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