]> 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 #13635 from fbuihuu/no-aliases-with-enable
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #if HAVE_BLKID
4 #include <blkid.h>
5 #endif
6 #include <errno.h>
7 #include <getopt.h>
8 #include <grp.h>
9 #include <linux/fs.h>
10 #include <linux/loop.h>
11 #include <pwd.h>
12 #include <sched.h>
13 #if HAVE_SELINUX
14 #include <selinux/selinux.h>
15 #endif
16 #include <signal.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 #include <sys/file.h>
21 #include <sys/personality.h>
22 #include <sys/prctl.h>
23 #include <sys/types.h>
24 #include <sys/wait.h>
25 #include <unistd.h>
26
27 #include "sd-bus.h"
28 #include "sd-daemon.h"
29 #include "sd-id128.h"
30
31 #include "alloc-util.h"
32 #include "barrier.h"
33 #include "base-filesystem.h"
34 #include "blkid-util.h"
35 #include "btrfs-util.h"
36 #include "bus-error.h"
37 #include "bus-util.h"
38 #include "cap-list.h"
39 #include "capability-util.h"
40 #include "cgroup-util.h"
41 #include "copy.h"
42 #include "cpu-set-util.h"
43 #include "dev-setup.h"
44 #include "dissect-image.h"
45 #include "env-util.h"
46 #include "fd-util.h"
47 #include "fdset.h"
48 #include "fileio.h"
49 #include "format-util.h"
50 #include "fs-util.h"
51 #include "gpt.h"
52 #include "hexdecoct.h"
53 #include "hostname-util.h"
54 #include "id128-util.h"
55 #include "log.h"
56 #include "loop-util.h"
57 #include "loopback-setup.h"
58 #include "machine-image.h"
59 #include "macro.h"
60 #include "main-func.h"
61 #include "missing.h"
62 #include "mkdir.h"
63 #include "mount-util.h"
64 #include "mountpoint-util.h"
65 #include "namespace-util.h"
66 #include "netlink-util.h"
67 #include "nspawn-cgroup.h"
68 #include "nspawn-def.h"
69 #include "nspawn-expose-ports.h"
70 #include "nspawn-mount.h"
71 #include "nspawn-network.h"
72 #include "nspawn-oci.h"
73 #include "nspawn-patch-uid.h"
74 #include "nspawn-register.h"
75 #include "nspawn-seccomp.h"
76 #include "nspawn-settings.h"
77 #include "nspawn-setuid.h"
78 #include "nspawn-stub-pid1.h"
79 #include "nulstr-util.h"
80 #include "os-util.h"
81 #include "pager.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "pretty-print.h"
85 #include "process-util.h"
86 #include "ptyfwd.h"
87 #include "random-util.h"
88 #include "raw-clone.h"
89 #include "rlimit-util.h"
90 #include "rm-rf.h"
91 #if HAVE_SECCOMP
92 #include "seccomp-util.h"
93 #endif
94 #include "selinux-util.h"
95 #include "signal-util.h"
96 #include "socket-util.h"
97 #include "stat-util.h"
98 #include "stdio-util.h"
99 #include "string-table.h"
100 #include "string-util.h"
101 #include "strv.h"
102 #include "sysctl-util.h"
103 #include "terminal-util.h"
104 #include "tmpfile-util.h"
105 #include "umask-util.h"
106 #include "user-util.h"
107 #include "util.h"
108
109 #if HAVE_SPLIT_USR
110 #define STATIC_RESOLV_CONF "/lib/systemd/resolv.conf"
111 #else
112 #define STATIC_RESOLV_CONF "/usr/lib/systemd/resolv.conf"
113 #endif
114
115 /* nspawn is listening on the socket at the path in the constant nspawn_notify_socket_path
116 * nspawn_notify_socket_path is relative to the container
117 * the init process in the container pid can send messages to nspawn following the sd_notify(3) protocol */
118 #define NSPAWN_NOTIFY_SOCKET_PATH "/run/systemd/nspawn/notify"
119
120 #define EXIT_FORCE_RESTART 133
121
122 typedef enum ContainerStatus {
123 CONTAINER_TERMINATED,
124 CONTAINER_REBOOTED,
125 } ContainerStatus;
126
127 static char *arg_directory = NULL;
128 static char *arg_template = NULL;
129 static char *arg_chdir = NULL;
130 static char *arg_pivot_root_new = NULL;
131 static char *arg_pivot_root_old = NULL;
132 static char *arg_user = NULL;
133 static uid_t arg_uid = UID_INVALID;
134 static gid_t arg_gid = GID_INVALID;
135 static gid_t* arg_supplementary_gids = NULL;
136 static size_t arg_n_supplementary_gids = 0;
137 static sd_id128_t arg_uuid = {};
138 static char *arg_machine = NULL; /* The name used by the host to refer to this */
139 static char *arg_hostname = NULL; /* The name the payload sees by default */
140 static const char *arg_selinux_context = NULL;
141 static const char *arg_selinux_apifs_context = NULL;
142 static char *arg_slice = NULL;
143 static bool arg_private_network = false;
144 static bool arg_read_only = false;
145 static StartMode arg_start_mode = START_PID1;
146 static bool arg_ephemeral = false;
147 static LinkJournal arg_link_journal = LINK_AUTO;
148 static bool arg_link_journal_try = false;
149 static uint64_t arg_caps_retain =
150 (1ULL << CAP_AUDIT_CONTROL) |
151 (1ULL << CAP_AUDIT_WRITE) |
152 (1ULL << CAP_CHOWN) |
153 (1ULL << CAP_DAC_OVERRIDE) |
154 (1ULL << CAP_DAC_READ_SEARCH) |
155 (1ULL << CAP_FOWNER) |
156 (1ULL << CAP_FSETID) |
157 (1ULL << CAP_IPC_OWNER) |
158 (1ULL << CAP_KILL) |
159 (1ULL << CAP_LEASE) |
160 (1ULL << CAP_LINUX_IMMUTABLE) |
161 (1ULL << CAP_MKNOD) |
162 (1ULL << CAP_NET_BIND_SERVICE) |
163 (1ULL << CAP_NET_BROADCAST) |
164 (1ULL << CAP_NET_RAW) |
165 (1ULL << CAP_SETFCAP) |
166 (1ULL << CAP_SETGID) |
167 (1ULL << CAP_SETPCAP) |
168 (1ULL << CAP_SETUID) |
169 (1ULL << CAP_SYS_ADMIN) |
170 (1ULL << CAP_SYS_BOOT) |
171 (1ULL << CAP_SYS_CHROOT) |
172 (1ULL << CAP_SYS_NICE) |
173 (1ULL << CAP_SYS_PTRACE) |
174 (1ULL << CAP_SYS_RESOURCE) |
175 (1ULL << CAP_SYS_TTY_CONFIG);
176 static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
177 static CustomMount *arg_custom_mounts = NULL;
178 static size_t arg_n_custom_mounts = 0;
179 static char **arg_setenv = NULL;
180 static bool arg_quiet = false;
181 static bool arg_register = true;
182 static bool arg_keep_unit = false;
183 static char **arg_network_interfaces = NULL;
184 static char **arg_network_macvlan = NULL;
185 static char **arg_network_ipvlan = NULL;
186 static bool arg_network_veth = false;
187 static char **arg_network_veth_extra = NULL;
188 static char *arg_network_bridge = NULL;
189 static char *arg_network_zone = NULL;
190 static char *arg_network_namespace_path = NULL;
191 static PagerFlags arg_pager_flags = 0;
192 static unsigned long arg_personality = PERSONALITY_INVALID;
193 static char *arg_image = NULL;
194 static char *arg_oci_bundle = NULL;
195 static VolatileMode arg_volatile_mode = VOLATILE_NO;
196 static ExposePort *arg_expose_ports = NULL;
197 static char **arg_property = NULL;
198 static sd_bus_message *arg_property_message = NULL;
199 static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
200 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
201 static bool arg_userns_chown = false;
202 static int arg_kill_signal = 0;
203 static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
204 static SettingsMask arg_settings_mask = 0;
205 static int arg_settings_trusted = -1;
206 static char **arg_parameters = NULL;
207 static const char *arg_container_service_name = "systemd-nspawn";
208 static bool arg_notify_ready = false;
209 static bool arg_use_cgns = true;
210 static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
211 static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
212 static void *arg_root_hash = NULL;
213 static size_t arg_root_hash_size = 0;
214 static char **arg_syscall_whitelist = NULL;
215 static char **arg_syscall_blacklist = NULL;
216 #if HAVE_SECCOMP
217 static scmp_filter_ctx arg_seccomp = NULL;
218 #endif
219 static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
220 static bool arg_no_new_privileges = false;
221 static int arg_oom_score_adjust = 0;
222 static bool arg_oom_score_adjust_set = false;
223 static CPUSet arg_cpu_set = {};
224 static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
225 static TimezoneMode arg_timezone = TIMEZONE_AUTO;
226 static unsigned arg_console_width = (unsigned) -1, arg_console_height = (unsigned) -1;
227 static DeviceNode* arg_extra_nodes = NULL;
228 static size_t arg_n_extra_nodes = 0;
229 static char **arg_sysctl = NULL;
230 static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
231
232 STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
233 STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
234 STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
235 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
236 STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
237 STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
238 STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
239 STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
240 STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
241 STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
242 STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
243 STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
244 STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
245 STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
246 STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
247 STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
248 STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
249 STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
250 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
251 STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
252 STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
253 STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
254 STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
255 STATIC_DESTRUCTOR_REGISTER(arg_root_hash, freep);
256 STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist, strv_freep);
257 STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist, strv_freep);
258 #if HAVE_SECCOMP
259 STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
260 #endif
261 STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
262 STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
263
264 static int handle_arg_console(const char *arg) {
265 if (streq(arg, "help")) {
266 puts("interactive\n"
267 "read-only\n"
268 "passive\n"
269 "pipe");
270 return 0;
271 }
272
273 if (streq(arg, "interactive"))
274 arg_console_mode = CONSOLE_INTERACTIVE;
275 else if (streq(arg, "read-only"))
276 arg_console_mode = CONSOLE_READ_ONLY;
277 else if (streq(arg, "passive"))
278 arg_console_mode = CONSOLE_PASSIVE;
279 else if (streq(arg, "pipe"))
280 arg_console_mode = CONSOLE_PIPE;
281 else
282 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
283
284 arg_settings_mask |= SETTING_CONSOLE_MODE;
285 return 1;
286 }
287
288 static int help(void) {
289 _cleanup_free_ char *link = NULL;
290 int r;
291
292 (void) pager_open(arg_pager_flags);
293
294 r = terminal_urlify_man("systemd-nspawn", "1", &link);
295 if (r < 0)
296 return log_oom();
297
298 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
299 "Spawn a command or OS in a light-weight container.\n\n"
300 " -h --help Show this help\n"
301 " --version Print version string\n"
302 " -q --quiet Do not show status information\n"
303 " --no-pager Do not pipe output into a pager\n"
304 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
305 "%3$sImage:%4$s\n"
306 " -D --directory=PATH Root directory for the container\n"
307 " --template=PATH Initialize root directory from template directory,\n"
308 " if missing\n"
309 " -x --ephemeral Run container with snapshot of root directory, and\n"
310 " remove it after exit\n"
311 " -i --image=PATH Root file system disk image (or device node) for\n"
312 " the container\n"
313 " --oci-bundle=PATH OCI bundle directory\n"
314 " --read-only Mount the root directory read-only\n"
315 " --volatile[=MODE] Run the system in volatile mode\n"
316 " --root-hash=HASH Specify verity root hash for root disk image\n"
317 " --pivot-root=PATH[:PATH]\n"
318 " Pivot root to given directory in the container\n\n"
319 "%3$sExecution:%4$s\n"
320 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
321 " -b --boot Boot up full system (i.e. invoke init)\n"
322 " --chdir=PATH Set working directory in the container\n"
323 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
324 " -u --user=USER Run the command under specified user or UID\n"
325 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
326 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
327 "%3$sSystem Identity:%4$s\n"
328 " -M --machine=NAME Set the machine name for the container\n"
329 " --hostname=NAME Override the hostname for the container\n"
330 " --uuid=UUID Set a specific machine UUID for the container\n\n"
331 "%3$sProperties:%4$s\n"
332 " -S --slice=SLICE Place the container in the specified slice\n"
333 " --property=NAME=VALUE Set scope unit property\n"
334 " --register=BOOLEAN Register container as machine\n"
335 " --keep-unit Do not register a scope for the machine, reuse\n"
336 " the service unit nspawn is running in\n\n"
337 "%3$sUser Namespacing:%4$s\n"
338 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
339 " --private-users[=UIDBASE[:NUIDS]]\n"
340 " Similar, but with user configured UID/GID range\n"
341 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
342 "%3$sNetworking:%4$s\n"
343 " --private-network Disable network in container\n"
344 " --network-interface=INTERFACE\n"
345 " Assign an existing network interface to the\n"
346 " container\n"
347 " --network-macvlan=INTERFACE\n"
348 " Create a macvlan network interface based on an\n"
349 " existing network interface to the container\n"
350 " --network-ipvlan=INTERFACE\n"
351 " Create a ipvlan network interface based on an\n"
352 " existing network interface to the container\n"
353 " -n --network-veth Add a virtual Ethernet connection between host\n"
354 " and container\n"
355 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
356 " Add an additional virtual Ethernet link between\n"
357 " host and container\n"
358 " --network-bridge=INTERFACE\n"
359 " Add a virtual Ethernet connection to the container\n"
360 " and attach it to an existing bridge on the host\n"
361 " --network-zone=NAME Similar, but attach the new interface to an\n"
362 " an automatically managed bridge interface\n"
363 " --network-namespace-path=PATH\n"
364 " Set network namespace to the one represented by\n"
365 " the specified kernel namespace file node\n"
366 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
367 " Expose a container IP port on the host\n\n"
368 "%3$sSecurity:%4$s\n"
369 " --capability=CAP In addition to the default, retain specified\n"
370 " capability\n"
371 " --drop-capability=CAP Drop the specified capability from the default set\n"
372 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
373 " --system-call-filter=LIST|~LIST\n"
374 " Permit/prohibit specific system calls\n"
375 " -Z --selinux-context=SECLABEL\n"
376 " Set the SELinux security context to be used by\n"
377 " processes in the container\n"
378 " -L --selinux-apifs-context=SECLABEL\n"
379 " Set the SELinux security context to be used by\n"
380 " API/tmpfs file systems in the container\n\n"
381 "%3$sResources:%4$s\n"
382 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
383 " --oom-score-adjust=VALUE\n"
384 " Adjust the OOM score value for the payload\n"
385 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
386 " --personality=ARCH Pick personality for this container\n\n"
387 "%3$sIntegration:%4$s\n"
388 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
389 " --timezone=MODE Select mode of /etc/localtime initialization\n"
390 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
391 " host, try-guest, try-host\n"
392 " -j Equivalent to --link-journal=try-guest\n\n"
393 "%3$sMounts:%4$s\n"
394 " --bind=PATH[:PATH[:OPTIONS]]\n"
395 " Bind mount a file or directory from the host into\n"
396 " the container\n"
397 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
398 " Similar, but creates a read-only bind mount\n"
399 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
400 " it\n"
401 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
402 " --overlay=PATH[:PATH...]:PATH\n"
403 " Create an overlay mount from the host to \n"
404 " the container\n"
405 " --overlay-ro=PATH[:PATH...]:PATH\n"
406 " Similar, but creates a read-only overlay mount\n\n"
407 "%3$sInput/Output:%4$s\n"
408 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
409 " set up for the container.\n"
410 " -P --pipe Equivalent to --console=pipe\n"
411 "\nSee the %2$s for details.\n"
412 , program_invocation_short_name
413 , link
414 , ansi_underline(), ansi_normal());
415
416 return 0;
417 }
418
419 static int custom_mount_check_all(void) {
420 size_t i;
421
422 for (i = 0; i < arg_n_custom_mounts; i++) {
423 CustomMount *m = &arg_custom_mounts[i];
424
425 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
426 if (arg_userns_chown)
427 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
428 "--private-users-chown may not be combined with custom root mounts.");
429 else if (arg_uid_shift == UID_INVALID)
430 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
431 "--private-users with automatic UID shift may not be combined with custom root mounts.");
432 }
433 }
434
435 return 0;
436 }
437
438 static int detect_unified_cgroup_hierarchy_from_environment(void) {
439 const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
440 int r;
441
442 /* Allow the user to control whether the unified hierarchy is used */
443
444 e = getenv(var);
445 if (!e) {
446 static bool warned = false;
447
448 var = "UNIFIED_CGROUP_HIERARCHY";
449 e = getenv(var);
450 if (e && !warned) {
451 log_info("$UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY.");
452 warned = true;
453 }
454 }
455
456 if (!isempty(e)) {
457 r = parse_boolean(e);
458 if (r < 0)
459 return log_error_errno(r, "Failed to parse $%s: %m", var);
460 if (r > 0)
461 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
462 else
463 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
464 }
465
466 return 0;
467 }
468
469 static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
470 int r;
471
472 /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
473 * in the image actually supports. */
474 r = cg_all_unified();
475 if (r < 0)
476 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
477 if (r > 0) {
478 /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
479 * routine only detects 231, so we'll have a false negative here for 230. */
480 r = systemd_installation_has_version(directory, 230);
481 if (r < 0)
482 return log_error_errno(r, "Failed to determine systemd version in container: %m");
483 if (r > 0)
484 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
485 else
486 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
487 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
488 /* Mixed cgroup hierarchy support was added in 233 */
489 r = systemd_installation_has_version(directory, 233);
490 if (r < 0)
491 return log_error_errno(r, "Failed to determine systemd version in container: %m");
492 if (r > 0)
493 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
494 else
495 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
496 } else
497 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
498
499 log_debug("Using %s hierarchy for container.",
500 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
501 arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
502
503 return 0;
504 }
505
506 static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
507 int r;
508
509 r = getenv_bool(name);
510 if (r == -ENXIO)
511 return 0;
512 if (r < 0)
513 return log_error_errno(r, "Failed to parse $%s: %m", name);
514
515 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
516 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
517 return 0;
518 }
519
520 static int parse_mount_settings_env(void) {
521 const char *e;
522 int r;
523
524 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
525 if (r < 0 && r != -ENXIO)
526 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
527 if (r >= 0)
528 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
529
530 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
531 if (streq_ptr(e, "network"))
532 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
533
534 else if (e) {
535 r = parse_boolean(e);
536 if (r < 0)
537 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
538
539 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
540 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
541 }
542
543 return 0;
544 }
545
546 static int parse_environment(void) {
547 const char *e;
548 int r;
549
550 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
551 if (r < 0)
552 return r;
553 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
554 if (r < 0)
555 return r;
556 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
557 if (r < 0)
558 return r;
559 r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
560 if (r < 0)
561 return r;
562
563 r = parse_mount_settings_env();
564 if (r < 0)
565 return r;
566
567 /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
568 * even if it is supported. If not supported, it has no effect. */
569 if (!cg_ns_supported())
570 arg_use_cgns = false;
571 else {
572 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
573 if (r < 0) {
574 if (r != -ENXIO)
575 return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
576
577 arg_use_cgns = true;
578 } else {
579 arg_use_cgns = r > 0;
580 arg_settings_mask |= SETTING_USE_CGNS;
581 }
582 }
583
584 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
585 if (e)
586 arg_container_service_name = e;
587
588 return detect_unified_cgroup_hierarchy_from_environment();
589 }
590
591 static int parse_argv(int argc, char *argv[]) {
592 enum {
593 ARG_VERSION = 0x100,
594 ARG_PRIVATE_NETWORK,
595 ARG_UUID,
596 ARG_READ_ONLY,
597 ARG_CAPABILITY,
598 ARG_DROP_CAPABILITY,
599 ARG_LINK_JOURNAL,
600 ARG_BIND,
601 ARG_BIND_RO,
602 ARG_TMPFS,
603 ARG_OVERLAY,
604 ARG_OVERLAY_RO,
605 ARG_INACCESSIBLE,
606 ARG_SHARE_SYSTEM,
607 ARG_REGISTER,
608 ARG_KEEP_UNIT,
609 ARG_NETWORK_INTERFACE,
610 ARG_NETWORK_MACVLAN,
611 ARG_NETWORK_IPVLAN,
612 ARG_NETWORK_BRIDGE,
613 ARG_NETWORK_ZONE,
614 ARG_NETWORK_VETH_EXTRA,
615 ARG_NETWORK_NAMESPACE_PATH,
616 ARG_PERSONALITY,
617 ARG_VOLATILE,
618 ARG_TEMPLATE,
619 ARG_PROPERTY,
620 ARG_PRIVATE_USERS,
621 ARG_KILL_SIGNAL,
622 ARG_SETTINGS,
623 ARG_CHDIR,
624 ARG_PIVOT_ROOT,
625 ARG_PRIVATE_USERS_CHOWN,
626 ARG_NOTIFY_READY,
627 ARG_ROOT_HASH,
628 ARG_SYSTEM_CALL_FILTER,
629 ARG_RLIMIT,
630 ARG_HOSTNAME,
631 ARG_NO_NEW_PRIVILEGES,
632 ARG_OOM_SCORE_ADJUST,
633 ARG_CPU_AFFINITY,
634 ARG_RESOLV_CONF,
635 ARG_TIMEZONE,
636 ARG_CONSOLE,
637 ARG_PIPE,
638 ARG_OCI_BUNDLE,
639 ARG_NO_PAGER,
640 };
641
642 static const struct option options[] = {
643 { "help", no_argument, NULL, 'h' },
644 { "version", no_argument, NULL, ARG_VERSION },
645 { "directory", required_argument, NULL, 'D' },
646 { "template", required_argument, NULL, ARG_TEMPLATE },
647 { "ephemeral", no_argument, NULL, 'x' },
648 { "user", required_argument, NULL, 'u' },
649 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
650 { "as-pid2", no_argument, NULL, 'a' },
651 { "boot", no_argument, NULL, 'b' },
652 { "uuid", required_argument, NULL, ARG_UUID },
653 { "read-only", no_argument, NULL, ARG_READ_ONLY },
654 { "capability", required_argument, NULL, ARG_CAPABILITY },
655 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
656 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
657 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
658 { "bind", required_argument, NULL, ARG_BIND },
659 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
660 { "tmpfs", required_argument, NULL, ARG_TMPFS },
661 { "overlay", required_argument, NULL, ARG_OVERLAY },
662 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
663 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
664 { "machine", required_argument, NULL, 'M' },
665 { "hostname", required_argument, NULL, ARG_HOSTNAME },
666 { "slice", required_argument, NULL, 'S' },
667 { "setenv", required_argument, NULL, 'E' },
668 { "selinux-context", required_argument, NULL, 'Z' },
669 { "selinux-apifs-context", required_argument, NULL, 'L' },
670 { "quiet", no_argument, NULL, 'q' },
671 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
672 { "register", required_argument, NULL, ARG_REGISTER },
673 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
674 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
675 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
676 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
677 { "network-veth", no_argument, NULL, 'n' },
678 { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
679 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
680 { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
681 { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
682 { "personality", required_argument, NULL, ARG_PERSONALITY },
683 { "image", required_argument, NULL, 'i' },
684 { "volatile", optional_argument, NULL, ARG_VOLATILE },
685 { "port", required_argument, NULL, 'p' },
686 { "property", required_argument, NULL, ARG_PROPERTY },
687 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
688 { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN },
689 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
690 { "settings", required_argument, NULL, ARG_SETTINGS },
691 { "chdir", required_argument, NULL, ARG_CHDIR },
692 { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
693 { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
694 { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
695 { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
696 { "rlimit", required_argument, NULL, ARG_RLIMIT },
697 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
698 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
699 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
700 { "timezone", required_argument, NULL, ARG_TIMEZONE },
701 { "console", required_argument, NULL, ARG_CONSOLE },
702 { "pipe", no_argument, NULL, ARG_PIPE },
703 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
704 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
705 {}
706 };
707
708 int c, r;
709 const char *p;
710 uint64_t plus = 0, minus = 0;
711 bool mask_all_settings = false, mask_no_settings = false;
712
713 assert(argc >= 0);
714 assert(argv);
715
716 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
717 switch (c) {
718
719 case 'h':
720 return help();
721
722 case ARG_VERSION:
723 return version();
724
725 case 'D':
726 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
727 if (r < 0)
728 return r;
729
730 arg_settings_mask |= SETTING_DIRECTORY;
731 break;
732
733 case ARG_TEMPLATE:
734 r = parse_path_argument_and_warn(optarg, false, &arg_template);
735 if (r < 0)
736 return r;
737
738 arg_settings_mask |= SETTING_DIRECTORY;
739 break;
740
741 case 'i':
742 r = parse_path_argument_and_warn(optarg, false, &arg_image);
743 if (r < 0)
744 return r;
745
746 arg_settings_mask |= SETTING_DIRECTORY;
747 break;
748
749 case ARG_OCI_BUNDLE:
750 r = parse_path_argument_and_warn(optarg, false, &arg_oci_bundle);
751 if (r < 0)
752 return r;
753
754 break;
755
756 case 'x':
757 arg_ephemeral = true;
758 arg_settings_mask |= SETTING_EPHEMERAL;
759 break;
760
761 case 'u':
762 r = free_and_strdup(&arg_user, optarg);
763 if (r < 0)
764 return log_oom();
765
766 arg_settings_mask |= SETTING_USER;
767 break;
768
769 case ARG_NETWORK_ZONE: {
770 char *j;
771
772 j = strjoin("vz-", optarg);
773 if (!j)
774 return log_oom();
775
776 if (!ifname_valid(j)) {
777 log_error("Network zone name not valid: %s", j);
778 free(j);
779 return -EINVAL;
780 }
781
782 free_and_replace(arg_network_zone, j);
783
784 arg_network_veth = true;
785 arg_private_network = true;
786 arg_settings_mask |= SETTING_NETWORK;
787 break;
788 }
789
790 case ARG_NETWORK_BRIDGE:
791
792 if (!ifname_valid(optarg))
793 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
794 "Bridge interface name not valid: %s", optarg);
795
796 r = free_and_strdup(&arg_network_bridge, optarg);
797 if (r < 0)
798 return log_oom();
799
800 _fallthrough_;
801 case 'n':
802 arg_network_veth = true;
803 arg_private_network = true;
804 arg_settings_mask |= SETTING_NETWORK;
805 break;
806
807 case ARG_NETWORK_VETH_EXTRA:
808 r = veth_extra_parse(&arg_network_veth_extra, optarg);
809 if (r < 0)
810 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
811
812 arg_private_network = true;
813 arg_settings_mask |= SETTING_NETWORK;
814 break;
815
816 case ARG_NETWORK_INTERFACE:
817 if (!ifname_valid(optarg))
818 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
819 "Network interface name not valid: %s", optarg);
820
821 if (strv_extend(&arg_network_interfaces, optarg) < 0)
822 return log_oom();
823
824 arg_private_network = true;
825 arg_settings_mask |= SETTING_NETWORK;
826 break;
827
828 case ARG_NETWORK_MACVLAN:
829
830 if (!ifname_valid(optarg))
831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
832 "MACVLAN network interface name not valid: %s", optarg);
833
834 if (strv_extend(&arg_network_macvlan, optarg) < 0)
835 return log_oom();
836
837 arg_private_network = true;
838 arg_settings_mask |= SETTING_NETWORK;
839 break;
840
841 case ARG_NETWORK_IPVLAN:
842
843 if (!ifname_valid(optarg))
844 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
845 "IPVLAN network interface name not valid: %s", optarg);
846
847 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
848 return log_oom();
849
850 _fallthrough_;
851 case ARG_PRIVATE_NETWORK:
852 arg_private_network = true;
853 arg_settings_mask |= SETTING_NETWORK;
854 break;
855
856 case ARG_NETWORK_NAMESPACE_PATH:
857 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
858 if (r < 0)
859 return r;
860
861 arg_settings_mask |= SETTING_NETWORK;
862 break;
863
864 case 'b':
865 if (arg_start_mode == START_PID2)
866 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
867 "--boot and --as-pid2 may not be combined.");
868
869 arg_start_mode = START_BOOT;
870 arg_settings_mask |= SETTING_START_MODE;
871 break;
872
873 case 'a':
874 if (arg_start_mode == START_BOOT)
875 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
876 "--boot and --as-pid2 may not be combined.");
877
878 arg_start_mode = START_PID2;
879 arg_settings_mask |= SETTING_START_MODE;
880 break;
881
882 case ARG_UUID:
883 r = sd_id128_from_string(optarg, &arg_uuid);
884 if (r < 0)
885 return log_error_errno(r, "Invalid UUID: %s", optarg);
886
887 if (sd_id128_is_null(arg_uuid))
888 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
889 "Machine UUID may not be all zeroes.");
890
891 arg_settings_mask |= SETTING_MACHINE_ID;
892 break;
893
894 case 'S':
895 r = free_and_strdup(&arg_slice, optarg);
896 if (r < 0)
897 return log_oom();
898
899 arg_settings_mask |= SETTING_SLICE;
900 break;
901
902 case 'M':
903 if (isempty(optarg))
904 arg_machine = mfree(arg_machine);
905 else {
906 if (!machine_name_is_valid(optarg))
907 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
908 "Invalid machine name: %s", optarg);
909
910 r = free_and_strdup(&arg_machine, optarg);
911 if (r < 0)
912 return log_oom();
913 }
914 break;
915
916 case ARG_HOSTNAME:
917 if (isempty(optarg))
918 arg_hostname = mfree(arg_hostname);
919 else {
920 if (!hostname_is_valid(optarg, false))
921 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
922 "Invalid hostname: %s", optarg);
923
924 r = free_and_strdup(&arg_hostname, optarg);
925 if (r < 0)
926 return log_oom();
927 }
928
929 arg_settings_mask |= SETTING_HOSTNAME;
930 break;
931
932 case 'Z':
933 arg_selinux_context = optarg;
934 break;
935
936 case 'L':
937 arg_selinux_apifs_context = optarg;
938 break;
939
940 case ARG_READ_ONLY:
941 arg_read_only = true;
942 arg_settings_mask |= SETTING_READ_ONLY;
943 break;
944
945 case ARG_CAPABILITY:
946 case ARG_DROP_CAPABILITY: {
947 p = optarg;
948 for (;;) {
949 _cleanup_free_ char *t = NULL;
950
951 r = extract_first_word(&p, &t, ",", 0);
952 if (r < 0)
953 return log_error_errno(r, "Failed to parse capability %s.", t);
954 if (r == 0)
955 break;
956
957 if (streq(t, "all")) {
958 if (c == ARG_CAPABILITY)
959 plus = (uint64_t) -1;
960 else
961 minus = (uint64_t) -1;
962 } else {
963 r = capability_from_name(t);
964 if (r < 0)
965 return log_error_errno(r, "Failed to parse capability %s.", t);
966
967 if (c == ARG_CAPABILITY)
968 plus |= 1ULL << r;
969 else
970 minus |= 1ULL << r;
971 }
972 }
973
974 arg_settings_mask |= SETTING_CAPABILITY;
975 break;
976 }
977
978 case ARG_NO_NEW_PRIVILEGES:
979 r = parse_boolean(optarg);
980 if (r < 0)
981 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
982
983 arg_no_new_privileges = r;
984 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
985 break;
986
987 case 'j':
988 arg_link_journal = LINK_GUEST;
989 arg_link_journal_try = true;
990 arg_settings_mask |= SETTING_LINK_JOURNAL;
991 break;
992
993 case ARG_LINK_JOURNAL:
994 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
995 if (r < 0)
996 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
997
998 arg_settings_mask |= SETTING_LINK_JOURNAL;
999 break;
1000
1001 case ARG_BIND:
1002 case ARG_BIND_RO:
1003 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
1004 if (r < 0)
1005 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
1006
1007 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1008 break;
1009
1010 case ARG_TMPFS:
1011 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1012 if (r < 0)
1013 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
1014
1015 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1016 break;
1017
1018 case ARG_OVERLAY:
1019 case ARG_OVERLAY_RO:
1020 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
1021 if (r == -EADDRNOTAVAIL)
1022 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
1023 if (r < 0)
1024 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
1025
1026 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1027 break;
1028
1029 case ARG_INACCESSIBLE:
1030 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
1031 if (r < 0)
1032 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
1033
1034 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
1035 break;
1036
1037 case 'E': {
1038 char **n;
1039
1040 if (!env_assignment_is_valid(optarg))
1041 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1042 "Environment variable assignment '%s' is not valid.", optarg);
1043
1044 n = strv_env_set(arg_setenv, optarg);
1045 if (!n)
1046 return log_oom();
1047
1048 strv_free_and_replace(arg_setenv, n);
1049 arg_settings_mask |= SETTING_ENVIRONMENT;
1050 break;
1051 }
1052
1053 case 'q':
1054 arg_quiet = true;
1055 break;
1056
1057 case ARG_SHARE_SYSTEM:
1058 /* We don't officially support this anymore, except for compat reasons. People should use the
1059 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
1060 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
1061 arg_clone_ns_flags = 0;
1062 break;
1063
1064 case ARG_REGISTER:
1065 r = parse_boolean(optarg);
1066 if (r < 0) {
1067 log_error("Failed to parse --register= argument: %s", optarg);
1068 return r;
1069 }
1070
1071 arg_register = r;
1072 break;
1073
1074 case ARG_KEEP_UNIT:
1075 arg_keep_unit = true;
1076 break;
1077
1078 case ARG_PERSONALITY:
1079
1080 arg_personality = personality_from_string(optarg);
1081 if (arg_personality == PERSONALITY_INVALID)
1082 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1083 "Unknown or unsupported personality '%s'.", optarg);
1084
1085 arg_settings_mask |= SETTING_PERSONALITY;
1086 break;
1087
1088 case ARG_VOLATILE:
1089
1090 if (!optarg)
1091 arg_volatile_mode = VOLATILE_YES;
1092 else if (streq(optarg, "help")) {
1093 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1094 return 0;
1095 } else {
1096 VolatileMode m;
1097
1098 m = volatile_mode_from_string(optarg);
1099 if (m < 0)
1100 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1101 "Failed to parse --volatile= argument: %s", optarg);
1102 else
1103 arg_volatile_mode = m;
1104 }
1105
1106 arg_settings_mask |= SETTING_VOLATILE_MODE;
1107 break;
1108
1109 case 'p':
1110 r = expose_port_parse(&arg_expose_ports, optarg);
1111 if (r == -EEXIST)
1112 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1113 if (r < 0)
1114 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
1115
1116 arg_settings_mask |= SETTING_EXPOSE_PORTS;
1117 break;
1118
1119 case ARG_PROPERTY:
1120 if (strv_extend(&arg_property, optarg) < 0)
1121 return log_oom();
1122
1123 break;
1124
1125 case ARG_PRIVATE_USERS: {
1126 int boolean = -1;
1127
1128 if (!optarg)
1129 boolean = true;
1130 else if (!in_charset(optarg, DIGITS))
1131 /* do *not* parse numbers as booleans */
1132 boolean = parse_boolean(optarg);
1133
1134 if (boolean == false) {
1135 /* no: User namespacing off */
1136 arg_userns_mode = USER_NAMESPACE_NO;
1137 arg_uid_shift = UID_INVALID;
1138 arg_uid_range = UINT32_C(0x10000);
1139 } else if (boolean == true) {
1140 /* yes: User namespacing on, UID range is read from root dir */
1141 arg_userns_mode = USER_NAMESPACE_FIXED;
1142 arg_uid_shift = UID_INVALID;
1143 arg_uid_range = UINT32_C(0x10000);
1144 } else if (streq(optarg, "pick")) {
1145 /* pick: User namespacing on, UID range is picked randomly */
1146 arg_userns_mode = USER_NAMESPACE_PICK;
1147 arg_uid_shift = UID_INVALID;
1148 arg_uid_range = UINT32_C(0x10000);
1149 } else {
1150 _cleanup_free_ char *buffer = NULL;
1151 const char *range, *shift;
1152
1153 /* anything else: User namespacing on, UID range is explicitly configured */
1154
1155 range = strchr(optarg, ':');
1156 if (range) {
1157 buffer = strndup(optarg, range - optarg);
1158 if (!buffer)
1159 return log_oom();
1160 shift = buffer;
1161
1162 range++;
1163 r = safe_atou32(range, &arg_uid_range);
1164 if (r < 0)
1165 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
1166 } else
1167 shift = optarg;
1168
1169 r = parse_uid(shift, &arg_uid_shift);
1170 if (r < 0)
1171 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
1172
1173 arg_userns_mode = USER_NAMESPACE_FIXED;
1174 }
1175
1176 if (arg_uid_range <= 0)
1177 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1178 "UID range cannot be 0.");
1179
1180 arg_settings_mask |= SETTING_USERNS;
1181 break;
1182 }
1183
1184 case 'U':
1185 if (userns_supported()) {
1186 arg_userns_mode = USER_NAMESPACE_PICK;
1187 arg_uid_shift = UID_INVALID;
1188 arg_uid_range = UINT32_C(0x10000);
1189
1190 arg_settings_mask |= SETTING_USERNS;
1191 }
1192
1193 break;
1194
1195 case ARG_PRIVATE_USERS_CHOWN:
1196 arg_userns_chown = true;
1197
1198 arg_settings_mask |= SETTING_USERNS;
1199 break;
1200
1201 case ARG_KILL_SIGNAL:
1202 if (streq(optarg, "help")) {
1203 DUMP_STRING_TABLE(signal, int, _NSIG);
1204 return 0;
1205 }
1206
1207 arg_kill_signal = signal_from_string(optarg);
1208 if (arg_kill_signal < 0)
1209 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1210 "Cannot parse signal: %s", optarg);
1211
1212 arg_settings_mask |= SETTING_KILL_SIGNAL;
1213 break;
1214
1215 case ARG_SETTINGS:
1216
1217 /* no → do not read files
1218 * yes → read files, do not override cmdline, trust only subset
1219 * override → read files, override cmdline, trust only subset
1220 * trusted → read files, do not override cmdline, trust all
1221 */
1222
1223 r = parse_boolean(optarg);
1224 if (r < 0) {
1225 if (streq(optarg, "trusted")) {
1226 mask_all_settings = false;
1227 mask_no_settings = false;
1228 arg_settings_trusted = true;
1229
1230 } else if (streq(optarg, "override")) {
1231 mask_all_settings = false;
1232 mask_no_settings = true;
1233 arg_settings_trusted = -1;
1234 } else
1235 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1236 } else if (r > 0) {
1237 /* yes */
1238 mask_all_settings = false;
1239 mask_no_settings = false;
1240 arg_settings_trusted = -1;
1241 } else {
1242 /* no */
1243 mask_all_settings = true;
1244 mask_no_settings = false;
1245 arg_settings_trusted = false;
1246 }
1247
1248 break;
1249
1250 case ARG_CHDIR:
1251 if (!path_is_absolute(optarg))
1252 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1253 "Working directory %s is not an absolute path.", optarg);
1254
1255 r = free_and_strdup(&arg_chdir, optarg);
1256 if (r < 0)
1257 return log_oom();
1258
1259 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1260 break;
1261
1262 case ARG_PIVOT_ROOT:
1263 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1264 if (r < 0)
1265 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1266
1267 arg_settings_mask |= SETTING_PIVOT_ROOT;
1268 break;
1269
1270 case ARG_NOTIFY_READY:
1271 r = parse_boolean(optarg);
1272 if (r < 0)
1273 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1274 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
1275 arg_notify_ready = r;
1276 arg_settings_mask |= SETTING_NOTIFY_READY;
1277 break;
1278
1279 case ARG_ROOT_HASH: {
1280 void *k;
1281 size_t l;
1282
1283 r = unhexmem(optarg, strlen(optarg), &k, &l);
1284 if (r < 0)
1285 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1286 if (l < sizeof(sd_id128_t)) {
1287 free(k);
1288 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128bit long: %s", optarg);
1289 }
1290
1291 free(arg_root_hash);
1292 arg_root_hash = k;
1293 arg_root_hash_size = l;
1294 break;
1295 }
1296
1297 case ARG_SYSTEM_CALL_FILTER: {
1298 bool negative;
1299 const char *items;
1300
1301 negative = optarg[0] == '~';
1302 items = negative ? optarg + 1 : optarg;
1303
1304 for (;;) {
1305 _cleanup_free_ char *word = NULL;
1306
1307 r = extract_first_word(&items, &word, NULL, 0);
1308 if (r == 0)
1309 break;
1310 if (r == -ENOMEM)
1311 return log_oom();
1312 if (r < 0)
1313 return log_error_errno(r, "Failed to parse system call filter: %m");
1314
1315 if (negative)
1316 r = strv_extend(&arg_syscall_blacklist, word);
1317 else
1318 r = strv_extend(&arg_syscall_whitelist, word);
1319 if (r < 0)
1320 return log_oom();
1321 }
1322
1323 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1324 break;
1325 }
1326
1327 case ARG_RLIMIT: {
1328 const char *eq;
1329 _cleanup_free_ char *name = NULL;
1330 int rl;
1331
1332 if (streq(optarg, "help")) {
1333 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1334 return 0;
1335 }
1336
1337 eq = strchr(optarg, '=');
1338 if (!eq)
1339 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1340 "--rlimit= expects an '=' assignment.");
1341
1342 name = strndup(optarg, eq - optarg);
1343 if (!name)
1344 return log_oom();
1345
1346 rl = rlimit_from_string_harder(name);
1347 if (rl < 0)
1348 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1349 "Unknown resource limit: %s", name);
1350
1351 if (!arg_rlimit[rl]) {
1352 arg_rlimit[rl] = new0(struct rlimit, 1);
1353 if (!arg_rlimit[rl])
1354 return log_oom();
1355 }
1356
1357 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1358 if (r < 0)
1359 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1360
1361 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1362 break;
1363 }
1364
1365 case ARG_OOM_SCORE_ADJUST:
1366 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1367 if (r < 0)
1368 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1369
1370 arg_oom_score_adjust_set = true;
1371 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1372 break;
1373
1374 case ARG_CPU_AFFINITY: {
1375 CPUSet cpuset;
1376
1377 r = parse_cpu_set(optarg, &cpuset);
1378 if (r < 0)
1379 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
1380
1381 cpu_set_reset(&arg_cpu_set);
1382 arg_cpu_set = cpuset;
1383 arg_settings_mask |= SETTING_CPU_AFFINITY;
1384 break;
1385 }
1386
1387 case ARG_RESOLV_CONF:
1388 if (streq(optarg, "help")) {
1389 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1390 return 0;
1391 }
1392
1393 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
1394 if (arg_resolv_conf < 0)
1395 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1396 "Failed to parse /etc/resolv.conf mode: %s", optarg);
1397
1398 arg_settings_mask |= SETTING_RESOLV_CONF;
1399 break;
1400
1401 case ARG_TIMEZONE:
1402 if (streq(optarg, "help")) {
1403 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1404 return 0;
1405 }
1406
1407 arg_timezone = timezone_mode_from_string(optarg);
1408 if (arg_timezone < 0)
1409 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1410 "Failed to parse /etc/localtime mode: %s", optarg);
1411
1412 arg_settings_mask |= SETTING_TIMEZONE;
1413 break;
1414
1415 case ARG_CONSOLE:
1416 r = handle_arg_console(optarg);
1417 if (r <= 0)
1418 return r;
1419 break;
1420
1421 case 'P':
1422 case ARG_PIPE:
1423 r = handle_arg_console("pipe");
1424 if (r <= 0)
1425 return r;
1426 break;
1427
1428 case ARG_NO_PAGER:
1429 arg_pager_flags |= PAGER_DISABLE;
1430 break;
1431
1432 case '?':
1433 return -EINVAL;
1434
1435 default:
1436 assert_not_reached("Unhandled option");
1437 }
1438
1439 if (argc > optind) {
1440 strv_free(arg_parameters);
1441 arg_parameters = strv_copy(argv + optind);
1442 if (!arg_parameters)
1443 return log_oom();
1444
1445 arg_settings_mask |= SETTING_START_MODE;
1446 }
1447
1448 if (arg_ephemeral && arg_template && !arg_directory)
1449 /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
1450 * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
1451 * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
1452 * --directory=". */
1453 arg_directory = TAKE_PTR(arg_template);
1454
1455 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0)) & ~minus;
1456
1457 /* Make sure to parse environment before we reset the settings mask below */
1458 r = parse_environment();
1459 if (r < 0)
1460 return r;
1461
1462 /* Load all settings from .nspawn files */
1463 if (mask_no_settings)
1464 arg_settings_mask = 0;
1465
1466 /* Don't load any settings from .nspawn files */
1467 if (mask_all_settings)
1468 arg_settings_mask = _SETTINGS_MASK_ALL;
1469
1470 return 1;
1471 }
1472
1473 static int verify_arguments(void) {
1474 int r;
1475
1476 if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
1477 /* If we are running the stub init in the container, we don't need to look at what the init
1478 * in the container supports, because we are not using it. Let's immediately pick the right
1479 * setting based on the host system configuration.
1480 *
1481 * We only do this, if the user didn't use an environment variable to override the detection.
1482 */
1483
1484 r = cg_all_unified();
1485 if (r < 0)
1486 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
1487 if (r > 0)
1488 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
1489 else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
1490 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
1491 else
1492 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
1493 }
1494
1495 if (arg_userns_mode != USER_NAMESPACE_NO)
1496 arg_mount_settings |= MOUNT_USE_USERNS;
1497
1498 if (arg_private_network)
1499 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1500
1501 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1502 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
1503 arg_register = false;
1504 if (arg_start_mode != START_PID1)
1505 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
1506 }
1507
1508 if (arg_userns_mode == USER_NAMESPACE_PICK)
1509 arg_userns_chown = true;
1510
1511 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1512 arg_kill_signal = SIGRTMIN+3;
1513
1514 if (arg_volatile_mode != VOLATILE_NO) /* Make sure all file systems contained in the image are mounted read-only if we are in volatile mode */
1515 arg_read_only = true;
1516
1517 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
1518 /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
1519 * The latter is not technically a user session, but we don't need to labour the point. */
1520 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
1521
1522 if (arg_directory && arg_image)
1523 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1524
1525 if (arg_template && arg_image)
1526 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
1527
1528 if (arg_template && !(arg_directory || arg_machine))
1529 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
1530
1531 if (arg_ephemeral && arg_template)
1532 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
1533
1534 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
1535 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
1536
1537 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
1538 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
1539
1540 if (arg_userns_chown && arg_read_only)
1541 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1542 "--read-only and --private-users-chown may not be combined.");
1543
1544 /* We don't support --private-users-chown together with any of the volatile modes since we couldn't
1545 * change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a massive
1546 * copy-up (in case of overlay) making the entire exercise pointless. */
1547 if (arg_userns_chown && arg_volatile_mode != VOLATILE_NO)
1548 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-chown may not be combined.");
1549
1550 /* If --network-namespace-path is given with any other network-related option, we need to error out,
1551 * to avoid conflicts between different network options. */
1552 if (arg_network_namespace_path &&
1553 (arg_network_interfaces || arg_network_macvlan ||
1554 arg_network_ipvlan || arg_network_veth_extra ||
1555 arg_network_bridge || arg_network_zone ||
1556 arg_network_veth || arg_private_network))
1557 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
1558
1559 if (arg_network_bridge && arg_network_zone)
1560 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1561 "--network-bridge= and --network-zone= may not be combined.");
1562
1563 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
1564 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
1565
1566 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
1567 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
1568
1569 if (arg_expose_ports && !arg_private_network)
1570 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
1571
1572 #if ! HAVE_LIBIPTC
1573 if (arg_expose_ports)
1574 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--port= is not supported, compiled without libiptc support.");
1575 #endif
1576
1577 r = custom_mount_check_all();
1578 if (r < 0)
1579 return r;
1580
1581 return 0;
1582 }
1583
1584 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1585 assert(p);
1586
1587 if (arg_userns_mode == USER_NAMESPACE_NO)
1588 return 0;
1589
1590 if (uid == UID_INVALID && gid == GID_INVALID)
1591 return 0;
1592
1593 if (uid != UID_INVALID) {
1594 uid += arg_uid_shift;
1595
1596 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1597 return -EOVERFLOW;
1598 }
1599
1600 if (gid != GID_INVALID) {
1601 gid += (gid_t) arg_uid_shift;
1602
1603 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1604 return -EOVERFLOW;
1605 }
1606
1607 if (lchown(p, uid, gid) < 0)
1608 return -errno;
1609
1610 return 0;
1611 }
1612
1613 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1614 const char *q;
1615 int r;
1616
1617 q = prefix_roota(root, path);
1618 r = mkdir_errno_wrapper(q, mode);
1619 if (r == -EEXIST)
1620 return 0;
1621 if (r < 0)
1622 return r;
1623
1624 return userns_lchown(q, uid, gid);
1625 }
1626
1627 static const char *timezone_from_path(const char *path) {
1628 return PATH_STARTSWITH_SET(
1629 path,
1630 "../usr/share/zoneinfo/",
1631 "/usr/share/zoneinfo/");
1632 }
1633
1634 static bool etc_writable(void) {
1635 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1636 }
1637
1638 static int setup_timezone(const char *dest) {
1639 _cleanup_free_ char *p = NULL, *etc = NULL;
1640 const char *where, *check;
1641 TimezoneMode m;
1642 int r;
1643
1644 assert(dest);
1645
1646 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1647 r = readlink_malloc("/etc/localtime", &p);
1648 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
1649 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1650 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
1651 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1652 else if (r < 0) {
1653 log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
1654 /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
1655 * file.
1656 *
1657 * Example:
1658 * ln -s /usr/share/zoneinfo/UTC /etc/localtime
1659 */
1660 return 0;
1661 } else if (arg_timezone == TIMEZONE_AUTO)
1662 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1663 else
1664 m = arg_timezone;
1665 } else
1666 m = arg_timezone;
1667
1668 if (m == TIMEZONE_OFF)
1669 return 0;
1670
1671 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1672 if (r < 0) {
1673 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1674 return 0;
1675 }
1676
1677 where = strjoina(etc, "/localtime");
1678
1679 switch (m) {
1680
1681 case TIMEZONE_DELETE:
1682 if (unlink(where) < 0)
1683 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1684
1685 return 0;
1686
1687 case TIMEZONE_SYMLINK: {
1688 _cleanup_free_ char *q = NULL;
1689 const char *z, *what;
1690
1691 z = timezone_from_path(p);
1692 if (!z) {
1693 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1694 return 0;
1695 }
1696
1697 r = readlink_malloc(where, &q);
1698 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1699 return 0; /* Already pointing to the right place? Then do nothing .. */
1700
1701 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1702 r = chase_symlinks(check, dest, 0, NULL, NULL);
1703 if (r < 0)
1704 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1705 else {
1706 if (unlink(where) < 0 && errno != ENOENT) {
1707 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
1708 errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
1709 return 0;
1710 }
1711
1712 what = strjoina("../usr/share/zoneinfo/", z);
1713 if (symlink(what, where) < 0) {
1714 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1715 errno, "Failed to correct timezone of container, ignoring: %m");
1716 return 0;
1717 }
1718
1719 break;
1720 }
1721
1722 _fallthrough_;
1723 }
1724
1725 case TIMEZONE_BIND: {
1726 _cleanup_free_ char *resolved = NULL;
1727 int found;
1728
1729 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
1730 if (found < 0) {
1731 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1732 return 0;
1733 }
1734
1735 if (found == 0) /* missing? */
1736 (void) touch(resolved);
1737
1738 r = mount_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1739 if (r >= 0)
1740 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1741
1742 _fallthrough_;
1743 }
1744
1745 case TIMEZONE_COPY:
1746 /* If mounting failed, try to copy */
1747 r = copy_file_atomic("/etc/localtime", where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
1748 if (r < 0) {
1749 log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1750 "Failed to copy /etc/localtime to %s, ignoring: %m", where);
1751 return 0;
1752 }
1753
1754 break;
1755
1756 default:
1757 assert_not_reached("unexpected mode");
1758 }
1759
1760 /* Fix permissions of the symlink or file copy we just created */
1761 r = userns_lchown(where, 0, 0);
1762 if (r < 0)
1763 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
1764
1765 return 0;
1766 }
1767
1768 static int have_resolv_conf(const char *path) {
1769 assert(path);
1770
1771 if (access(path, F_OK) < 0) {
1772 if (errno == ENOENT)
1773 return 0;
1774
1775 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1776 }
1777
1778 return 1;
1779 }
1780
1781 static int resolved_listening(void) {
1782 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1783 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
1784 _cleanup_free_ char *dns_stub_listener_mode = NULL;
1785 int r;
1786
1787 /* Check if resolved is listening */
1788
1789 r = sd_bus_open_system(&bus);
1790 if (r < 0)
1791 return log_debug_errno(r, "Failed to open system bus: %m");
1792
1793 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
1794 if (r < 0)
1795 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1796 if (r == 0)
1797 return 0;
1798
1799 r = sd_bus_get_property_string(bus,
1800 "org.freedesktop.resolve1",
1801 "/org/freedesktop/resolve1",
1802 "org.freedesktop.resolve1.Manager",
1803 "DNSStubListener",
1804 &error,
1805 &dns_stub_listener_mode);
1806 if (r < 0)
1807 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
1808
1809 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
1810 }
1811
1812 static int setup_resolv_conf(const char *dest) {
1813 _cleanup_free_ char *etc = NULL;
1814 const char *where, *what;
1815 ResolvConfMode m;
1816 int r;
1817
1818 assert(dest);
1819
1820 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
1821 if (arg_private_network)
1822 m = RESOLV_CONF_OFF;
1823 else if (have_resolv_conf(STATIC_RESOLV_CONF) > 0 && resolved_listening() > 0)
1824 m = etc_writable() ? RESOLV_CONF_COPY_STATIC : RESOLV_CONF_BIND_STATIC;
1825 else if (have_resolv_conf("/etc/resolv.conf") > 0)
1826 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
1827 else
1828 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
1829 } else
1830 m = arg_resolv_conf;
1831
1832 if (m == RESOLV_CONF_OFF)
1833 return 0;
1834
1835 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
1836 if (r < 0) {
1837 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1838 return 0;
1839 }
1840
1841 where = strjoina(etc, "/resolv.conf");
1842
1843 if (m == RESOLV_CONF_DELETE) {
1844 if (unlink(where) < 0)
1845 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1846
1847 return 0;
1848 }
1849
1850 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_COPY_STATIC))
1851 what = STATIC_RESOLV_CONF;
1852 else
1853 what = "/etc/resolv.conf";
1854
1855 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC)) {
1856 _cleanup_free_ char *resolved = NULL;
1857 int found;
1858
1859 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
1860 if (found < 0) {
1861 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1862 return 0;
1863 }
1864
1865 if (found == 0) /* missing? */
1866 (void) touch(resolved);
1867
1868 r = mount_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
1869 if (r >= 0)
1870 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1871 }
1872
1873 /* If that didn't work, let's copy the file */
1874 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, 0, COPY_REFLINK);
1875 if (r < 0) {
1876 /* If the file already exists as symlink, let's suppress the warning, under the assumption that
1877 * resolved or something similar runs inside and the symlink points there.
1878 *
1879 * If the disk image is read-only, there's also no point in complaining.
1880 */
1881 log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC) && IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
1882 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
1883 return 0;
1884 }
1885
1886 r = userns_lchown(where, 0, 0);
1887 if (r < 0)
1888 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
1889
1890 return 0;
1891 }
1892
1893 static int setup_boot_id(void) {
1894 _cleanup_(unlink_and_freep) char *from = NULL;
1895 _cleanup_free_ char *path = NULL;
1896 sd_id128_t rnd = SD_ID128_NULL;
1897 const char *to;
1898 int r;
1899
1900 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
1901
1902 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
1903 if (r < 0)
1904 return log_error_errno(r, "Failed to generate random boot ID path: %m");
1905
1906 r = sd_id128_randomize(&rnd);
1907 if (r < 0)
1908 return log_error_errno(r, "Failed to generate random boot id: %m");
1909
1910 r = id128_write(path, ID128_UUID, rnd, false);
1911 if (r < 0)
1912 return log_error_errno(r, "Failed to write boot id: %m");
1913
1914 from = TAKE_PTR(path);
1915 to = "/proc/sys/kernel/random/boot_id";
1916
1917 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
1918 if (r < 0)
1919 return r;
1920
1921 return mount_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1922 }
1923
1924 static int copy_devnodes(const char *dest) {
1925 static const char devnodes[] =
1926 "null\0"
1927 "zero\0"
1928 "full\0"
1929 "random\0"
1930 "urandom\0"
1931 "tty\0"
1932 "net/tun\0";
1933
1934 _cleanup_umask_ mode_t u;
1935 const char *d;
1936 int r = 0;
1937
1938 assert(dest);
1939
1940 u = umask(0000);
1941
1942 /* Create /dev/net, so that we can create /dev/net/tun in it */
1943 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1944 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1945
1946 NULSTR_FOREACH(d, devnodes) {
1947 _cleanup_free_ char *from = NULL, *to = NULL;
1948 struct stat st;
1949
1950 from = path_join("/dev/", d);
1951 if (!from)
1952 return log_oom();
1953
1954 to = path_join(dest, from);
1955 if (!to)
1956 return log_oom();
1957
1958 if (stat(from, &st) < 0) {
1959
1960 if (errno != ENOENT)
1961 return log_error_errno(errno, "Failed to stat %s: %m", from);
1962
1963 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
1964 return log_error_errno(SYNTHETIC_ERRNO(EIO),
1965 "%s is not a char or block device, cannot copy.", from);
1966 else {
1967 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
1968
1969 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1970 /* Explicitly warn the user when /dev is already populated. */
1971 if (errno == EEXIST)
1972 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);
1973 if (errno != EPERM)
1974 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1975
1976 /* Some systems abusively restrict mknod but allow bind mounts. */
1977 r = touch(to);
1978 if (r < 0)
1979 return log_error_errno(r, "touch (%s) failed: %m", to);
1980 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1981 if (r < 0)
1982 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
1983 }
1984
1985 r = userns_lchown(to, 0, 0);
1986 if (r < 0)
1987 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1988
1989 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
1990 if (!dn)
1991 return log_oom();
1992
1993 r = userns_mkdir(dest, dn, 0755, 0, 0);
1994 if (r < 0)
1995 return log_error_errno(r, "Failed to create '%s': %m", dn);
1996
1997 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
1998 return log_oom();
1999
2000 prefixed = path_join(dest, sl);
2001 if (!prefixed)
2002 return log_oom();
2003
2004 t = path_join("..", d);
2005 if (!t)
2006 return log_oom();
2007
2008 if (symlink(t, prefixed) < 0)
2009 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
2010 }
2011 }
2012
2013 return r;
2014 }
2015
2016 static int make_extra_nodes(const char *dest) {
2017 _cleanup_umask_ mode_t u;
2018 size_t i;
2019 int r;
2020
2021 u = umask(0000);
2022
2023 for (i = 0; i < arg_n_extra_nodes; i++) {
2024 _cleanup_free_ char *path = NULL;
2025 DeviceNode *n = arg_extra_nodes + i;
2026
2027 path = path_join(dest, n->path);
2028 if (!path)
2029 return log_oom();
2030
2031 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
2032 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
2033
2034 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
2035 if (r < 0)
2036 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
2037 }
2038
2039 return 0;
2040 }
2041
2042 static int setup_pts(const char *dest) {
2043 _cleanup_free_ char *options = NULL;
2044 const char *p;
2045 int r;
2046
2047 #if HAVE_SELINUX
2048 if (arg_selinux_apifs_context)
2049 (void) asprintf(&options,
2050 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
2051 arg_uid_shift + TTY_GID,
2052 arg_selinux_apifs_context);
2053 else
2054 #endif
2055 (void) asprintf(&options,
2056 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
2057 arg_uid_shift + TTY_GID);
2058
2059 if (!options)
2060 return log_oom();
2061
2062 /* Mount /dev/pts itself */
2063 p = prefix_roota(dest, "/dev/pts");
2064 r = mkdir_errno_wrapper(p, 0755);
2065 if (r < 0)
2066 return log_error_errno(r, "Failed to create /dev/pts: %m");
2067
2068 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2069 if (r < 0)
2070 return r;
2071 r = userns_lchown(p, 0, 0);
2072 if (r < 0)
2073 return log_error_errno(r, "Failed to chown /dev/pts: %m");
2074
2075 /* Create /dev/ptmx symlink */
2076 p = prefix_roota(dest, "/dev/ptmx");
2077 if (symlink("pts/ptmx", p) < 0)
2078 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
2079 r = userns_lchown(p, 0, 0);
2080 if (r < 0)
2081 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
2082
2083 /* And fix /dev/pts/ptmx ownership */
2084 p = prefix_roota(dest, "/dev/pts/ptmx");
2085 r = userns_lchown(p, 0, 0);
2086 if (r < 0)
2087 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
2088
2089 return 0;
2090 }
2091
2092 static int setup_stdio_as_dev_console(void) {
2093 int terminal;
2094 int r;
2095
2096 terminal = open_terminal("/dev/console", O_RDWR);
2097 if (terminal < 0)
2098 return log_error_errno(terminal, "Failed to open console: %m");
2099
2100 /* Make sure we can continue logging to the original stderr, even if
2101 * stderr points elsewhere now */
2102 r = log_dup_console();
2103 if (r < 0)
2104 return log_error_errno(r, "Failed to duplicate stderr: %m");
2105
2106 /* invalidates 'terminal' on success and failure */
2107 r = rearrange_stdio(terminal, terminal, terminal);
2108 if (r < 0)
2109 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2110
2111 return 0;
2112 }
2113
2114 static int setup_dev_console(const char *console) {
2115 _cleanup_free_ char *p = NULL;
2116 int r;
2117
2118 /* Create /dev/console symlink */
2119 r = path_make_relative("/dev", console, &p);
2120 if (r < 0)
2121 return log_error_errno(r, "Failed to create relative path: %m");
2122
2123 if (symlink(p, "/dev/console") < 0)
2124 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
2125
2126 return 0;
2127 }
2128
2129 static int setup_keyring(void) {
2130 key_serial_t keyring;
2131
2132 /* Allocate a new session keyring for the container. This makes sure the keyring of the session systemd-nspawn
2133 * was invoked from doesn't leak into the container. Note that by default we block keyctl() and request_key()
2134 * anyway via seccomp so doing this operation isn't strictly necessary, but in case people explicitly whitelist
2135 * these system calls let's make sure we don't leak anything into the container. */
2136
2137 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2138 if (keyring == -1) {
2139 if (errno == ENOSYS)
2140 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2141 else if (IN_SET(errno, EACCES, EPERM))
2142 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2143 else
2144 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2145 }
2146
2147 return 0;
2148 }
2149
2150 static int setup_kmsg(int kmsg_socket) {
2151 _cleanup_(unlink_and_freep) char *from = NULL;
2152 _cleanup_free_ char *fifo = NULL;
2153 _cleanup_close_ int fd = -1;
2154 _cleanup_umask_ mode_t u;
2155 int r;
2156
2157 assert(kmsg_socket >= 0);
2158
2159 u = umask(0000);
2160
2161 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
2162 * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
2163 * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
2164 * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
2165
2166 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
2167 if (r < 0)
2168 return log_error_errno(r, "Failed to generate kmsg path: %m");
2169
2170 if (mkfifo(fifo, 0600) < 0)
2171 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2172
2173 from = TAKE_PTR(fifo);
2174
2175 r = mount_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
2176 if (r < 0)
2177 return r;
2178
2179 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
2180 if (fd < 0)
2181 return log_error_errno(errno, "Failed to open fifo: %m");
2182
2183 /* Store away the fd in the socket, so that it stays open as long as we run the child */
2184 r = send_one_fd(kmsg_socket, fd, 0);
2185 if (r < 0)
2186 return log_error_errno(r, "Failed to send FIFO fd: %m");
2187
2188 return 0;
2189 }
2190
2191 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2192 union in_addr_union *exposed = userdata;
2193
2194 assert(rtnl);
2195 assert(m);
2196 assert(exposed);
2197
2198 expose_port_execute(rtnl, arg_expose_ports, exposed);
2199 return 0;
2200 }
2201
2202 static int setup_hostname(void) {
2203 int r;
2204
2205 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
2206 return 0;
2207
2208 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2209 if (r < 0)
2210 return log_error_errno(r, "Failed to set hostname: %m");
2211
2212 return 0;
2213 }
2214
2215 static int setup_journal(const char *directory) {
2216 _cleanup_free_ char *d = NULL;
2217 const char *dirname, *p, *q;
2218 sd_id128_t this_id;
2219 char id[33];
2220 bool try;
2221 int r;
2222
2223 /* Don't link journals in ephemeral mode */
2224 if (arg_ephemeral)
2225 return 0;
2226
2227 if (arg_link_journal == LINK_NO)
2228 return 0;
2229
2230 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2231
2232 r = sd_id128_get_machine(&this_id);
2233 if (r < 0)
2234 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2235
2236 if (sd_id128_equal(arg_uuid, this_id)) {
2237 log_full(try ? LOG_WARNING : LOG_ERR,
2238 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
2239 if (try)
2240 return 0;
2241 return -EEXIST;
2242 }
2243
2244 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2245 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2246 if (r < 0) {
2247 bool ignore = r == -EROFS && try;
2248 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2249 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2250 return ignore ? 0 : r;
2251 }
2252 }
2253
2254 (void) sd_id128_to_string(arg_uuid, id);
2255
2256 p = strjoina("/var/log/journal/", id);
2257 q = prefix_roota(directory, p);
2258
2259 if (path_is_mount_point(p, NULL, 0) > 0) {
2260 if (try)
2261 return 0;
2262
2263 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2264 "%s: already a mount point, refusing to use for journal", p);
2265 }
2266
2267 if (path_is_mount_point(q, NULL, 0) > 0) {
2268 if (try)
2269 return 0;
2270
2271 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2272 "%s: already a mount point, refusing to use for journal", q);
2273 }
2274
2275 r = readlink_and_make_absolute(p, &d);
2276 if (r >= 0) {
2277 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
2278 path_equal(d, q)) {
2279
2280 r = userns_mkdir(directory, p, 0755, 0, 0);
2281 if (r < 0)
2282 log_warning_errno(r, "Failed to create directory %s: %m", q);
2283 return 0;
2284 }
2285
2286 if (unlink(p) < 0)
2287 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2288 } else if (r == -EINVAL) {
2289
2290 if (arg_link_journal == LINK_GUEST &&
2291 rmdir(p) < 0) {
2292
2293 if (errno == ENOTDIR) {
2294 log_error("%s already exists and is neither a symlink nor a directory", p);
2295 return r;
2296 } else
2297 return log_error_errno(errno, "Failed to remove %s: %m", p);
2298 }
2299 } else if (r != -ENOENT)
2300 return log_error_errno(r, "readlink(%s) failed: %m", p);
2301
2302 if (arg_link_journal == LINK_GUEST) {
2303
2304 if (symlink(q, p) < 0) {
2305 if (try) {
2306 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2307 return 0;
2308 } else
2309 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2310 }
2311
2312 r = userns_mkdir(directory, p, 0755, 0, 0);
2313 if (r < 0)
2314 log_warning_errno(r, "Failed to create directory %s: %m", q);
2315 return 0;
2316 }
2317
2318 if (arg_link_journal == LINK_HOST) {
2319 /* don't create parents here — if the host doesn't have
2320 * permanent journal set up, don't force it here */
2321
2322 r = mkdir_errno_wrapper(p, 0755);
2323 if (r < 0 && r != -EEXIST) {
2324 if (try) {
2325 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
2326 return 0;
2327 } else
2328 return log_error_errno(r, "Failed to create %s: %m", p);
2329 }
2330
2331 } else if (access(p, F_OK) < 0)
2332 return 0;
2333
2334 if (dir_is_empty(q) == 0)
2335 log_warning("%s is not empty, proceeding anyway.", q);
2336
2337 r = userns_mkdir(directory, p, 0755, 0, 0);
2338 if (r < 0)
2339 return log_error_errno(r, "Failed to create %s: %m", q);
2340
2341 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2342 if (r < 0)
2343 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2344
2345 return 0;
2346 }
2347
2348 static int drop_capabilities(uid_t uid) {
2349 CapabilityQuintet q;
2350
2351 /* Let's initialize all five capability sets to something valid. If the quintet was configured via
2352 * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
2353 * arg_caps_retain. */
2354
2355 if (capability_quintet_is_set(&arg_full_capabilities)) {
2356 q = arg_full_capabilities;
2357
2358 if (q.bounding == (uint64_t) -1)
2359 q.bounding = uid == 0 ? arg_caps_retain : 0;
2360
2361 if (q.effective == (uint64_t) -1)
2362 q.effective = uid == 0 ? q.bounding : 0;
2363
2364 if (q.inheritable == (uint64_t) -1)
2365 q.inheritable = uid == 0 ? q.bounding : 0;
2366
2367 if (q.permitted == (uint64_t) -1)
2368 q.permitted = uid == 0 ? q.bounding : 0;
2369
2370 if (q.ambient == (uint64_t) -1 && ambient_capabilities_supported())
2371 q.ambient = 0;
2372
2373 if (capability_quintet_mangle(&q))
2374 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2375
2376 } else {
2377 q = (CapabilityQuintet) {
2378 .bounding = arg_caps_retain,
2379 .effective = uid == 0 ? arg_caps_retain : 0,
2380 .inheritable = uid == 0 ? arg_caps_retain : 0,
2381 .permitted = uid == 0 ? arg_caps_retain : 0,
2382 .ambient = ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2383 };
2384
2385 /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
2386 * in order to maintain the same behavior as systemd < 242. */
2387 if (capability_quintet_mangle(&q))
2388 log_warning("Some capabilities will not be set because they are not in the current bounding set.");
2389
2390 }
2391
2392 return capability_quintet_enforce(&q);
2393 }
2394
2395 static int reset_audit_loginuid(void) {
2396 _cleanup_free_ char *p = NULL;
2397 int r;
2398
2399 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
2400 return 0;
2401
2402 r = read_one_line_file("/proc/self/loginuid", &p);
2403 if (r == -ENOENT)
2404 return 0;
2405 if (r < 0)
2406 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2407
2408 /* Already reset? */
2409 if (streq(p, "4294967295"))
2410 return 0;
2411
2412 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
2413 if (r < 0) {
2414 log_error_errno(r,
2415 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2416 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2417 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2418 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2419 "using systemd-nspawn. Sleeping for 5s... (%m)");
2420
2421 sleep(5);
2422 }
2423
2424 return 0;
2425 }
2426
2427 static int setup_propagate(const char *root) {
2428 const char *p, *q;
2429 int r;
2430
2431 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2432 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2433 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2434 (void) mkdir_p(p, 0600);
2435
2436 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
2437 if (r < 0)
2438 return log_error_errno(r, "Failed to create /run/systemd: %m");
2439
2440 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
2441 if (r < 0)
2442 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
2443
2444 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2445 if (r < 0)
2446 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
2447
2448 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
2449 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2450 if (r < 0)
2451 return r;
2452
2453 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2454 if (r < 0)
2455 return r;
2456
2457 /* machined will MS_MOVE into that directory, and that's only
2458 * supported for non-shared mounts. */
2459 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
2460 }
2461
2462 static int setup_machine_id(const char *directory) {
2463 const char *etc_machine_id;
2464 sd_id128_t id;
2465 int r;
2466
2467 /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
2468 * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
2469 * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
2470 * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
2471 * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
2472 * container behaves nicely). */
2473
2474 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2475
2476 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
2477 if (r < 0) {
2478 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2479 return log_error_errno(r, "Failed to read machine ID from container image: %m");
2480
2481 if (sd_id128_is_null(arg_uuid)) {
2482 r = sd_id128_randomize(&arg_uuid);
2483 if (r < 0)
2484 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2485 }
2486 } else {
2487 if (sd_id128_is_null(id))
2488 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2489 "Machine ID in container image is zero, refusing.");
2490
2491 arg_uuid = id;
2492 }
2493
2494 return 0;
2495 }
2496
2497 static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2498 int r;
2499
2500 assert(directory);
2501
2502 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
2503 return 0;
2504
2505 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2506 if (r == -EOPNOTSUPP)
2507 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.");
2508 if (r == -EBADE)
2509 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2510 if (r < 0)
2511 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2512 if (r == 0)
2513 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2514 else
2515 log_debug("Patched directory tree to match UID/GID range.");
2516
2517 return r;
2518 }
2519
2520 /*
2521 * Return values:
2522 * < 0 : wait_for_terminate() failed to get the state of the
2523 * container, the container was terminated by a signal, or
2524 * failed for an unknown reason. No change is made to the
2525 * container argument.
2526 * > 0 : The program executed in the container terminated with an
2527 * error. The exit code of the program executed in the
2528 * container is returned. The container argument has been set
2529 * to CONTAINER_TERMINATED.
2530 * 0 : The container is being rebooted, has been shut down or exited
2531 * successfully. The container argument has been set to either
2532 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
2533 *
2534 * That is, success is indicated by a return value of zero, and an
2535 * error is indicated by a non-zero value.
2536 */
2537 static int wait_for_container(pid_t pid, ContainerStatus *container) {
2538 siginfo_t status;
2539 int r;
2540
2541 r = wait_for_terminate(pid, &status);
2542 if (r < 0)
2543 return log_warning_errno(r, "Failed to wait for container: %m");
2544
2545 switch (status.si_code) {
2546
2547 case CLD_EXITED:
2548 if (status.si_status == 0)
2549 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
2550 else
2551 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
2552
2553 *container = CONTAINER_TERMINATED;
2554 return status.si_status;
2555
2556 case CLD_KILLED:
2557 if (status.si_status == SIGINT) {
2558 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
2559 *container = CONTAINER_TERMINATED;
2560 return 0;
2561
2562 } else if (status.si_status == SIGHUP) {
2563 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
2564 *container = CONTAINER_REBOOTED;
2565 return 0;
2566 }
2567
2568 _fallthrough_;
2569 case CLD_DUMPED:
2570 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2571 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
2572
2573 default:
2574 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2575 "Container %s failed due to unknown reason.", arg_machine);
2576 }
2577 }
2578
2579 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2580 pid_t pid;
2581
2582 pid = PTR_TO_PID(userdata);
2583 if (pid > 0) {
2584 if (kill(pid, arg_kill_signal) >= 0) {
2585 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2586 sd_event_source_set_userdata(s, NULL);
2587 return 0;
2588 }
2589 }
2590
2591 sd_event_exit(sd_event_source_get_event(s), 0);
2592 return 0;
2593 }
2594
2595 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
2596 pid_t pid;
2597
2598 assert(s);
2599 assert(ssi);
2600
2601 pid = PTR_TO_PID(userdata);
2602
2603 for (;;) {
2604 siginfo_t si = {};
2605
2606 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2607 return log_error_errno(errno, "Failed to waitid(): %m");
2608 if (si.si_pid == 0) /* No pending children. */
2609 break;
2610 if (si.si_pid == pid) {
2611 /* The main process we care for has exited. Return from
2612 * signal handler but leave the zombie. */
2613 sd_event_exit(sd_event_source_get_event(s), 0);
2614 break;
2615 }
2616
2617 /* Reap all other children. */
2618 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2619 }
2620
2621 return 0;
2622 }
2623
2624 static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2625 pid_t pid;
2626
2627 assert(m);
2628
2629 pid = PTR_TO_PID(userdata);
2630
2631 if (arg_kill_signal > 0) {
2632 log_info("Container termination requested. Attempting to halt container.");
2633 (void) kill(pid, arg_kill_signal);
2634 } else {
2635 log_info("Container termination requested. Exiting.");
2636 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2637 }
2638
2639 return 0;
2640 }
2641
2642 static int determine_names(void) {
2643 int r;
2644
2645 if (arg_template && !arg_directory && arg_machine) {
2646
2647 /* If --template= was specified then we should not
2648 * search for a machine, but instead create a new one
2649 * in /var/lib/machine. */
2650
2651 arg_directory = path_join("/var/lib/machines", arg_machine);
2652 if (!arg_directory)
2653 return log_oom();
2654 }
2655
2656 if (!arg_image && !arg_directory) {
2657 if (arg_machine) {
2658 _cleanup_(image_unrefp) Image *i = NULL;
2659
2660 r = image_find(IMAGE_MACHINE, arg_machine, &i);
2661 if (r == -ENOENT)
2662 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
2663 if (r < 0)
2664 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
2665
2666 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
2667 r = free_and_strdup(&arg_image, i->path);
2668 else
2669 r = free_and_strdup(&arg_directory, i->path);
2670 if (r < 0)
2671 return log_oom();
2672
2673 if (!arg_ephemeral)
2674 arg_read_only = arg_read_only || i->read_only;
2675 } else {
2676 r = safe_getcwd(&arg_directory);
2677 if (r < 0)
2678 return log_error_errno(r, "Failed to determine current directory: %m");
2679 }
2680
2681 if (!arg_directory && !arg_image)
2682 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
2683 }
2684
2685 if (!arg_machine) {
2686 if (arg_directory && path_equal(arg_directory, "/"))
2687 arg_machine = gethostname_malloc();
2688 else {
2689 if (arg_image) {
2690 char *e;
2691
2692 arg_machine = strdup(basename(arg_image));
2693
2694 /* Truncate suffix if there is one */
2695 e = endswith(arg_machine, ".raw");
2696 if (e)
2697 *e = 0;
2698 } else
2699 arg_machine = strdup(basename(arg_directory));
2700 }
2701 if (!arg_machine)
2702 return log_oom();
2703
2704 hostname_cleanup(arg_machine);
2705 if (!machine_name_is_valid(arg_machine))
2706 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
2707
2708 if (arg_ephemeral) {
2709 char *b;
2710
2711 /* Add a random suffix when this is an
2712 * ephemeral machine, so that we can run many
2713 * instances at once without manually having
2714 * to specify -M each time. */
2715
2716 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2717 return log_oom();
2718
2719 free(arg_machine);
2720 arg_machine = b;
2721 }
2722 }
2723
2724 return 0;
2725 }
2726
2727 static int chase_symlinks_and_update(char **p, unsigned flags) {
2728 char *chased;
2729 int r;
2730
2731 assert(p);
2732
2733 if (!*p)
2734 return 0;
2735
2736 r = chase_symlinks(*p, NULL, flags, &chased, NULL);
2737 if (r < 0)
2738 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2739
2740 return free_and_replace(*p, chased);
2741 }
2742
2743 static int determine_uid_shift(const char *directory) {
2744 int r;
2745
2746 if (arg_userns_mode == USER_NAMESPACE_NO) {
2747 arg_uid_shift = 0;
2748 return 0;
2749 }
2750
2751 if (arg_uid_shift == UID_INVALID) {
2752 struct stat st;
2753
2754 r = stat(directory, &st);
2755 if (r < 0)
2756 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
2757
2758 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2759
2760 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
2761 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2762 "UID and GID base of %s don't match.", directory);
2763
2764 arg_uid_range = UINT32_C(0x10000);
2765 }
2766
2767 if (arg_uid_shift > (uid_t) -1 - arg_uid_range)
2768 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2769 "UID base too high for UID range.");
2770
2771 return 0;
2772 }
2773
2774 static unsigned long effective_clone_ns_flags(void) {
2775 unsigned long flags = arg_clone_ns_flags;
2776
2777 if (arg_private_network)
2778 flags |= CLONE_NEWNET;
2779 if (arg_use_cgns)
2780 flags |= CLONE_NEWCGROUP;
2781 if (arg_userns_mode != USER_NAMESPACE_NO)
2782 flags |= CLONE_NEWUSER;
2783
2784 return flags;
2785 }
2786
2787 static int patch_sysctl(void) {
2788
2789 /* This table is inspired by runc's sysctl() function */
2790 static const struct {
2791 const char *key;
2792 bool prefix;
2793 unsigned long clone_flags;
2794 } safe_sysctl[] = {
2795 { "kernel.hostname", false, CLONE_NEWUTS },
2796 { "kernel.domainname", false, CLONE_NEWUTS },
2797 { "kernel.msgmax", false, CLONE_NEWIPC },
2798 { "kernel.msgmnb", false, CLONE_NEWIPC },
2799 { "kernel.msgmni", false, CLONE_NEWIPC },
2800 { "kernel.sem", false, CLONE_NEWIPC },
2801 { "kernel.shmall", false, CLONE_NEWIPC },
2802 { "kernel.shmmax", false, CLONE_NEWIPC },
2803 { "kernel.shmmni", false, CLONE_NEWIPC },
2804 { "fs.mqueue.", true, CLONE_NEWIPC },
2805 { "net.", true, CLONE_NEWNET },
2806 };
2807
2808 unsigned long flags;
2809 char **k, **v;
2810 int r;
2811
2812 flags = effective_clone_ns_flags();
2813
2814 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
2815 bool good = false;
2816 size_t i;
2817
2818 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
2819
2820 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
2821 continue;
2822
2823 if (safe_sysctl[i].prefix)
2824 good = startswith(*k, safe_sysctl[i].key);
2825 else
2826 good = streq(*k, safe_sysctl[i].key);
2827
2828 if (good)
2829 break;
2830 }
2831
2832 if (!good)
2833 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
2834
2835 r = sysctl_write(*k, *v);
2836 if (r < 0)
2837 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
2838 }
2839
2840 return 0;
2841 }
2842
2843 static int inner_child(
2844 Barrier *barrier,
2845 const char *directory,
2846 bool secondary,
2847 int kmsg_socket,
2848 int rtnl_socket,
2849 int master_pty_socket,
2850 FDSet *fds) {
2851
2852 _cleanup_free_ char *home = NULL;
2853 char as_uuid[37];
2854 size_t n_env = 1;
2855 const char *envp[] = {
2856 "PATH=" DEFAULT_PATH_COMPAT,
2857 NULL, /* container */
2858 NULL, /* TERM */
2859 NULL, /* HOME */
2860 NULL, /* USER */
2861 NULL, /* LOGNAME */
2862 NULL, /* container_uuid */
2863 NULL, /* LISTEN_FDS */
2864 NULL, /* LISTEN_PID */
2865 NULL, /* NOTIFY_SOCKET */
2866 NULL
2867 };
2868 const char *exec_target;
2869 _cleanup_strv_free_ char **env_use = NULL;
2870 int r, which_failed;
2871
2872 /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
2873 * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
2874 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
2875 * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
2876 * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
2877 * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
2878 * namespace.
2879 *
2880 * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
2881 * unshare(). See below. */
2882
2883 assert(barrier);
2884 assert(directory);
2885 assert(kmsg_socket >= 0);
2886
2887 log_debug("Inner child is initializing.");
2888
2889 if (arg_userns_mode != USER_NAMESPACE_NO) {
2890 /* Tell the parent, that it now can write the UID map. */
2891 (void) barrier_place(barrier); /* #1 */
2892
2893 /* Wait until the parent wrote the UID map */
2894 if (!barrier_place_and_sync(barrier)) /* #2 */
2895 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2896 "Parent died too early");
2897 }
2898
2899 r = reset_uid_gid();
2900 if (r < 0)
2901 return log_error_errno(r, "Couldn't become new root: %m");
2902
2903 r = mount_all(NULL,
2904 arg_mount_settings | MOUNT_IN_USERNS,
2905 arg_uid_shift,
2906 arg_selinux_apifs_context);
2907 if (r < 0)
2908 return r;
2909
2910 if (!arg_network_namespace_path && arg_private_network) {
2911 r = unshare(CLONE_NEWNET);
2912 if (r < 0)
2913 return log_error_errno(errno, "Failed to unshare network namespace: %m");
2914
2915 /* Tell the parent that it can setup network interfaces. */
2916 (void) barrier_place(barrier); /* #3 */
2917 }
2918
2919 r = mount_sysfs(NULL, arg_mount_settings);
2920 if (r < 0)
2921 return r;
2922
2923 /* Wait until we are cgroup-ified, so that we
2924 * can mount the right cgroup path writable */
2925 if (!barrier_place_and_sync(barrier)) /* #4 */
2926 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2927 "Parent died too early");
2928
2929 if (arg_use_cgns) {
2930 r = unshare(CLONE_NEWCGROUP);
2931 if (r < 0)
2932 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
2933 r = mount_cgroups(
2934 "",
2935 arg_unified_cgroup_hierarchy,
2936 arg_userns_mode != USER_NAMESPACE_NO,
2937 arg_uid_shift,
2938 arg_uid_range,
2939 arg_selinux_apifs_context,
2940 true);
2941 if (r < 0)
2942 return r;
2943 } else {
2944 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2945 if (r < 0)
2946 return r;
2947 }
2948
2949 r = setup_boot_id();
2950 if (r < 0)
2951 return r;
2952
2953 r = setup_kmsg(kmsg_socket);
2954 if (r < 0)
2955 return r;
2956 kmsg_socket = safe_close(kmsg_socket);
2957
2958 r = mount_custom(
2959 "/",
2960 arg_custom_mounts,
2961 arg_n_custom_mounts,
2962 false,
2963 0,
2964 0,
2965 arg_selinux_apifs_context,
2966 true);
2967 if (r < 0)
2968 return r;
2969
2970 if (setsid() < 0)
2971 return log_error_errno(errno, "setsid() failed: %m");
2972
2973 if (arg_private_network)
2974 loopback_setup();
2975
2976 if (arg_expose_ports) {
2977 r = expose_port_send_rtnl(rtnl_socket);
2978 if (r < 0)
2979 return r;
2980 rtnl_socket = safe_close(rtnl_socket);
2981 }
2982
2983 if (arg_console_mode != CONSOLE_PIPE) {
2984 _cleanup_close_ int master = -1;
2985 _cleanup_free_ char *console = NULL;
2986
2987 /* Allocate a pty and make it available as /dev/console. */
2988 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
2989 if (master < 0)
2990 return log_error_errno(master, "Failed to allocate a pty: %m");
2991
2992 r = setup_dev_console(console);
2993 if (r < 0)
2994 return log_error_errno(r, "Failed to setup /dev/console: %m");
2995
2996 r = send_one_fd(master_pty_socket, master, 0);
2997 if (r < 0)
2998 return log_error_errno(r, "Failed to send master fd: %m");
2999 master_pty_socket = safe_close(master_pty_socket);
3000
3001 r = setup_stdio_as_dev_console();
3002 if (r < 0)
3003 return r;
3004 }
3005
3006 r = patch_sysctl();
3007 if (r < 0)
3008 return r;
3009
3010 if (arg_oom_score_adjust_set) {
3011 r = set_oom_score_adjust(arg_oom_score_adjust);
3012 if (r < 0)
3013 return log_error_errno(r, "Failed to adjust OOM score: %m");
3014 }
3015
3016 if (arg_cpu_set.set)
3017 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
3018 return log_error_errno(errno, "Failed to set CPU affinity: %m");
3019
3020 (void) setup_hostname();
3021
3022 if (arg_personality != PERSONALITY_INVALID) {
3023 r = safe_personality(arg_personality);
3024 if (r < 0)
3025 return log_error_errno(r, "personality() failed: %m");
3026 } else if (secondary) {
3027 r = safe_personality(PER_LINUX32);
3028 if (r < 0)
3029 return log_error_errno(r, "personality() failed: %m");
3030 }
3031
3032 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
3033 if (r < 0)
3034 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
3035
3036 #if HAVE_SECCOMP
3037 if (arg_seccomp) {
3038
3039 if (is_seccomp_available()) {
3040
3041 r = seccomp_load(arg_seccomp);
3042 if (ERRNO_IS_SECCOMP_FATAL(r))
3043 return log_error_errno(r, "Failed to install seccomp filter: %m");
3044 if (r < 0)
3045 log_debug_errno(r, "Failed to install seccomp filter: %m");
3046 }
3047 } else
3048 #endif
3049 {
3050 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
3051 if (r < 0)
3052 return r;
3053 }
3054
3055 #if HAVE_SELINUX
3056 if (arg_selinux_context)
3057 if (setexeccon(arg_selinux_context) < 0)
3058 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3059 #endif
3060
3061 /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
3062 * if we need to later on. */
3063 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3064 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3065
3066 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3067 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids);
3068 else
3069 r = change_uid_gid(arg_user, &home);
3070 if (r < 0)
3071 return r;
3072
3073 r = drop_capabilities(getuid());
3074 if (r < 0)
3075 return log_error_errno(r, "Dropping capabilities failed: %m");
3076
3077 if (arg_no_new_privileges)
3078 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3079 return log_error_errno(errno, "Failed to disable new privileges: %m");
3080
3081 /* LXC sets container=lxc, so follow the scheme here */
3082 envp[n_env++] = strjoina("container=", arg_container_service_name);
3083
3084 envp[n_env] = strv_find_prefix(environ, "TERM=");
3085 if (envp[n_env])
3086 n_env++;
3087
3088 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3089 if (asprintf((char**)(envp + n_env++), "HOME=%s", home ?: "/root") < 0)
3090 return log_oom();
3091
3092 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3093 if (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ?: "root") < 0 ||
3094 asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3095 return log_oom();
3096
3097 assert(!sd_id128_is_null(arg_uuid));
3098
3099 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
3100 return log_oom();
3101
3102 if (fdset_size(fds) > 0) {
3103 r = fdset_cloexec(fds, false);
3104 if (r < 0)
3105 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3106
3107 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3108 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3109 return log_oom();
3110 }
3111 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3112 return log_oom();
3113
3114 env_use = strv_env_merge(2, envp, arg_setenv);
3115 if (!env_use)
3116 return log_oom();
3117
3118 /* Let the parent know that we are ready and
3119 * wait until the parent is ready with the
3120 * setup, too... */
3121 if (!barrier_place_and_sync(barrier)) /* #5 */
3122 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3123 "Parent died too early");
3124
3125 if (arg_chdir)
3126 if (chdir(arg_chdir) < 0)
3127 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3128
3129 if (arg_start_mode == START_PID2) {
3130 r = stub_pid1(arg_uuid);
3131 if (r < 0)
3132 return r;
3133 }
3134
3135 log_debug("Inner child completed, invoking payload.");
3136
3137 /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
3138 * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
3139 * it again. Note that the other fds closed here are at least the locking and barrier fds. */
3140 log_close();
3141 log_set_open_when_needed(true);
3142
3143 (void) fdset_close_others(fds);
3144
3145 if (arg_start_mode == START_BOOT) {
3146 char **a;
3147 size_t m;
3148
3149 /* Automatically search for the init system */
3150
3151 m = strv_length(arg_parameters);
3152 a = newa(char*, m + 2);
3153 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3154 a[1 + m] = NULL;
3155
3156 a[0] = (char*) "/usr/lib/systemd/systemd";
3157 execve(a[0], a, env_use);
3158
3159 a[0] = (char*) "/lib/systemd/systemd";
3160 execve(a[0], a, env_use);
3161
3162 a[0] = (char*) "/sbin/init";
3163 execve(a[0], a, env_use);
3164
3165 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
3166 } else if (!strv_isempty(arg_parameters)) {
3167 const char *dollar_path;
3168
3169 exec_target = arg_parameters[0];
3170
3171 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3172 * binary. */
3173 dollar_path = strv_env_get(env_use, "PATH");
3174 if (dollar_path) {
3175 if (putenv((char*) dollar_path) != 0)
3176 return log_error_errno(errno, "Failed to update $PATH: %m");
3177 }
3178
3179 execvpe(arg_parameters[0], arg_parameters, env_use);
3180 } else {
3181 if (!arg_chdir)
3182 /* If we cannot change the directory, we'll end up in /, that is expected. */
3183 (void) chdir(home ?: "/root");
3184
3185 execle("/bin/bash", "-bash", NULL, env_use);
3186 execle("/bin/sh", "-sh", NULL, env_use);
3187
3188 exec_target = "/bin/bash, /bin/sh";
3189 }
3190
3191 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
3192 }
3193
3194 static int setup_sd_notify_child(void) {
3195 _cleanup_close_ int fd = -1;
3196 union sockaddr_union sa = {
3197 .un.sun_family = AF_UNIX,
3198 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
3199 };
3200 int r;
3201
3202 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3203 if (fd < 0)
3204 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3205
3206 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
3207 (void) sockaddr_un_unlink(&sa.un);
3208
3209 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
3210 if (r < 0)
3211 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
3212
3213 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
3214 if (r < 0)
3215 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
3216
3217 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
3218 if (r < 0)
3219 return log_error_errno(r, "SO_PASSCRED failed: %m");
3220
3221 return TAKE_FD(fd);
3222 }
3223
3224 static int outer_child(
3225 Barrier *barrier,
3226 const char *directory,
3227 DissectedImage *dissected_image,
3228 bool secondary,
3229 int pid_socket,
3230 int uuid_socket,
3231 int notify_socket,
3232 int kmsg_socket,
3233 int rtnl_socket,
3234 int uid_shift_socket,
3235 int master_pty_socket,
3236 int unified_cgroup_hierarchy_socket,
3237 FDSet *fds,
3238 int netns_fd) {
3239
3240 _cleanup_close_ int fd = -1;
3241 pid_t pid;
3242 ssize_t l;
3243 int r;
3244
3245 /* This is the "outer" child process, i.e the one forked off by the container manager itself. It already has
3246 * its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in the host's CLONE_NEWPID,
3247 * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET namespaces. After it completed a number of
3248 * initializations a second child (the "inner" one) is forked off it, and it exits. */
3249
3250 assert(barrier);
3251 assert(directory);
3252 assert(pid_socket >= 0);
3253 assert(uuid_socket >= 0);
3254 assert(notify_socket >= 0);
3255 assert(master_pty_socket >= 0);
3256 assert(kmsg_socket >= 0);
3257
3258 log_debug("Outer child is initializing.");
3259
3260 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3261 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3262
3263 r = reset_audit_loginuid();
3264 if (r < 0)
3265 return r;
3266
3267 /* Mark everything as slave, so that we still
3268 * receive mounts from the real root, but don't
3269 * propagate mounts to the real root. */
3270 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3271 if (r < 0)
3272 return r;
3273
3274 if (dissected_image) {
3275 /* If we are operating on a disk image, then mount its root directory now, but leave out the rest. We
3276 * can read the UID shift from it if we need to. Further down we'll mount the rest, but then with the
3277 * uid shift known. That way we can mount VFAT file systems shifted to the right place right away. This
3278 * makes sure ESP partitions and userns are compatible. */
3279
3280 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3281 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
3282 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0)|
3283 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
3284 if (r < 0)
3285 return r;
3286 }
3287
3288 r = determine_uid_shift(directory);
3289 if (r < 0)
3290 return r;
3291
3292 if (arg_userns_mode != USER_NAMESPACE_NO) {
3293 /* Let the parent know which UID shift we read from the image */
3294 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3295 if (l < 0)
3296 return log_error_errno(errno, "Failed to send UID shift: %m");
3297 if (l != sizeof(arg_uid_shift))
3298 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3299 "Short write while sending UID shift.");
3300
3301 if (arg_userns_mode == USER_NAMESPACE_PICK) {
3302 /* When we are supposed to pick the UID shift, the parent will check now whether the UID shift
3303 * we just read from the image is available. If yes, it will send the UID shift back to us, if
3304 * not it will pick a different one, and send it back to us. */
3305
3306 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3307 if (l < 0)
3308 return log_error_errno(errno, "Failed to recv UID shift: %m");
3309 if (l != sizeof(arg_uid_shift))
3310 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3311 "Short read while receiving UID shift.");
3312 }
3313
3314 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
3315 "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3316 }
3317
3318 if (path_equal(directory, "/")) {
3319 /* If the directory we shall boot is the host, let's operate on a bind mount at a different
3320 * place, so that we can make changes to its mount structure (for example, to implement
3321 * --volatile=) without this interfering with our ability to access files such as
3322 * /etc/localtime to copy into the container. Note that we use a fixed place for this
3323 * (instead of a temporary directory, since we are living in our own mount namspace here
3324 * already, and thus don't need to be afraid of colliding with anyone else's mounts).*/
3325 (void) mkdir_p("/run/systemd/nspawn-root", 0755);
3326
3327 r = mount_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
3328 if (r < 0)
3329 return r;
3330
3331 directory = "/run/systemd/nspawn-root";
3332
3333 } else if (!dissected_image) {
3334 /* Turn directory into bind mount (we need that so that we can move the bind mount to root
3335 * later on). */
3336 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
3337 if (r < 0)
3338 return r;
3339 }
3340
3341 r = setup_pivot_root(
3342 directory,
3343 arg_pivot_root_new,
3344 arg_pivot_root_old);
3345 if (r < 0)
3346 return r;
3347
3348 r = setup_volatile_mode(
3349 directory,
3350 arg_volatile_mode,
3351 arg_userns_mode != USER_NAMESPACE_NO,
3352 arg_uid_shift,
3353 arg_uid_range,
3354 arg_selinux_apifs_context);
3355 if (r < 0)
3356 return r;
3357
3358 if (dissected_image) {
3359 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3360 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3361 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
3362 if (r < 0)
3363 return r;
3364 }
3365
3366 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3367 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3368
3369 r = detect_unified_cgroup_hierarchy_from_image(directory);
3370 if (r < 0)
3371 return r;
3372
3373 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3374 if (l < 0)
3375 return log_error_errno(errno, "Failed to send cgroup mode: %m");
3376 if (l != sizeof(arg_unified_cgroup_hierarchy))
3377 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3378 "Short write while sending cgroup mode.");
3379
3380 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3381 }
3382
3383 /* Mark everything as shared so our mounts get propagated down. This is
3384 * required to make new bind mounts available in systemd services
3385 * inside the container that create a new mount namespace.
3386 * See https://github.com/systemd/systemd/issues/3860
3387 * Further submounts (such as /dev) done after this will inherit the
3388 * shared propagation mode. */
3389 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3390 if (r < 0)
3391 return r;
3392
3393 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3394 if (r < 0)
3395 return r;
3396
3397 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3398 if (r < 0)
3399 return r;
3400
3401 if (arg_read_only && arg_volatile_mode == VOLATILE_NO) {
3402 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
3403 if (r < 0)
3404 return log_error_errno(r, "Failed to make tree read-only: %m");
3405 }
3406
3407 r = mount_all(directory,
3408 arg_mount_settings,
3409 arg_uid_shift,
3410 arg_selinux_apifs_context);
3411 if (r < 0)
3412 return r;
3413
3414 r = copy_devnodes(directory);
3415 if (r < 0)
3416 return r;
3417
3418 r = make_extra_nodes(directory);
3419 if (r < 0)
3420 return r;
3421
3422 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3423 (void) make_inaccessible_nodes(directory, arg_uid_shift, arg_uid_shift);
3424
3425 r = setup_pts(directory);
3426 if (r < 0)
3427 return r;
3428
3429 r = setup_propagate(directory);
3430 if (r < 0)
3431 return r;
3432
3433 r = setup_keyring();
3434 if (r < 0)
3435 return r;
3436
3437 r = setup_timezone(directory);
3438 if (r < 0)
3439 return r;
3440
3441 r = setup_resolv_conf(directory);
3442 if (r < 0)
3443 return r;
3444
3445 r = setup_machine_id(directory);
3446 if (r < 0)
3447 return r;
3448
3449 r = setup_journal(directory);
3450 if (r < 0)
3451 return r;
3452
3453 r = mount_custom(
3454 directory,
3455 arg_custom_mounts,
3456 arg_n_custom_mounts,
3457 arg_userns_mode != USER_NAMESPACE_NO,
3458 arg_uid_shift,
3459 arg_uid_range,
3460 arg_selinux_apifs_context,
3461 false);
3462 if (r < 0)
3463 return r;
3464
3465 if (!arg_use_cgns) {
3466 r = mount_cgroups(
3467 directory,
3468 arg_unified_cgroup_hierarchy,
3469 arg_userns_mode != USER_NAMESPACE_NO,
3470 arg_uid_shift,
3471 arg_uid_range,
3472 arg_selinux_apifs_context,
3473 false);
3474 if (r < 0)
3475 return r;
3476 }
3477
3478 r = mount_move_root(directory);
3479 if (r < 0)
3480 return log_error_errno(r, "Failed to move root directory: %m");
3481
3482 fd = setup_sd_notify_child();
3483 if (fd < 0)
3484 return fd;
3485
3486 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3487 arg_clone_ns_flags |
3488 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
3489 if (pid < 0)
3490 return log_error_errno(errno, "Failed to fork inner child: %m");
3491 if (pid == 0) {
3492 pid_socket = safe_close(pid_socket);
3493 uuid_socket = safe_close(uuid_socket);
3494 notify_socket = safe_close(notify_socket);
3495 uid_shift_socket = safe_close(uid_shift_socket);
3496
3497 /* The inner child has all namespaces that are
3498 * requested, so that we all are owned by the user if
3499 * user namespaces are turned on. */
3500
3501 if (arg_network_namespace_path) {
3502 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3503 if (r < 0)
3504 return log_error_errno(r, "Failed to join network namespace: %m");
3505 }
3506
3507 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds);
3508 if (r < 0)
3509 _exit(EXIT_FAILURE);
3510
3511 _exit(EXIT_SUCCESS);
3512 }
3513
3514 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3515 if (l < 0)
3516 return log_error_errno(errno, "Failed to send PID: %m");
3517 if (l != sizeof(pid))
3518 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3519 "Short write while sending PID.");
3520
3521 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3522 if (l < 0)
3523 return log_error_errno(errno, "Failed to send machine ID: %m");
3524 if (l != sizeof(arg_uuid))
3525 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3526 "Short write while sending machine ID.");
3527
3528 l = send_one_fd(notify_socket, fd, 0);
3529 if (l < 0)
3530 return log_error_errno(l, "Failed to send notify fd: %m");
3531
3532 pid_socket = safe_close(pid_socket);
3533 uuid_socket = safe_close(uuid_socket);
3534 notify_socket = safe_close(notify_socket);
3535 master_pty_socket = safe_close(master_pty_socket);
3536 kmsg_socket = safe_close(kmsg_socket);
3537 rtnl_socket = safe_close(rtnl_socket);
3538 netns_fd = safe_close(netns_fd);
3539
3540 return 0;
3541 }
3542
3543 static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
3544 bool tried_hashed = false;
3545 unsigned n_tries = 100;
3546 uid_t candidate;
3547 int r;
3548
3549 assert(shift);
3550 assert(ret_lock_file);
3551 assert(arg_userns_mode == USER_NAMESPACE_PICK);
3552 assert(arg_uid_range == 0x10000U);
3553
3554 candidate = *shift;
3555
3556 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3557
3558 for (;;) {
3559 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
3560 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
3561
3562 if (--n_tries <= 0)
3563 return -EBUSY;
3564
3565 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
3566 goto next;
3567 if ((candidate & UINT32_C(0xFFFF)) != 0)
3568 goto next;
3569
3570 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3571 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3572 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3573 goto next;
3574 if (r < 0)
3575 return r;
3576
3577 /* Make some superficial checks whether the range is currently known in the user database */
3578 if (getpwuid(candidate))
3579 goto next;
3580 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3581 goto next;
3582 if (getgrgid(candidate))
3583 goto next;
3584 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3585 goto next;
3586
3587 *ret_lock_file = lf;
3588 lf = (struct LockFile) LOCK_FILE_INIT;
3589 *shift = candidate;
3590 return 0;
3591
3592 next:
3593 if (arg_machine && !tried_hashed) {
3594 /* Try to hash the base from the container name */
3595
3596 static const uint8_t hash_key[] = {
3597 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3598 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3599 };
3600
3601 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3602
3603 tried_hashed = true;
3604 } else
3605 random_bytes(&candidate, sizeof(candidate));
3606
3607 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
3608 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3609 }
3610 }
3611
3612 static int setup_uid_map(pid_t pid) {
3613 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3614 int r;
3615
3616 assert(pid > 1);
3617
3618 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3619 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3620 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3621 if (r < 0)
3622 return log_error_errno(r, "Failed to write UID map: %m");
3623
3624 /* We always assign the same UID and GID ranges */
3625 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3626 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
3627 if (r < 0)
3628 return log_error_errno(r, "Failed to write GID map: %m");
3629
3630 return 0;
3631 }
3632
3633 static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3634 char buf[NOTIFY_BUFFER_MAX+1];
3635 char *p = NULL;
3636 struct iovec iovec = {
3637 .iov_base = buf,
3638 .iov_len = sizeof(buf)-1,
3639 };
3640 union {
3641 struct cmsghdr cmsghdr;
3642 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
3643 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
3644 } control = {};
3645 struct msghdr msghdr = {
3646 .msg_iov = &iovec,
3647 .msg_iovlen = 1,
3648 .msg_control = &control,
3649 .msg_controllen = sizeof(control),
3650 };
3651 struct cmsghdr *cmsg;
3652 struct ucred *ucred = NULL;
3653 ssize_t n;
3654 pid_t inner_child_pid;
3655 _cleanup_strv_free_ char **tags = NULL;
3656
3657 assert(userdata);
3658
3659 inner_child_pid = PTR_TO_PID(userdata);
3660
3661 if (revents != EPOLLIN) {
3662 log_warning("Got unexpected poll event for notify fd.");
3663 return 0;
3664 }
3665
3666 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3667 if (n < 0) {
3668 if (IN_SET(errno, EAGAIN, EINTR))
3669 return 0;
3670
3671 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3672 }
3673 cmsg_close_all(&msghdr);
3674
3675 CMSG_FOREACH(cmsg, &msghdr) {
3676 if (cmsg->cmsg_level == SOL_SOCKET &&
3677 cmsg->cmsg_type == SCM_CREDENTIALS &&
3678 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3679
3680 ucred = (struct ucred*) CMSG_DATA(cmsg);
3681 }
3682 }
3683
3684 if (!ucred || ucred->pid != inner_child_pid) {
3685 log_debug("Received notify message without valid credentials. Ignoring.");
3686 return 0;
3687 }
3688
3689 if ((size_t) n >= sizeof(buf)) {
3690 log_warning("Received notify message exceeded maximum size. Ignoring.");
3691 return 0;
3692 }
3693
3694 buf[n] = 0;
3695 tags = strv_split(buf, "\n\r");
3696 if (!tags)
3697 return log_oom();
3698
3699 if (strv_find(tags, "READY=1"))
3700 (void) sd_notifyf(false, "READY=1\n");
3701
3702 p = strv_find_startswith(tags, "STATUS=");
3703 if (p)
3704 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
3705
3706 return 0;
3707 }
3708
3709 static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
3710 int r;
3711
3712 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
3713 if (r < 0)
3714 return log_error_errno(r, "Failed to allocate notify event source: %m");
3715
3716 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
3717
3718 return 0;
3719 }
3720
3721 static int merge_settings(Settings *settings, const char *path) {
3722 int rl;
3723
3724 assert(settings);
3725 assert(path);
3726
3727 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3728 * that this steals the fields of the Settings* structure, and hence modifies it. */
3729
3730 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3731 settings->start_mode >= 0) {
3732 arg_start_mode = settings->start_mode;
3733 strv_free_and_replace(arg_parameters, settings->parameters);
3734 }
3735
3736 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
3737 arg_ephemeral = settings->ephemeral;
3738
3739 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
3740 settings->root) {
3741
3742 if (!arg_settings_trusted)
3743 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
3744 else
3745 free_and_replace(arg_directory, settings->root);
3746 }
3747
3748 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3749 settings->pivot_root_new) {
3750 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3751 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3752 }
3753
3754 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
3755 settings->working_directory)
3756 free_and_replace(arg_chdir, settings->working_directory);
3757
3758 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
3759 settings->environment)
3760 strv_free_and_replace(arg_setenv, settings->environment);
3761
3762 if ((arg_settings_mask & SETTING_USER) == 0) {
3763
3764 if (settings->user)
3765 free_and_replace(arg_user, settings->user);
3766
3767 if (uid_is_valid(settings->uid))
3768 arg_uid = settings->uid;
3769 if (gid_is_valid(settings->gid))
3770 arg_gid = settings->gid;
3771 if (settings->n_supplementary_gids > 0) {
3772 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
3773 arg_n_supplementary_gids = settings->n_supplementary_gids;
3774 }
3775 }
3776
3777 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
3778 uint64_t plus, minus;
3779
3780 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3781 * Settings structure */
3782
3783 plus = settings->capability;
3784 minus = settings->drop_capability;
3785
3786 if ((arg_settings_mask & SETTING_NETWORK) == 0) {
3787 if (settings_private_network(settings))
3788 plus |= UINT64_C(1) << CAP_NET_ADMIN;
3789 else
3790 minus |= UINT64_C(1) << CAP_NET_ADMIN;
3791 }
3792
3793 if (!arg_settings_trusted && plus != 0) {
3794 if (settings->capability != 0)
3795 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
3796 } else
3797 arg_caps_retain |= plus;
3798
3799 arg_caps_retain &= ~minus;
3800
3801 /* Copy the full capabilities over too */
3802 if (capability_quintet_is_set(&settings->full_capabilities)) {
3803 if (!arg_settings_trusted)
3804 log_warning("Ignoring capability settings, file %s is not trusted.", path);
3805 else
3806 arg_full_capabilities = settings->full_capabilities;
3807 }
3808 }
3809
3810 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3811 settings->kill_signal > 0)
3812 arg_kill_signal = settings->kill_signal;
3813
3814 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3815 settings->personality != PERSONALITY_INVALID)
3816 arg_personality = settings->personality;
3817
3818 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3819 !sd_id128_is_null(settings->machine_id)) {
3820
3821 if (!arg_settings_trusted)
3822 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
3823 else
3824 arg_uuid = settings->machine_id;
3825 }
3826
3827 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3828 settings->read_only >= 0)
3829 arg_read_only = settings->read_only;
3830
3831 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3832 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3833 arg_volatile_mode = settings->volatile_mode;
3834
3835 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3836 settings->n_custom_mounts > 0) {
3837
3838 if (!arg_settings_trusted)
3839 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
3840 else {
3841 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
3842 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
3843 arg_n_custom_mounts = settings->n_custom_mounts;
3844 settings->n_custom_mounts = 0;
3845 }
3846 }
3847
3848 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3849 (settings->private_network >= 0 ||
3850 settings->network_veth >= 0 ||
3851 settings->network_bridge ||
3852 settings->network_zone ||
3853 settings->network_interfaces ||
3854 settings->network_macvlan ||
3855 settings->network_ipvlan ||
3856 settings->network_veth_extra ||
3857 settings->network_namespace_path)) {
3858
3859 if (!arg_settings_trusted)
3860 log_warning("Ignoring network settings, file %s is not trusted.", path);
3861 else {
3862 arg_network_veth = settings_network_veth(settings);
3863 arg_private_network = settings_private_network(settings);
3864
3865 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
3866 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
3867 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
3868 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
3869
3870 free_and_replace(arg_network_bridge, settings->network_bridge);
3871 free_and_replace(arg_network_zone, settings->network_zone);
3872
3873 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
3874 }
3875 }
3876
3877 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3878 settings->expose_ports) {
3879
3880 if (!arg_settings_trusted)
3881 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
3882 else {
3883 expose_port_free_all(arg_expose_ports);
3884 arg_expose_ports = TAKE_PTR(settings->expose_ports);
3885 }
3886 }
3887
3888 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3889 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3890
3891 if (!arg_settings_trusted)
3892 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
3893 else {
3894 arg_userns_mode = settings->userns_mode;
3895 arg_uid_shift = settings->uid_shift;
3896 arg_uid_range = settings->uid_range;
3897 arg_userns_chown = settings->userns_chown;
3898 }
3899 }
3900
3901 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3902 arg_notify_ready = settings->notify_ready;
3903
3904 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3905
3906 if (!arg_settings_trusted && !strv_isempty(settings->syscall_whitelist))
3907 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
3908 else {
3909 strv_free_and_replace(arg_syscall_whitelist, settings->syscall_whitelist);
3910 strv_free_and_replace(arg_syscall_blacklist, settings->syscall_blacklist);
3911 }
3912
3913 #if HAVE_SECCOMP
3914 if (!arg_settings_trusted && settings->seccomp)
3915 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
3916 else {
3917 seccomp_release(arg_seccomp);
3918 arg_seccomp = TAKE_PTR(settings->seccomp);
3919 }
3920 #endif
3921 }
3922
3923 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
3924 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
3925 continue;
3926
3927 if (!settings->rlimit[rl])
3928 continue;
3929
3930 if (!arg_settings_trusted) {
3931 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
3932 continue;
3933 }
3934
3935 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
3936 }
3937
3938 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
3939 settings->hostname)
3940 free_and_replace(arg_hostname, settings->hostname);
3941
3942 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
3943 settings->no_new_privileges >= 0)
3944 arg_no_new_privileges = settings->no_new_privileges;
3945
3946 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
3947 settings->oom_score_adjust_set) {
3948
3949 if (!arg_settings_trusted)
3950 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
3951 else {
3952 arg_oom_score_adjust = settings->oom_score_adjust;
3953 arg_oom_score_adjust_set = true;
3954 }
3955 }
3956
3957 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
3958 settings->cpu_set.set) {
3959
3960 if (!arg_settings_trusted)
3961 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
3962 else {
3963 cpu_set_reset(&arg_cpu_set);
3964 arg_cpu_set = settings->cpu_set;
3965 settings->cpu_set = (CPUSet) {};
3966 }
3967 }
3968
3969 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
3970 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
3971 arg_resolv_conf = settings->resolv_conf;
3972
3973 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
3974 settings->link_journal != _LINK_JOURNAL_INVALID) {
3975
3976 if (!arg_settings_trusted)
3977 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
3978 else {
3979 arg_link_journal = settings->link_journal;
3980 arg_link_journal_try = settings->link_journal_try;
3981 }
3982 }
3983
3984 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
3985 settings->timezone != _TIMEZONE_MODE_INVALID)
3986 arg_timezone = settings->timezone;
3987
3988 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
3989 settings->slice) {
3990
3991 if (!arg_settings_trusted)
3992 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
3993 else
3994 free_and_replace(arg_slice, settings->slice);
3995 }
3996
3997 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
3998 settings->use_cgns >= 0) {
3999
4000 if (!arg_settings_trusted)
4001 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
4002 else
4003 arg_use_cgns = settings->use_cgns;
4004 }
4005
4006 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
4007 settings->clone_ns_flags != (unsigned long) -1) {
4008
4009 if (!arg_settings_trusted)
4010 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
4011 else
4012 arg_clone_ns_flags = settings->clone_ns_flags;
4013 }
4014
4015 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
4016 settings->console_mode >= 0) {
4017
4018 if (!arg_settings_trusted)
4019 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
4020 else
4021 arg_console_mode = settings->console_mode;
4022 }
4023
4024 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
4025 * don't consult arg_settings_mask for them. */
4026
4027 sd_bus_message_unref(arg_property_message);
4028 arg_property_message = TAKE_PTR(settings->properties);
4029
4030 arg_console_width = settings->console_width;
4031 arg_console_height = settings->console_height;
4032
4033 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
4034 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
4035 arg_n_extra_nodes = settings->n_extra_nodes;
4036
4037 return 0;
4038 }
4039
4040 static int load_settings(void) {
4041 _cleanup_(settings_freep) Settings *settings = NULL;
4042 _cleanup_fclose_ FILE *f = NULL;
4043 _cleanup_free_ char *p = NULL;
4044 const char *fn, *i;
4045 int r;
4046
4047 if (arg_oci_bundle)
4048 return 0;
4049
4050 /* If all settings are masked, there's no point in looking for
4051 * the settings file */
4052 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
4053 return 0;
4054
4055 fn = strjoina(arg_machine, ".nspawn");
4056
4057 /* We first look in the admin's directories in /etc and /run */
4058 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4059 _cleanup_free_ char *j = NULL;
4060
4061 j = path_join(i, fn);
4062 if (!j)
4063 return log_oom();
4064
4065 f = fopen(j, "re");
4066 if (f) {
4067 p = TAKE_PTR(j);
4068
4069 /* By default, we trust configuration from /etc and /run */
4070 if (arg_settings_trusted < 0)
4071 arg_settings_trusted = true;
4072
4073 break;
4074 }
4075
4076 if (errno != ENOENT)
4077 return log_error_errno(errno, "Failed to open %s: %m", j);
4078 }
4079
4080 if (!f) {
4081 /* After that, let's look for a file next to the
4082 * actual image we shall boot. */
4083
4084 if (arg_image) {
4085 p = file_in_same_dir(arg_image, fn);
4086 if (!p)
4087 return log_oom();
4088 } else if (arg_directory && !path_equal(arg_directory, "/")) {
4089 p = file_in_same_dir(arg_directory, fn);
4090 if (!p)
4091 return log_oom();
4092 }
4093
4094 if (p) {
4095 f = fopen(p, "re");
4096 if (!f && errno != ENOENT)
4097 return log_error_errno(errno, "Failed to open %s: %m", p);
4098
4099 /* By default, we do not trust configuration from /var/lib/machines */
4100 if (arg_settings_trusted < 0)
4101 arg_settings_trusted = false;
4102 }
4103 }
4104
4105 if (!f)
4106 return 0;
4107
4108 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4109
4110 r = settings_load(f, p, &settings);
4111 if (r < 0)
4112 return r;
4113
4114 return merge_settings(settings, p);
4115 }
4116
4117 static int load_oci_bundle(void) {
4118 _cleanup_(settings_freep) Settings *settings = NULL;
4119 int r;
4120
4121 if (!arg_oci_bundle)
4122 return 0;
4123
4124 /* By default let's trust OCI bundles */
4125 if (arg_settings_trusted < 0)
4126 arg_settings_trusted = true;
4127
4128 r = oci_load(NULL, arg_oci_bundle, &settings);
4129 if (r < 0)
4130 return r;
4131
4132 return merge_settings(settings, arg_oci_bundle);
4133 }
4134
4135 static int run_container(
4136 DissectedImage *dissected_image,
4137 bool secondary,
4138 FDSet *fds,
4139 char veth_name[IFNAMSIZ], bool *veth_created,
4140 union in_addr_union *exposed,
4141 int *master, pid_t *pid, int *ret) {
4142
4143 static const struct sigaction sa = {
4144 .sa_handler = nop_signal_handler,
4145 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
4146 };
4147
4148 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
4149 _cleanup_close_ int etc_passwd_lock = -1;
4150 _cleanup_close_pair_ int
4151 kmsg_socket_pair[2] = { -1, -1 },
4152 rtnl_socket_pair[2] = { -1, -1 },
4153 pid_socket_pair[2] = { -1, -1 },
4154 uuid_socket_pair[2] = { -1, -1 },
4155 notify_socket_pair[2] = { -1, -1 },
4156 uid_shift_socket_pair[2] = { -1, -1 },
4157 master_pty_socket_pair[2] = { -1, -1 },
4158 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4159
4160 _cleanup_close_ int notify_socket = -1;
4161 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4162 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
4163 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4164 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4165 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
4166 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
4167 ContainerStatus container_status = 0;
4168 int ifi = 0, r;
4169 ssize_t l;
4170 sigset_t mask_chld;
4171 _cleanup_close_ int netns_fd = -1;
4172
4173 assert_se(sigemptyset(&mask_chld) == 0);
4174 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4175
4176 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4177 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4178 * check with getpwuid() if the specific user already exists. Note that /etc might be
4179 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4180 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4181 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4182 * really ours. */
4183
4184 etc_passwd_lock = take_etc_passwd_lock(NULL);
4185 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4186 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4187 }
4188
4189 r = barrier_create(&barrier);
4190 if (r < 0)
4191 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4192
4193 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4194 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4195
4196 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4197 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4198
4199 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4200 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4201
4202 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4203 return log_error_errno(errno, "Failed to create id socket pair: %m");
4204
4205 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4206 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4207
4208 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
4209 return log_error_errno(errno, "Failed to create console socket pair: %m");
4210
4211 if (arg_userns_mode != USER_NAMESPACE_NO)
4212 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4213 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4214
4215 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4216 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4217 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4218
4219 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4220 * parent's blocking calls and give it a chance to call wait() and terminate. */
4221 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4222 if (r < 0)
4223 return log_error_errno(errno, "Failed to change the signal mask: %m");
4224
4225 r = sigaction(SIGCHLD, &sa, NULL);
4226 if (r < 0)
4227 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4228
4229 if (arg_network_namespace_path) {
4230 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4231 if (netns_fd < 0)
4232 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4233
4234 r = fd_is_network_ns(netns_fd);
4235 if (r == -EUCLEAN)
4236 log_debug_errno(r, "Cannot determine if passed network namespace path '%s' really refers to a network namespace, assuming it does.", arg_network_namespace_path);
4237 else if (r < 0)
4238 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
4239 else if (r == 0)
4240 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4241 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
4242 }
4243
4244 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4245 if (*pid < 0)
4246 return log_error_errno(errno, "clone() failed%s: %m",
4247 errno == EINVAL ?
4248 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4249
4250 if (*pid == 0) {
4251 /* The outer child only has a file system namespace. */
4252 barrier_set_role(&barrier, BARRIER_CHILD);
4253
4254 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4255 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4256 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4257 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4258 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
4259 master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
4260 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4261 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
4262
4263 (void) reset_all_signal_handlers();
4264 (void) reset_signal_mask();
4265
4266 r = outer_child(&barrier,
4267 arg_directory,
4268 dissected_image,
4269 secondary,
4270 pid_socket_pair[1],
4271 uuid_socket_pair[1],
4272 notify_socket_pair[1],
4273 kmsg_socket_pair[1],
4274 rtnl_socket_pair[1],
4275 uid_shift_socket_pair[1],
4276 master_pty_socket_pair[1],
4277 unified_cgroup_hierarchy_socket_pair[1],
4278 fds,
4279 netns_fd);
4280 if (r < 0)
4281 _exit(EXIT_FAILURE);
4282
4283 _exit(EXIT_SUCCESS);
4284 }
4285
4286 barrier_set_role(&barrier, BARRIER_PARENT);
4287
4288 fdset_close(fds);
4289
4290 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4291 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4292 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4293 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4294 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
4295 master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
4296 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
4297 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
4298
4299 if (arg_userns_mode != USER_NAMESPACE_NO) {
4300 /* The child just let us know the UID shift it might have read from the image. */
4301 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4302 if (l < 0)
4303 return log_error_errno(errno, "Failed to read UID shift: %m");
4304 if (l != sizeof arg_uid_shift)
4305 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
4306
4307 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4308 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4309 * image, but if that's already in use, pick a new one, and report back to the child,
4310 * which one we now picked. */
4311
4312 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4313 if (r < 0)
4314 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4315
4316 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4317 if (l < 0)
4318 return log_error_errno(errno, "Failed to send UID shift: %m");
4319 if (l != sizeof arg_uid_shift)
4320 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
4321 }
4322 }
4323
4324 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4325 /* The child let us know the support cgroup mode it might have read from the image. */
4326 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4327 if (l < 0)
4328 return log_error_errno(errno, "Failed to read cgroup mode: %m");
4329 if (l != sizeof(arg_unified_cgroup_hierarchy))
4330 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zu bytes).%s",
4331 l, l == 0 ? " The child is most likely dead." : "");
4332 }
4333
4334 /* Wait for the outer child. */
4335 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4336 if (r < 0)
4337 return r;
4338 if (r != EXIT_SUCCESS)
4339 return -EIO;
4340
4341 /* And now retrieve the PID of the inner child. */
4342 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4343 if (l < 0)
4344 return log_error_errno(errno, "Failed to read inner child PID: %m");
4345 if (l != sizeof *pid)
4346 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
4347
4348 /* We also retrieve container UUID in case it was generated by outer child */
4349 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4350 if (l < 0)
4351 return log_error_errno(errno, "Failed to read container machine ID: %m");
4352 if (l != sizeof(arg_uuid))
4353 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
4354
4355 /* We also retrieve the socket used for notifications generated by outer child */
4356 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4357 if (notify_socket < 0)
4358 return log_error_errno(notify_socket,
4359 "Failed to receive notification socket from the outer child: %m");
4360
4361 log_debug("Init process invoked as PID "PID_FMT, *pid);
4362
4363 if (arg_userns_mode != USER_NAMESPACE_NO) {
4364 if (!barrier_place_and_sync(&barrier)) /* #1 */
4365 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4366
4367 r = setup_uid_map(*pid);
4368 if (r < 0)
4369 return r;
4370
4371 (void) barrier_place(&barrier); /* #2 */
4372 }
4373
4374 if (arg_private_network) {
4375 if (!arg_network_namespace_path) {
4376 /* Wait until the child has unshared its network namespace. */
4377 if (!barrier_place_and_sync(&barrier)) /* #3 */
4378 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
4379 }
4380
4381 r = move_network_interfaces(*pid, arg_network_interfaces);
4382 if (r < 0)
4383 return r;
4384
4385 if (arg_network_veth) {
4386 r = setup_veth(arg_machine, *pid, veth_name,
4387 arg_network_bridge || arg_network_zone);
4388 if (r < 0)
4389 return r;
4390 else if (r > 0)
4391 ifi = r;
4392
4393 if (arg_network_bridge) {
4394 /* Add the interface to a bridge */
4395 r = setup_bridge(veth_name, arg_network_bridge, false);
4396 if (r < 0)
4397 return r;
4398 if (r > 0)
4399 ifi = r;
4400 } else if (arg_network_zone) {
4401 /* Add the interface to a bridge, possibly creating it */
4402 r = setup_bridge(veth_name, arg_network_zone, true);
4403 if (r < 0)
4404 return r;
4405 if (r > 0)
4406 ifi = r;
4407 }
4408 }
4409
4410 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
4411 if (r < 0)
4412 return r;
4413
4414 /* We created the primary and extra veth links now; let's remember this, so that we know to
4415 remove them later on. Note that we don't bother with removing veth links that were created
4416 here when their setup failed half-way, because in that case the kernel should be able to
4417 remove them on its own, since they cannot be referenced by anything yet. */
4418 *veth_created = true;
4419
4420 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
4421 if (r < 0)
4422 return r;
4423
4424 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
4425 if (r < 0)
4426 return r;
4427 }
4428
4429 if (arg_register || !arg_keep_unit) {
4430 r = sd_bus_default_system(&bus);
4431 if (r < 0)
4432 return log_error_errno(r, "Failed to open system bus: %m");
4433
4434 r = sd_bus_set_close_on_exit(bus, false);
4435 if (r < 0)
4436 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
4437 }
4438
4439 if (!arg_keep_unit) {
4440 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4441 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4442 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4443
4444 r = sd_bus_match_signal_async(
4445 bus,
4446 NULL,
4447 "org.freedesktop.systemd1",
4448 NULL,
4449 "org.freedesktop.systemd1.Scope",
4450 "RequestStop",
4451 on_request_stop, NULL, PID_TO_PTR(*pid));
4452 if (r < 0)
4453 return log_error_errno(r, "Failed to request RequestStop match: %m");
4454 }
4455
4456 if (arg_register) {
4457 r = register_machine(
4458 bus,
4459 arg_machine,
4460 *pid,
4461 arg_directory,
4462 arg_uuid,
4463 ifi,
4464 arg_slice,
4465 arg_custom_mounts, arg_n_custom_mounts,
4466 arg_kill_signal,
4467 arg_property,
4468 arg_property_message,
4469 arg_keep_unit,
4470 arg_container_service_name);
4471 if (r < 0)
4472 return r;
4473
4474 } else if (!arg_keep_unit) {
4475 r = allocate_scope(
4476 bus,
4477 arg_machine,
4478 *pid,
4479 arg_slice,
4480 arg_custom_mounts, arg_n_custom_mounts,
4481 arg_kill_signal,
4482 arg_property,
4483 arg_property_message);
4484 if (r < 0)
4485 return r;
4486
4487 } else if (arg_slice || arg_property)
4488 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
4489
4490 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
4491 if (r < 0)
4492 return r;
4493
4494 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4495 if (r < 0)
4496 return r;
4497
4498 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
4499 if (r < 0)
4500 return r;
4501
4502 /* Notify the child that the parent is ready with all
4503 * its setup (including cgroup-ification), and that
4504 * the child can now hand over control to the code to
4505 * run inside the container. */
4506 (void) barrier_place(&barrier); /* #4 */
4507
4508 /* Block SIGCHLD here, before notifying child.
4509 * process_pty() will handle it with the other signals. */
4510 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4511
4512 /* Reset signal to default */
4513 r = default_signals(SIGCHLD, -1);
4514 if (r < 0)
4515 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4516
4517 r = sd_event_new(&event);
4518 if (r < 0)
4519 return log_error_errno(r, "Failed to get default event source: %m");
4520
4521 (void) sd_event_set_watchdog(event, true);
4522
4523 if (bus) {
4524 r = sd_bus_attach_event(bus, event, 0);
4525 if (r < 0)
4526 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4527 }
4528
4529 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
4530 if (r < 0)
4531 return r;
4532
4533 /* Let the child know that we are ready and wait that the child is completely ready now. */
4534 if (!barrier_place_and_sync(&barrier)) /* #5 */
4535 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
4536
4537 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4538 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4539 etc_passwd_lock = safe_close(etc_passwd_lock);
4540
4541 (void) sd_notifyf(false,
4542 "STATUS=Container running.\n"
4543 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
4544 if (!arg_notify_ready)
4545 (void) sd_notify(false, "READY=1\n");
4546
4547 if (arg_kill_signal > 0) {
4548 /* Try to kill the init system on SIGINT or SIGTERM */
4549 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4550 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
4551 } else {
4552 /* Immediately exit */
4553 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4554 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4555 }
4556
4557 /* Exit when the child exits */
4558 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
4559
4560 if (arg_expose_ports) {
4561 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
4562 if (r < 0)
4563 return r;
4564
4565 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
4566 }
4567
4568 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4569
4570 if (arg_console_mode != CONSOLE_PIPE) {
4571 _cleanup_close_ int fd = -1;
4572 PTYForwardFlags flags = 0;
4573
4574 /* Retrieve the master pty allocated by inner child */
4575 fd = receive_one_fd(master_pty_socket_pair[0], 0);
4576 if (fd < 0)
4577 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
4578
4579 switch (arg_console_mode) {
4580
4581 case CONSOLE_READ_ONLY:
4582 flags |= PTY_FORWARD_READ_ONLY;
4583
4584 _fallthrough_;
4585
4586 case CONSOLE_INTERACTIVE:
4587 flags |= PTY_FORWARD_IGNORE_VHANGUP;
4588
4589 r = pty_forward_new(event, fd, flags, &forward);
4590 if (r < 0)
4591 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4592
4593 if (arg_console_width != (unsigned) -1 || arg_console_height != (unsigned) -1)
4594 (void) pty_forward_set_width_height(forward,
4595 arg_console_width,
4596 arg_console_height);
4597 break;
4598
4599 default:
4600 assert(arg_console_mode == CONSOLE_PASSIVE);
4601 }
4602
4603 *master = TAKE_FD(fd);
4604 }
4605
4606 r = sd_event_loop(event);
4607 if (r < 0)
4608 return log_error_errno(r, "Failed to run event loop: %m");
4609
4610 if (forward) {
4611 char last_char = 0;
4612
4613 (void) pty_forward_get_last_char(forward, &last_char);
4614 forward = pty_forward_free(forward);
4615
4616 if (!arg_quiet && last_char != '\n')
4617 putc('\n', stdout);
4618 }
4619
4620 /* Kill if it is not dead yet anyway */
4621 if (bus) {
4622 if (arg_register)
4623 terminate_machine(bus, arg_machine);
4624 else if (!arg_keep_unit)
4625 terminate_scope(bus, arg_machine);
4626 }
4627
4628 /* Normally redundant, but better safe than sorry */
4629 (void) kill(*pid, SIGKILL);
4630
4631 r = wait_for_container(*pid, &container_status);
4632 *pid = 0;
4633
4634 if (r < 0)
4635 /* We failed to wait for the container, or the container exited abnormally. */
4636 return r;
4637 if (r > 0 || container_status == CONTAINER_TERMINATED) {
4638 /* r > 0 → The container exited with a non-zero status.
4639 * As a special case, we need to replace 133 with a different value,
4640 * because 133 is special-cased in the service file to reboot the container.
4641 * otherwise → The container exited with zero status and a reboot was not requested.
4642 */
4643 if (r == EXIT_FORCE_RESTART)
4644 r = EXIT_FAILURE; /* replace 133 with the general failure code */
4645 *ret = r;
4646 return 0; /* finito */
4647 }
4648
4649 /* CONTAINER_REBOOTED, loop again */
4650
4651 if (arg_keep_unit) {
4652 /* Special handling if we are running as a service: instead of simply
4653 * restarting the machine we want to restart the entire service, so let's
4654 * inform systemd about this with the special exit code 133. The service
4655 * file uses RestartForceExitStatus=133 so that this results in a full
4656 * nspawn restart. This is necessary since we might have cgroup parameters
4657 * set we want to have flushed out. */
4658 *ret = EXIT_FORCE_RESTART;
4659 return 0; /* finito */
4660 }
4661
4662 expose_port_flush(arg_expose_ports, exposed);
4663
4664 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4665 *veth_created = false;
4666 return 1; /* loop again */
4667 }
4668
4669 static int initialize_rlimits(void) {
4670 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4671 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4672 * container execution environments. */
4673
4674 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
4675 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
4676 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
4677 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
4678 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
4679 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
4680 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
4681 [RLIMIT_MEMLOCK] = { 65536, 65536 },
4682 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
4683 [RLIMIT_NICE] = { 0, 0 },
4684 [RLIMIT_NOFILE] = { 1024, 4096 },
4685 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
4686 [RLIMIT_RTPRIO] = { 0, 0 },
4687 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
4688 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
4689
4690 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4691 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4692 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4693 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4694 * that PID 1 changes a number of other resource limits during early initialization which is why we
4695 * don't read the other limits from PID 1 but prefer the static table above. */
4696 };
4697
4698 int rl;
4699
4700 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
4701 /* Let's only fill in what the user hasn't explicitly configured anyway */
4702 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
4703 const struct rlimit *v;
4704 struct rlimit buffer;
4705
4706 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
4707 /* For these two let's read the limits off PID 1. See above for an explanation. */
4708
4709 if (prlimit(1, rl, NULL, &buffer) < 0)
4710 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
4711
4712 v = &buffer;
4713 } else
4714 v = kernel_defaults + rl;
4715
4716 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
4717 if (!arg_rlimit[rl])
4718 return log_oom();
4719 }
4720
4721 if (DEBUG_LOGGING) {
4722 _cleanup_free_ char *k = NULL;
4723
4724 (void) rlimit_format(arg_rlimit[rl], &k);
4725 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
4726 }
4727 }
4728
4729 return 0;
4730 }
4731
4732 static int run(int argc, char *argv[]) {
4733 bool secondary = false, remove_directory = false, remove_image = false,
4734 veth_created = false, remove_tmprootdir = false;
4735 _cleanup_close_ int master = -1;
4736 _cleanup_fdset_free_ FDSet *fds = NULL;
4737 int r, n_fd_passed, ret = EXIT_SUCCESS;
4738 char veth_name[IFNAMSIZ] = "";
4739 union in_addr_union exposed = {};
4740 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4741 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
4742 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
4743 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4744 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
4745 pid_t pid = 0;
4746
4747 log_parse_environment();
4748 log_open();
4749
4750 r = parse_argv(argc, argv);
4751 if (r <= 0)
4752 goto finish;
4753
4754 r = must_be_root();
4755 if (r < 0)
4756 goto finish;
4757
4758 r = initialize_rlimits();
4759 if (r < 0)
4760 goto finish;
4761
4762 r = load_oci_bundle();
4763 if (r < 0)
4764 goto finish;
4765
4766 r = determine_names();
4767 if (r < 0)
4768 goto finish;
4769
4770 r = load_settings();
4771 if (r < 0)
4772 goto finish;
4773
4774 r = cg_unified();
4775 if (r < 0) {
4776 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4777 goto finish;
4778 }
4779
4780 r = verify_arguments();
4781 if (r < 0)
4782 goto finish;
4783
4784 /* Reapply environment settings. */
4785 (void) detect_unified_cgroup_hierarchy_from_environment();
4786
4787 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4788 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4789 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4790 (void) ignore_signals(SIGPIPE, -1);
4791
4792 n_fd_passed = sd_listen_fds(false);
4793 if (n_fd_passed > 0) {
4794 r = fdset_new_listen_fds(&fds, false);
4795 if (r < 0) {
4796 log_error_errno(r, "Failed to collect file descriptors: %m");
4797 goto finish;
4798 }
4799 }
4800
4801 /* The "default" umask. This is appropriate for most file and directory
4802 * operations performed by nspawn, and is the umask that will be used for
4803 * the child. Functions like copy_devnodes() change the umask temporarily. */
4804 umask(0022);
4805
4806 if (arg_directory) {
4807 assert(!arg_image);
4808
4809 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4810 * /var from the host will propagate into container dynamically (because bad things happen if
4811 * two systems write to the same /var). Let's allow it for the special cases where /var is
4812 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4813 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
4814 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
4815 r = -EINVAL;
4816 goto finish;
4817 }
4818
4819 if (arg_ephemeral) {
4820 _cleanup_free_ char *np = NULL;
4821
4822 r = chase_symlinks_and_update(&arg_directory, 0);
4823 if (r < 0)
4824 goto finish;
4825
4826 /* If the specified path is a mount point we generate the new snapshot immediately
4827 * inside it under a random name. However if the specified is not a mount point we
4828 * create the new snapshot in the parent directory, just next to it. */
4829 r = path_is_mount_point(arg_directory, NULL, 0);
4830 if (r < 0) {
4831 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4832 goto finish;
4833 }
4834 if (r > 0)
4835 r = tempfn_random_child(arg_directory, "machine.", &np);
4836 else
4837 r = tempfn_random(arg_directory, "machine.", &np);
4838 if (r < 0) {
4839 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
4840 goto finish;
4841 }
4842
4843 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4844 * only owned by us and noone else. */
4845 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
4846 if (r < 0) {
4847 log_error_errno(r, "Failed to lock %s: %m", np);
4848 goto finish;
4849 }
4850
4851 {
4852 BLOCK_SIGNALS(SIGINT);
4853 r = btrfs_subvol_snapshot(arg_directory, np,
4854 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4855 BTRFS_SNAPSHOT_FALLBACK_COPY |
4856 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4857 BTRFS_SNAPSHOT_RECURSIVE |
4858 BTRFS_SNAPSHOT_QUOTA |
4859 BTRFS_SNAPSHOT_SIGINT);
4860 }
4861 if (r == -EINTR) {
4862 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
4863 goto finish;
4864 }
4865 if (r < 0) {
4866 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4867 goto finish;
4868 }
4869
4870 free_and_replace(arg_directory, np);
4871 remove_directory = true;
4872 } else {
4873 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
4874 if (r < 0)
4875 goto finish;
4876
4877 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4878 if (r == -EBUSY) {
4879 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4880 goto finish;
4881 }
4882 if (r < 0) {
4883 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4884 goto finish;
4885 }
4886
4887 if (arg_template) {
4888 r = chase_symlinks_and_update(&arg_template, 0);
4889 if (r < 0)
4890 goto finish;
4891
4892 {
4893 BLOCK_SIGNALS(SIGINT);
4894 r = btrfs_subvol_snapshot(arg_template, arg_directory,
4895 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4896 BTRFS_SNAPSHOT_FALLBACK_COPY |
4897 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4898 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
4899 BTRFS_SNAPSHOT_RECURSIVE |
4900 BTRFS_SNAPSHOT_QUOTA |
4901 BTRFS_SNAPSHOT_SIGINT);
4902 }
4903 if (r == -EEXIST)
4904 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4905 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4906 else if (r == -EINTR) {
4907 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
4908 goto finish;
4909 } else if (r < 0) {
4910 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4911 goto finish;
4912 } else
4913 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4914 "Populated %s from template %s.", arg_directory, arg_template);
4915 }
4916 }
4917
4918 if (arg_start_mode == START_BOOT) {
4919 const char *p;
4920
4921 if (arg_pivot_root_new)
4922 p = prefix_roota(arg_directory, arg_pivot_root_new);
4923 else
4924 p = arg_directory;
4925
4926 if (path_is_os_tree(p) <= 0) {
4927 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
4928 r = -EINVAL;
4929 goto finish;
4930 }
4931 } else {
4932 const char *p, *q;
4933
4934 if (arg_pivot_root_new)
4935 p = prefix_roota(arg_directory, arg_pivot_root_new);
4936 else
4937 p = arg_directory;
4938
4939 q = strjoina(p, "/usr/");
4940
4941 if (laccess(q, F_OK) < 0) {
4942 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
4943 r = -EINVAL;
4944 goto finish;
4945 }
4946 }
4947
4948 } else {
4949 assert(arg_image);
4950 assert(!arg_template);
4951
4952 r = chase_symlinks_and_update(&arg_image, 0);
4953 if (r < 0)
4954 goto finish;
4955
4956 if (arg_ephemeral) {
4957 _cleanup_free_ char *np = NULL;
4958
4959 r = tempfn_random(arg_image, "machine.", &np);
4960 if (r < 0) {
4961 log_error_errno(r, "Failed to generate name for image snapshot: %m");
4962 goto finish;
4963 }
4964
4965 /* Always take an exclusive lock on our own ephemeral copy. */
4966 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
4967 if (r < 0) {
4968 r = log_error_errno(r, "Failed to create image lock: %m");
4969 goto finish;
4970 }
4971
4972 {
4973 BLOCK_SIGNALS(SIGINT);
4974 r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME|COPY_SIGINT);
4975 }
4976 if (r == -EINTR) {
4977 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
4978 goto finish;
4979 }
4980 if (r < 0) {
4981 r = log_error_errno(r, "Failed to copy image file: %m");
4982 goto finish;
4983 }
4984
4985 free_and_replace(arg_image, np);
4986 remove_image = true;
4987 } else {
4988 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4989 if (r == -EBUSY) {
4990 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4991 goto finish;
4992 }
4993 if (r < 0) {
4994 r = log_error_errno(r, "Failed to create image lock: %m");
4995 goto finish;
4996 }
4997
4998 if (!arg_root_hash) {
4999 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
5000 if (r < 0) {
5001 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
5002 goto finish;
5003 }
5004 }
5005 }
5006
5007 if (!mkdtemp(tmprootdir)) {
5008 r = log_error_errno(errno, "Failed to create temporary directory: %m");
5009 goto finish;
5010 }
5011
5012 remove_tmprootdir = true;
5013
5014 arg_directory = strdup(tmprootdir);
5015 if (!arg_directory) {
5016 r = log_oom();
5017 goto finish;
5018 }
5019
5020 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
5021 if (r < 0) {
5022 log_error_errno(r, "Failed to set up loopback block device: %m");
5023 goto finish;
5024 }
5025
5026 r = dissect_image_and_warn(
5027 loop->fd,
5028 arg_image,
5029 arg_root_hash, arg_root_hash_size,
5030 DISSECT_IMAGE_REQUIRE_ROOT,
5031 &dissected_image);
5032 if (r == -ENOPKG) {
5033 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
5034 log_notice("Note that the disk image needs to\n"
5035 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
5036 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
5037 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
5038 " d) or contain a file system without a partition table\n"
5039 "in order to be bootable with systemd-nspawn.");
5040 goto finish;
5041 }
5042 if (r < 0)
5043 goto finish;
5044
5045 if (!arg_root_hash && dissected_image->can_verity)
5046 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
5047
5048 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
5049 if (r < 0)
5050 goto finish;
5051
5052 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
5053 if (remove_image && unlink(arg_image) >= 0)
5054 remove_image = false;
5055 }
5056
5057 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5058 if (r < 0)
5059 goto finish;
5060
5061 if (arg_console_mode < 0)
5062 arg_console_mode =
5063 isatty(STDIN_FILENO) > 0 &&
5064 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
5065
5066 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5067 arg_quiet = true;
5068
5069 if (!arg_quiet)
5070 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5071 arg_machine, arg_image ?: arg_directory);
5072
5073 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
5074
5075 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
5076 r = log_error_errno(errno, "Failed to become subreaper: %m");
5077 goto finish;
5078 }
5079
5080 for (;;) {
5081 r = run_container(dissected_image,
5082 secondary,
5083 fds,
5084 veth_name, &veth_created,
5085 &exposed, &master,
5086 &pid, &ret);
5087 if (r <= 0)
5088 break;
5089 }
5090
5091 finish:
5092 (void) sd_notify(false,
5093 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5094 "STOPPING=1\nSTATUS=Terminating...");
5095
5096 if (pid > 0)
5097 (void) kill(pid, SIGKILL);
5098
5099 /* Try to flush whatever is still queued in the pty */
5100 if (master >= 0) {
5101 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
5102 master = safe_close(master);
5103 }
5104
5105 if (pid > 0)
5106 (void) wait_for_terminate(pid, NULL);
5107
5108 pager_close();
5109
5110 if (remove_directory && arg_directory) {
5111 int k;
5112
5113 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
5114 if (k < 0)
5115 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
5116 }
5117
5118 if (remove_image && arg_image) {
5119 if (unlink(arg_image) < 0)
5120 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5121 }
5122
5123 if (remove_tmprootdir) {
5124 if (rmdir(tmprootdir) < 0)
5125 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5126 }
5127
5128 if (arg_machine) {
5129 const char *p;
5130
5131 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5132 (void) rm_rf(p, REMOVE_ROOT);
5133 }
5134
5135 expose_port_flush(arg_expose_ports, &exposed);
5136
5137 if (veth_created)
5138 (void) remove_veth_links(veth_name, arg_network_veth_extra);
5139 (void) remove_bridge(arg_network_zone);
5140
5141 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5142 expose_port_free_all(arg_expose_ports);
5143 rlimit_free_all(arg_rlimit);
5144 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
5145
5146 if (r < 0)
5147 return r;
5148
5149 return ret;
5150 }
5151
5152 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.