 } else if (r != -ENOENT)
1816 return log_error_errno(r, "readlink(%s) failed: %m", p);
1817
1818 if (arg_link_journal == LINK_GUEST) {
1819
1820 if (symlink(q, p) < 0) {
1821 if (try) {
1822 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
1823 return 0;
1824 } else
1825 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
1826 }
1827
1828 r = userns_mkdir(directory, p, 0755, 0, 0);
1829 if (r < 0)
1830 log_warning_errno(r, "Failed to create directory %s: %m", q);
1831 return 0;
1832 }
1833
1834 if (arg_link_journal == LINK_HOST) {
1835 /* don't create parents here — if the host doesn't have
1836 * permanent journal set up, don't force it here */
1837
1838 r = mkdir_errno_wrapper(p, 0755);
1839 if (r < 0 && r != -EEXIST) {
1840 if (try) {
1841 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
1842 return 0;
1843 } else
1844 return log_error_errno(r, "Failed to create %s: %m", p);
1845 }
1846
1847 } else if (access(p, F_OK) < 0)
1848 return 0;
1849
1850 if (dir_is_empty(q) == 0)
1851 log_warning("%s is not empty, proceeding anyway.", q);
1852
1853 r = userns_mkdir(directory, p, 0755, 0, 0);
1854 if (r < 0)
1855 return log_error_errno(r, "Failed to create %s: %m", q);
1856
1857 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
1858 if (r < 0)
1859 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
1860
1861 return 0;
1862 }
1863
1864 static int drop_capabilities(void) {
1865 return capability_bounding_set_drop(arg_caps_retain, false);
1866 }
1867
1868 static int reset_audit_loginuid(void) {
1869 _cleanup_free_ char *p = NULL;
1870 int r;
1871
1872 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
1873 return 0;
1874
1875 r = read_one_line_file("/proc/self/loginuid", &p);
1876 if (r == -ENOENT)
1877 return 0;
1878 if (r < 0)
1879 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
1880
1881 /* Already reset? */
1882 if (streq(p, "4294967295"))
1883 return 0;
1884
1885 r = write_string_file("/proc/self/loginuid", "4294967295", 0);
1886 if (r < 0) {
1887 log_error_errno(r,
1888 "Failed to reset audit login UID. This probably means that your kernel is too\n"
1889 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1890 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1891 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1892 "using systemd-nspawn. Sleeping for 5s... (%m)");
1893
1894 sleep(5);
1895 }
1896
1897 return 0;
1898 }
1899
1900
1901 static int setup_propagate(const char *root) {
1902 const char *p, *q;
1903 int r;
1904
1905 (void) mkdir_p("/run/systemd/nspawn/", 0755);
1906 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
1907 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
1908 (void) mkdir_p(p, 0600);
1909
1910 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
1911 if (r < 0)
1912 return log_error_errno(r, "Failed to create /run/systemd: %m");
1913
1914 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
1915 if (r < 0)
1916 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
1917
1918 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
1919 if (r < 0)
1920 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
1921
1922 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
1923 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
1924 if (r < 0)
1925 return r;
1926
1927 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
1928 if (r < 0)
1929 return r;
1930
1931 /* machined will MS_MOVE into that directory, and that's only
1932 * supported for non-shared mounts. */
1933 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
1934 }
1935
1936 static int setup_machine_id(const char *directory) {
1937 const char *etc_machine_id;
1938 sd_id128_t id;
1939 int r;
1940
1941 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
1942 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
1943 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
1944 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
1945 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
1946 * container behaves nicely). */
1947
1948 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
1949
1950 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
1951 if (r < 0) {
1952 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
1953 return log_error_errno(r, "Failed to read machine ID from container image: %m");
1954
1955 if (sd_id128_is_null(arg_uuid)) {
1956 r = sd_id128_randomize(&arg_uuid);
1957 if (r < 0)
1958 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
1959 }
1960 } else {
1961 if (sd_id128_is_null(id)) {
1962 log_error("Machine ID in container image is zero, refusing.");
1963 return -EINVAL;
1964 }
1965
1966 arg_uuid = id;
1967 }
1968
1969 return 0;
1970 }
1971
1972 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
1973 int r;
1974
1975 assert(directory);
1976
1977 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
1978 return 0;
1979
1980 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
1981 if (r == -EOPNOTSUPP)
1982 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.");
1983 if (r == -EBADE)
1984 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
1985 if (r < 0)
1986 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
1987 if (r == 0)
1988 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
1989 else
1990 log_debug("Patched directory tree to match UID/GID range.");
1991
1992 return r;
1993 }
1994
1995 /*
1996 * Return values:
1997 * < 0 : wait_for_terminate() failed to get the state of the
1998 * container, the container was terminated by a signal, or
1999 * failed for an unknown reason. No change is made to the
2000 * container argument.
2001 * > 0 : The program executed in the container terminated with an
2002 * error. The exit code of the program executed in the
2003 * container is returned. The container argument has been set
2004 * to CONTAINER_TERMINATED.
2005 * 0 : The container is being rebooted, has been shut down or exited
2006 * successfully. The container argument has been set to either
2007 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2008 *
2009 * That is, success is indicated by a return value of zero, and an
2010 * error is indicated by a non-zero value.
2011 */
2012 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2013 siginfo_t status;
2014 int r;
2015
2016 r = wait_for_terminate(pid, &status);
2017 if (r < 0)
2018 return log_warning_errno(r, "Failed to wait for container: %m");
2019
2020 switch (status.si_code) {
2021
2022 case CLD_EXITED:
2023 if (status.si_status == 0)
2024 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2025 else
2026 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2027
2028 *container = CONTAINER_TERMINATED;
2029 return status.si_status;
2030
2031 case CLD_KILLED:
2032 if (status.si_status == SIGINT) {
2033 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2034 *container = CONTAINER_TERMINATED;
2035 return 0;
2036
2037 } else if (status.si_status == SIGHUP) {
2038 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2039 *container = CONTAINER_REBOOTED;
2040 return 0;
2041 }
2042
2043 _fallthrough_;
2044 case CLD_DUMPED:
2045 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2046 return -EIO;
2047
2048 default:
2049 log_error("Container %s failed due to unknown reason.", arg_machine);
2050 return -EIO;
2051 }
2052 }
2053
2054 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2055 pid_t pid;
2056
2057 pid = PTR_TO_PID(userdata);
2058 if (pid > 0) {
2059 if (kill(pid, arg_kill_signal) >= 0) {
2060 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2061 sd_event_source_set_userdata(s, NULL);
2062 return 0;
2063 }
2064 }
2065
2066 sd_event_exit(sd_event_source_get_event(s), 0);
2067 return 0;
2068 }
2069
2070 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2071 pid_t pid;
2072
2073 assert(s);
2074 assert(ssi);
2075
2076 pid = PTR_TO_PID(userdata);
2077
2078 for (;;) {
2079 siginfo_t si = {};
2080
2081 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2082 return log_error_errno(errno, "Failed to waitid(): %m");
2083 if (si.si_pid == 0) /* No pending children. */
2084 break;
2085 if (si.si_pid == pid) {
2086 /* The main process we care for has exited. Return from
2087 * signal handler but leave the zombie. */
2088 sd_event_exit(sd_event_source_get_event(s), 0);
2089 break;
2090 }
2091
2092 /* Reap all other children. */
2093 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2094 }
2095
2096 return 0;
2097 }
2098
2099 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2100 pid_t pid;
2101
2102 assert(m);
2103
2104 pid = PTR_TO_PID(userdata);
2105
2106 if (arg_kill_signal > 0) {
2107 log_info("Container termination requested. Attempting to halt container.");
2108 (void) kill(pid, arg_kill_signal);
2109 } else {
2110 log_info("Container termination requested. Exiting.");
2111 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2112 }
2113
2114 return 0;
2115 }
2116
2117 static int determine_names(void) {
2118 int r;
2119
2120 if (arg_template && !arg_directory && arg_machine) {
2121
2122 /* If --template= was specified then we should not
2123 * search for a machine, but instead create a new one
2124 * in /var/lib/machine. */
2125
2126 arg_directory = strjoin("/var/lib/machines/", arg_machine);
2127 if (!arg_directory)
2128 return log_oom();
2129 }
2130
2131 if (!arg_image && !arg_directory) {
2132 if (arg_machine) {
2133 _cleanup_(image_unrefp) Image *i = NULL;
2134
2135 r = image_find(arg_machine, &i);
2136 if (r < 0)
2137 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2138 if (r == 0) {
2139 log_error("No image for machine '%s'.", arg_machine);
2140 return -ENOENT;
2141 }
2142
2143 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2144 r = free_and_strdup(&arg_image, i->path);
2145 else
2146 r = free_and_strdup(&arg_directory, i->path);
2147 if (r < 0)
2148 return log_oom();
2149
2150 if (!arg_ephemeral)
2151 arg_read_only = arg_read_only || i->read_only;
2152 } else {
2153 r = safe_getcwd(&arg_directory);
2154 if (r < 0)
2155 return log_error_errno(r, "Failed to determine current directory: %m");
2156 }
2157
2158 if (!arg_directory && !arg_image) {
2159 log_error("Failed to determine path, please use -D or -i.");
2160 return -EINVAL;
2161 }
2162 }
2163
2164 if (!arg_machine) {
2165
2166 if (arg_directory && path_equal(arg_directory, "/"))
2167 arg_machine = gethostname_malloc();
2168 else {
2169 if (arg_image) {
2170 char *e;
2171
2172 arg_machine = strdup(basename(arg_image));
2173
2174 /* Truncate suffix if there is one */
2175 e = endswith(arg_machine, ".raw");
2176 if (e)
2177 *e = 0;
2178 } else
2179 arg_machine = strdup(basename(arg_directory));
2180 }
2181 if (!arg_machine)
2182 return log_oom();
2183
2184 hostname_cleanup(arg_machine);
2185 if (!machine_name_is_valid(arg_machine)) {
2186 log_error("Failed to determine machine name automatically, please use -M.");
2187 return -EINVAL;
2188 }
2189
2190 if (arg_ephemeral) {
2191 char *b;
2192
2193 /* Add a random suffix when this is an
2194 * ephemeral machine, so that we can run many
2195 * instances at once without manually having
2196 * to specify -M each time. */
2197
2198 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2199 return log_oom();
2200
2201 free(arg_machine);
2202 arg_machine = b;
2203 }
2204 }
2205
2206 return 0;
2207 }
2208
2209 static int chase_symlinks_and_update(char **p, unsigned flags) {
2210 char *chased;
2211 int r;
2212
2213 assert(p);
2214
2215 if (!*p)
2216 return 0;
2217
2218 r = chase_symlinks(*p, NULL, flags, &chased);
2219 if (r < 0)
2220 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2221
2222 free_and_replace(*p, chased);
2223 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
2224 }
2225
2226 static int determine_uid_shift(const char *directory) {
2227 int r;
2228
2229 if (arg_userns_mode == USER_NAMESPACE_NO) {
2230 arg_uid_shift = 0;
2231 return 0;
2232 }
2233
2234 if (arg_uid_shift == UID_INVALID) {
2235 struct stat st;
2236
2237 r = stat(directory, &st);
2238 if (r < 0)
2239 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2240
2241 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2242
2243 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
2244 log_error("UID and GID base of %s don't match.", directory);
2245 return -EINVAL;
2246 }
2247
2248 arg_uid_range = UINT32_C(0x10000);
2249 }
2250
2251 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
2252 log_error("UID base too high for UID range.");
2253 return -EINVAL;
2254 }
2255
2256 return 0;
2257 }
2258
2259 static int inner_child(
2260 Barrier *barrier,
2261 const char *directory,
2262 bool secondary,
2263 int kmsg_socket,
2264 int rtnl_socket,
2265 FDSet *fds) {
2266
2267 _cleanup_free_ char *home = NULL;
2268 char as_uuid[37];
2269 unsigned n_env = 1;
2270 const char *envp[] = {
2271 "PATH=" DEFAULT_PATH_SPLIT_USR,
2272 NULL, /* container */
2273 NULL, /* TERM */
2274 NULL, /* HOME */
2275 NULL, /* USER */
2276 NULL, /* LOGNAME */
2277 NULL, /* container_uuid */
2278 NULL, /* LISTEN_FDS */
2279 NULL, /* LISTEN_PID */
2280 NULL, /* NOTIFY_SOCKET */
2281 NULL
2282 };
2283 const char *exec_target;
2284
2285 _cleanup_strv_free_ char **env_use = NULL;
2286 int r;
2287
2288 assert(barrier);
2289 assert(directory);
2290 assert(kmsg_socket >= 0);
2291
2292 if (arg_userns_mode != USER_NAMESPACE_NO) {
2293 /* Tell the parent, that it now can write the UID map. */
2294 (void) barrier_place(barrier); /* #1 */
2295
2296 /* Wait until the parent wrote the UID map */
2297 if (!barrier_place_and_sync(barrier)) { /* #2 */
2298 log_error("Parent died too early");
2299 return -ESRCH;
2300 }
2301 }
2302
2303 r = reset_uid_gid();
2304 if (r < 0)
2305 return log_error_errno(r, "Couldn't become new root: %m");
2306
2307 r = mount_all(NULL,
2308 arg_mount_settings | MOUNT_IN_USERNS,
2309 arg_uid_shift,
2310 arg_uid_range,
2311 arg_selinux_apifs_context);
2312 if (r < 0)
2313 return r;
2314
2315 if (!arg_network_namespace_path && arg_private_network) {
2316 r = unshare(CLONE_NEWNET);
2317 if (r < 0)
2318 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2319
2320 /* Tell the parent that it can setup network interfaces. */
2321 (void) barrier_place(barrier); /* #3 */
2322 }
2323
2324 r = mount_sysfs(NULL, arg_mount_settings);
2325 if (r < 0)
2326 return r;
2327
2328 /* Wait until we are cgroup-ified, so that we
2329 * can mount the right cgroup path writable */
2330 if (!barrier_place_and_sync(barrier)) { /* #4 */
2331 log_error("Parent died too early");
2332 return -ESRCH;
2333 }
2334
2335 if (arg_use_cgns && cg_ns_supported()) {
2336 r = unshare(CLONE_NEWCGROUP);
2337 if (r < 0)
2338 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2339 r = mount_cgroups(
2340 "",
2341 arg_unified_cgroup_hierarchy,
2342 arg_userns_mode != USER_NAMESPACE_NO,
2343 arg_uid_shift,
2344 arg_uid_range,
2345 arg_selinux_apifs_context,
2346 true);
2347 if (r < 0)
2348 return r;
2349 } else {
2350 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2351 if (r < 0)
2352 return r;
2353 }
2354
2355 r = setup_boot_id(NULL);
2356 if (r < 0)
2357 return r;
2358
2359 r = setup_kmsg(NULL, kmsg_socket);
2360 if (r < 0)
2361 return r;
2362 kmsg_socket = safe_close(kmsg_socket);
2363
2364 umask(0022);
2365
2366 if (setsid() < 0)
2367 return log_error_errno(errno, "setsid() failed: %m");
2368
2369 if (arg_private_network)
2370 loopback_setup();
2371
2372 if (arg_expose_ports) {
2373 r = expose_port_send_rtnl(rtnl_socket);
2374 if (r < 0)
2375 return r;
2376 rtnl_socket = safe_close(rtnl_socket);
2377 }
2378
2379 r = drop_capabilities();
2380 if (r < 0)
2381 return log_error_errno(r, "drop_capabilities() failed: %m");
2382
2383 setup_hostname();
2384
2385 if (arg_personality != PERSONALITY_INVALID) {
2386 r = safe_personality(arg_personality);
2387 if (r < 0)
2388 return log_error_errno(r, "personality() failed: %m");
2389 } else if (secondary) {
2390 r = safe_personality(PER_LINUX32);
2391 if (r < 0)
2392 return log_error_errno(r, "personality() failed: %m");
2393 }
2394
2395 #if HAVE_SELINUX
2396 if (arg_selinux_context)
2397 if (setexeccon(arg_selinux_context) < 0)
2398 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
2399 #endif
2400
2401 r = change_uid_gid(arg_user, &home);
2402 if (r < 0)
2403 return r;
2404
2405 /* LXC sets container=lxc, so follow the scheme here */
2406 envp[n_env++] = strjoina("container=", arg_container_service_name);
2407
2408 envp[n_env] = strv_find_prefix(environ, "TERM=");
2409 if (envp[n_env])
2410 n_env++;
2411
2412 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
2413 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
2414 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
2415 return log_oom();
2416
2417 assert(!sd_id128_is_null(arg_uuid));
2418
2419 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
2420 return log_oom();
2421
2422 if (fdset_size(fds) > 0) {
2423 r = fdset_cloexec(fds, false);
2424 if (r < 0)
2425 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
2426
2427 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
2428 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
2429 return log_oom();
2430 }
2431 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
2432 return log_oom();
2433
2434 env_use = strv_env_merge(2, envp, arg_setenv);
2435 if (!env_use)
2436 return log_oom();
2437
2438 /* Let the parent know that we are ready and
2439 * wait until the parent is ready with the
2440 * setup, too... */
2441 if (!barrier_place_and_sync(barrier)) { /* #5 */
2442 log_error("Parent died too early");
2443 return -ESRCH;
2444 }
2445
2446 if (arg_chdir)
2447 if (chdir(arg_chdir) < 0)
2448 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
2449
2450 if (arg_start_mode == START_PID2) {
2451 r = stub_pid1(arg_uuid);
2452 if (r < 0)
2453 return r;
2454 }
2455
2456 /* Now, explicitly close the log, so that we
2457 * then can close all remaining fds. Closing
2458 * the log explicitly first has the benefit
2459 * that the logging subsystem knows about it,
2460 * and is thus ready to be reopened should we
2461 * need it again. Note that the other fds
2462 * closed here are at least the locking and
2463 * barrier fds. */
2464 log_close();
2465 (void) fdset_close_others(fds);
2466
2467 if (arg_start_mode == START_BOOT) {
2468 char **a;
2469 size_t m;
2470
2471 /* Automatically search for the init system */
2472
2473 m = strv_length(arg_parameters);
2474 a = newa(char*, m + 2);
2475 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
2476 a[1 + m] = NULL;
2477
2478 a[0] = (char*) "/usr/lib/systemd/systemd";
2479 execve(a[0], a, env_use);
2480
2481 a[0] = (char*) "/lib/systemd/systemd";
2482 execve(a[0], a, env_use);
2483
2484 a[0] = (char*) "/sbin/init";
2485 execve(a[0], a, env_use);
2486
2487 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
2488 } else if (!strv_isempty(arg_parameters)) {
2489 exec_target = arg_parameters[0];
2490 execvpe(arg_parameters[0], arg_parameters, env_use);
2491 } else {
2492 if (!arg_chdir)
2493 /* If we cannot change the directory, we'll end up in /, that is expected. */
2494 (void) chdir(home ?: "/root");
2495
2496 execle("/bin/bash", "-bash", NULL, env_use);
2497 execle("/bin/sh", "-sh", NULL, env_use);
2498
2499 exec_target = "/bin/bash, /bin/sh";
2500 }
2501
2502 r = -errno;
2503 (void) log_open();
2504 return log_error_errno(r, "execv(%s) failed: %m", exec_target);
2505 }
2506
2507 static int setup_sd_notify_child(void) {
2508 static const int one = 1;
2509 int fd = -1;
2510 union sockaddr_union sa = {
2511 .sa.sa_family = AF_UNIX,
2512 };
2513 int r;
2514
2515 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2516 if (fd < 0)
2517 return log_error_errno(errno, "Failed to allocate notification socket: %m");
2518
2519 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
2520 (void) unlink(NSPAWN_NOTIFY_SOCKET_PATH);
2521
2522 strncpy(sa.un.sun_path, NSPAWN_NOTIFY_SOCKET_PATH, sizeof(sa.un.sun_path)-1);
2523 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
2524 if (r < 0) {
2525 safe_close(fd);
2526 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
2527 }
2528
2529 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
2530 if (r < 0) {
2531 safe_close(fd);
2532 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
2533 }
2534
2535 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
2536 if (r < 0) {
2537 safe_close(fd);
2538 return log_error_errno(errno, "SO_PASSCRED failed: %m");
2539 }
2540
2541 return fd;
2542 }
2543
2544 static int outer_child(
2545 Barrier *barrier,
2546 const char *directory,
2547 const char *console,
2548 DissectedImage *dissected_image,
2549 bool interactive,
2550 bool secondary,
2551 int pid_socket,
2552 int uuid_socket,
2553 int notify_socket,
2554 int kmsg_socket,
2555 int rtnl_socket,
2556 int uid_shift_socket,
2557 int unified_cgroup_hierarchy_socket,
2558 FDSet *fds,
2559 int netns_fd) {
2560
2561 pid_t pid;
2562 ssize_t l;
2563 int r;
2564 _cleanup_close_ int fd = -1;
2565
2566 assert(barrier);
2567 assert(directory);
2568 assert(console);
2569 assert(pid_socket >= 0);
2570 assert(uuid_socket >= 0);
2571 assert(notify_socket >= 0);
2572 assert(kmsg_socket >= 0);
2573
2574 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
2575 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
2576
2577 if (interactive) {
2578 int terminal;
2579
2580 terminal = open_terminal(console, O_RDWR);
2581 if (terminal < 0)
2582 return log_error_errno(terminal, "Failed to open console: %m");
2583
2584 r = rearrange_stdio(terminal, terminal, terminal); /* invalidates 'terminal' on success and failure */
2585 if (r < 0)
2586 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2587 }
2588
2589 r = reset_audit_loginuid();
2590 if (r < 0)
2591 return r;
2592
2593 /* Mark everything as slave, so that we still
2594 * receive mounts from the real root, but don't
2595 * propagate mounts to the real root. */
2596 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
2597 if (r < 0)
2598 return r;
2599
2600 if (dissected_image) {
2601 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
2602 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
2603 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
2604 * makes sure ESP partitions and userns are compatible. */
2605
2606 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2607 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2608 if (r < 0)
2609 return r;
2610 }
2611
2612 r = determine_uid_shift(directory);
2613 if (r < 0)
2614 return r;
2615
2616 if (arg_userns_mode != USER_NAMESPACE_NO) {
2617 /* Let the parent know which UID shift we read from the image */
2618 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
2619 if (l < 0)
2620 return log_error_errno(errno, "Failed to send UID shift: %m");
2621 if (l != sizeof(arg_uid_shift)) {
2622 log_error("Short write while sending UID shift.");
2623 return -EIO;
2624 }
2625
2626 if (arg_userns_mode == USER_NAMESPACE_PICK) {
2627 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
2628 * we just read from the image is available. If yes, it will send the UID shift back to us, if
2629 * not it will pick a different one, and send it back to us. */
2630
2631 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
2632 if (l < 0)
2633 return log_error_errno(errno, "Failed to recv UID shift: %m");
2634 if (l != sizeof(arg_uid_shift)) {
2635 log_error("Short read while receiving UID shift.");
2636 return -EIO;
2637 }
2638 }
2639
2640 log_info("Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
2641 }
2642
2643 if (dissected_image) {
2644 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
2645 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
2646 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
2647 if (r < 0)
2648 return r;
2649 }
2650
2651 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
2652 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
2653
2654 r = detect_unified_cgroup_hierarchy_from_image(directory);
2655 if (r < 0)
2656 return r;
2657
2658 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
2659 if (l < 0)
2660 return log_error_errno(errno, "Failed to send cgroup mode: %m");
2661 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
2662 log_error("Short write while sending cgroup mode: %m");
2663 return -EIO;
2664 }
2665
2666 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
2667 }
2668
2669 /* Turn directory into bind mount */
2670 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
2671 if (r < 0)
2672 return r;
2673
2674 r = setup_pivot_root(
2675 directory,
2676 arg_pivot_root_new,
2677 arg_pivot_root_old);
2678 if (r < 0)
2679 return r;
2680
2681 r = setup_volatile(
2682 directory,
2683 arg_volatile_mode,
2684 arg_userns_mode != USER_NAMESPACE_NO,
2685 arg_uid_shift,
2686 arg_uid_range,
2687 arg_selinux_context);
2688 if (r < 0)
2689 return r;
2690
2691 r = setup_volatile_state(
2692 directory,
2693 arg_volatile_mode,
2694 arg_userns_mode != USER_NAMESPACE_NO,
2695 arg_uid_shift,
2696 arg_uid_range,
2697 arg_selinux_context);
2698 if (r < 0)
2699 return r;
2700
2701 /* Mark everything as shared so our mounts get propagated down. This is
2702 * required to make new bind mounts available in systemd services
2703 * inside the containter that create a new mount namespace.
2704 * See https://github.com/systemd/systemd/issues/3860
2705 * Further submounts (such as /dev) done after this will inherit the
2706 * shared propagation mode. */
2707 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
2708 if (r < 0)
2709 return r;
2710
2711 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
2712 if (r < 0)
2713 return r;
2714
2715 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
2716 if (r < 0)
2717 return r;
2718
2719 if (arg_read_only) {
2720 r = bind_remount_recursive(directory, true, NULL);
2721 if (r < 0)
2722 return log_error_errno(r, "Failed to make tree read-only: %m");
2723 }
2724
2725 r = mount_all(directory,
2726 arg_mount_settings,
2727 arg_uid_shift,
2728 arg_uid_range,
2729 arg_selinux_apifs_context);
2730 if (r < 0)
2731 return r;
2732
2733 r = copy_devnodes(directory);
2734 if (r < 0)
2735 return r;
2736
2737 dev_setup(directory, arg_uid_shift, arg_uid_shift);
2738
2739 r = setup_pts(directory);
2740 if (r < 0)
2741 return r;
2742
2743 r = setup_propagate(directory);
2744 if (r < 0)
2745 return r;
2746
2747 r = setup_dev_console(directory, console);
2748 if (r < 0)
2749 return r;
2750
2751 r = setup_keyring();
2752 if (r < 0)
2753 return r;
2754
2755 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
2756 if (r < 0)
2757 return r;
2758
2759 r = setup_timezone(directory);
2760 if (r < 0)
2761 return r;
2762
2763 r = setup_resolv_conf(directory);
2764 if (r < 0)
2765 return r;
2766
2767 r = setup_machine_id(directory);
2768 if (r < 0)
2769 return r;
2770
2771 r = setup_journal(directory);
2772 if (r < 0)
2773 return r;
2774
2775 r = mount_custom(
2776 directory,
2777 arg_custom_mounts,
2778 arg_n_custom_mounts,
2779 arg_userns_mode != USER_NAMESPACE_NO,
2780 arg_uid_shift,
2781 arg_uid_range,
2782 arg_selinux_apifs_context);
2783 if (r < 0)
2784 return r;
2785
2786 if (!arg_use_cgns || !cg_ns_supported()) {
2787 r = mount_cgroups(
2788 directory,
2789 arg_unified_cgroup_hierarchy,
2790 arg_userns_mode != USER_NAMESPACE_NO,
2791 arg_uid_shift,
2792 arg_uid_range,
2793 arg_selinux_apifs_context,
2794 false);
2795 if (r < 0)
2796 return r;
2797 }
2798
2799 r = mount_move_root(directory);
2800 if (r < 0)
2801 return log_error_errno(r, "Failed to move root directory: %m");
2802
2803 fd = setup_sd_notify_child();
2804 if (fd < 0)
2805 return fd;
2806
2807 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
2808 arg_clone_ns_flags |
2809 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
2810 if (pid < 0)
2811 return log_error_errno(errno, "Failed to fork inner child: %m");
2812 if (pid == 0) {
2813 pid_socket = safe_close(pid_socket);
2814 uuid_socket = safe_close(uuid_socket);
2815 notify_socket = safe_close(notify_socket);
2816 uid_shift_socket = safe_close(uid_shift_socket);
2817
2818 /* The inner child has all namespaces that are
2819 * requested, so that we all are owned by the user if
2820 * user namespaces are turned on. */
2821
2822 if (arg_network_namespace_path) {
2823 r = namespace_enter(-1, -1, netns_fd, -1, -1);
2824 if (r < 0)
2825 return r;
2826 }
2827
2828 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
2829 if (r < 0)
2830 _exit(EXIT_FAILURE);
2831
2832 _exit(EXIT_SUCCESS);
2833 }
2834
2835 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
2836 if (l < 0)
2837 return log_error_errno(errno, "Failed to send PID: %m");
2838 if (l != sizeof(pid)) {
2839 log_error("Short write while sending PID.");
2840 return -EIO;
2841 }
2842
2843 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
2844 if (l < 0)
2845 return log_error_errno(errno, "Failed to send machine ID: %m");
2846 if (l != sizeof(arg_uuid)) {
2847 log_error("Short write while sending machine ID.");
2848 return -EIO;
2849 }
2850
2851 l = send_one_fd(notify_socket, fd, 0);
2852 if (l < 0)
2853 return log_error_errno(errno, "Failed to send notify fd: %m");
2854
2855 pid_socket = safe_close(pid_socket);
2856 uuid_socket = safe_close(uuid_socket);
2857 notify_socket = safe_close(notify_socket);
2858 kmsg_socket = safe_close(kmsg_socket);
2859 rtnl_socket = safe_close(rtnl_socket);
2860 netns_fd = safe_close(netns_fd);
2861
2862 return 0;
2863 }
2864
2865 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
2866 bool tried_hashed = false;
2867 unsigned n_tries = 100;
2868 uid_t candidate;
2869 int r;
2870
2871 assert(shift);
2872 assert(ret_lock_file);
2873 assert(arg_userns_mode == USER_NAMESPACE_PICK);
2874 assert(arg_uid_range == 0x10000U);
2875
2876 candidate = *shift;
2877
2878 (void) mkdir("/run/systemd/nspawn-uid", 0755);
2879
2880 for (;;) {
2881 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
2882 _cleanup_release_lock_file_ LockFile lf = LOCK_FILE_INIT;
2883
2884 if (--n_tries <= 0)
2885 return -EBUSY;
2886
2887 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
2888 goto next;
2889 if ((candidate & UINT32_C(0xFFFF)) != 0)
2890 goto next;
2891
2892 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
2893 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
2894 if (r == -EBUSY) /* Range already taken by another nspawn instance */
2895 goto next;
2896 if (r < 0)
2897 return r;
2898
2899 /* Make some superficial checks whether the range is currently known in the user database */
2900 if (getpwuid(candidate))
2901 goto next;
2902 if (getpwuid(candidate + UINT32_C(0xFFFE)))
2903 goto next;
2904 if (getgrgid(candidate))
2905 goto next;
2906 if (getgrgid(candidate + UINT32_C(0xFFFE)))
2907 goto next;
2908
2909 *ret_lock_file = lf;
2910 lf = (struct LockFile) LOCK_FILE_INIT;
2911 *shift = candidate;
2912 return 0;
2913
2914 next:
2915 if (arg_machine && !tried_hashed) {
2916 /* Try to hash the base from the container name */
2917
2918 static const uint8_t hash_key[] = {
2919 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
2920 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
2921 };
2922
2923 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
2924
2925 tried_hashed = true;
2926 } else
2927 random_bytes(&candidate, sizeof(candidate));
2928
2929 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
2930 candidate &= (uid_t) UINT32_C(0xFFFF0000);
2931 }
2932 }
2933
2934 static int setup_uid_map(pid_t pid) {
2935 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
2936 int r;
2937
2938 assert(pid > 1);
2939
2940 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
2941 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
2942 r = write_string_file(uid_map, line, 0);
2943 if (r < 0)
2944 return log_error_errno(r, "Failed to write UID map: %m");
2945
2946 /* We always assign the same UID and GID ranges */
2947 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
2948 r = write_string_file(uid_map, line, 0);
2949 if (r < 0)
2950 return log_error_errno(r, "Failed to write GID map: %m");
2951
2952 return 0;
2953 }
2954
2955 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2956 char buf[NOTIFY_BUFFER_MAX+1];
2957 char *p = NULL;
2958 struct iovec iovec = {
2959 .iov_base = buf,
2960 .iov_len = sizeof(buf)-1,
2961 };
2962 union {
2963 struct cmsghdr cmsghdr;
2964 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
2965 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
2966 } control = {};
2967 struct msghdr msghdr = {
2968 .msg_iov = &iovec,
2969 .msg_iovlen = 1,
2970 .msg_control = &control,
2971 .msg_controllen = sizeof(control),
2972 };
2973 struct cmsghdr *cmsg;
2974 struct ucred *ucred = NULL;
2975 ssize_t n;
2976 pid_t inner_child_pid;
2977 _cleanup_strv_free_ char **tags = NULL;
2978
2979 assert(userdata);
2980
2981 inner_child_pid = PTR_TO_PID(userdata);
2982
2983 if (revents != EPOLLIN) {
2984 log_warning("Got unexpected poll event for notify fd.");
2985 return 0;
2986 }
2987
2988 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
2989 if (n < 0) {
2990 if (IN_SET(errno, EAGAIN, EINTR))
2991 return 0;
2992
2993 return log_warning_errno(errno, "Couldn't read notification socket: %m");
2994 }
2995 cmsg_close_all(&msghdr);
2996
2997 CMSG_FOREACH(cmsg, &msghdr) {
2998 if (cmsg->cmsg_level == SOL_SOCKET &&
2999 cmsg->cmsg_type == SCM_CREDENTIALS &&
3000 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3001
3002 ucred = (struct ucred*) CMSG_DATA(cmsg);
3003 }
3004 }
3005
3006 if (!ucred || ucred->pid != inner_child_pid) {
3007 log_debug("Received notify message without valid credentials. Ignoring.");
3008 return 0;
3009 }
3010
3011 if ((size_t) n >= sizeof(buf)) {
3012 log_warning("Received notify message exceeded maximum size. Ignoring.");
3013 return 0;
3014 }
3015
3016 buf[n] = 0;
3017 tags = strv_split(buf, "\n\r");
3018 if (!tags)
3019 return log_oom();
3020
3021 if (strv_find(tags, "READY=1"))
3022 sd_notifyf(false, "READY=1\n");
3023
3024 p = strv_find_startswith(tags, "STATUS=");
3025 if (p)
3026 sd_notifyf(false, "STATUS=Container running: %s", p);
3027
3028 return 0;
3029 }
3030
3031 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3032 int r;
3033
3034 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3035 if (r < 0)
3036 return log_error_errno(r, "Failed to allocate notify event source: %m");
3037
3038 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3039
3040 return 0;
3041 }
3042
3043 static int load_settings(void) {
3044 _cleanup_(settings_freep) Settings *settings = NULL;
3045 _cleanup_fclose_ FILE *f = NULL;
3046 _cleanup_free_ char *p = NULL;
3047 const char *fn, *i;
3048 int r;
3049
3050 /* If all settings are masked, there's no point in looking for
3051 * the settings file */
3052 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3053 return 0;
3054
3055 fn = strjoina(arg_machine, ".nspawn");
3056
3057 /* We first look in the admin's directories in /etc and /run */
3058 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3059 _cleanup_free_ char *j = NULL;
3060
3061 j = strjoin(i, "/", fn);
3062 if (!j)
3063 return log_oom();
3064
3065 f = fopen(j, "re");
3066 if (f) {
3067 p = TAKE_PTR(j);
3068
3069 /* By default, we trust configuration from /etc and /run */
3070 if (arg_settings_trusted < 0)
3071 arg_settings_trusted = true;
3072
3073 break;
3074 }
3075
3076 if (errno != ENOENT)
3077 return log_error_errno(errno, "Failed to open %s: %m", j);
3078 }
3079
3080 if (!f) {
3081 /* After that, let's look for a file next to the
3082 * actual image we shall boot. */
3083
3084 if (arg_image) {
3085 p = file_in_same_dir(arg_image, fn);
3086 if (!p)
3087 return log_oom();
3088 } else if (arg_directory) {
3089 p = file_in_same_dir(arg_directory, fn);
3090 if (!p)
3091 return log_oom();
3092 }
3093
3094 if (p) {
3095 f = fopen(p, "re");
3096 if (!f && errno != ENOENT)
3097 return log_error_errno(errno, "Failed to open %s: %m", p);
3098
3099 /* By default, we do not trust configuration from /var/lib/machines */
3100 if (arg_settings_trusted < 0)
3101 arg_settings_trusted = false;
3102 }
3103 }
3104
3105 if (!f)
3106 return 0;
3107
3108 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
3109
3110 r = settings_load(f, p, &settings);
3111 if (r < 0)
3112 return r;
3113
3114 /* Copy over bits from the settings, unless they have been
3115 * explicitly masked by command line switches. */
3116
3117 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3118 settings->start_mode >= 0) {
3119 arg_start_mode = settings->start_mode;
3120
3121 strv_free(arg_parameters);
3122 arg_parameters = TAKE_PTR(settings->parameters);
3123 }
3124
3125 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3126 settings->pivot_root_new) {
3127 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3128 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3129 }
3130
3131 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3132 settings->working_directory)
3133 free_and_replace(arg_chdir, settings->working_directory);
3134
3135 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3136 settings->environment) {
3137 strv_free(arg_setenv);
3138 arg_setenv = TAKE_PTR(settings->environment);
3139 }
3140
3141 if ((arg_settings_mask & SETTING_USER) == 0 &&
3142 settings->user)
3143 free_and_replace(arg_user, settings->user);
3144
3145 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3146 uint64_t plus;
3147
3148 plus = settings->capability;
3149 if (settings_private_network(settings))
3150 plus |= (1ULL << CAP_NET_ADMIN);
3151
3152 if (!arg_settings_trusted && plus != 0) {
3153 if (settings->capability != 0)
3154 log_warning("Ignoring Capability= setting, file %s is not trusted.", p);
3155 } else
3156 arg_caps_retain |= plus;
3157
3158 arg_caps_retain &= ~settings->drop_capability;
3159 }
3160
3161 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3162 settings->kill_signal > 0)
3163 arg_kill_signal = settings->kill_signal;
3164
3165 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3166 settings->personality != PERSONALITY_INVALID)
3167 arg_personality = settings->personality;
3168
3169 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3170 !sd_id128_is_null(settings->machine_id)) {
3171
3172 if (!arg_settings_trusted)
3173 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p);
3174 else
3175 arg_uuid = settings->machine_id;
3176 }
3177
3178 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3179 settings->read_only >= 0)
3180 arg_read_only = settings->read_only;
3181
3182 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3183 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3184 arg_volatile_mode = settings->volatile_mode;
3185
3186 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3187 settings->n_custom_mounts > 0) {
3188
3189 if (!arg_settings_trusted)
3190 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p);
3191 else {
3192 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3193 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3194 arg_n_custom_mounts = settings->n_custom_mounts;
3195 settings->n_custom_mounts = 0;
3196 }
3197 }
3198
3199 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3200 (settings->private_network >= 0 ||
3201 settings->network_veth >= 0 ||
3202 settings->network_bridge ||
3203 settings->network_zone ||
3204 settings->network_interfaces ||
3205 settings->network_macvlan ||
3206 settings->network_ipvlan ||
3207 settings->network_veth_extra)) {
3208
3209 if (!arg_settings_trusted)
3210 log_warning("Ignoring network settings, file %s is not trusted.", p);
3211 else {
3212 arg_network_veth = settings_network_veth(settings);
3213 arg_private_network = settings_private_network(settings);
3214
3215 strv_free(arg_network_interfaces);
3216 arg_network_interfaces = TAKE_PTR(settings->network_interfaces);
3217
3218 strv_free(arg_network_macvlan);
3219 arg_network_macvlan = TAKE_PTR(settings->network_macvlan);
3220
3221 strv_free(arg_network_ipvlan);
3222 arg_network_ipvlan = TAKE_PTR(settings->network_ipvlan);
3223
3224 strv_free(arg_network_veth_extra);
3225 arg_network_veth_extra = TAKE_PTR(settings->network_veth_extra);
3226
3227 free_and_replace(arg_network_bridge, settings->network_bridge);
3228 free_and_replace(arg_network_zone, settings->network_zone);
3229 }
3230 }
3231
3232 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3233 settings->expose_ports) {
3234
3235 if (!arg_settings_trusted)
3236 log_warning("Ignoring Port= setting, file %s is not trusted.", p);
3237 else {
3238 expose_port_free_all(arg_expose_ports);
3239 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3240 }
3241 }
3242
3243 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3244 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3245
3246 if (!arg_settings_trusted)
3247 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", p);
3248 else {
3249 arg_userns_mode = settings->userns_mode;
3250 arg_uid_shift = settings->uid_shift;
3251 arg_uid_range = settings->uid_range;
3252 arg_userns_chown = settings->userns_chown;
3253 }
3254 }
3255
3256 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3257 arg_notify_ready = settings->notify_ready;
3258
3259 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3260
3261 if (!arg_settings_trusted && !strv_isempty(arg_syscall_whitelist))
3262 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", p);
3263 else {
3264 strv_free(arg_syscall_whitelist);
3265 strv_free(arg_syscall_blacklist);
3266
3267 arg_syscall_whitelist = TAKE_PTR(settings->syscall_whitelist);
3268 arg_syscall_blacklist = TAKE_PTR(settings->syscall_blacklist);
3269 }
3270 }
3271
3272 return 0;
3273 }
3274
3275 static int run(int master,
3276 const char* console,
3277 DissectedImage *dissected_image,
3278 bool interactive,
3279 bool secondary,
3280 FDSet *fds,
3281 char veth_name[IFNAMSIZ], bool *veth_created,
3282 union in_addr_union *exposed,
3283 pid_t *pid, int *ret) {
3284
3285 static const struct sigaction sa = {
3286 .sa_handler = nop_signal_handler,
3287 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
3288 };
3289
3290 _cleanup_release_lock_file_ LockFile uid_shift_lock = LOCK_FILE_INIT;
3291 _cleanup_close_ int etc_passwd_lock = -1;
3292 _cleanup_close_pair_ int
3293 kmsg_socket_pair[2] = { -1, -1 },
3294 rtnl_socket_pair[2] = { -1, -1 },
3295 pid_socket_pair[2] = { -1, -1 },
3296 uuid_socket_pair[2] = { -1, -1 },
3297 notify_socket_pair[2] = { -1, -1 },
3298 uid_shift_socket_pair[2] = { -1, -1 },
3299 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
3300
3301 _cleanup_close_ int notify_socket= -1;
3302 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
3303 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
3304 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
3305 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
3306 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
3307 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
3308 ContainerStatus container_status = 0;
3309 char last_char = 0;
3310 int ifi = 0, r;
3311 ssize_t l;
3312 sigset_t mask_chld;
3313 _cleanup_close_ int netns_fd = -1;
3314
3315 assert_se(sigemptyset(&mask_chld) == 0);
3316 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
3317
3318 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3319 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
3320 * check with getpwuid() if the specific user already exists. Note that /etc might be
3321 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
3322 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
3323 * really just an extra safety net. We kinda assume that the UID range we allocate from is
3324 * really ours. */
3325
3326 etc_passwd_lock = take_etc_passwd_lock(NULL);
3327 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
3328 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
3329 }
3330
3331 r = barrier_create(&barrier);
3332 if (r < 0)
3333 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
3334
3335 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
3336 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
3337
3338 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
3339 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
3340
3341 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
3342 return log_error_errno(errno, "Failed to create pid socket pair: %m");
3343
3344 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
3345 return log_error_errno(errno, "Failed to create id socket pair: %m");
3346
3347 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
3348 return log_error_errno(errno, "Failed to create notify socket pair: %m");
3349
3350 if (arg_userns_mode != USER_NAMESPACE_NO)
3351 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
3352 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
3353
3354 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
3355 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
3356 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
3357
3358 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
3359 * parent's blocking calls and give it a chance to call wait() and terminate. */
3360 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
3361 if (r < 0)
3362 return log_error_errno(errno, "Failed to change the signal mask: %m");
3363
3364 r = sigaction(SIGCHLD, &sa, NULL);
3365 if (r < 0)
3366 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
3367
3368 if (arg_network_namespace_path) {
3369 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
3370 if (netns_fd < 0)
3371 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
3372
3373 r = fd_is_network_ns(netns_fd);
3374 if (r < 0 && r != -ENOTTY)
3375 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
3376 if (r == 0) {
3377 log_error("Path %s doesn't refer to a network namespace", arg_network_namespace_path);
3378 return -EINVAL;
3379 }
3380 }
3381
3382 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
3383 if (*pid < 0)
3384 return log_error_errno(errno, "clone() failed%s: %m",
3385 errno == EINVAL ?
3386 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
3387
3388 if (*pid == 0) {
3389 /* The outer child only has a file system namespace. */
3390 barrier_set_role(&barrier, BARRIER_CHILD);
3391
3392 master = safe_close(master);
3393
3394 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
3395 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3396 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
3397 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
3398 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
3399 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
3400 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
3401
3402 (void) reset_all_signal_handlers();
3403 (void) reset_signal_mask();
3404
3405 r = outer_child(&barrier,
3406 arg_directory,
3407 console,
3408 dissected_image,
3409 interactive,
3410 secondary,
3411 pid_socket_pair[1],
3412 uuid_socket_pair[1],
3413 notify_socket_pair[1],
3414 kmsg_socket_pair[1],
3415 rtnl_socket_pair[1],
3416 uid_shift_socket_pair[1],
3417 unified_cgroup_hierarchy_socket_pair[1],
3418 fds,
3419 netns_fd);
3420 if (r < 0)
3421 _exit(EXIT_FAILURE);
3422
3423 _exit(EXIT_SUCCESS);
3424 }
3425
3426 barrier_set_role(&barrier, BARRIER_PARENT);
3427
3428 fds = fdset_free(fds);
3429
3430 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
3431 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
3432 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
3433 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
3434 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
3435 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
3436 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
3437
3438 if (arg_userns_mode != USER_NAMESPACE_NO) {
3439 /* The child just let us know the UID shift it might have read from the image. */
3440 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
3441 if (l < 0)
3442 return log_error_errno(errno, "Failed to read UID shift: %m");
3443 if (l != sizeof arg_uid_shift) {
3444 log_error("Short read while reading UID shift.");
3445 return -EIO;
3446 }
3447
3448 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3449 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
3450 * image, but if that's already in use, pick a new one, and report back to the child,
3451 * which one we now picked. */
3452
3453 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
3454 if (r < 0)
3455 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
3456
3457 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
3458 if (l < 0)
3459 return log_error_errno(errno, "Failed to send UID shift: %m");
3460 if (l != sizeof arg_uid_shift) {
3461 log_error("Short write while writing UID shift.");
3462 return -EIO;
3463 }
3464 }
3465 }
3466
3467 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3468 /* The child let us know the support cgroup mode it might have read from the image. */
3469 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
3470 if (l < 0)
3471 return log_error_errno(errno, "Failed to read cgroup mode: %m");
3472 if (l != sizeof(arg_unified_cgroup_hierarchy)) {
3473 log_error("Short read while reading cgroup mode.");
3474 return -EIO;
3475 }
3476 }
3477
3478 /* Wait for the outer child. */
3479 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
3480 if (r < 0)
3481 return r;
3482 if (r != EXIT_SUCCESS)
3483 return -EIO;
3484
3485 /* And now retrieve the PID of the inner child. */
3486 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
3487 if (l < 0)
3488 return log_error_errno(errno, "Failed to read inner child PID: %m");
3489 if (l != sizeof *pid) {
3490 log_error("Short read while reading inner child PID.");
3491 return -EIO;
3492 }
3493
3494 /* We also retrieve container UUID in case it was generated by outer child */
3495 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
3496 if (l < 0)
3497 return log_error_errno(errno, "Failed to read container machine ID: %m");
3498 if (l != sizeof(arg_uuid)) {
3499 log_error("Short read while reading container machined ID.");
3500 return -EIO;
3501 }
3502
3503 /* We also retrieve the socket used for notifications generated by outer child */
3504 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
3505 if (notify_socket < 0)
3506 return log_error_errno(notify_socket,
3507 "Failed to receive notification socket from the outer child: %m");
3508
3509 log_debug("Init process invoked as PID "PID_FMT, *pid);
3510
3511 if (arg_userns_mode != USER_NAMESPACE_NO) {
3512 if (!barrier_place_and_sync(&barrier)) { /* #1 */
3513 log_error("Child died too early.");
3514 return -ESRCH;
3515 }
3516
3517 r = setup_uid_map(*pid);
3518 if (r < 0)
3519 return r;
3520
3521 (void) barrier_place(&barrier); /* #2 */
3522 }
3523
3524 if (arg_private_network) {
3525
3526 if (!arg_network_namespace_path) {
3527 /* Wait until the child has unshared its network namespace. */
3528 if (!barrier_place_and_sync(&barrier)) { /* #3 */
3529 log_error("Child died too early");
3530 return -ESRCH;
3531 }
3532 }
3533
3534 r = move_network_interfaces(*pid, arg_network_interfaces);
3535 if (r < 0)
3536 return r;
3537
3538 if (arg_network_veth) {
3539 r = setup_veth(arg_machine, *pid, veth_name,
3540 arg_network_bridge || arg_network_zone);
3541 if (r < 0)
3542 return r;
3543 else if (r > 0)
3544 ifi = r;
3545
3546 if (arg_network_bridge) {
3547 /* Add the interface to a bridge */
3548 r = setup_bridge(veth_name, arg_network_bridge, false);
3549 if (r < 0)
3550 return r;
3551 if (r > 0)
3552 ifi = r;
3553 } else if (arg_network_zone) {
3554 /* Add the interface to a bridge, possibly creating it */
3555 r = setup_bridge(veth_name, arg_network_zone, true);
3556 if (r < 0)
3557 return r;
3558 if (r > 0)
3559 ifi = r;
3560 }
3561 }
3562
3563 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
3564 if (r < 0)
3565 return r;
3566
3567 /* We created the primary and extra veth links now; let's remember this, so that we know to
3568 remove them later on. Note that we don't bother with removing veth links that were created
3569 here when their setup failed half-way, because in that case the kernel should be able to
3570 remove them on its own, since they cannot be referenced by anything yet. */
3571 *veth_created = true;
3572
3573 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
3574 if (r < 0)
3575 return r;
3576
3577 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
3578 if (r < 0)
3579 return r;
3580 }
3581
3582 if (arg_register || !arg_keep_unit) {
3583 r = sd_bus_default_system(&bus);
3584 if (r < 0)
3585 return log_error_errno(r, "Failed to open system bus: %m");
3586 }
3587
3588 if (!arg_keep_unit) {
3589 /* When a new scope is created for this container, then we'll be registered as its controller, in which
3590 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
3591 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
3592
3593 r = sd_bus_match_signal_async(
3594 bus,
3595 NULL,
3596 "org.freedesktop.systemd1",
3597 NULL,
3598 "org.freedesktop.systemd1.Scope",
3599 "RequestStop",
3600 on_request_stop, NULL, PID_TO_PTR(*pid));
3601 if (r < 0)
3602 return log_error_errno(r, "Failed to request RequestStop match: %m");
3603 }
3604
3605 if (arg_register) {
3606
3607 r = register_machine(
3608 bus,
3609 arg_machine,
3610 *pid,
3611 arg_directory,
3612 arg_uuid,
3613 ifi,
3614 arg_slice,
3615 arg_custom_mounts, arg_n_custom_mounts,
3616 arg_kill_signal,
3617 arg_property,
3618 arg_keep_unit,
3619 arg_container_service_name);
3620 if (r < 0)
3621 return r;
3622
3623 } else if (!arg_keep_unit) {
3624
3625 r = allocate_scope(
3626 bus,
3627 arg_machine,
3628 *pid,
3629 arg_slice,
3630 arg_custom_mounts, arg_n_custom_mounts,
3631 arg_kill_signal,
3632 arg_property);
3633 if (r < 0)
3634 return r;
3635
3636 } else if (arg_slice || arg_property)
3637 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
3638
3639 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3640 if (r < 0)
3641 return r;
3642
3643 if (arg_keep_unit) {
3644 r = create_subcgroup(*pid, arg_unified_cgroup_hierarchy);
3645 if (r < 0)
3646 return r;
3647 }
3648
3649 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
3650 if (r < 0)
3651 return r;
3652
3653 /* Notify the child that the parent is ready with all
3654 * its setup (including cgroup-ification), and that
3655 * the child can now hand over control to the code to
3656 * run inside the container. */
3657 (void) barrier_place(&barrier); /* #4 */
3658
3659 /* Block SIGCHLD here, before notifying child.
3660 * process_pty() will handle it with the other signals. */
3661 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
3662
3663 /* Reset signal to default */
3664 r = default_signals(SIGCHLD, -1);
3665 if (r < 0)
3666 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
3667
3668 r = sd_event_new(&event);
3669 if (r < 0)
3670 return log_error_errno(r, "Failed to get default event source: %m");
3671
3672 (void) sd_event_set_watchdog(event, true);
3673
3674 if (bus) {
3675 r = sd_bus_attach_event(bus, event, 0);
3676 if (r < 0)
3677 return log_error_errno(r, "Failed to attach bus to event loop: %m");
3678 }
3679
3680 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
3681 if (r < 0)
3682 return r;
3683
3684 /* Let the child know that we are ready and wait that the child is completely ready now. */
3685 if (!barrier_place_and_sync(&barrier)) { /* #5 */
3686 log_error("Child died too early.");
3687 return -ESRCH;
3688 }
3689
3690 /* At this point we have made use of the UID we picked, and thus nss-mymachines
3691 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
3692 etc_passwd_lock = safe_close(etc_passwd_lock);
3693
3694 sd_notifyf(false,
3695 "STATUS=Container running.\n"
3696 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
3697 if (!arg_notify_ready)
3698 sd_notify(false, "READY=1\n");
3699
3700 if (arg_kill_signal > 0) {
3701 /* Try to kill the init system on SIGINT or SIGTERM */
3702 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
3703 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
3704 } else {
3705 /* Immediately exit */
3706 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
3707 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
3708 }
3709
3710 /* Exit when the child exits */
3711 sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
3712
3713 if (arg_expose_ports) {
3714 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
3715 if (r < 0)
3716 return r;
3717
3718 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
3719 }
3720
3721 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3722
3723 r = pty_forward_new(event, master,
3724 PTY_FORWARD_IGNORE_VHANGUP | (interactive ? 0 : PTY_FORWARD_READ_ONLY),
3725 &forward);
3726 if (r < 0)
3727 return log_error_errno(r, "Failed to create PTY forwarder: %m");
3728
3729 r = sd_event_loop(event);
3730 if (r < 0)
3731 return log_error_errno(r, "Failed to run event loop: %m");
3732
3733 pty_forward_get_last_char(forward, &last_char);
3734
3735 forward = pty_forward_free(forward);
3736
3737 if (!arg_quiet && last_char != '\n')
3738 putc('\n', stdout);
3739
3740 /* Kill if it is not dead yet anyway */
3741 if (arg_register && !arg_keep_unit && bus)
3742 terminate_machine(bus, *pid);
3743
3744 /* Normally redundant, but better safe than sorry */
3745 (void) kill(*pid, SIGKILL);
3746
3747 r = wait_for_container(*pid, &container_status);
3748 *pid = 0;
3749
3750 if (r < 0)
3751 /* We failed to wait for the container, or the container exited abnormally. */
3752 return r;
3753 if (r > 0 || container_status == CONTAINER_TERMINATED) {
3754 /* r > 0 → The container exited with a non-zero status.
3755 * As a special case, we need to replace 133 with a different value,
3756 * because 133 is special-cased in the service file to reboot the container.
3757 * otherwise → The container exited with zero status and a reboot was not requested.
3758 */
3759 if (r == EXIT_FORCE_RESTART)
3760 r = EXIT_FAILURE; /* replace 133 with the general failure code */
3761 *ret = r;
3762 return 0; /* finito */
3763 }
3764
3765 /* CONTAINER_REBOOTED, loop again */
3766
3767 if (arg_keep_unit) {
3768 /* Special handling if we are running as a service: instead of simply
3769 * restarting the machine we want to restart the entire service, so let's
3770 * inform systemd about this with the special exit code 133. The service
3771 * file uses RestartForceExitStatus=133 so that this results in a full
3772 * nspawn restart. This is necessary since we might have cgroup parameters
3773 * set we want to have flushed out. */
3774 *ret = EXIT_FORCE_RESTART;
3775 return 0; /* finito */
3776 }
3777
3778 expose_port_flush(arg_expose_ports, exposed);
3779
3780 (void) remove_veth_links(veth_name, arg_network_veth_extra);
3781 *veth_created = false;
3782 return 1; /* loop again */
3783 }
3784
3785 int main(int argc, char *argv[]) {
3786
3787 _cleanup_free_ char *console = NULL;
3788 _cleanup_close_ int master = -1;
3789 _cleanup_fdset_free_ FDSet *fds = NULL;
3790 int r, n_fd_passed, ret = EXIT_SUCCESS;
3791 char veth_name[IFNAMSIZ] = "";
3792 bool secondary = false, remove_directory = false, remove_image = false;
3793 pid_t pid = 0;
3794 union in_addr_union exposed = {};
3795 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
3796 bool interactive, veth_created = false, remove_tmprootdir = false;
3797 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
3798 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
3799 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3800 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3801
3802 log_parse_environment();
3803 log_open();
3804
3805 /* Make sure rename_process() in the stub init process can work */
3806 saved_argv = argv;
3807 saved_argc = argc;
3808
3809 r = parse_argv(argc, argv);
3810 if (r <= 0)
3811 goto finish;
3812
3813 r = must_be_root();
3814 if (r < 0)
3815 goto finish;
3816
3817 r = determine_names();
3818 if (r < 0)
3819 goto finish;
3820
3821 r = load_settings();
3822 if (r < 0)
3823 goto finish;
3824
3825 r = verify_arguments();
3826 if (r < 0)
3827 goto finish;
3828
3829 r = detect_unified_cgroup_hierarchy_from_environment();
3830 if (r < 0)
3831 goto finish;
3832
3833 n_fd_passed = sd_listen_fds(false);
3834 if (n_fd_passed > 0) {
3835 r = fdset_new_listen_fds(&fds, false);
3836 if (r < 0) {
3837 log_error_errno(r, "Failed to collect file descriptors: %m");
3838 goto finish;
3839 }
3840 }
3841
3842 if (arg_directory) {
3843 assert(!arg_image);
3844
3845 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
3846 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3847 r = -EINVAL;
3848 goto finish;
3849 }
3850
3851 if (arg_ephemeral) {
3852 _cleanup_free_ char *np = NULL;
3853
3854 r = chase_symlinks_and_update(&arg_directory, 0);
3855 if (r < 0)
3856 goto finish;
3857
3858 /* If the specified path is a mount point we
3859 * generate the new snapshot immediately
3860 * inside it under a random name. However if
3861 * the specified is not a mount point we
3862 * create the new snapshot in the parent
3863 * directory, just next to it. */
3864 r = path_is_mount_point(arg_directory, NULL, 0);
3865 if (r < 0) {
3866 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
3867 goto finish;
3868 }
3869 if (r > 0)
3870 r = tempfn_random_child(arg_directory, "machine.", &np);
3871 else
3872 r = tempfn_random(arg_directory, "machine.", &np);
3873 if (r < 0) {
3874 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
3875 goto finish;
3876 }
3877
3878 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3879 if (r < 0) {
3880 log_error_errno(r, "Failed to lock %s: %m", np);
3881 goto finish;
3882 }
3883
3884 r = btrfs_subvol_snapshot(arg_directory, np,
3885 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3886 BTRFS_SNAPSHOT_FALLBACK_COPY |
3887 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3888 BTRFS_SNAPSHOT_RECURSIVE |
3889 BTRFS_SNAPSHOT_QUOTA);
3890 if (r < 0) {
3891 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
3892 goto finish;
3893 }
3894
3895 free_and_replace(arg_directory, np);
3896
3897 remove_directory = true;
3898
3899 } else {
3900 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
3901 if (r < 0)
3902 goto finish;
3903
3904 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3905 if (r == -EBUSY) {
3906 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
3907 goto finish;
3908 }
3909 if (r < 0) {
3910 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
3911 goto finish;
3912 }
3913
3914 if (arg_template) {
3915 r = chase_symlinks_and_update(&arg_template, 0);
3916 if (r < 0)
3917 goto finish;
3918
3919 r = btrfs_subvol_snapshot(arg_template, arg_directory,
3920 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
3921 BTRFS_SNAPSHOT_FALLBACK_COPY |
3922 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
3923 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
3924 BTRFS_SNAPSHOT_RECURSIVE |
3925 BTRFS_SNAPSHOT_QUOTA);
3926 if (r == -EEXIST) {
3927 if (!arg_quiet)
3928 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
3929 } else if (r < 0) {
3930 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
3931 goto finish;
3932 } else {
3933 if (!arg_quiet)
3934 log_info("Populated %s from template %s.", arg_directory, arg_template);
3935 }
3936 }
3937 }
3938
3939 if (arg_start_mode == START_BOOT) {
3940 if (path_is_os_tree(arg_directory) <= 0) {
3941 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
3942 r = -EINVAL;
3943 goto finish;
3944 }
3945 } else {
3946 const char *p;
3947
3948 p = strjoina(arg_directory, "/usr/");
3949 if (laccess(p, F_OK) < 0) {
3950 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", arg_directory);
3951 r = -EINVAL;
3952 goto finish;
3953 }
3954 }
3955
3956 } else {
3957 assert(arg_image);
3958 assert(!arg_template);
3959
3960 r = chase_symlinks_and_update(&arg_image, 0);
3961 if (r < 0)
3962 goto finish;
3963
3964 if (arg_ephemeral) {
3965 _cleanup_free_ char *np = NULL;
3966
3967 r = tempfn_random(arg_image, "machine.", &np);
3968 if (r < 0) {
3969 log_error_errno(r, "Failed to generate name for image snapshot: %m");
3970 goto finish;
3971 }
3972
3973 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3974 if (r < 0) {
3975 r = log_error_errno(r, "Failed to create image lock: %m");
3976 goto finish;
3977 }
3978
3979 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, COPY_REFLINK);
3980 if (r < 0) {
3981 r = log_error_errno(r, "Failed to copy image file: %m");
3982 goto finish;
3983 }
3984
3985 free_and_replace(arg_image, np);
3986
3987 remove_image = true;
3988 } else {
3989 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3990 if (r == -EBUSY) {
3991 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
3992 goto finish;
3993 }
3994 if (r < 0) {
3995 r = log_error_errno(r, "Failed to create image lock: %m");
3996 goto finish;
3997 }
3998
3999 if (!arg_root_hash) {
4000 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4001 if (r < 0) {
4002 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4003 goto finish;
4004 }
4005 }
4006 }
4007
4008 if (!mkdtemp(tmprootdir)) {
4009 r = log_error_errno(errno, "Failed to create temporary directory: %m");
4010 goto finish;
4011 }
4012
4013 remove_tmprootdir = true;
4014
4015 arg_directory = strdup(tmprootdir);
4016 if (!arg_directory) {
4017 r = log_oom();
4018 goto finish;
4019 }
4020
4021 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4022 if (r < 0) {
4023 log_error_errno(r, "Failed to set up loopback block device: %m");
4024 goto finish;
4025 }
4026
4027 r = dissect_image_and_warn(
4028 loop->fd,
4029 arg_image,
4030 arg_root_hash, arg_root_hash_size,
4031 DISSECT_IMAGE_REQUIRE_ROOT,
4032 &dissected_image);
4033 if (r == -ENOPKG) {
4034 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
4035 log_notice("Note that the disk image needs to\n"
4036 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4037 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4038 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4039 " d) or contain a file system without a partition table\n"
4040 "in order to be bootable with systemd-nspawn.");
4041 goto finish;
4042 }
4043 if (r < 0)
4044 goto finish;
4045
4046 if (!arg_root_hash && dissected_image->can_verity)
4047 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4048
4049 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
4050 if (r < 0)
4051 goto finish;
4052
4053 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4054 if (remove_image && unlink(arg_image) >= 0)
4055 remove_image = false;
4056 }
4057
4058 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
4059 if (r < 0)
4060 goto finish;
4061
4062 interactive =
4063 isatty(STDIN_FILENO) > 0 &&
4064 isatty(STDOUT_FILENO) > 0;
4065
4066 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
4067 if (master < 0) {
4068 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4069 goto finish;
4070 }
4071
4072 r = ptsname_malloc(master, &console);
4073 if (r < 0) {
4074 r = log_error_errno(r, "Failed to determine tty name: %m");
4075 goto finish;
4076 }
4077
4078 if (arg_selinux_apifs_context) {
4079 r = mac_selinux_apply(console, arg_selinux_apifs_context);
4080 if (r < 0)
4081 goto finish;
4082 }
4083
4084 if (unlockpt(master) < 0) {
4085 r = log_error_errno(errno, "Failed to unlock tty: %m");
4086 goto finish;
4087 }
4088
4089 if (!arg_quiet)
4090 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4091 arg_machine, arg_image ?: arg_directory);
4092
4093 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
4094
4095 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
4096 r = log_error_errno(errno, "Failed to become subreaper: %m");
4097 goto finish;
4098 }
4099
4100 for (;;) {
4101 r = run(master,
4102 console,
4103 dissected_image,
4104 interactive, secondary,
4105 fds,
4106 veth_name, &veth_created,
4107 &exposed,
4108 &pid, &ret);
4109 if (r <= 0)
4110 break;
4111 }
4112
4113 finish:
4114 sd_notify(false,
4115 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
4116 "STOPPING=1\nSTATUS=Terminating...");
4117
4118 if (pid > 0)
4119 (void) kill(pid, SIGKILL);
4120
4121 /* Try to flush whatever is still queued in the pty */
4122 if (master >= 0) {
4123 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
4124 master = safe_close(master);
4125 }
4126
4127 if (pid > 0)
4128 (void) wait_for_terminate(pid, NULL);
4129
4130 if (remove_directory && arg_directory) {
4131 int k;
4132
4133 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
4134 if (k < 0)
4135 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
4136 }
4137
4138 if (remove_image && arg_image) {
4139 if (unlink(arg_image) < 0)
4140 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
4141 }
4142
4143 if (remove_tmprootdir) {
4144 if (rmdir(tmprootdir) < 0)
4145 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
4146 }
4147
4148 if (arg_machine) {
4149 const char *p;
4150
4151 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
4152 (void) rm_rf(p, REMOVE_ROOT);
4153 }
4154
4155 expose_port_flush(arg_expose_ports, &exposed);
4156
4157 if (veth_created)
4158 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4159 (void) remove_bridge(arg_network_zone);
4160
4161 free(arg_directory);
4162 free(arg_template);
4163 free(arg_image);
4164 free(arg_machine);
4165 free(arg_user);
4166 free(arg_pivot_root_new);
4167 free(arg_pivot_root_old);
4168 free(arg_chdir);
4169 strv_free(arg_setenv);
4170 free(arg_network_bridge);
4171 strv_free(arg_network_interfaces);
4172 strv_free(arg_network_macvlan);
4173 strv_free(arg_network_ipvlan);
4174 strv_free(arg_network_veth_extra);
4175 strv_free(arg_parameters);
4176 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4177 expose_port_free_all(arg_expose_ports);
4178 free(arg_root_hash);
4179
4180 return r < 0 ? EXIT_FAILURE : ret;
4181 }
Unnamed repository; edit this file 'description' to name the repository.