]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
nspawn: always take exclusive locks of ephemeral OS tree copies
[thirdparty/systemd.git] / src / nspawn / nspawn.c
CommitLineData
53e1b683 1/* SPDX-License-Identifier: LGPL-2.1+ */
88213476 2
349cc4a5 3#if HAVE_BLKID
6b5cf3ea 4#include <blkid.h>
8fe0087e 5#endif
88213476 6#include <errno.h>
88213476 7#include <getopt.h>
0e7ac751 8#include <grp.h>
503f480f 9#include <linux/fs.h>
1b9e5b12 10#include <linux/loop.h>
0e7ac751 11#include <pwd.h>
8fe0087e 12#include <sched.h>
349cc4a5 13#if HAVE_SELINUX
8fe0087e 14#include <selinux/selinux.h>
1b9e5b12 15#endif
8fe0087e
LP		 16#include <signal.h>
17#include <stdio.h>
18#include <stdlib.h>
19#include <string.h>
20#include <sys/file.h>
8fe0087e
LP		 21#include <sys/personality.h>
22#include <sys/prctl.h>
23#include <sys/types.h>
6916b164 24#include <sys/wait.h>
8fe0087e 25#include <unistd.h>
1b9e5b12 26
b053cd5f 27#include "sd-bus.h"
1f0cd86b 28#include "sd-daemon.h"
1f0cd86b 29#include "sd-id128.h"
8fe0087e 30
b5efdb8a 31#include "alloc-util.h"
8fe0087e
LP		 32#include "barrier.h"
33#include "base-filesystem.h"
34#include "blkid-util.h"
35#include "btrfs-util.h"
b8ea7a6e 36#include "bus-error.h"
b053cd5f 37#include "bus-util.h"
8fe0087e 38#include "cap-list.h"
430f0182 39#include "capability-util.h"
04d391da 40#include "cgroup-util.h"
8fe0087e 41#include "copy.h"
d107bb7d 42#include "cpu-set-util.h"
4fc9982c 43#include "dev-setup.h"
2d845785 44#include "dissect-image.h"
8fe0087e 45#include "env-util.h"
3ffd4af2 46#include "fd-util.h"
842f3b0f 47#include "fdset.h"
a5c32cff 48#include "fileio.h"
f97b34a6 49#include "format-util.h"
f4f15635 50#include "fs-util.h"
1b9e5b12 51#include "gpt.h"
4623e8e6 52#include "hexdecoct.h"
8fe0087e 53#include "hostname-util.h"
910fd145 54#include "id128-util.h"
8fe0087e 55#include "log.h"
2d845785 56#include "loop-util.h"
8fe0087e 57#include "loopback-setup.h"
1b9cebf6 58#include "machine-image.h"
8fe0087e 59#include "macro.h"
44dbef90 60#include "main-func.h"
8fe0087e
LP		 61#include "missing.h"
62#include "mkdir.h"
4349cd7c 63#include "mount-util.h"
049af8ad 64#include "mountpoint-util.h"
0cb8e3d1 65#include "namespace-util.h"
8fe0087e 66#include "netlink-util.h"
07630cea 67#include "nspawn-cgroup.h"
3603efde 68#include "nspawn-def.h"
07630cea
LP		 69#include "nspawn-expose-ports.h"
70#include "nspawn-mount.h"
71#include "nspawn-network.h"
de40a303 72#include "nspawn-oci.h"
7336138e 73#include "nspawn-patch-uid.h"
07630cea 74#include "nspawn-register.h"
910fd145 75#include "nspawn-seccomp.h"
07630cea
LP		 76#include "nspawn-settings.h"
77#include "nspawn-setuid.h"
7732f92b 78#include "nspawn-stub-pid1.h"
d8b4d14d 79#include "nulstr-util.h"
d58ad743 80#include "os-util.h"
50ebcf6c 81#include "pager.h"
6bedfcbb 82#include "parse-util.h"
8fe0087e 83#include "path-util.h"
294bf0c3 84#include "pretty-print.h"
0b452006 85#include "process-util.h"
8fe0087e
LP		 86#include "ptyfwd.h"
87#include "random-util.h"
8869a0b4 88#include "raw-clone.h"
bf428efb 89#include "rlimit-util.h"
8fe0087e 90#include "rm-rf.h"
de40a303
LP		 91#if HAVE_SECCOMP
92#include "seccomp-util.h"
93#endif
68b02049 94#include "selinux-util.h"
8fe0087e 95#include "signal-util.h"
2583fbea 96#include "socket-util.h"
8fcde012 97#include "stat-util.h"
15a5e950 98#include "stdio-util.h"
5c828e66 99#include "string-table.h"
07630cea 100#include "string-util.h"
8fe0087e 101#include "strv.h"
de40a303 102#include "sysctl-util.h"
8fe0087e 103#include "terminal-util.h"
e4de7287 104#include "tmpfile-util.h"
affb60b1 105#include "umask-util.h"
b1d4f8e1 106#include "user-util.h"
8fe0087e 107#include "util.h"
e9642be2 108
62b1e758
YW		 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
9c1e04d0
AP		 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"
0e7ac751 119
2a49b612
ZJS		 120#define EXIT_FORCE_RESTART 133
121
113cea80
DH		 122typedef enum ContainerStatus {
123 CONTAINER_TERMINATED,
6145bb4f 124 CONTAINER_REBOOTED,
113cea80
DH		 125} ContainerStatus;
126
88213476 127static char *arg_directory = NULL;
ec16945e 128static char *arg_template = NULL;
5f932eb9 129static char *arg_chdir = NULL;
b53ede69
PW		 130static char *arg_pivot_root_new = NULL;
131static char *arg_pivot_root_old = NULL;
687d0825 132static char *arg_user = NULL;
de40a303
LP		 133static uid_t arg_uid = UID_INVALID;
134static gid_t arg_gid = GID_INVALID;
135static gid_t* arg_supplementary_gids = NULL;
136static size_t arg_n_supplementary_gids = 0;
9444b1f2 137static sd_id128_t arg_uuid = {};
3a9530e5
LP		 138static char *arg_machine = NULL; /* The name used by the host to refer to this */
139static char *arg_hostname = NULL; /* The name the payload sees by default */
c74e630d
LP		 140static const char *arg_selinux_context = NULL;
141static const char *arg_selinux_apifs_context = NULL;
de40a303 142static char *arg_slice = NULL;
ff01d048 143static bool arg_private_network = false;
bc2f673e 144static bool arg_read_only = false;
7732f92b 145static StartMode arg_start_mode = START_PID1;
ec16945e 146static bool arg_ephemeral = false;
57fb9fb5 147static LinkJournal arg_link_journal = LINK_AUTO;
574edc90 148static bool arg_link_journal_try = false;
520e0d54 149static uint64_t arg_caps_retain =
50b52222
LP		 150 (1ULL << CAP_AUDIT_CONTROL) |
151 (1ULL << CAP_AUDIT_WRITE) |
5076f0cc
LP		 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) |
50b52222 161 (1ULL << CAP_MKNOD) |
5076f0cc
LP		 162 (1ULL << CAP_NET_BIND_SERVICE) |
163 (1ULL << CAP_NET_BROADCAST) |
164 (1ULL << CAP_NET_RAW) |
5076f0cc 165 (1ULL << CAP_SETFCAP) |
50b52222 166 (1ULL << CAP_SETGID) |
5076f0cc
LP		 167 (1ULL << CAP_SETPCAP) |
168 (1ULL << CAP_SETUID) |
169 (1ULL << CAP_SYS_ADMIN) |
50b52222 170 (1ULL << CAP_SYS_BOOT) |
5076f0cc
LP		 171 (1ULL << CAP_SYS_CHROOT) |
172 (1ULL << CAP_SYS_NICE) |
173 (1ULL << CAP_SYS_PTRACE) |
d87be9b0 174 (1ULL << CAP_SYS_RESOURCE) |
50b52222 175 (1ULL << CAP_SYS_TTY_CONFIG);
de40a303 176static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
5a8af538 177static CustomMount *arg_custom_mounts = NULL;
88614c8a 178static size_t arg_n_custom_mounts = 0;
f4889f65 179static char **arg_setenv = NULL;
284c0b91 180static bool arg_quiet = false;
eb91eb18 181static bool arg_register = true;
89f7c846 182static bool arg_keep_unit = false;
aa28aefe 183static char **arg_network_interfaces = NULL;
c74e630d 184static char **arg_network_macvlan = NULL;
4bbfe7ad 185static char **arg_network_ipvlan = NULL;
69c79d3c 186static bool arg_network_veth = false;
f6d6bad1 187static char **arg_network_veth_extra = NULL;
f757855e 188static char *arg_network_bridge = NULL;
22b28dfd 189static char *arg_network_zone = NULL;
d7bea6b6 190static char *arg_network_namespace_path = NULL;
bb068de0 191static PagerFlags arg_pager_flags = 0;
050f7277 192static unsigned long arg_personality = PERSONALITY_INVALID;
ec16945e 193static char *arg_image = NULL;
de40a303 194static char *arg_oci_bundle = NULL;
f757855e 195static VolatileMode arg_volatile_mode = VOLATILE_NO;
6d0b55c2 196static ExposePort *arg_expose_ports = NULL;
f36933fe 197static char **arg_property = NULL;
de40a303 198static sd_bus_message *arg_property_message = NULL;
0de7acce 199static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
6dac160c 200static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
0de7acce 201static bool arg_userns_chown = false;
c6c8f6e2 202static int arg_kill_signal = 0;
5da38d07 203static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
f757855e
LP		 204static SettingsMask arg_settings_mask = 0;
205static int arg_settings_trusted = -1;
206static char **arg_parameters = NULL;
6aadfa4c 207static const char *arg_container_service_name = "systemd-nspawn";
9c1e04d0 208static bool arg_notify_ready = false;
5a8ff0e6 209static bool arg_use_cgns = true;
0c582db0 210static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
1099ceeb 211static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
4623e8e6
LP		 212static void *arg_root_hash = NULL;
213static size_t arg_root_hash_size = 0;
960e4569
LP		 214static char **arg_syscall_whitelist = NULL;
215static char **arg_syscall_blacklist = NULL;
de40a303
LP		 216#if HAVE_SECCOMP
217static scmp_filter_ctx arg_seccomp = NULL;
218#endif
bf428efb 219static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
66edd963 220static bool arg_no_new_privileges = false;
81f345df
LP		 221static int arg_oom_score_adjust = 0;
222static bool arg_oom_score_adjust_set = false;
0985c7c4 223static CPUSet arg_cpu_set = {};
09d423e9 224static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
1688841f 225static TimezoneMode arg_timezone = TIMEZONE_AUTO;
de40a303
LP		 226static unsigned arg_console_width = (unsigned) -1, arg_console_height = (unsigned) -1;
227static DeviceNode* arg_extra_nodes = NULL;
228static size_t arg_n_extra_nodes = 0;
229static char **arg_sysctl = NULL;
230static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
88213476 231
6145bb4f
LP		 232STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
233STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
234STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
235STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
236STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
237STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
238STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
239STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
240STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
241STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
242STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
243STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
244STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
245STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
246STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
247STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
248STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
249STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
250STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
251STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
252STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
253STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
254STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
255STATIC_DESTRUCTOR_REGISTER(arg_root_hash, freep);
256STATIC_DESTRUCTOR_REGISTER(arg_syscall_whitelist, strv_freep);
257STATIC_DESTRUCTOR_REGISTER(arg_syscall_blacklist, strv_freep);
258#if HAVE_SECCOMP
259STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
260#endif
0985c7c4 261STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
6145bb4f
LP		 262STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
263
37ec0fdd
LP		 264static int help(void) {
265 _cleanup_free_ char *link = NULL;
266 int r;
267
bb068de0 268 (void) pager_open(arg_pager_flags);
50ebcf6c 269
37ec0fdd
LP		 270 r = terminal_urlify_man("systemd-nspawn", "1", &link);
271 if (r < 0)
272 return log_oom();
273
25148653 274 printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
a7e2e50d 275 "Spawn a command or OS in a light-weight container.\n\n"
a8828ed9
DW		 276 " -h --help Show this help\n"
277 " --version Print version string\n"
69c79d3c 278 " -q --quiet Do not show status information\n"
bb068de0 279 " --no-pager Do not pipe output into a pager\n"
25148653
LP		 280 " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
281 "%3$sImage:%4$s\n"
1b9e5b12 282 " -D --directory=PATH Root directory for the container\n"
ec16945e
LP		 283 " --template=PATH Initialize root directory from template directory,\n"
284 " if missing\n"
285 " -x --ephemeral Run container with snapshot of root directory, and\n"
286 " remove it after exit\n"
25e68fd3
LP		 287 " -i --image=PATH Root file system disk image (or device node) for\n"
288 " the container\n"
de40a303 289 " --oci-bundle=PATH OCI bundle directory\n"
25148653
LP		 290 " --read-only Mount the root directory read-only\n"
291 " --volatile[=MODE] Run the system in volatile mode\n"
25e68fd3 292 " --root-hash=HASH Specify verity root hash for root disk image\n"
25148653
LP		 293 " --pivot-root=PATH[:PATH]\n"
294 " Pivot root to given directory in the container\n\n"
295 "%3$sExecution:%4$s\n"
7732f92b 296 " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
a8828ed9 297 " -b --boot Boot up full system (i.e. invoke init)\n"
5f932eb9 298 " --chdir=PATH Set working directory in the container\n"
25148653
LP		 299 " -E --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
300 " -u --user=USER Run the command under specified user or UID\n"
301 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
302 " --notify-ready=BOOLEAN Receive notifications from the child init process\n\n"
303 "%3$sSystem Identity:%4$s\n"
a8828ed9 304 " -M --machine=NAME Set the machine name for the container\n"
3a9530e5 305 " --hostname=NAME Override the hostname for the container\n"
25148653
LP		 306 " --uuid=UUID Set a specific machine UUID for the container\n\n"
307 "%3$sProperties:%4$s\n"
a8828ed9 308 " -S --slice=SLICE Place the container in the specified slice\n"
f36933fe 309 " --property=NAME=VALUE Set scope unit property\n"
25148653
LP		 310 " --register=BOOLEAN Register container as machine\n"
311 " --keep-unit Do not register a scope for the machine, reuse\n"
312 " the service unit nspawn is running in\n\n"
313 "%3$sUser Namespacing:%4$s\n"
90b4a64d 314 " -U --private-users=pick Run within user namespace, autoselect UID/GID range\n"
03cfe0d5 315 " --private-users[=UIDBASE[:NUIDS]]\n"
90b4a64d 316 " Similar, but with user configured UID/GID range\n"
25148653
LP		 317 " --private-users-chown Adjust OS tree ownership to private UID/GID range\n\n"
318 "%3$sNetworking:%4$s\n"
69c79d3c
LP		 319 " --private-network Disable network in container\n"
320 " --network-interface=INTERFACE\n"
321 " Assign an existing network interface to the\n"
322 " container\n"
c74e630d
LP		 323 " --network-macvlan=INTERFACE\n"
324 " Create a macvlan network interface based on an\n"
325 " existing network interface to the container\n"
4bbfe7ad
TG		 326 " --network-ipvlan=INTERFACE\n"
327 " Create a ipvlan network interface based on an\n"
328 " existing network interface to the container\n"
a8eaaee7 329 " -n --network-veth Add a virtual Ethernet connection between host\n"
69c79d3c 330 " and container\n"
f6d6bad1
LP		 331 " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
332 " Add an additional virtual Ethernet link between\n"
333 " host and container\n"
ab046dde 334 " --network-bridge=INTERFACE\n"
90b4a64d
ZJS		 335 " Add a virtual Ethernet connection to the container\n"
336 " and attach it to an existing bridge on the host\n"
337 " --network-zone=NAME Similar, but attach the new interface to an\n"
338 " an automatically managed bridge interface\n"
d7bea6b6
DP		 339 " --network-namespace-path=PATH\n"
340 " Set network namespace to the one represented by\n"
341 " the specified kernel namespace file node\n"
6d0b55c2 342 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
25148653
LP		 343 " Expose a container IP port on the host\n\n"
344 "%3$sSecurity:%4$s\n"
a8828ed9
DW		 345 " --capability=CAP In addition to the default, retain specified\n"
346 " capability\n"
347 " --drop-capability=CAP Drop the specified capability from the default set\n"
f4e803c8 348 " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
960e4569
LP		 349 " --system-call-filter=LIST|~LIST\n"
350 " Permit/prohibit specific system calls\n"
25148653
LP		 351 " -Z --selinux-context=SECLABEL\n"
352 " Set the SELinux security context to be used by\n"
353 " processes in the container\n"
354 " -L --selinux-apifs-context=SECLABEL\n"
355 " Set the SELinux security context to be used by\n"
356 " API/tmpfs file systems in the container\n\n"
357 "%3$sResources:%4$s\n"
bf428efb 358 " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
81f345df
LP		 359 " --oom-score-adjust=VALUE\n"
360 " Adjust the OOM score value for the payload\n"
f4e803c8
LP		 361 " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
362 " --personality=ARCH Pick personality for this container\n\n"
25148653 363 "%3$sIntegration:%4$s\n"
09d423e9 364 " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
1688841f 365 " --timezone=MODE Select mode of /etc/localtime initialization\n"
25148653
LP		 366 " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
367 " host, try-guest, try-host\n"
368 " -j Equivalent to --link-journal=try-guest\n\n"
369 "%3$sMounts:%4$s\n"
5e5bfa6e
EY		 370 " --bind=PATH[:PATH[:OPTIONS]]\n"
371 " Bind mount a file or directory from the host into\n"
a8828ed9 372 " the container\n"
5e5bfa6e
EY		 373 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
374 " Similar, but creates a read-only bind mount\n"
de40a303
LP		 375 " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
376 " it\n"
06c17c39 377 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
5a8af538
LP		 378 " --overlay=PATH[:PATH...]:PATH\n"
379 " Create an overlay mount from the host to \n"
380 " the container\n"
381 " --overlay-ro=PATH[:PATH...]:PATH\n"
25148653
LP		 382 " Similar, but creates a read-only overlay mount\n\n"
383 "%3$sInput/Output:%4$s\n"
de40a303
LP		 384 " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
385 " set up for the container.\n"
386 " -P --pipe Equivalent to --console=pipe\n"
25148653 387 "\nSee the %2$s for details.\n"
37ec0fdd
LP		 388 , program_invocation_short_name
389 , link
25148653 390 , ansi_underline(), ansi_normal());
37ec0fdd
LP		 391
392 return 0;
88213476
LP		 393}
394
86c0dd4a 395static int custom_mount_check_all(void) {
88614c8a 396 size_t i;
5a8af538 397
5a8af538
LP		 398 for (i = 0; i < arg_n_custom_mounts; i++) {
399 CustomMount *m = &arg_custom_mounts[i];
400
0de7acce 401 if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
baaa35ad
ZJS		 402 if (arg_userns_chown)
403 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
404 "--private-users-chown may not be combined with custom root mounts.");
405 else if (arg_uid_shift == UID_INVALID)
406 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
407 "--private-users with automatic UID shift may not be combined with custom root mounts.");
825d5287 408 }
5a8af538
LP		 409 }
410
411 return 0;
412}
413
8199d554 414static int detect_unified_cgroup_hierarchy_from_environment(void) {
efdb0237 415 const char *e;
415fc41c 416 int r;
5da38d07 417
efdb0237
LP		 418 /* Allow the user to control whether the unified hierarchy is used */
419 e = getenv("UNIFIED_CGROUP_HIERARCHY");
420 if (e) {
421 r = parse_boolean(e);
422 if (r < 0)
423 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
5da38d07
TH		 424 if (r > 0)
425 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
426 else
427 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
efdb0237
LP		 428 }
429
8199d554
LP		 430 return 0;
431}
432
433static 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. */
b4cccbc1
LP		 438 r = cg_all_unified();
439 if (r < 0)
440 return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
441 if (r > 0) {
a8725a06
ZJS		 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 r = systemd_installation_has_version(directory, 230);
445 if (r < 0)
446 return log_error_errno(r, "Failed to determine systemd version in container: %m");
447 if (r > 0)
448 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
449 else
450 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
c22800e4 451 } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
2977724b
TH		 452 /* Mixed cgroup hierarchy support was added in 233 */
453 r = systemd_installation_has_version(directory, 233);
0fd9563f
ZJS		 454 if (r < 0)
455 return log_error_errno(r, "Failed to determine systemd version in container: %m");
456 if (r > 0)
457 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
458 else
459 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
460 } else
5da38d07 461 arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
efdb0237 462
8199d554
LP		 463 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
efdb0237
LP		 467 return 0;
468}
469
0c582db0
LB		 470static void parse_share_ns_env(const char *name, unsigned long ns_flag) {
471 int r;
472
473 r = getenv_bool(name);
474 if (r == -ENXIO)
475 return;
476 if (r < 0)
477 log_warning_errno(r, "Failed to parse %s from environment, defaulting to false.", name);
de40a303 478
0c582db0 479 arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
de40a303 480 arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
0c582db0
LB		 481}
482
4f086aab 483static void parse_mount_settings_env(void) {
4f086aab 484 const char *e;
1099ceeb
LP		 485 int r;
486
487 r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
488 if (r >= 0)
489 SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
490 else if (r != -ENXIO)
491 log_warning_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP, ignoring: %m");
4f086aab
SU		 492
493 e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
494 if (!e)
495 return;
496
497 if (streq(e, "network")) {
498 arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
499 return;
500 }
501
502 r = parse_boolean(e);
503 if (r < 0) {
504 log_warning_errno(r, "Failed to parse SYSTEMD_NSPAWN_API_VFS_WRITABLE from environment, ignoring.");
505 return;
ab8ee0f2 506 }
4f086aab 507
ab8ee0f2
ZJS		 508 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
509 SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
4f086aab
SU		 510}
511
d5455d2f
LP		 512static void parse_environment(void) {
513 const char *e;
514 int r;
515
516 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
517 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
518 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
519 parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
520
521 parse_mount_settings_env();
522
489fae52
ZJS		 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. */
de40a303 525 if (!cg_ns_supported())
489fae52 526 arg_use_cgns = false;
de40a303
LP		 527 else {
528 r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
529 if (r < 0) {
530 if (r != -ENXIO)
531 log_warning_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS, ignoring: %m");
532
533 arg_use_cgns = true;
534 } else {
535 arg_use_cgns = r > 0;
536 arg_settings_mask |= SETTING_USE_CGNS;
537 }
538 }
d5455d2f
LP		 539
540 e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
541 if (e)
542 arg_container_service_name = e;
543
544 detect_unified_cgroup_hierarchy_from_environment();
545}
546
88213476 547static int parse_argv(int argc, char *argv[]) {
a41fe3a2 548 enum {
acbeb427
ZJS		 549 ARG_VERSION = 0x100,
550 ARG_PRIVATE_NETWORK,
bc2f673e 551 ARG_UUID,
5076f0cc 552 ARG_READ_ONLY,
57fb9fb5 553 ARG_CAPABILITY,
420c7379 554 ARG_DROP_CAPABILITY,
17fe0523
LP		 555 ARG_LINK_JOURNAL,
556 ARG_BIND,
f4889f65 557 ARG_BIND_RO,
06c17c39 558 ARG_TMPFS,
5a8af538
LP		 559 ARG_OVERLAY,
560 ARG_OVERLAY_RO,
de40a303 561 ARG_INACCESSIBLE,
eb91eb18 562 ARG_SHARE_SYSTEM,
89f7c846 563 ARG_REGISTER,
aa28aefe 564 ARG_KEEP_UNIT,
69c79d3c 565 ARG_NETWORK_INTERFACE,
c74e630d 566 ARG_NETWORK_MACVLAN,
4bbfe7ad 567 ARG_NETWORK_IPVLAN,
ab046dde 568 ARG_NETWORK_BRIDGE,
22b28dfd 569 ARG_NETWORK_ZONE,
f6d6bad1 570 ARG_NETWORK_VETH_EXTRA,
d7bea6b6 571 ARG_NETWORK_NAMESPACE_PATH,
6afc95b7 572 ARG_PERSONALITY,
4d9f07b4 573 ARG_VOLATILE,
ec16945e 574 ARG_TEMPLATE,
f36933fe 575 ARG_PROPERTY,
6dac160c 576 ARG_PRIVATE_USERS,
c6c8f6e2 577 ARG_KILL_SIGNAL,
f757855e 578 ARG_SETTINGS,
5f932eb9 579 ARG_CHDIR,
b53ede69 580 ARG_PIVOT_ROOT,
7336138e 581 ARG_PRIVATE_USERS_CHOWN,
9c1e04d0 582 ARG_NOTIFY_READY,
4623e8e6 583 ARG_ROOT_HASH,
960e4569 584 ARG_SYSTEM_CALL_FILTER,
bf428efb 585 ARG_RLIMIT,
3a9530e5 586 ARG_HOSTNAME,
66edd963 587 ARG_NO_NEW_PRIVILEGES,
81f345df 588 ARG_OOM_SCORE_ADJUST,
d107bb7d 589 ARG_CPU_AFFINITY,
09d423e9 590 ARG_RESOLV_CONF,
1688841f 591 ARG_TIMEZONE,
de40a303
LP		 592 ARG_CONSOLE,
593 ARG_PIPE,
594 ARG_OCI_BUNDLE,
bb068de0 595 ARG_NO_PAGER,
a41fe3a2
LP		 596 };
597
88213476 598 static const struct option options[] = {
d7bea6b6
DP		 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 },
66edd963 612 { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
d7bea6b6
DP		 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 },
de40a303 619 { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
d7bea6b6 620 { "machine", required_argument, NULL, 'M' },
3a9530e5 621 { "hostname", required_argument, NULL, ARG_HOSTNAME },
d7bea6b6
DP		 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 },
bf428efb 652 { "rlimit", required_argument, NULL, ARG_RLIMIT },
81f345df 653 { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
d107bb7d 654 { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
09d423e9 655 { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
1688841f 656 { "timezone", required_argument, NULL, ARG_TIMEZONE },
de40a303
LP		 657 { "console", required_argument, NULL, ARG_CONSOLE },
658 { "pipe", no_argument, NULL, ARG_PIPE },
659 { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
bb068de0 660 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
eb9da376 661 {}
88213476
LP		 662 };
663
9444b1f2 664 int c, r;
d5455d2f 665 const char *p;
a42c8b54 666 uint64_t plus = 0, minus = 0;
f757855e 667 bool mask_all_settings = false, mask_no_settings = false;
88213476
LP		 668
669 assert(argc >= 0);
670 assert(argv);
671
de40a303 672 while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
88213476
LP		 673 switch (c) {
674
675 case 'h':
37ec0fdd 676 return help();
88213476 677
acbeb427 678 case ARG_VERSION:
3f6fd1ba 679 return version();
acbeb427 680
88213476 681 case 'D':
0f03c2a4 682 r = parse_path_argument_and_warn(optarg, false, &arg_directory);
ec16945e 683 if (r < 0)
0f03c2a4 684 return r;
de40a303
LP		 685
686 arg_settings_mask |= SETTING_DIRECTORY;
ec16945e
LP		 687 break;
688
689 case ARG_TEMPLATE:
0f03c2a4 690 r = parse_path_argument_and_warn(optarg, false, &arg_template);
ec16945e 691 if (r < 0)
0f03c2a4 692 return r;
de40a303
LP		 693
694 arg_settings_mask |= SETTING_DIRECTORY;
88213476
LP		 695 break;
696
1b9e5b12 697 case 'i':
0f03c2a4 698 r = parse_path_argument_and_warn(optarg, false, &arg_image);
ec16945e 699 if (r < 0)
0f03c2a4 700 return r;
de40a303
LP		 701
702 arg_settings_mask |= SETTING_DIRECTORY;
703 break;
704
705 case ARG_OCI_BUNDLE:
706 r = parse_path_argument_and_warn(optarg, false, &arg_oci_bundle);
707 if (r < 0)
708 return r;
709
ec16945e
LP		 710 break;
711
712 case 'x':
713 arg_ephemeral = true;
a2f577fc 714 arg_settings_mask |= SETTING_EPHEMERAL;
1b9e5b12
LP		 715 break;
716
687d0825 717 case 'u':
2fc09a9c
DM		 718 r = free_and_strdup(&arg_user, optarg);
719 if (r < 0)
7027ff61 720 return log_oom();
687d0825 721
f757855e 722 arg_settings_mask |= SETTING_USER;
687d0825
MV		 723 break;
724
22b28dfd
LP		 725 case ARG_NETWORK_ZONE: {
726 char *j;
727
b910cc72 728 j = strjoin("vz-", optarg);
22b28dfd
LP		 729 if (!j)
730 return log_oom();
731
732 if (!ifname_valid(j)) {
733 log_error("Network zone name not valid: %s", j);
734 free(j);
735 return -EINVAL;
736 }
737
df1fac6d 738 free_and_replace(arg_network_zone, j);
22b28dfd
LP		 739
740 arg_network_veth = true;
741 arg_private_network = true;
742 arg_settings_mask |= SETTING_NETWORK;
743 break;
744 }
745
ab046dde 746 case ARG_NETWORK_BRIDGE:
ef76dff2 747
baaa35ad
ZJS		 748 if (!ifname_valid(optarg))
749 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
750 "Bridge interface name not valid: %s", optarg);
ef76dff2 751
f757855e
LP		 752 r = free_and_strdup(&arg_network_bridge, optarg);
753 if (r < 0)
754 return log_oom();
ab046dde 755
4831981d 756 _fallthrough_;
0dfaa006 757 case 'n':
69c79d3c
LP		 758 arg_network_veth = true;
759 arg_private_network = true;
f757855e 760 arg_settings_mask |= SETTING_NETWORK;
69c79d3c
LP		 761 break;
762
f6d6bad1
LP		 763 case ARG_NETWORK_VETH_EXTRA:
764 r = veth_extra_parse(&arg_network_veth_extra, optarg);
765 if (r < 0)
766 return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
767
768 arg_private_network = true;
769 arg_settings_mask |= SETTING_NETWORK;
770 break;
771
aa28aefe 772 case ARG_NETWORK_INTERFACE:
baaa35ad
ZJS		 773 if (!ifname_valid(optarg))
774 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
775 "Network interface name not valid: %s", optarg);
ef76dff2 776
c74e630d
LP		 777 if (strv_extend(&arg_network_interfaces, optarg) < 0)
778 return log_oom();
779
780 arg_private_network = true;
f757855e 781 arg_settings_mask |= SETTING_NETWORK;
c74e630d
LP		 782 break;
783
784 case ARG_NETWORK_MACVLAN:
ef76dff2 785
baaa35ad
ZJS		 786 if (!ifname_valid(optarg))
787 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
788 "MACVLAN network interface name not valid: %s", optarg);
ef76dff2 789
c74e630d 790 if (strv_extend(&arg_network_macvlan, optarg) < 0)
aa28aefe
LP		 791 return log_oom();
792
4bbfe7ad 793 arg_private_network = true;
f757855e 794 arg_settings_mask |= SETTING_NETWORK;
4bbfe7ad
TG		 795 break;
796
797 case ARG_NETWORK_IPVLAN:
ef76dff2 798
baaa35ad
ZJS		 799 if (!ifname_valid(optarg))
800 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
801 "IPVLAN network interface name not valid: %s", optarg);
ef76dff2 802
4bbfe7ad
TG		 803 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
804 return log_oom();
805
4831981d 806 _fallthrough_;
ff01d048
LP		 807 case ARG_PRIVATE_NETWORK:
808 arg_private_network = true;
f757855e 809 arg_settings_mask |= SETTING_NETWORK;
a41fe3a2
LP		 810 break;
811
d7bea6b6
DP		 812 case ARG_NETWORK_NAMESPACE_PATH:
813 r = parse_path_argument_and_warn(optarg, false, &arg_network_namespace_path);
814 if (r < 0)
815 return r;
816
de40a303 817 arg_settings_mask |= SETTING_NETWORK;
d7bea6b6
DP		 818 break;
819
0f0dbc46 820 case 'b':
baaa35ad
ZJS		 821 if (arg_start_mode == START_PID2)
822 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
823 "--boot and --as-pid2 may not be combined.");
7732f92b
LP		 824
825 arg_start_mode = START_BOOT;
826 arg_settings_mask |= SETTING_START_MODE;
827 break;
828
829 case 'a':
baaa35ad
ZJS		 830 if (arg_start_mode == START_BOOT)
831 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
832 "--boot and --as-pid2 may not be combined.");
7732f92b
LP		 833
834 arg_start_mode = START_PID2;
835 arg_settings_mask |= SETTING_START_MODE;
0f0dbc46
LP		 836 break;
837
144f0fc0 838 case ARG_UUID:
9444b1f2 839 r = sd_id128_from_string(optarg, &arg_uuid);
317feb4d
LP		 840 if (r < 0)
841 return log_error_errno(r, "Invalid UUID: %s", optarg);
842
baaa35ad
ZJS		 843 if (sd_id128_is_null(arg_uuid))
844 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
845 "Machine UUID may not be all zeroes.");
f757855e
LP		 846
847 arg_settings_mask |= SETTING_MACHINE_ID;
9444b1f2 848 break;
aa96c6cb 849
9444b1f2 850 case 'S':
de40a303
LP		 851 r = free_and_strdup(&arg_slice, optarg);
852 if (r < 0)
853 return log_oom();
854
855 arg_settings_mask |= SETTING_SLICE;
144f0fc0
LP		 856 break;
857
7027ff61 858 case 'M':
c1521918 859 if (isempty(optarg))
97b11eed 860 arg_machine = mfree(arg_machine);
c1521918 861 else {
baaa35ad
ZJS		 862 if (!machine_name_is_valid(optarg))
863 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
864 "Invalid machine name: %s", optarg);
7027ff61 865
0c3c4284
LP		 866 r = free_and_strdup(&arg_machine, optarg);
867 if (r < 0)
eb91eb18 868 return log_oom();
eb91eb18 869 }
9ce6d1b3 870 break;
7027ff61 871
3a9530e5
LP		 872 case ARG_HOSTNAME:
873 if (isempty(optarg))
874 arg_hostname = mfree(arg_hostname);
875 else {
baaa35ad
ZJS		 876 if (!hostname_is_valid(optarg, false))
877 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
878 "Invalid hostname: %s", optarg);
3a9530e5
LP		 879
880 r = free_and_strdup(&arg_hostname, optarg);
881 if (r < 0)
882 return log_oom();
883 }
884
885 arg_settings_mask |= SETTING_HOSTNAME;
886 break;
887
82adf6af
LP		 888 case 'Z':
889 arg_selinux_context = optarg;
a8828ed9
DW		 890 break;
891
82adf6af
LP		 892 case 'L':
893 arg_selinux_apifs_context = optarg;
a8828ed9
DW		 894 break;
895
bc2f673e
LP		 896 case ARG_READ_ONLY:
897 arg_read_only = true;
f757855e 898 arg_settings_mask |= SETTING_READ_ONLY;
bc2f673e
LP		 899 break;
900
420c7379
LP		 901 case ARG_CAPABILITY:
902 case ARG_DROP_CAPABILITY: {
6cbe4ed1 903 p = optarg;
9ed794a3 904 for (;;) {
6cbe4ed1 905 _cleanup_free_ char *t = NULL;
5076f0cc 906
6cbe4ed1
SS		 907 r = extract_first_word(&p, &t, ",", 0);
908 if (r < 0)
909 return log_error_errno(r, "Failed to parse capability %s.", t);
6cbe4ed1
SS		 910 if (r == 0)
911 break;
5076f0cc 912
39ed67d1
LP		 913 if (streq(t, "all")) {
914 if (c == ARG_CAPABILITY)
a42c8b54 915 plus = (uint64_t) -1;
39ed67d1 916 else
a42c8b54 917 minus = (uint64_t) -1;
39ed67d1 918 } else {
acf4d158
YW		 919 r = capability_from_name(t);
920 if (r < 0)
921 return log_error_errno(r, "Failed to parse capability %s.", t);
39ed67d1
LP		 922
923 if (c == ARG_CAPABILITY)
acf4d158 924 plus |= 1ULL << r;
39ed67d1 925 else
acf4d158 926 minus |= 1ULL << r;
5076f0cc 927 }
5076f0cc
LP		 928 }
929
f757855e 930 arg_settings_mask |= SETTING_CAPABILITY;
5076f0cc
LP		 931 break;
932 }
933
66edd963
LP		 934 case ARG_NO_NEW_PRIVILEGES:
935 r = parse_boolean(optarg);
936 if (r < 0)
937 return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
938
939 arg_no_new_privileges = r;
940 arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
941 break;
942
57fb9fb5
LP		 943 case 'j':
944 arg_link_journal = LINK_GUEST;
574edc90 945 arg_link_journal_try = true;
4e1d6aa9 946 arg_settings_mask |= SETTING_LINK_JOURNAL;
57fb9fb5
LP		 947 break;
948
949 case ARG_LINK_JOURNAL:
4e1d6aa9 950 r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
c6147113
LP		 951 if (r < 0)
952 return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
57fb9fb5 953
4e1d6aa9 954 arg_settings_mask |= SETTING_LINK_JOURNAL;
57fb9fb5
LP		 955 break;
956
17fe0523 957 case ARG_BIND:
f757855e
LP		 958 case ARG_BIND_RO:
959 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
960 if (r < 0)
961 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
17fe0523 962
f757855e 963 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
17fe0523 964 break;
06c17c39 965
f757855e
LP		 966 case ARG_TMPFS:
967 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
968 if (r < 0)
969 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
5a8af538 970
f757855e 971 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
5a8af538 972 break;
5a8af538
LP		 973
974 case ARG_OVERLAY:
ad85779a
LP		 975 case ARG_OVERLAY_RO:
976 r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
977 if (r == -EADDRNOTAVAIL)
978 return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
979 if (r < 0)
980 return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
06c17c39 981
f757855e 982 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
06c17c39 983 break;
06c17c39 984
de40a303
LP		 985 case ARG_INACCESSIBLE:
986 r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
987 if (r < 0)
988 return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
989
990 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
991 break;
992
a5f1cb3b 993 case 'E': {
f4889f65
LP		 994 char **n;
995
baaa35ad
ZJS		 996 if (!env_assignment_is_valid(optarg))
997 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
998 "Environment variable assignment '%s' is not valid.", optarg);
f4889f65
LP		 999
1000 n = strv_env_set(arg_setenv, optarg);
1001 if (!n)
1002 return log_oom();
1003
130d3d22 1004 strv_free_and_replace(arg_setenv, n);
f757855e 1005 arg_settings_mask |= SETTING_ENVIRONMENT;
f4889f65
LP		 1006 break;
1007 }
1008
284c0b91
LP		 1009 case 'q':
1010 arg_quiet = true;
1011 break;
1012
8a96d94e 1013 case ARG_SHARE_SYSTEM:
a6b5216c 1014 /* We don't officially support this anymore, except for compat reasons. People should use the
0c582db0 1015 * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
63d1c29f 1016 log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
0c582db0 1017 arg_clone_ns_flags = 0;
8a96d94e
LP		 1018 break;
1019
eb91eb18
LP		 1020 case ARG_REGISTER:
1021 r = parse_boolean(optarg);
1022 if (r < 0) {
1023 log_error("Failed to parse --register= argument: %s", optarg);
1024 return r;
1025 }
1026
1027 arg_register = r;
1028 break;
1029
89f7c846
LP		 1030 case ARG_KEEP_UNIT:
1031 arg_keep_unit = true;
1032 break;
1033
6afc95b7
LP		 1034 case ARG_PERSONALITY:
1035
ac45f971 1036 arg_personality = personality_from_string(optarg);
baaa35ad
ZJS		 1037 if (arg_personality == PERSONALITY_INVALID)
1038 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1039 "Unknown or unsupported personality '%s'.", optarg);
6afc95b7 1040
f757855e 1041 arg_settings_mask |= SETTING_PERSONALITY;
6afc95b7
LP		 1042 break;
1043
4d9f07b4
LP		 1044 case ARG_VOLATILE:
1045
1046 if (!optarg)
f757855e 1047 arg_volatile_mode = VOLATILE_YES;
5c828e66
LP		 1048 else if (streq(optarg, "help")) {
1049 DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
1050 return 0;
1051 } else {
f757855e 1052 VolatileMode m;
4d9f07b4 1053
f757855e 1054 m = volatile_mode_from_string(optarg);
baaa35ad
ZJS		 1055 if (m < 0)
1056 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1057 "Failed to parse --volatile= argument: %s", optarg);
1058 else
f757855e 1059 arg_volatile_mode = m;
6d0b55c2
LP		 1060 }
1061
f757855e
LP		 1062 arg_settings_mask |= SETTING_VOLATILE_MODE;
1063 break;
6d0b55c2 1064
f757855e
LP		 1065 case 'p':
1066 r = expose_port_parse(&arg_expose_ports, optarg);
1067 if (r == -EEXIST)
1068 return log_error_errno(r, "Duplicate port specification: %s", optarg);
1069 if (r < 0)
1070 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
6d0b55c2 1071
f757855e 1072 arg_settings_mask |= SETTING_EXPOSE_PORTS;
6d0b55c2 1073 break;
6d0b55c2 1074
f36933fe
LP		 1075 case ARG_PROPERTY:
1076 if (strv_extend(&arg_property, optarg) < 0)
1077 return log_oom();
1078
1079 break;
1080
ae209204
ZJS		 1081 case ARG_PRIVATE_USERS: {
1082 int boolean = -1;
0de7acce 1083
ae209204
ZJS		 1084 if (!optarg)
1085 boolean = true;
1086 else if (!in_charset(optarg, DIGITS))
1087 /* do *not* parse numbers as booleans */
1088 boolean = parse_boolean(optarg);
1089
1090 if (boolean == false) {
0de7acce
LP		 1091 /* no: User namespacing off */
1092 arg_userns_mode = USER_NAMESPACE_NO;
1093 arg_uid_shift = UID_INVALID;
1094 arg_uid_range = UINT32_C(0x10000);
ae209204 1095 } else if (boolean == true) {
0de7acce
LP		 1096 /* yes: User namespacing on, UID range is read from root dir */
1097 arg_userns_mode = USER_NAMESPACE_FIXED;
1098 arg_uid_shift = UID_INVALID;
1099 arg_uid_range = UINT32_C(0x10000);
1100 } else if (streq(optarg, "pick")) {
1101 /* pick: User namespacing on, UID range is picked randomly */
1102 arg_userns_mode = USER_NAMESPACE_PICK;
1103 arg_uid_shift = UID_INVALID;
1104 arg_uid_range = UINT32_C(0x10000);
1105 } else {
6c2058b3 1106 _cleanup_free_ char *buffer = NULL;
6dac160c
LP		 1107 const char *range, *shift;
1108
0de7acce
LP		 1109 /* anything else: User namespacing on, UID range is explicitly configured */
1110
6dac160c
LP		 1111 range = strchr(optarg, ':');
1112 if (range) {
6c2058b3
ZJS		 1113 buffer = strndup(optarg, range - optarg);
1114 if (!buffer)
1115 return log_oom();
1116 shift = buffer;
6dac160c
LP		 1117
1118 range++;
bfd292ec
ZJS		 1119 r = safe_atou32(range, &arg_uid_range);
1120 if (r < 0)
be715731 1121 return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
6dac160c
LP		 1122 } else
1123 shift = optarg;
1124
be715731
ZJS		 1125 r = parse_uid(shift, &arg_uid_shift);
1126 if (r < 0)
1127 return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
0de7acce
LP		 1128
1129 arg_userns_mode = USER_NAMESPACE_FIXED;
6dac160c
LP		 1130 }
1131
baaa35ad
ZJS		 1132 if (arg_uid_range <= 0)
1133 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1134 "UID range cannot be 0.");
be715731 1135
0de7acce 1136 arg_settings_mask |= SETTING_USERNS;
6dac160c 1137 break;
ae209204 1138 }
6dac160c 1139
0de7acce 1140 case 'U':
ccabee0d
LP		 1141 if (userns_supported()) {
1142 arg_userns_mode = USER_NAMESPACE_PICK;
1143 arg_uid_shift = UID_INVALID;
1144 arg_uid_range = UINT32_C(0x10000);
1145
1146 arg_settings_mask |= SETTING_USERNS;
6dac160c
LP		 1147 }
1148
7336138e
LP		 1149 break;
1150
0de7acce 1151 case ARG_PRIVATE_USERS_CHOWN:
19aac838 1152 arg_userns_chown = true;
0de7acce
LP		 1153
1154 arg_settings_mask |= SETTING_USERNS;
6dac160c
LP		 1155 break;
1156
c6c8f6e2 1157 case ARG_KILL_SIGNAL:
5c828e66
LP		 1158 if (streq(optarg, "help")) {
1159 DUMP_STRING_TABLE(signal, int, _NSIG);
1160 return 0;
1161 }
1162
29a3db75 1163 arg_kill_signal = signal_from_string(optarg);
baaa35ad
ZJS		 1164 if (arg_kill_signal < 0)
1165 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1166 "Cannot parse signal: %s", optarg);
c6c8f6e2 1167
f757855e
LP		 1168 arg_settings_mask |= SETTING_KILL_SIGNAL;
1169 break;
1170
1171 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 r = parse_boolean(optarg);
1180 if (r < 0) {
1181 if (streq(optarg, "trusted")) {
1182 mask_all_settings = false;
1183 mask_no_settings = false;
1184 arg_settings_trusted = true;
1185
1186 } else if (streq(optarg, "override")) {
1187 mask_all_settings = false;
1188 mask_no_settings = true;
1189 arg_settings_trusted = -1;
1190 } else
1191 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
1192 } else if (r > 0) {
1193 /* yes */
1194 mask_all_settings = false;
1195 mask_no_settings = false;
1196 arg_settings_trusted = -1;
1197 } else {
1198 /* no */
1199 mask_all_settings = true;
1200 mask_no_settings = false;
1201 arg_settings_trusted = false;
1202 }
1203
c6c8f6e2
LP		 1204 break;
1205
5f932eb9 1206 case ARG_CHDIR:
baaa35ad
ZJS		 1207 if (!path_is_absolute(optarg))
1208 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1209 "Working directory %s is not an absolute path.", optarg);
5f932eb9
LP		 1210
1211 r = free_and_strdup(&arg_chdir, optarg);
1212 if (r < 0)
1213 return log_oom();
1214
1215 arg_settings_mask |= SETTING_WORKING_DIRECTORY;
1216 break;
1217
b53ede69
PW		 1218 case ARG_PIVOT_ROOT:
1219 r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
1220 if (r < 0)
1221 return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
1222
1223 arg_settings_mask |= SETTING_PIVOT_ROOT;
1224 break;
1225
9c1e04d0
AP		 1226 case ARG_NOTIFY_READY:
1227 r = parse_boolean(optarg);
baaa35ad
ZJS		 1228 if (r < 0)
1229 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1230 "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
9c1e04d0
AP		 1231 arg_notify_ready = r;
1232 arg_settings_mask |= SETTING_NOTIFY_READY;
1233 break;
1234
4623e8e6
LP		 1235 case ARG_ROOT_HASH: {
1236 void *k;
1237 size_t l;
1238
1239 r = unhexmem(optarg, strlen(optarg), &k, &l);
1240 if (r < 0)
1241 return log_error_errno(r, "Failed to parse root hash: %s", optarg);
1242 if (l < sizeof(sd_id128_t)) {
4623e8e6 1243 free(k);
c6147113 1244 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128bit long: %s", optarg);
4623e8e6
LP		 1245 }
1246
1247 free(arg_root_hash);
1248 arg_root_hash = k;
1249 arg_root_hash_size = l;
1250 break;
1251 }
1252
960e4569
LP		 1253 case ARG_SYSTEM_CALL_FILTER: {
1254 bool negative;
1255 const char *items;
1256
1257 negative = optarg[0] == '~';
1258 items = negative ? optarg + 1 : optarg;
1259
1260 for (;;) {
1261 _cleanup_free_ char *word = NULL;
1262
1263 r = extract_first_word(&items, &word, NULL, 0);
1264 if (r == 0)
1265 break;
1266 if (r == -ENOMEM)
1267 return log_oom();
1268 if (r < 0)
1269 return log_error_errno(r, "Failed to parse system call filter: %m");
1270
1271 if (negative)
1272 r = strv_extend(&arg_syscall_blacklist, word);
1273 else
1274 r = strv_extend(&arg_syscall_whitelist, word);
1275 if (r < 0)
1276 return log_oom();
1277 }
1278
1279 arg_settings_mask |= SETTING_SYSCALL_FILTER;
1280 break;
1281 }
1282
bf428efb
LP		 1283 case ARG_RLIMIT: {
1284 const char *eq;
622ecfa8 1285 _cleanup_free_ char *name = NULL;
bf428efb
LP		 1286 int rl;
1287
5c828e66
LP		 1288 if (streq(optarg, "help")) {
1289 DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
1290 return 0;
1291 }
1292
bf428efb 1293 eq = strchr(optarg, '=');
baaa35ad
ZJS		 1294 if (!eq)
1295 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1296 "--rlimit= expects an '=' assignment.");
bf428efb
LP		 1297
1298 name = strndup(optarg, eq - optarg);
1299 if (!name)
1300 return log_oom();
1301
1302 rl = rlimit_from_string_harder(name);
baaa35ad
ZJS		 1303 if (rl < 0)
1304 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1305 "Unknown resource limit: %s", name);
bf428efb
LP		 1306
1307 if (!arg_rlimit[rl]) {
1308 arg_rlimit[rl] = new0(struct rlimit, 1);
1309 if (!arg_rlimit[rl])
1310 return log_oom();
1311 }
1312
1313 r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
1314 if (r < 0)
1315 return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
1316
1317 arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
1318 break;
1319 }
1320
81f345df
LP		 1321 case ARG_OOM_SCORE_ADJUST:
1322 r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
1323 if (r < 0)
1324 return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
1325
1326 arg_oom_score_adjust_set = true;
1327 arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
1328 break;
1329
d107bb7d 1330 case ARG_CPU_AFFINITY: {
0985c7c4 1331 CPUSet cpuset;
d107bb7d
LP		 1332
1333 r = parse_cpu_set(optarg, &cpuset);
1334 if (r < 0)
0985c7c4 1335 return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
d107bb7d 1336
0985c7c4
ZJS		 1337 cpu_set_reset(&arg_cpu_set);
1338 arg_cpu_set = cpuset;
d107bb7d
LP		 1339 arg_settings_mask |= SETTING_CPU_AFFINITY;
1340 break;
1341 }
1342
09d423e9
LP		 1343 case ARG_RESOLV_CONF:
1344 if (streq(optarg, "help")) {
1345 DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
1346 return 0;
1347 }
1348
1349 arg_resolv_conf = resolv_conf_mode_from_string(optarg);
baaa35ad
ZJS		 1350 if (arg_resolv_conf < 0)
1351 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1352 "Failed to parse /etc/resolv.conf mode: %s", optarg);
09d423e9
LP		 1353
1354 arg_settings_mask |= SETTING_RESOLV_CONF;
1355 break;
1356
1688841f
LP		 1357 case ARG_TIMEZONE:
1358 if (streq(optarg, "help")) {
1359 DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
1360 return 0;
1361 }
1362
1363 arg_timezone = timezone_mode_from_string(optarg);
baaa35ad
ZJS		 1364 if (arg_timezone < 0)
1365 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1366 "Failed to parse /etc/localtime mode: %s", optarg);
1688841f
LP		 1367
1368 arg_settings_mask |= SETTING_TIMEZONE;
1369 break;
1370
de40a303
LP		 1371 case ARG_CONSOLE:
1372 if (streq(optarg, "interactive"))
1373 arg_console_mode = CONSOLE_INTERACTIVE;
1374 else if (streq(optarg, "read-only"))
1375 arg_console_mode = CONSOLE_READ_ONLY;
1376 else if (streq(optarg, "passive"))
1377 arg_console_mode = CONSOLE_PASSIVE;
1378 else if (streq(optarg, "pipe"))
1379 arg_console_mode = CONSOLE_PIPE;
1380 else if (streq(optarg, "help"))
1381 puts("interactive\n"
1382 "read-only\n"
1383 "passive\n"
1384 "pipe");
c6147113
LP		 1385 else
1386 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
de40a303
LP		 1387
1388 arg_settings_mask |= SETTING_CONSOLE_MODE;
1389 break;
1390
1391 case 'P':
1392 case ARG_PIPE:
1393 arg_console_mode = CONSOLE_PIPE;
1394 arg_settings_mask |= SETTING_CONSOLE_MODE;
1395 break;
1396
bb068de0
ZJS		 1397 case ARG_NO_PAGER:
1398 arg_pager_flags |= PAGER_DISABLE;
1399 break;
1400
88213476
LP		 1401 case '?':
1402 return -EINVAL;
1403
1404 default:
eb9da376 1405 assert_not_reached("Unhandled option");
88213476 1406 }
88213476 1407
60f1ec13
LP		 1408 if (argc > optind) {
1409 strv_free(arg_parameters);
1410 arg_parameters = strv_copy(argv + optind);
1411 if (!arg_parameters)
1412 return log_oom();
d7bea6b6 1413
60f1ec13
LP		 1414 arg_settings_mask |= SETTING_START_MODE;
1415 }
1416
1417 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 arg_directory = TAKE_PTR(arg_template);
1423
bd4b15f2 1424 arg_caps_retain = (arg_caps_retain | plus | (arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0)) & ~minus;
60f1ec13 1425
de40a303
LP		 1426 /* Make sure to parse environment before we reset the settings mask below */
1427 parse_environment();
1428
60f1ec13
LP		 1429 /* Load all settings from .nspawn files */
1430 if (mask_no_settings)
1431 arg_settings_mask = 0;
1432
1433 /* Don't load any settings from .nspawn files */
1434 if (mask_all_settings)
1435 arg_settings_mask = _SETTINGS_MASK_ALL;
1436
1437 return 1;
1438}
1439
1440static int verify_arguments(void) {
1441 int r;
a6b5216c 1442
4f086aab
SU		 1443 if (arg_userns_mode != USER_NAMESPACE_NO)
1444 arg_mount_settings |= MOUNT_USE_USERNS;
1445
1446 if (arg_private_network)
1447 arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
1448
48a8d337
LB		 1449 if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
1450 !(arg_clone_ns_flags & CLONE_NEWUTS)) {
eb91eb18 1451 arg_register = false;
baaa35ad 1452 if (arg_start_mode != START_PID1)
60f1ec13 1453 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
0c582db0 1454 }
eb91eb18 1455
0de7acce 1456 if (arg_userns_mode == USER_NAMESPACE_PICK)
0e7ac751
LP		 1457 arg_userns_chown = true;
1458
60f1ec13
LP		 1459 if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
1460 arg_kill_signal = SIGRTMIN+3;
1461
e5a4bb0d
LP		 1462 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 arg_read_only = true;
1464
baaa35ad 1465 if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
8d9c2bca
AJ		 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. */
60f1ec13 1468 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
89f7c846 1469
baaa35ad 1470 if (arg_directory && arg_image)
60f1ec13 1471 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
1b9e5b12 1472
baaa35ad 1473 if (arg_template && arg_image)
60f1ec13 1474 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
8cd328d8 1475
baaa35ad 1476 if (arg_template && !(arg_directory || arg_machine))
60f1ec13 1477 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
ec16945e 1478
baaa35ad 1479 if (arg_ephemeral && arg_template)
60f1ec13 1480 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
ec16945e 1481
baaa35ad 1482 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
60f1ec13 1483 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
df9a75e4 1484
baaa35ad 1485 if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
60f1ec13 1486 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
7336138e 1487
baaa35ad 1488 if (arg_userns_chown && arg_read_only)
de40a303
LP		 1489 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1490 "--read-only and --private-users-chown may not be combined.");
f757855e 1491
e5a4bb0d
LP		 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
5238e957 1494 * copy-up (in case of overlay) making the entire exercise pointless. */
e5a4bb0d
LP		 1495 if (arg_userns_chown && arg_volatile_mode != VOLATILE_NO)
1496 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-chown may not be combined.");
1497
de40a303
LP		 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. */
60f1ec13
LP		 1500 if (arg_network_namespace_path &&
1501 (arg_network_interfaces || arg_network_macvlan ||
1502 arg_network_ipvlan || arg_network_veth_extra ||
1503 arg_network_bridge || arg_network_zone ||
1504 arg_network_veth || arg_private_network))
de40a303 1505 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
86c0dd4a 1506
60f1ec13 1507 if (arg_network_bridge && arg_network_zone)
de40a303
LP		 1508 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1509 "--network-bridge= and --network-zone= may not be combined.");
f757855e 1510
baaa35ad 1511 if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
60f1ec13 1512 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
4f086aab 1513
baaa35ad 1514 if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
60f1ec13 1515 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
f757855e 1516
baaa35ad 1517 if (arg_expose_ports && !arg_private_network)
60f1ec13 1518 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
6d0b55c2 1519
349cc4a5 1520#if ! HAVE_LIBIPTC
baaa35ad 1521 if (arg_expose_ports)
60f1ec13 1522 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--port= is not supported, compiled without libiptc support.");
1c1ea217
EV		 1523#endif
1524
60f1ec13
LP		 1525 r = custom_mount_check_all();
1526 if (r < 0)
1527 return r;
c6c8f6e2 1528
f757855e 1529 return 0;
88213476
LP		 1530}
1531
03cfe0d5
LP		 1532static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1533 assert(p);
1534
0de7acce 1535 if (arg_userns_mode == USER_NAMESPACE_NO)
03cfe0d5
LP		 1536 return 0;
1537
1538 if (uid == UID_INVALID && gid == GID_INVALID)
1539 return 0;
1540
1541 if (uid != UID_INVALID) {
1542 uid += arg_uid_shift;
1543
1544 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1545 return -EOVERFLOW;
1546 }
1547
1548 if (gid != GID_INVALID) {
1549 gid += (gid_t) arg_uid_shift;
1550
1551 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1552 return -EOVERFLOW;
1553 }
1554
1555 if (lchown(p, uid, gid) < 0)
1556 return -errno;
b12afc8c
LP		 1557
1558 return 0;
1559}
1560
03cfe0d5
LP		 1561static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1562 const char *q;
dae8b82e 1563 int r;
03cfe0d5
LP		 1564
1565 q = prefix_roota(root, path);
dae8b82e
ZJS		 1566 r = mkdir_errno_wrapper(q, mode);
1567 if (r == -EEXIST)
1568 return 0;
1569 if (r < 0)
1570 return r;
03cfe0d5
LP		 1571
1572 return userns_lchown(q, uid, gid);
1573}
1574
1688841f 1575static const char *timezone_from_path(const char *path) {
da9fc98d
LP		 1576 return PATH_STARTSWITH_SET(
1577 path,
1578 "../usr/share/zoneinfo/",
1579 "/usr/share/zoneinfo/");
1688841f
LP		 1580}
1581
83205269
LP		 1582static bool etc_writable(void) {
1583 return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
1584}
1585
e58a1277 1586static int setup_timezone(const char *dest) {
1688841f
LP		 1587 _cleanup_free_ char *p = NULL, *etc = NULL;
1588 const char *where, *check;
1589 TimezoneMode m;
d4036145 1590 int r;
f8440af5 1591
e58a1277
LP		 1592 assert(dest);
1593
1688841f 1594 if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
1688841f
LP		 1595 r = readlink_malloc("/etc/localtime", &p);
1596 if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
83205269 1597 m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
1688841f 1598 else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
83205269 1599 m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
1688841f
LP		 1600 else if (r < 0) {
1601 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 return 0;
1609 } else if (arg_timezone == TIMEZONE_AUTO)
83205269 1610 m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
1688841f
LP		 1611 else
1612 m = arg_timezone;
1613 } else
1614 m = arg_timezone;
1615
1616 if (m == TIMEZONE_OFF)
1617 return 0;
1618
1619 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
d4036145 1620 if (r < 0) {
1688841f 1621 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
d4036145
LP		 1622 return 0;
1623 }
1624
1688841f
LP		 1625 where = strjoina(etc, "/localtime");
1626
1627 switch (m) {
1628
1629 case TIMEZONE_DELETE:
1630 if (unlink(where) < 0)
1631 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1632
d4036145 1633 return 0;
d4036145 1634
1688841f
LP		 1635 case TIMEZONE_SYMLINK: {
1636 _cleanup_free_ char *q = NULL;
1637 const char *z, *what;
4d1c38b8 1638
1688841f
LP		 1639 z = timezone_from_path(p);
1640 if (!z) {
1641 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
d4036145 1642 return 0;
1688841f 1643 }
d4036145 1644
1688841f
LP		 1645 r = readlink_malloc(where, &q);
1646 if (r >= 0 && streq_ptr(timezone_from_path(q), z))
1647 return 0; /* Already pointing to the right place? Then do nothing .. */
1648
1649 check = strjoina(dest, "/usr/share/zoneinfo/", z);
1650 r = chase_symlinks(check, dest, 0, NULL);
1651 if (r < 0)
1652 log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
1653 else {
1654 if (unlink(where) < 0 && errno != ENOENT) {
1655 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 return 0;
1658 }
1659
1660 what = strjoina("../usr/share/zoneinfo/", z);
1661 if (symlink(what, where) < 0) {
1662 log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
1663 errno, "Failed to correct timezone of container, ignoring: %m");
1664 return 0;
1665 }
1666
1667 break;
1668 }
1669
1670 _fallthrough_;
d4036145 1671 }
68fb0892 1672
1688841f
LP		 1673 case TIMEZONE_BIND: {
1674 _cleanup_free_ char *resolved = NULL;
1675 int found;
1676
1677 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1678 if (found < 0) {
1679 log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
1680 return 0;
1681 }
1682
1683 if (found == 0) /* missing? */
1684 (void) touch(resolved);
1685
1686 r = mount_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
1687 if (r >= 0)
1688 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
1689
1690 _fallthrough_;
79d80fc1 1691 }
4d9f07b4 1692
1688841f
LP		 1693 case TIMEZONE_COPY:
1694 /* If mounting failed, try to copy */
8a016c74 1695 r = copy_file_atomic("/etc/localtime", where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
1688841f
LP		 1696 if (r < 0) {
1697 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 return 0;
1700 }
1701
1702 break;
1703
1704 default:
1705 assert_not_reached("unexpected mode");
d4036145 1706 }
e58a1277 1707
1688841f 1708 /* Fix permissions of the symlink or file copy we just created */
03cfe0d5
LP		 1709 r = userns_lchown(where, 0, 0);
1710 if (r < 0)
1688841f 1711 log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
03cfe0d5 1712
e58a1277 1713 return 0;
88213476
LP		 1714}
1715
09d423e9
LP		 1716static int have_resolv_conf(const char *path) {
1717 assert(path);
1718
1719 if (access(path, F_OK) < 0) {
1720 if (errno == ENOENT)
1721 return 0;
1722
1723 return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
1724 }
1725
1726 return 1;
1727}
1728
7357272e 1729static int resolved_listening(void) {
b8ea7a6e 1730 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
b053cd5f 1731 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
7357272e 1732 _cleanup_free_ char *dns_stub_listener_mode = NULL;
b053cd5f
LP		 1733 int r;
1734
7357272e 1735 /* Check if resolved is listening */
b053cd5f
LP		 1736
1737 r = sd_bus_open_system(&bus);
1738 if (r < 0)
b8ea7a6e 1739 return log_debug_errno(r, "Failed to open system bus: %m");
b053cd5f 1740
7357272e 1741 r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
b8ea7a6e
LP		 1742 if (r < 0)
1743 return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
1744 if (r == 0)
1745 return 0;
7357272e
DM		 1746
1747 r = sd_bus_get_property_string(bus,
1748 "org.freedesktop.resolve1",
1749 "/org/freedesktop/resolve1",
1750 "org.freedesktop.resolve1.Manager",
1751 "DNSStubListener",
b8ea7a6e 1752 &error,
7357272e
DM		 1753 &dns_stub_listener_mode);
1754 if (r < 0)
b8ea7a6e 1755 return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
7357272e
DM		 1756
1757 return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
b053cd5f
LP		 1758}
1759
2547bb41 1760static int setup_resolv_conf(const char *dest) {
09d423e9
LP		 1761 _cleanup_free_ char *etc = NULL;
1762 const char *where, *what;
1763 ResolvConfMode m;
1764 int r;
2547bb41
LP		 1765
1766 assert(dest);
1767
09d423e9
LP		 1768 if (arg_resolv_conf == RESOLV_CONF_AUTO) {
1769 if (arg_private_network)
1770 m = RESOLV_CONF_OFF;
1771 else if (have_resolv_conf(STATIC_RESOLV_CONF) > 0 && resolved_listening() > 0)
83205269 1772 m = etc_writable() ? RESOLV_CONF_COPY_STATIC : RESOLV_CONF_BIND_STATIC;
09d423e9 1773 else if (have_resolv_conf("/etc/resolv.conf") > 0)
83205269 1774 m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
09d423e9 1775 else
83205269 1776 m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
09d423e9
LP		 1777 } else
1778 m = arg_resolv_conf;
1779
1780 if (m == RESOLV_CONF_OFF)
2547bb41
LP		 1781 return 0;
1782
87447ae4
LP		 1783 r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc);
1784 if (r < 0) {
1785 log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
1786 return 0;
1787 }
1788
1789 where = strjoina(etc, "/resolv.conf");
09d423e9
LP		 1790
1791 if (m == RESOLV_CONF_DELETE) {
1792 if (unlink(where) < 0)
1793 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
1794
87447ae4
LP		 1795 return 0;
1796 }
79d80fc1 1797
09d423e9
LP		 1798 if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_COPY_STATIC))
1799 what = STATIC_RESOLV_CONF;
1800 else
1801 what = "/etc/resolv.conf";
87447ae4 1802
09d423e9
LP		 1803 if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC)) {
1804 _cleanup_free_ char *resolved = NULL;
1805 int found;
1806
1807 found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved);
1808 if (found < 0) {
1809 log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
1810 return 0;
1811 }
3539724c 1812
87447ae4
LP		 1813 if (found == 0) /* missing? */
1814 (void) touch(resolved);
5367354d 1815
09d423e9 1816 r = mount_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
60e76d48 1817 if (r >= 0)
87447ae4 1818 return mount_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
3539724c
LP		 1819 }
1820
1821 /* If that didn't work, let's copy the file */
8a016c74 1822 r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, 0, COPY_REFLINK);
79d80fc1 1823 if (r < 0) {
3539724c
LP		 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.
68a313c5 1826 *
3539724c 1827 * If the disk image is read-only, there's also no point in complaining.
68a313c5 1828 */
09d423e9 1829 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,
3539724c 1830 "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
79d80fc1
TG		 1831 return 0;
1832 }
2547bb41 1833
03cfe0d5
LP		 1834 r = userns_lchown(where, 0, 0);
1835 if (r < 0)
3539724c 1836 log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
03cfe0d5 1837
2547bb41
LP		 1838 return 0;
1839}
1840
1e4f1671 1841static int setup_boot_id(void) {
cdde6ba6
LP		 1842 _cleanup_(unlink_and_freep) char *from = NULL;
1843 _cleanup_free_ char *path = NULL;
3bbaff3e 1844 sd_id128_t rnd = SD_ID128_NULL;
cdde6ba6 1845 const char *to;
04bc4a3f
LP		 1846 int r;
1847
1eacc470 1848 /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
04bc4a3f 1849
1eacc470 1850 r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
cdde6ba6
LP		 1851 if (r < 0)
1852 return log_error_errno(r, "Failed to generate random boot ID path: %m");
04bc4a3f
LP		 1853
1854 r = sd_id128_randomize(&rnd);
f647962d
MS		 1855 if (r < 0)
1856 return log_error_errno(r, "Failed to generate random boot id: %m");
04bc4a3f 1857
cdde6ba6 1858 r = id128_write(path, ID128_UUID, rnd, false);
f647962d
MS		 1859 if (r < 0)
1860 return log_error_errno(r, "Failed to write boot id: %m");
04bc4a3f 1861
cdde6ba6
LP		 1862 from = TAKE_PTR(path);
1863 to = "/proc/sys/kernel/random/boot_id";
1864
60e76d48 1865 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
cdde6ba6
LP		 1866 if (r < 0)
1867 return r;
04bc4a3f 1868
cdde6ba6 1869 return mount_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
04bc4a3f
LP		 1870}
1871
e58a1277 1872static int copy_devnodes(const char *dest) {
88213476
LP		 1873 static const char devnodes[] =
1874 "null\0"
1875 "zero\0"
1876 "full\0"
1877 "random\0"
1878 "urandom\0"
85614d66
TG		 1879 "tty\0"
1880 "net/tun\0";
88213476 1881
de40a303 1882 _cleanup_umask_ mode_t u;
88213476 1883 const char *d;
e58a1277 1884 int r = 0;
a258bf26
LP		 1885
1886 assert(dest);
124640f1
LP		 1887
1888 u = umask(0000);
88213476 1889
03cfe0d5
LP		 1890 /* Create /dev/net, so that we can create /dev/net/tun in it */
1891 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1892 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1893
88213476 1894 NULSTR_FOREACH(d, devnodes) {
7fd1b19b 1895 _cleanup_free_ char *from = NULL, *to = NULL;
7f112f50 1896 struct stat st;
88213476 1897
c6134d3e 1898 from = path_join("/dev/", d);
8967f291
LP		 1899 if (!from)
1900 return log_oom();
1901
c6134d3e 1902 to = path_join(dest, from);
8967f291
LP		 1903 if (!to)
1904 return log_oom();
88213476
LP		 1905
1906 if (stat(from, &st) < 0) {
1907
4a62c710
MS		 1908 if (errno != ENOENT)
1909 return log_error_errno(errno, "Failed to stat %s: %m", from);
88213476 1910
baaa35ad
ZJS		 1911 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
1912 return log_error_errno(SYNTHETIC_ERRNO(EIO),
1913 "%s is not a char or block device, cannot copy.", from);
1914 else {
8dfce114
LP		 1915 _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
1916
81f5049b 1917 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
8dbf71ec 1918 /* Explicitly warn the user when /dev is already populated. */
41eb4362 1919 if (errno == EEXIST)
8dbf71ec 1920 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);
81f5049b
AC		 1921 if (errno != EPERM)
1922 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1923
8dfce114 1924 /* Some systems abusively restrict mknod but allow bind mounts. */
81f5049b
AC		 1925 r = touch(to);
1926 if (r < 0)
1927 return log_error_errno(r, "touch (%s) failed: %m", to);
60e76d48
ZJS		 1928 r = mount_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
1929 if (r < 0)
1930 return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
81f5049b 1931 }
6278cf60 1932
03cfe0d5
LP		 1933 r = userns_lchown(to, 0, 0);
1934 if (r < 0)
1935 return log_error_errno(r, "chown() of device node %s failed: %m", to);
8dfce114 1936
657ee2d8 1937 dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
8dfce114
LP		 1938 if (!dn)
1939 return log_oom();
1940
1941 r = userns_mkdir(dest, dn, 0755, 0, 0);
1942 if (r < 0)
1943 return log_error_errno(r, "Failed to create '%s': %m", dn);
1944
1945 if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
1946 return log_oom();
1947
c6134d3e 1948 prefixed = path_join(dest, sl);
8dfce114
LP		 1949 if (!prefixed)
1950 return log_oom();
1951
2d9b74ba 1952 t = path_join("..", d);
8dfce114
LP		 1953 if (!t)
1954 return log_oom();
1955
1956 if (symlink(t, prefixed) < 0)
1957 log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
88213476 1958 }
88213476
LP		 1959 }
1960
e58a1277
LP		 1961 return r;
1962}
88213476 1963
de40a303
LP		 1964static int make_extra_nodes(const char *dest) {
1965 _cleanup_umask_ mode_t u;
1966 size_t i;
1967 int r;
1968
1969 u = umask(0000);
1970
1971 for (i = 0; i < arg_n_extra_nodes; i++) {
1972 _cleanup_free_ char *path = NULL;
1973 DeviceNode *n = arg_extra_nodes + i;
1974
c6134d3e 1975 path = path_join(dest, n->path);
de40a303
LP		 1976 if (!path)
1977 return log_oom();
1978
1979 if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
1980 return log_error_errno(errno, "Failed to create device node '%s': %m", path);
1981
1982 r = chmod_and_chown(path, n->mode, n->uid, n->gid);
1983 if (r < 0)
1984 return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
1985 }
1986
1987 return 0;
1988}
1989
03cfe0d5
LP		 1990static int setup_pts(const char *dest) {
1991 _cleanup_free_ char *options = NULL;
1992 const char *p;
709f6e46 1993 int r;
03cfe0d5 1994
349cc4a5 1995#if HAVE_SELINUX
03cfe0d5
LP		 1996 if (arg_selinux_apifs_context)
1997 (void) asprintf(&options,
3dce8915 1998 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
03cfe0d5
LP		 1999 arg_uid_shift + TTY_GID,
2000 arg_selinux_apifs_context);
2001 else
2002#endif
2003 (void) asprintf(&options,
3dce8915 2004 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
03cfe0d5 2005 arg_uid_shift + TTY_GID);
f2d88580 2006
03cfe0d5 2007 if (!options)
f2d88580
LP		 2008 return log_oom();
2009
03cfe0d5 2010 /* Mount /dev/pts itself */
cc9fce65 2011 p = prefix_roota(dest, "/dev/pts");
dae8b82e
ZJS		 2012 r = mkdir_errno_wrapper(p, 0755);
2013 if (r < 0)
2014 return log_error_errno(r, "Failed to create /dev/pts: %m");
2015
60e76d48
ZJS		 2016 r = mount_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
2017 if (r < 0)
2018 return r;
709f6e46
MS		 2019 r = userns_lchown(p, 0, 0);
2020 if (r < 0)
2021 return log_error_errno(r, "Failed to chown /dev/pts: %m");
03cfe0d5
LP		 2022
2023 /* Create /dev/ptmx symlink */
2024 p = prefix_roota(dest, "/dev/ptmx");
4a62c710
MS		 2025 if (symlink("pts/ptmx", p) < 0)
2026 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
709f6e46
MS		 2027 r = userns_lchown(p, 0, 0);
2028 if (r < 0)
2029 return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
f2d88580 2030
03cfe0d5
LP		 2031 /* And fix /dev/pts/ptmx ownership */
2032 p = prefix_roota(dest, "/dev/pts/ptmx");
709f6e46
MS		 2033 r = userns_lchown(p, 0, 0);
2034 if (r < 0)
2035 return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
6278cf60 2036
f2d88580
LP		 2037 return 0;
2038}
2039
3acc84eb
FB		 2040static int setup_stdio_as_dev_console(void) {
2041 int terminal;
e58a1277 2042 int r;
e58a1277 2043
3acc84eb
FB		 2044 terminal = open_terminal("/dev/console", O_RDWR);
2045 if (terminal < 0)
2046 return log_error_errno(terminal, "Failed to open console: %m");
e58a1277 2047
3acc84eb
FB		 2048 /* Make sure we can continue logging to the original stderr, even if
2049 * stderr points elsewhere now */
2050 r = log_dup_console();
2051 if (r < 0)
2052 return log_error_errno(r, "Failed to duplicate stderr: %m");
de40a303 2053
3acc84eb
FB		 2054 /* invalidates 'terminal' on success and failure */
2055 r = rearrange_stdio(terminal, terminal, terminal);
f647962d 2056 if (r < 0)
3acc84eb
FB		 2057 return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
2058
2059 return 0;
2060}
88213476 2061
3acc84eb
FB		 2062static int setup_dev_console(const char *console) {
2063 _cleanup_free_ char *p = NULL;
2064 int r;
a258bf26 2065
3acc84eb
FB		 2066 /* Create /dev/console symlink */
2067 r = path_make_relative("/dev", console, &p);
81f5049b 2068 if (r < 0)
3acc84eb
FB		 2069 return log_error_errno(r, "Failed to create relative path: %m");
2070
2071 if (symlink(p, "/dev/console") < 0)
2072 return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
a258bf26 2073
3acc84eb 2074 return 0;
e58a1277
LP		 2075}
2076
8e5430c4
LP		 2077static 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 keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2086 if (keyring == -1) {
2087 if (errno == ENOSYS)
2088 log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2089 else if (IN_SET(errno, EACCES, EPERM))
2090 log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2091 else
2092 return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2093 }
2094
2095 return 0;
2096}
2097
1e4f1671 2098static int setup_kmsg(int kmsg_socket) {
9ec5a93c
LP		 2099 _cleanup_(unlink_and_freep) char *from = NULL;
2100 _cleanup_free_ char *fifo = NULL;
2101 _cleanup_close_ int fd = -1;
7fd1b19b 2102 _cleanup_umask_ mode_t u;
9ec5a93c 2103 int r;
e58a1277 2104
e58a1277 2105 assert(kmsg_socket >= 0);
a258bf26 2106
e58a1277 2107 u = umask(0000);
a258bf26 2108
1eacc470 2109 /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
9ec5a93c
LP		 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
1eacc470 2114 r = tempfn_random_child("/run", "proc-kmsg", &fifo);
9ec5a93c
LP		 2115 if (r < 0)
2116 return log_error_errno(r, "Failed to generate kmsg path: %m");
e58a1277 2117
9ec5a93c 2118 if (mkfifo(fifo, 0600) < 0)
03cfe0d5 2119 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
9ec5a93c
LP		 2120
2121 from = TAKE_PTR(fifo);
9ec5a93c 2122
1eacc470 2123 r = mount_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
60e76d48
ZJS		 2124 if (r < 0)
2125 return r;
e58a1277 2126
669fc4e5 2127 fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
4a62c710
MS		 2128 if (fd < 0)
2129 return log_error_errno(errno, "Failed to open fifo: %m");
e58a1277 2130
9ec5a93c 2131 /* Store away the fd in the socket, so that it stays open as long as we run the child */
3ee897d6 2132 r = send_one_fd(kmsg_socket, fd, 0);
d9603714
DH		 2133 if (r < 0)
2134 return log_error_errno(r, "Failed to send FIFO fd: %m");
a258bf26 2135
25ea79fe 2136 return 0;
88213476
LP		 2137}
2138
1c4baffc 2139static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
6d0b55c2
LP		 2140 union in_addr_union *exposed = userdata;
2141
2142 assert(rtnl);
2143 assert(m);
2144 assert(exposed);
2145
7a8f6325 2146 expose_port_execute(rtnl, arg_expose_ports, exposed);
6d0b55c2
LP		 2147 return 0;
2148}
2149
3a74cea5 2150static int setup_hostname(void) {
c818eef1 2151 int r;
3a74cea5 2152
0c582db0 2153 if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
eb91eb18
LP		 2154 return 0;
2155
c818eef1
LP		 2156 r = sethostname_idempotent(arg_hostname ?: arg_machine);
2157 if (r < 0)
2158 return log_error_errno(r, "Failed to set hostname: %m");
3a74cea5 2159
7027ff61 2160 return 0;
3a74cea5
LP		 2161}
2162
57fb9fb5 2163static int setup_journal(const char *directory) {
0f5e1382 2164 _cleanup_free_ char *d = NULL;
b2238e38
LP		 2165 const char *dirname, *p, *q;
2166 sd_id128_t this_id;
2167 char id[33];
8054d749 2168 bool try;
57fb9fb5
LP		 2169 int r;
2170
df9a75e4
LP		 2171 /* Don't link journals in ephemeral mode */
2172 if (arg_ephemeral)
2173 return 0;
2174
8054d749
LP		 2175 if (arg_link_journal == LINK_NO)
2176 return 0;
2177
2178 try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
2179
4d680aee 2180 r = sd_id128_get_machine(&this_id);
f647962d
MS		 2181 if (r < 0)
2182 return log_error_errno(r, "Failed to retrieve machine ID: %m");
4d680aee 2183
e01ff70a 2184 if (sd_id128_equal(arg_uuid, this_id)) {
8054d749 2185 log_full(try ? LOG_WARNING : LOG_ERR,
e192a281 2186 "Host and machine ids are equal (%s): refusing to link journals", sd_id128_to_string(arg_uuid, id));
8054d749 2187 if (try)
4d680aee 2188 return 0;
df9a75e4 2189 return -EEXIST;
4d680aee
ZJS		 2190 }
2191
369ca6da
ZJS		 2192 FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
2193 r = userns_mkdir(directory, dirname, 0755, 0, 0);
2194 if (r < 0) {
2195 bool ignore = r == -EROFS && try;
2196 log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
2197 "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
2198 return ignore ? 0 : r;
2199 }
2200 }
03cfe0d5 2201
e01ff70a
MS		 2202 (void) sd_id128_to_string(arg_uuid, id);
2203
03cfe0d5
LP		 2204 p = strjoina("/var/log/journal/", id);
2205 q = prefix_roota(directory, p);
27407a01 2206
e1873695 2207 if (path_is_mount_point(p, NULL, 0) > 0) {
8054d749
LP		 2208 if (try)
2209 return 0;
27407a01 2210
baaa35ad
ZJS		 2211 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2212 "%s: already a mount point, refusing to use for journal", p);
57fb9fb5
LP		 2213 }
2214
e1873695 2215 if (path_is_mount_point(q, NULL, 0) > 0) {
8054d749
LP		 2216 if (try)
2217 return 0;
57fb9fb5 2218
baaa35ad
ZJS		 2219 return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
2220 "%s: already a mount point, refusing to use for journal", q);
57fb9fb5
LP		 2221 }
2222
2223 r = readlink_and_make_absolute(p, &d);
2224 if (r >= 0) {
3742095b 2225 if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
57fb9fb5
LP		 2226 path_equal(d, q)) {
2227
03cfe0d5 2228 r = userns_mkdir(directory, p, 0755, 0, 0);
27407a01 2229 if (r < 0)
709f6e46 2230 log_warning_errno(r, "Failed to create directory %s: %m", q);
27407a01 2231 return 0;
57fb9fb5
LP		 2232 }
2233
4a62c710
MS		 2234 if (unlink(p) < 0)
2235 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
57fb9fb5
LP		 2236 } else if (r == -EINVAL) {
2237
2238 if (arg_link_journal == LINK_GUEST &&
2239 rmdir(p) < 0) {
2240
27407a01
ZJS		 2241 if (errno == ENOTDIR) {
2242 log_error("%s already exists and is neither a symlink nor a directory", p);
2243 return r;
4314d33f
MS		 2244 } else
2245 return log_error_errno(errno, "Failed to remove %s: %m", p);
57fb9fb5 2246 }
4314d33f
MS		 2247 } else if (r != -ENOENT)
2248 return log_error_errno(r, "readlink(%s) failed: %m", p);
57fb9fb5
LP		 2249
2250 if (arg_link_journal == LINK_GUEST) {
2251
2252 if (symlink(q, p) < 0) {
8054d749 2253 if (try) {
56f64d95 2254 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
574edc90 2255 return 0;
4314d33f
MS		 2256 } else
2257 return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
57fb9fb5
LP		 2258 }
2259
03cfe0d5 2260 r = userns_mkdir(directory, p, 0755, 0, 0);
27407a01 2261 if (r < 0)
709f6e46 2262 log_warning_errno(r, "Failed to create directory %s: %m", q);
27407a01 2263 return 0;
57fb9fb5
LP		 2264 }
2265
2266 if (arg_link_journal == LINK_HOST) {
ccddd104 2267 /* don't create parents here — if the host doesn't have
574edc90 2268 * permanent journal set up, don't force it here */
ba8e6c4d 2269
dae8b82e
ZJS		 2270 r = mkdir_errno_wrapper(p, 0755);
2271 if (r < 0 && r != -EEXIST) {
8054d749 2272 if (try) {
dae8b82e 2273 log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
574edc90 2274 return 0;
4314d33f 2275 } else
dae8b82e 2276 return log_error_errno(r, "Failed to create %s: %m", p);
57fb9fb5
LP		 2277 }
2278
27407a01
ZJS		 2279 } else if (access(p, F_OK) < 0)
2280 return 0;
57fb9fb5 2281
cdb2b9d0
LP		 2282 if (dir_is_empty(q) == 0)
2283 log_warning("%s is not empty, proceeding anyway.", q);
2284
03cfe0d5 2285 r = userns_mkdir(directory, p, 0755, 0, 0);
709f6e46
MS		 2286 if (r < 0)
2287 return log_error_errno(r, "Failed to create %s: %m", q);
57fb9fb5 2288
60e76d48
ZJS		 2289 r = mount_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
2290 if (r < 0)
4a62c710 2291 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
57fb9fb5 2292
27407a01 2293 return 0;
57fb9fb5
LP		 2294}
2295
de40a303
LP		 2296static 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 if (capability_quintet_is_set(&arg_full_capabilities)) {
2304 q = arg_full_capabilities;
2305
2306 if (q.bounding == (uint64_t) -1)
2307 q.bounding = uid == 0 ? arg_caps_retain : 0;
2308
2309 if (q.effective == (uint64_t) -1)
2310 q.effective = uid == 0 ? q.bounding : 0;
2311
2312 if (q.inheritable == (uint64_t) -1)
2313 q.inheritable = uid == 0 ? q.bounding : 0;
2314
2315 if (q.permitted == (uint64_t) -1)
2316 q.permitted = uid == 0 ? q.bounding : 0;
2317
2318 if (q.ambient == (uint64_t) -1 && ambient_capabilities_supported())
2319 q.ambient = 0;
f66ad460
AZ		 2320
2321 if (capability_quintet_mangle(&q))
2322 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
2323
2324 } else {
de40a303
LP		 2325 q = (CapabilityQuintet) {
2326 .bounding = arg_caps_retain,
2327 .effective = uid == 0 ? arg_caps_retain : 0,
2328 .inheritable = uid == 0 ? arg_caps_retain : 0,
2329 .permitted = uid == 0 ? arg_caps_retain : 0,
2330 .ambient = ambient_capabilities_supported() ? 0 : (uint64_t) -1,
2331 };
2332
f66ad460
AZ		 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 if (capability_quintet_mangle(&q))
2336 log_warning("Some capabilities will not be set because they are not in the current bounding set.");
2337
2338 }
2339
de40a303 2340 return capability_quintet_enforce(&q);
88213476
LP		 2341}
2342
db999e0f
LP		 2343static int reset_audit_loginuid(void) {
2344 _cleanup_free_ char *p = NULL;
2345 int r;
2346
0c582db0 2347 if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
db999e0f
LP		 2348 return 0;
2349
2350 r = read_one_line_file("/proc/self/loginuid", &p);
13e8ceb8 2351 if (r == -ENOENT)
db999e0f 2352 return 0;
f647962d
MS		 2353 if (r < 0)
2354 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
db999e0f
LP		 2355
2356 /* Already reset? */
2357 if (streq(p, "4294967295"))
2358 return 0;
2359
57512c89 2360 r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
db999e0f 2361 if (r < 0) {
10a87006
LP		 2362 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)");
77b6e194 2368
db999e0f 2369 sleep(5);
77b6e194 2370 }
db999e0f
LP		 2371
2372 return 0;
77b6e194
LP		 2373}
2374
785890ac
LP		 2375static int setup_propagate(const char *root) {
2376 const char *p, *q;
709f6e46 2377 int r;
785890ac
LP		 2378
2379 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2380 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
63c372cb 2381 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
785890ac
LP		 2382 (void) mkdir_p(p, 0600);
2383
709f6e46
MS		 2384 r = userns_mkdir(root, "/run/systemd", 0755, 0, 0);
2385 if (r < 0)
2386 return log_error_errno(r, "Failed to create /run/systemd: %m");
03cfe0d5 2387
709f6e46
MS		 2388 r = userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0);
2389 if (r < 0)
2390 return log_error_errno(r, "Failed to create /run/systemd/nspawn: %m");
03cfe0d5 2391
709f6e46
MS		 2392 r = userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0);
2393 if (r < 0)
2394 return log_error_errno(r, "Failed to create /run/systemd/nspawn/incoming: %m");
785890ac 2395
03cfe0d5 2396 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
60e76d48
ZJS		 2397 r = mount_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
2398 if (r < 0)
2399 return r;
785890ac 2400
60e76d48
ZJS		 2401 r = mount_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
2402 if (r < 0)
2403 return r;
785890ac 2404
19caffac
AC		 2405 /* machined will MS_MOVE into that directory, and that's only
2406 * supported for non-shared mounts. */
60e76d48 2407 return mount_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
785890ac
LP		 2408}
2409
317feb4d 2410static int setup_machine_id(const char *directory) {
691675ba
LP		 2411 const char *etc_machine_id;
2412 sd_id128_t id;
3bbaff3e 2413 int r;
e01ff70a 2414
317feb4d
LP		 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
e01ff70a
MS		 2422 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2423
691675ba 2424 r = id128_read(etc_machine_id, ID128_PLAIN, &id);
317feb4d
LP		 2425 if (r < 0) {
2426 if (!IN_SET(r, -ENOENT, -ENOMEDIUM)) /* If the file is missing or empty, we don't mind */
2427 return log_error_errno(r, "Failed to read machine ID from container image: %m");
691675ba 2428
317feb4d
LP		 2429 if (sd_id128_is_null(arg_uuid)) {
2430 r = sd_id128_randomize(&arg_uuid);
2431 if (r < 0)
2432 return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
2433 }
2434 } else {
baaa35ad
ZJS		 2435 if (sd_id128_is_null(id))
2436 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2437 "Machine ID in container image is zero, refusing.");
e01ff70a 2438
317feb4d
LP		 2439 arg_uuid = id;
2440 }
691675ba 2441
e01ff70a
MS		 2442 return 0;
2443}
2444
7336138e
LP		 2445static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
2446 int r;
2447
2448 assert(directory);
2449
0de7acce 2450 if (arg_userns_mode == USER_NAMESPACE_NO || !arg_userns_chown)
7336138e
LP		 2451 return 0;
2452
2453 r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
2454 if (r == -EOPNOTSUPP)
2455 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 if (r == -EBADE)
2457 return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
2458 if (r < 0)
2459 return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
2460 if (r == 0)
2461 log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
2462 else
2463 log_debug("Patched directory tree to match UID/GID range.");
2464
2465 return r;
2466}
2467
113cea80 2468/*
6d416b9c
LS		 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
919699ec
LP		 2476 * container is returned. The container argument has been set
2477 * to CONTAINER_TERMINATED.
6d416b9c
LS		 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.
113cea80 2481 *
6d416b9c
LS		 2482 * That is, success is indicated by a return value of zero, and an
2483 * error is indicated by a non-zero value.
113cea80
DH		 2484 */
2485static int wait_for_container(pid_t pid, ContainerStatus *container) {
113cea80 2486 siginfo_t status;
919699ec 2487 int r;
113cea80
DH		 2488
2489 r = wait_for_terminate(pid, &status);
f647962d
MS		 2490 if (r < 0)
2491 return log_warning_errno(r, "Failed to wait for container: %m");
113cea80
DH		 2492
2493 switch (status.si_code) {
fddbb89c 2494
113cea80 2495 case CLD_EXITED:
b5a2179b 2496 if (status.si_status == 0)
919699ec 2497 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
b5a2179b 2498 else
919699ec 2499 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
fddbb89c 2500
919699ec
LP		 2501 *container = CONTAINER_TERMINATED;
2502 return status.si_status;
113cea80
DH		 2503
2504 case CLD_KILLED:
2505 if (status.si_status == SIGINT) {
919699ec 2506 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
113cea80 2507 *container = CONTAINER_TERMINATED;
919699ec
LP		 2508 return 0;
2509
113cea80 2510 } else if (status.si_status == SIGHUP) {
919699ec 2511 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
113cea80 2512 *container = CONTAINER_REBOOTED;
919699ec 2513 return 0;
113cea80 2514 }
919699ec 2515
4831981d 2516 _fallthrough_;
113cea80 2517 case CLD_DUMPED:
baaa35ad
ZJS		 2518 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2519 "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
113cea80
DH		 2520
2521 default:
baaa35ad
ZJS		 2522 return log_error_errno(SYNTHETIC_ERRNO(EIO),
2523 "Container %s failed due to unknown reason.", arg_machine);
113cea80 2524 }
113cea80
DH		 2525}
2526
023fb90b
LP		 2527static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
2528 pid_t pid;
2529
4a0b58c4 2530 pid = PTR_TO_PID(userdata);
023fb90b 2531 if (pid > 0) {
c6c8f6e2 2532 if (kill(pid, arg_kill_signal) >= 0) {
023fb90b
LP		 2533 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
2534 sd_event_source_set_userdata(s, NULL);
2535 return 0;
2536 }
2537 }
2538
2539 sd_event_exit(sd_event_source_get_event(s), 0);
2540 return 0;
2541}
2542
6916b164 2543static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
abdb9b08
LP		 2544 pid_t pid;
2545
2546 assert(s);
2547 assert(ssi);
2548
2549 pid = PTR_TO_PID(userdata);
2550
6916b164
AU		 2551 for (;;) {
2552 siginfo_t si = {};
abdb9b08 2553
6916b164
AU		 2554 if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
2555 return log_error_errno(errno, "Failed to waitid(): %m");
2556 if (si.si_pid == 0) /* No pending children. */
2557 break;
abdb9b08 2558 if (si.si_pid == pid) {
6916b164
AU		 2559 /* The main process we care for has exited. Return from
2560 * signal handler but leave the zombie. */
2561 sd_event_exit(sd_event_source_get_event(s), 0);
2562 break;
2563 }
abdb9b08 2564
6916b164
AU		 2565 /* Reap all other children. */
2566 (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
2567 }
2568
2569 return 0;
2570}
2571
abdb9b08
LP		 2572static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
2573 pid_t pid;
2574
2575 assert(m);
2576
2577 pid = PTR_TO_PID(userdata);
2578
2579 if (arg_kill_signal > 0) {
2580 log_info("Container termination requested. Attempting to halt container.");
2581 (void) kill(pid, arg_kill_signal);
2582 } else {
2583 log_info("Container termination requested. Exiting.");
2584 sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
2585 }
2586
2587 return 0;
2588}
2589
ec16945e 2590static int determine_names(void) {
1b9cebf6 2591 int r;
ec16945e 2592
c1521918
LP		 2593 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
657ee2d8 2599 arg_directory = path_join("/var/lib/machines", arg_machine);
c1521918
LP		 2600 if (!arg_directory)
2601 return log_oom();
2602 }
2603
ec16945e 2604 if (!arg_image && !arg_directory) {
1b9cebf6
LP		 2605 if (arg_machine) {
2606 _cleanup_(image_unrefp) Image *i = NULL;
2607
5ef46e5f 2608 r = image_find(IMAGE_MACHINE, arg_machine, &i);
3a6ce860
LP		 2609 if (r == -ENOENT)
2610 return log_error_errno(r, "No image for machine '%s'.", arg_machine);
1b9cebf6
LP		 2611 if (r < 0)
2612 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
1b9cebf6 2613
eb38edce 2614 if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
0f03c2a4 2615 r = free_and_strdup(&arg_image, i->path);
1b9cebf6 2616 else
0f03c2a4 2617 r = free_and_strdup(&arg_directory, i->path);
1b9cebf6 2618 if (r < 0)
0f3be6ca 2619 return log_oom();
1b9cebf6 2620
aee327b8
LP		 2621 if (!arg_ephemeral)
2622 arg_read_only = arg_read_only || i->read_only;
d7249575
LP		 2623 } else {
2624 r = safe_getcwd(&arg_directory);
2625 if (r < 0)
2626 return log_error_errno(r, "Failed to determine current directory: %m");
2627 }
ec16945e 2628
c6147113
LP		 2629 if (!arg_directory && !arg_image)
2630 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
ec16945e
LP		 2631 }
2632
2633 if (!arg_machine) {
b9ba4dab
LP		 2634 if (arg_directory && path_equal(arg_directory, "/"))
2635 arg_machine = gethostname_malloc();
4827ab48
LP		 2636 else {
2637 if (arg_image) {
2638 char *e;
2639
2640 arg_machine = strdup(basename(arg_image));
2641
2642 /* Truncate suffix if there is one */
2643 e = endswith(arg_machine, ".raw");
2644 if (e)
2645 *e = 0;
2646 } else
2647 arg_machine = strdup(basename(arg_directory));
2648 }
ec16945e
LP		 2649 if (!arg_machine)
2650 return log_oom();
2651
ae691c1d 2652 hostname_cleanup(arg_machine);
c6147113
LP		 2653 if (!machine_name_is_valid(arg_machine))
2654 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
b9ba4dab
LP		 2655
2656 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 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
2665 return log_oom();
2666
2667 free(arg_machine);
2668 arg_machine = b;
2669 }
ec16945e
LP		 2670 }
2671
2672 return 0;
2673}
2674
8d4aa2bb 2675static int chase_symlinks_and_update(char **p, unsigned flags) {
3f342ec4
LP		 2676 char *chased;
2677 int r;
2678
2679 assert(p);
2680
2681 if (!*p)
2682 return 0;
2683
8d4aa2bb 2684 r = chase_symlinks(*p, NULL, flags, &chased);
3f342ec4
LP		 2685 if (r < 0)
2686 return log_error_errno(r, "Failed to resolve path %s: %m", *p);
2687
8405dcf7
ZJS		 2688 free_and_replace(*p, chased);
2689 return r; /* r might be an fd here in case we ever use CHASE_OPEN in flags */
3f342ec4
LP		 2690}
2691
03cfe0d5 2692static int determine_uid_shift(const char *directory) {
6dac160c
LP		 2693 int r;
2694
0de7acce 2695 if (arg_userns_mode == USER_NAMESPACE_NO) {
03cfe0d5 2696 arg_uid_shift = 0;
6dac160c 2697 return 0;
03cfe0d5 2698 }
6dac160c
LP		 2699
2700 if (arg_uid_shift == UID_INVALID) {
2701 struct stat st;
2702
03cfe0d5 2703 r = stat(directory, &st);
6dac160c 2704 if (r < 0)
03cfe0d5 2705 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
6dac160c
LP		 2706
2707 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
2708
baaa35ad
ZJS		 2709 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
2710 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2711 "UID and GID base of %s don't match.", directory);
6dac160c
LP		 2712
2713 arg_uid_range = UINT32_C(0x10000);
2714 }
2715
baaa35ad
ZJS		 2716 if (arg_uid_shift > (uid_t) -1 - arg_uid_range)
2717 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2718 "UID base too high for UID range.");
6dac160c 2719
6dac160c
LP		 2720 return 0;
2721}
2722
de40a303
LP		 2723static unsigned long effective_clone_ns_flags(void) {
2724 unsigned long flags = arg_clone_ns_flags;
2725
2726 if (arg_private_network)
2727 flags |= CLONE_NEWNET;
2728 if (arg_use_cgns)
2729 flags |= CLONE_NEWCGROUP;
2730 if (arg_userns_mode != USER_NAMESPACE_NO)
2731 flags |= CLONE_NEWUSER;
2732
2733 return flags;
2734}
2735
2736static 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 flags = effective_clone_ns_flags();
2762
2763 STRV_FOREACH_PAIR(k, v, arg_sysctl) {
2764 bool good = false;
2765 size_t i;
2766
2767 for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
2768
2769 if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
2770 continue;
2771
2772 if (safe_sysctl[i].prefix)
2773 good = startswith(*k, safe_sysctl[i].key);
2774 else
2775 good = streq(*k, safe_sysctl[i].key);
2776
2777 if (good)
2778 break;
2779 }
2780
c6147113
LP		 2781 if (!good)
2782 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
de40a303
LP		 2783
2784 r = sysctl_write(*k, *v);
2785 if (r < 0)
2786 return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
2787 }
2788
2789 return 0;
2790}
2791
03cfe0d5
LP		 2792static int inner_child(
2793 Barrier *barrier,
2794 const char *directory,
2795 bool secondary,
2796 int kmsg_socket,
2797 int rtnl_socket,
3acc84eb 2798 int master_pty_socket,
f757855e 2799 FDSet *fds) {
69c79d3c 2800
03cfe0d5 2801 _cleanup_free_ char *home = NULL;
e01ff70a 2802 char as_uuid[37];
88614c8a 2803 size_t n_env = 1;
03cfe0d5 2804 const char *envp[] = {
0c300adf 2805 "PATH=" DEFAULT_PATH_COMPAT,
6aadfa4c 2806 NULL, /* container */
03cfe0d5
LP		 2807 NULL, /* TERM */
2808 NULL, /* HOME */
2809 NULL, /* USER */
2810 NULL, /* LOGNAME */
2811 NULL, /* container_uuid */
2812 NULL, /* LISTEN_FDS */
2813 NULL, /* LISTEN_PID */
9c1e04d0 2814 NULL, /* NOTIFY_SOCKET */
03cfe0d5
LP		 2815 NULL
2816 };
1a68e1e5 2817 const char *exec_target;
2371271c 2818 _cleanup_strv_free_ char **env_use = NULL;
de40a303 2819 int r, which_failed;
88213476 2820
b37469d7
LP		 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
03cfe0d5
LP		 2832 assert(barrier);
2833 assert(directory);
2834 assert(kmsg_socket >= 0);
88213476 2835
de40a303
LP		 2836 log_debug("Inner child is initializing.");
2837
0de7acce 2838 if (arg_userns_mode != USER_NAMESPACE_NO) {
03cfe0d5
LP		 2839 /* Tell the parent, that it now can write the UID map. */
2840 (void) barrier_place(barrier); /* #1 */
7027ff61 2841
03cfe0d5 2842 /* Wait until the parent wrote the UID map */
baaa35ad
ZJS		 2843 if (!barrier_place_and_sync(barrier)) /* #2 */
2844 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2845 "Parent died too early");
88213476
LP		 2846 }
2847
6d66bd3b
EV		 2848 r = reset_uid_gid();
2849 if (r < 0)
2850 return log_error_errno(r, "Couldn't become new root: %m");
2851
0de7acce 2852 r = mount_all(NULL,
4f086aab 2853 arg_mount_settings | MOUNT_IN_USERNS,
0de7acce 2854 arg_uid_shift,
0de7acce 2855 arg_selinux_apifs_context);
03cfe0d5
LP		 2856 if (r < 0)
2857 return r;
2858
04413780
ZJS		 2859 if (!arg_network_namespace_path && arg_private_network) {
2860 r = unshare(CLONE_NEWNET);
2861 if (r < 0)
2862 return log_error_errno(errno, "Failed to unshare network namespace: %m");
75116558
PS		 2863
2864 /* Tell the parent that it can setup network interfaces. */
2865 (void) barrier_place(barrier); /* #3 */
04413780
ZJS		 2866 }
2867
4f086aab 2868 r = mount_sysfs(NULL, arg_mount_settings);
d8fc6a00
LP		 2869 if (r < 0)
2870 return r;
2871
03cfe0d5
LP		 2872 /* Wait until we are cgroup-ified, so that we
2873 * can mount the right cgroup path writable */
baaa35ad
ZJS		 2874 if (!barrier_place_and_sync(barrier)) /* #4 */
2875 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
2876 "Parent died too early");
88213476 2877
489fae52 2878 if (arg_use_cgns) {
0996ef00
CB		 2879 r = unshare(CLONE_NEWCGROUP);
2880 if (r < 0)
04413780 2881 return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
0996ef00
CB		 2882 r = mount_cgroups(
2883 "",
2884 arg_unified_cgroup_hierarchy,
2885 arg_userns_mode != USER_NAMESPACE_NO,
2886 arg_uid_shift,
2887 arg_uid_range,
5a8ff0e6 2888 arg_selinux_apifs_context,
ada54120 2889 true);
0996ef00
CB		 2890 if (r < 0)
2891 return r;
2892 } else {
2893 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
2894 if (r < 0)
2895 return r;
2896 }
ec16945e 2897
1e4f1671 2898 r = setup_boot_id();
03cfe0d5
LP		 2899 if (r < 0)
2900 return r;
ec16945e 2901
1e4f1671 2902 r = setup_kmsg(kmsg_socket);
03cfe0d5
LP		 2903 if (r < 0)
2904 return r;
2905 kmsg_socket = safe_close(kmsg_socket);
ec16945e 2906
de40a303
LP		 2907 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 if (r < 0)
2917 return r;
2918
03cfe0d5
LP		 2919 if (setsid() < 0)
2920 return log_error_errno(errno, "setsid() failed: %m");
2921
2922 if (arg_private_network)
2923 loopback_setup();
2924
7a8f6325
LP		 2925 if (arg_expose_ports) {
2926 r = expose_port_send_rtnl(rtnl_socket);
2927 if (r < 0)
2928 return r;
2929 rtnl_socket = safe_close(rtnl_socket);
2930 }
03cfe0d5 2931
3acc84eb 2932 if (arg_console_mode != CONSOLE_PIPE) {
cd132992 2933 _cleanup_close_ int master = -1;
3acc84eb
FB		 2934 _cleanup_free_ char *console = NULL;
2935
2936 /* Allocate a pty and make it available as /dev/console. */
dc98caea 2937 master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
3acc84eb 2938 if (master < 0)
dc98caea 2939 return log_error_errno(master, "Failed to allocate a pty: %m");
3acc84eb
FB		 2940
2941 r = setup_dev_console(console);
2942 if (r < 0)
2943 return log_error_errno(r, "Failed to setup /dev/console: %m");
2944
2945 r = send_one_fd(master_pty_socket, master, 0);
2946 if (r < 0)
2947 return log_error_errno(r, "Failed to send master fd: %m");
2948 master_pty_socket = safe_close(master_pty_socket);
2949
2950 r = setup_stdio_as_dev_console();
2951 if (r < 0)
2952 return r;
2953 }
2954
de40a303
LP		 2955 r = patch_sysctl();
2956 if (r < 0)
2957 return r;
2958
81f345df
LP		 2959 if (arg_oom_score_adjust_set) {
2960 r = set_oom_score_adjust(arg_oom_score_adjust);
2961 if (r < 0)
2962 return log_error_errno(r, "Failed to adjust OOM score: %m");
2963 }
2964
0985c7c4
ZJS		 2965 if (arg_cpu_set.set)
2966 if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
d107bb7d
LP		 2967 return log_error_errno(errno, "Failed to set CPU affinity: %m");
2968
c818eef1 2969 (void) setup_hostname();
03cfe0d5 2970
050f7277 2971 if (arg_personality != PERSONALITY_INVALID) {
21022b9d
LP		 2972 r = safe_personality(arg_personality);
2973 if (r < 0)
2974 return log_error_errno(r, "personality() failed: %m");
03cfe0d5 2975 } else if (secondary) {
21022b9d
LP		 2976 r = safe_personality(PER_LINUX32);
2977 if (r < 0)
2978 return log_error_errno(r, "personality() failed: %m");
03cfe0d5
LP		 2979 }
2980
de40a303
LP		 2981 r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
2982 if (r < 0)
2983 return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
2984
2985#if HAVE_SECCOMP
2986 if (arg_seccomp) {
2987
2988 if (is_seccomp_available()) {
2989
2990 r = seccomp_load(arg_seccomp);
7bc5e0b1 2991 if (ERRNO_IS_SECCOMP_FATAL(r))
de40a303
LP		 2992 return log_error_errno(r, "Failed to install seccomp filter: %m");
2993 if (r < 0)
2994 log_debug_errno(r, "Failed to install seccomp filter: %m");
2995 }
2996 } else
2997#endif
2998 {
2999 r = setup_seccomp(arg_caps_retain, arg_syscall_whitelist, arg_syscall_blacklist);
3000 if (r < 0)
3001 return r;
3002 }
3003
349cc4a5 3004#if HAVE_SELINUX
03cfe0d5 3005 if (arg_selinux_context)
2ed96880 3006 if (setexeccon(arg_selinux_context) < 0)
03cfe0d5
LP		 3007 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3008#endif
3009
de40a303
LP		 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 if (prctl(PR_SET_KEEPCAPS, 1) < 0)
3013 return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
3014
3015 if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
3016 r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids);
3017 else
3018 r = change_uid_gid(arg_user, &home);
03cfe0d5
LP		 3019 if (r < 0)
3020 return r;
3021
de40a303
LP		 3022 r = drop_capabilities(getuid());
3023 if (r < 0)
3024 return log_error_errno(r, "Dropping capabilities failed: %m");
3025
66edd963
LP		 3026 if (arg_no_new_privileges)
3027 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
3028 return log_error_errno(errno, "Failed to disable new privileges: %m");
3029
6aadfa4c
ILG		 3030 /* LXC sets container=lxc, so follow the scheme here */
3031 envp[n_env++] = strjoina("container=", arg_container_service_name);
3032
03cfe0d5
LP		 3033 envp[n_env] = strv_find_prefix(environ, "TERM=");
3034 if (envp[n_env])
313cefa1 3035 n_env++;
03cfe0d5 3036
de40a303
LP		 3037 if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
3038 if (asprintf((char**)(envp + n_env++), "HOME=%s", home ?: "/root") < 0)
3039 return log_oom();
3040
3041 if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
3042 if (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ?: "root") < 0 ||
3043 asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
3044 return log_oom();
03cfe0d5 3045
3bbaff3e 3046 assert(!sd_id128_is_null(arg_uuid));
03cfe0d5 3047
691675ba 3048 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_to_uuid_string(arg_uuid, as_uuid)) < 0)
e01ff70a 3049 return log_oom();
03cfe0d5
LP		 3050
3051 if (fdset_size(fds) > 0) {
3052 r = fdset_cloexec(fds, false);
3053 if (r < 0)
3054 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3055
3056 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3057 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3058 return log_oom();
3059 }
9c1e04d0
AP		 3060 if (asprintf((char **)(envp + n_env++), "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
3061 return log_oom();
03cfe0d5 3062
2371271c
TG		 3063 env_use = strv_env_merge(2, envp, arg_setenv);
3064 if (!env_use)
3065 return log_oom();
03cfe0d5
LP		 3066
3067 /* Let the parent know that we are ready and
3068 * wait until the parent is ready with the
3069 * setup, too... */
baaa35ad
ZJS		 3070 if (!barrier_place_and_sync(barrier)) /* #5 */
3071 return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
3072 "Parent died too early");
03cfe0d5 3073
5f932eb9
LP		 3074 if (arg_chdir)
3075 if (chdir(arg_chdir) < 0)
3076 return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
3077
7732f92b 3078 if (arg_start_mode == START_PID2) {
75bf701f 3079 r = stub_pid1(arg_uuid);
7732f92b
LP		 3080 if (r < 0)
3081 return r;
3082 }
3083
de40a303
LP		 3084 log_debug("Inner child completed, invoking payload.");
3085
8ca082b4
LP		 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. */
03cfe0d5 3089 log_close();
8ca082b4
LP		 3090 log_set_open_when_needed(true);
3091
03cfe0d5
LP		 3092 (void) fdset_close_others(fds);
3093
7732f92b 3094 if (arg_start_mode == START_BOOT) {
03cfe0d5
LP		 3095 char **a;
3096 size_t m;
3097
3098 /* Automatically search for the init system */
3099
75f32f04
ZJS		 3100 m = strv_length(arg_parameters);
3101 a = newa(char*, m + 2);
3102 memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
3103 a[1 + m] = NULL;
03cfe0d5 3104
ced58da7 3105 a[0] = (char*) "/usr/lib/systemd/systemd";
03cfe0d5
LP		 3106 execve(a[0], a, env_use);
3107
ced58da7 3108 a[0] = (char*) "/lib/systemd/systemd";
03cfe0d5
LP		 3109 execve(a[0], a, env_use);
3110
ced58da7 3111 a[0] = (char*) "/sbin/init";
03cfe0d5 3112 execve(a[0], a, env_use);
ced58da7
LP		 3113
3114 exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
1a68e1e5 3115 } else if (!strv_isempty(arg_parameters)) {
b6b180b7
LP		 3116 const char *dollar_path;
3117
1a68e1e5 3118 exec_target = arg_parameters[0];
b6b180b7
LP		 3119
3120 /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
3121 * binary. */
3122 dollar_path = strv_env_get(env_use, "PATH");
3123 if (dollar_path) {
3124 if (putenv((char*) dollar_path) != 0)
3125 return log_error_errno(errno, "Failed to update $PATH: %m");
3126 }
3127
f757855e 3128 execvpe(arg_parameters[0], arg_parameters, env_use);
1a68e1e5 3129 } else {
5f932eb9 3130 if (!arg_chdir)
d929b0f9
ZJS		 3131 /* If we cannot change the directory, we'll end up in /, that is expected. */
3132 (void) chdir(home ?: "/root");
5f932eb9 3133
03cfe0d5
LP		 3134 execle("/bin/bash", "-bash", NULL, env_use);
3135 execle("/bin/sh", "-sh", NULL, env_use);
ced58da7
LP		 3136
3137 exec_target = "/bin/bash, /bin/sh";
03cfe0d5
LP		 3138 }
3139
8ca082b4 3140 return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
03cfe0d5
LP		 3141}
3142
9c1e04d0 3143static int setup_sd_notify_child(void) {
271f518f 3144 _cleanup_close_ int fd = -1;
9c1e04d0 3145 union sockaddr_union sa = {
44ed5214
LP		 3146 .un.sun_family = AF_UNIX,
3147 .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
9c1e04d0
AP		 3148 };
3149 int r;
3150
3151 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3152 if (fd < 0)
3153 return log_error_errno(errno, "Failed to allocate notification socket: %m");
3154
3155 (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
fbda85b0 3156 (void) sockaddr_un_unlink(&sa.un);
9c1e04d0 3157
9c1e04d0 3158 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
271f518f 3159 if (r < 0)
44ed5214 3160 return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
9c1e04d0 3161
adc7d9f0 3162 r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
271f518f 3163 if (r < 0)
adc7d9f0 3164 return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
adc7d9f0 3165
2ff48e98 3166 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
271f518f 3167 if (r < 0)
2ff48e98 3168 return log_error_errno(r, "SO_PASSCRED failed: %m");
9c1e04d0 3169
271f518f 3170 return TAKE_FD(fd);
9c1e04d0
AP		 3171}
3172
03cfe0d5
LP		 3173static int outer_child(
3174 Barrier *barrier,
3175 const char *directory,
2d845785 3176 DissectedImage *dissected_image,
03cfe0d5
LP		 3177 bool secondary,
3178 int pid_socket,
e01ff70a 3179 int uuid_socket,
9c1e04d0 3180 int notify_socket,
03cfe0d5
LP		 3181 int kmsg_socket,
3182 int rtnl_socket,
825d5287 3183 int uid_shift_socket,
3acc84eb 3184 int master_pty_socket,
8199d554 3185 int unified_cgroup_hierarchy_socket,
d7bea6b6
DP		 3186 FDSet *fds,
3187 int netns_fd) {
03cfe0d5 3188
bf428efb 3189 _cleanup_close_ int fd = -1;
03cfe0d5
LP		 3190 pid_t pid;
3191 ssize_t l;
de40a303 3192 int r;
03cfe0d5 3193
b37469d7
LP		 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
03cfe0d5
LP		 3199 assert(barrier);
3200 assert(directory);
03cfe0d5 3201 assert(pid_socket >= 0);
e01ff70a 3202 assert(uuid_socket >= 0);
9c1e04d0 3203 assert(notify_socket >= 0);
3acc84eb 3204 assert(master_pty_socket >= 0);
03cfe0d5
LP		 3205 assert(kmsg_socket >= 0);
3206
de40a303
LP		 3207 log_debug("Outer child is initializing.");
3208
03cfe0d5
LP		 3209 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3210 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3211
03cfe0d5
LP		 3212 r = reset_audit_loginuid();
3213 if (r < 0)
3214 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. */
60e76d48
ZJS		 3219 r = mount_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
3220 if (r < 0)
3221 return r;
03cfe0d5 3222
2d845785 3223 if (dissected_image) {
2d3a5a73
LP		 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 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
03bcb6d4
LP		 3230 DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|
3231 (arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0)|
3232 (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
2d845785
LP		 3233 if (r < 0)
3234 return r;
3235 }
03cfe0d5 3236
391567f4
LP		 3237 r = determine_uid_shift(directory);
3238 if (r < 0)
3239 return r;
3240
0de7acce 3241 if (arg_userns_mode != USER_NAMESPACE_NO) {
0e7ac751 3242 /* Let the parent know which UID shift we read from the image */
825d5287
RM		 3243 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3244 if (l < 0)
3245 return log_error_errno(errno, "Failed to send UID shift: %m");
baaa35ad
ZJS		 3246 if (l != sizeof(arg_uid_shift))
3247 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3248 "Short write while sending UID shift.");
0e7ac751 3249
0de7acce 3250 if (arg_userns_mode == USER_NAMESPACE_PICK) {
0e7ac751
LP		 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 l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
3256 if (l < 0)
3257 return log_error_errno(errno, "Failed to recv UID shift: %m");
baaa35ad
ZJS		 3258 if (l != sizeof(arg_uid_shift))
3259 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3260 "Short read while receiving UID shift.");
0e7ac751
LP		 3261 }
3262
ff6c6cc1
LP		 3263 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);
825d5287
RM		 3265 }
3266
e50cd82f
LP		 3267 if (!dissected_image) {
3268 /* Turn directory into bind mount */
3269 r = mount_verbose(LOG_ERR, directory, directory, NULL, MS_BIND|MS_REC, NULL);
3270 if (r < 0)
3271 return r;
3272 }
7d0ecdd6
LP		 3273
3274 r = setup_pivot_root(
3275 directory,
3276 arg_pivot_root_new,
3277 arg_pivot_root_old);
3278 if (r < 0)
3279 return r;
3280
3281 r = setup_volatile_mode(
3282 directory,
3283 arg_volatile_mode,
3284 arg_userns_mode != USER_NAMESPACE_NO,
3285 arg_uid_shift,
3286 arg_uid_range,
8f1ed04a 3287 arg_selinux_apifs_context);
7d0ecdd6
LP		 3288 if (r < 0)
3289 return r;
3290
2d3a5a73
LP		 3291 if (dissected_image) {
3292 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
3293 r = dissected_image_mount(dissected_image, directory, arg_uid_shift,
3294 DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|DISSECT_IMAGE_DISCARD_ON_LOOP|(arg_read_only ? DISSECT_IMAGE_READ_ONLY : 0));
3295 if (r < 0)
3296 return r;
3297 }
3298
8199d554
LP		 3299 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
3300 /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
3301
3302 r = detect_unified_cgroup_hierarchy_from_image(directory);
3303 if (r < 0)
3304 return r;
3305
3306 l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
3307 if (l < 0)
3308 return log_error_errno(errno, "Failed to send cgroup mode: %m");
baaa35ad
ZJS		 3309 if (l != sizeof(arg_unified_cgroup_hierarchy))
3310 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3311 "Short write while sending cgroup mode.");
8199d554
LP		 3312
3313 unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
3314 }
3315
4ad14eff
LP		 3316 /* Mark everything as shared so our mounts get propagated down. This is
3317 * required to make new bind mounts available in systemd services
5238e957 3318 * inside the container that create a new mount namespace.
4ad14eff
LP		 3319 * See https://github.com/systemd/systemd/issues/3860
3320 * Further submounts (such as /dev) done after this will inherit the
13e785f7 3321 * shared propagation mode. */
4ad14eff
LP		 3322 r = mount_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
3323 if (r < 0)
3324 return r;
3325
3326 r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
3327 if (r < 0)
3328 return r;
3329
03cfe0d5
LP		 3330 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3331 if (r < 0)
3332 return r;
3333
e5a4bb0d 3334 if (arg_read_only && arg_volatile_mode == VOLATILE_NO) {
64e82c19 3335 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
03cfe0d5
LP		 3336 if (r < 0)
3337 return log_error_errno(r, "Failed to make tree read-only: %m");
3338 }
3339
0de7acce 3340 r = mount_all(directory,
4f086aab 3341 arg_mount_settings,
0de7acce 3342 arg_uid_shift,
0de7acce 3343 arg_selinux_apifs_context);
03cfe0d5
LP		 3344 if (r < 0)
3345 return r;
3346
07fa00f9
LP		 3347 r = copy_devnodes(directory);
3348 if (r < 0)
03cfe0d5
LP		 3349 return r;
3350
de40a303
LP		 3351 r = make_extra_nodes(directory);
3352 if (r < 0)
3353 return r;
3354
3355 (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
3356 (void) make_inaccessible_nodes(directory, arg_uid_shift, arg_uid_shift);
03cfe0d5 3357
07fa00f9
LP		 3358 r = setup_pts(directory);
3359 if (r < 0)
03cfe0d5
LP		 3360 return r;
3361
3362 r = setup_propagate(directory);
3363 if (r < 0)
3364 return r;
3365
8e5430c4
LP		 3366 r = setup_keyring();
3367 if (r < 0)
3368 return r;
3369
03cfe0d5
LP		 3370 r = setup_timezone(directory);
3371 if (r < 0)
3372 return r;
3373
3374 r = setup_resolv_conf(directory);
3375 if (r < 0)
3376 return r;
3377
e01ff70a
MS		 3378 r = setup_machine_id(directory);
3379 if (r < 0)
3380 return r;
3381
03cfe0d5
LP		 3382 r = setup_journal(directory);
3383 if (r < 0)
3384 return r;
3385
0de7acce
LP		 3386 r = mount_custom(
3387 directory,
3388 arg_custom_mounts,
3389 arg_n_custom_mounts,
3390 arg_userns_mode != USER_NAMESPACE_NO,
3391 arg_uid_shift,
3392 arg_uid_range,
de40a303
LP		 3393 arg_selinux_apifs_context,
3394 false);
03cfe0d5
LP		 3395 if (r < 0)
3396 return r;
3397
489fae52 3398 if (!arg_use_cgns) {
0996ef00
CB		 3399 r = mount_cgroups(
3400 directory,
3401 arg_unified_cgroup_hierarchy,
3402 arg_userns_mode != USER_NAMESPACE_NO,
3403 arg_uid_shift,
3404 arg_uid_range,
5a8ff0e6 3405 arg_selinux_apifs_context,
ada54120 3406 false);
0996ef00
CB		 3407 if (r < 0)
3408 return r;
3409 }
03cfe0d5
LP		 3410
3411 r = mount_move_root(directory);
3412 if (r < 0)
3413 return log_error_errno(r, "Failed to move root directory: %m");
3414
9c1e04d0
AP		 3415 fd = setup_sd_notify_child();
3416 if (fd < 0)
3417 return fd;
3418
03cfe0d5 3419 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
0c582db0 3420 arg_clone_ns_flags |
8869a0b4 3421 (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
03cfe0d5
LP		 3422 if (pid < 0)
3423 return log_error_errno(errno, "Failed to fork inner child: %m");
03cfe0d5
LP		 3424 if (pid == 0) {
3425 pid_socket = safe_close(pid_socket);
e01ff70a 3426 uuid_socket = safe_close(uuid_socket);
9c1e04d0 3427 notify_socket = safe_close(notify_socket);
825d5287 3428 uid_shift_socket = safe_close(uid_shift_socket);
03cfe0d5
LP		 3429
3430 /* The inner child has all namespaces that are
3431 * requested, so that we all are owned by the user if
3432 * user namespaces are turned on. */
3433
d7bea6b6
DP		 3434 if (arg_network_namespace_path) {
3435 r = namespace_enter(-1, -1, netns_fd, -1, -1);
3436 if (r < 0)
e2d39e54 3437 return log_error_errno(r, "Failed to join network namespace: %m");
d7bea6b6
DP		 3438 }
3439
3acc84eb 3440 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds);
03cfe0d5
LP		 3441 if (r < 0)
3442 _exit(EXIT_FAILURE);
3443
3444 _exit(EXIT_SUCCESS);
3445 }
3446
3447 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3448 if (l < 0)
3449 return log_error_errno(errno, "Failed to send PID: %m");
baaa35ad
ZJS		 3450 if (l != sizeof(pid))
3451 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3452 "Short write while sending PID.");
03cfe0d5 3453
e01ff70a
MS		 3454 l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
3455 if (l < 0)
3456 return log_error_errno(errno, "Failed to send machine ID: %m");
baaa35ad
ZJS		 3457 if (l != sizeof(arg_uuid))
3458 return log_error_errno(SYNTHETIC_ERRNO(EIO),
3459 "Short write while sending machine ID.");
e01ff70a 3460
9c1e04d0
AP		 3461 l = send_one_fd(notify_socket, fd, 0);
3462 if (l < 0)
ba72801d 3463 return log_error_errno(l, "Failed to send notify fd: %m");
9c1e04d0 3464
03cfe0d5 3465 pid_socket = safe_close(pid_socket);
e01ff70a 3466 uuid_socket = safe_close(uuid_socket);
9c1e04d0 3467 notify_socket = safe_close(notify_socket);
3acc84eb 3468 master_pty_socket = safe_close(master_pty_socket);
327e26d6
KN		 3469 kmsg_socket = safe_close(kmsg_socket);
3470 rtnl_socket = safe_close(rtnl_socket);
d7bea6b6 3471 netns_fd = safe_close(netns_fd);
03cfe0d5
LP		 3472
3473 return 0;
3474}
3475
0e7ac751 3476static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
d381c8a6 3477 bool tried_hashed = false;
0e7ac751
LP		 3478 unsigned n_tries = 100;
3479 uid_t candidate;
3480 int r;
3481
3482 assert(shift);
3483 assert(ret_lock_file);
0de7acce 3484 assert(arg_userns_mode == USER_NAMESPACE_PICK);
0e7ac751
LP		 3485 assert(arg_uid_range == 0x10000U);
3486
3487 candidate = *shift;
3488
3489 (void) mkdir("/run/systemd/nspawn-uid", 0755);
3490
3491 for (;;) {
fbd0b64f 3492 char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
8e766630 3493 _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
0e7ac751
LP		 3494
3495 if (--n_tries <= 0)
3496 return -EBUSY;
3497
87d5e4f2 3498 if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
0e7ac751
LP		 3499 goto next;
3500 if ((candidate & UINT32_C(0xFFFF)) != 0)
3501 goto next;
3502
3503 xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
3504 r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
3505 if (r == -EBUSY) /* Range already taken by another nspawn instance */
3506 goto next;
3507 if (r < 0)
3508 return r;
3509
3510 /* Make some superficial checks whether the range is currently known in the user database */
3511 if (getpwuid(candidate))
3512 goto next;
3513 if (getpwuid(candidate + UINT32_C(0xFFFE)))
3514 goto next;
3515 if (getgrgid(candidate))
3516 goto next;
3517 if (getgrgid(candidate + UINT32_C(0xFFFE)))
3518 goto next;
3519
3520 *ret_lock_file = lf;
3521 lf = (struct LockFile) LOCK_FILE_INIT;
3522 *shift = candidate;
3523 return 0;
3524
3525 next:
d381c8a6
LP		 3526 if (arg_machine && !tried_hashed) {
3527 /* Try to hash the base from the container name */
3528
3529 static const uint8_t hash_key[] = {
3530 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
3531 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
3532 };
3533
3534 candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
3535
3536 tried_hashed = true;
3537 } else
3538 random_bytes(&candidate, sizeof(candidate));
3539
87d5e4f2 3540 candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
0e7ac751
LP		 3541 candidate &= (uid_t) UINT32_C(0xFFFF0000);
3542 }
3543}
3544
03cfe0d5 3545static int setup_uid_map(pid_t pid) {
fbd0b64f 3546 char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
03cfe0d5
LP		 3547 int r;
3548
3549 assert(pid > 1);
3550
3551 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3552 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
57512c89 3553 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
03cfe0d5
LP		 3554 if (r < 0)
3555 return log_error_errno(r, "Failed to write UID map: %m");
3556
3557 /* We always assign the same UID and GID ranges */
3558 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
57512c89 3559 r = write_string_file(uid_map, line, WRITE_STRING_FILE_DISABLE_BUFFER);
03cfe0d5
LP		 3560 if (r < 0)
3561 return log_error_errno(r, "Failed to write GID map: %m");
3562
3563 return 0;
3564}
3565
9c1e04d0 3566static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
9c1e04d0
AP		 3567 char buf[NOTIFY_BUFFER_MAX+1];
3568 char *p = NULL;
3569 struct iovec iovec = {
3570 .iov_base = buf,
3571 .iov_len = sizeof(buf)-1,
3572 };
3573 union {
3574 struct cmsghdr cmsghdr;
3575 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
3576 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
3577 } control = {};
3578 struct msghdr msghdr = {
3579 .msg_iov = &iovec,
3580 .msg_iovlen = 1,
3581 .msg_control = &control,
3582 .msg_controllen = sizeof(control),
3583 };
3584 struct cmsghdr *cmsg;
3585 struct ucred *ucred = NULL;
3586 ssize_t n;
3587 pid_t inner_child_pid;
3588 _cleanup_strv_free_ char **tags = NULL;
3589
3590 assert(userdata);
3591
3592 inner_child_pid = PTR_TO_PID(userdata);
3593
3594 if (revents != EPOLLIN) {
3595 log_warning("Got unexpected poll event for notify fd.");
3596 return 0;
3597 }
3598
3599 n = recvmsg(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
3600 if (n < 0) {
3742095b 3601 if (IN_SET(errno, EAGAIN, EINTR))
9c1e04d0
AP		 3602 return 0;
3603
3604 return log_warning_errno(errno, "Couldn't read notification socket: %m");
3605 }
3606 cmsg_close_all(&msghdr);
3607
3608 CMSG_FOREACH(cmsg, &msghdr) {
3609 if (cmsg->cmsg_level == SOL_SOCKET &&
3610 cmsg->cmsg_type == SCM_CREDENTIALS &&
3611 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
3612
3613 ucred = (struct ucred*) CMSG_DATA(cmsg);
3614 }
3615 }
3616
3617 if (!ucred || ucred->pid != inner_child_pid) {
8cb57430 3618 log_debug("Received notify message without valid credentials. Ignoring.");
9c1e04d0
AP		 3619 return 0;
3620 }
3621
3622 if ((size_t) n >= sizeof(buf)) {
3623 log_warning("Received notify message exceeded maximum size. Ignoring.");
3624 return 0;
3625 }
3626
3627 buf[n] = 0;
3628 tags = strv_split(buf, "\n\r");
3629 if (!tags)
3630 return log_oom();
3631
3632 if (strv_find(tags, "READY=1"))
04f590a4 3633 (void) sd_notifyf(false, "READY=1\n");
9c1e04d0
AP		 3634
3635 p = strv_find_startswith(tags, "STATUS=");
3636 if (p)
04f590a4 3637 (void) sd_notifyf(false, "STATUS=Container running: %s", p);
9c1e04d0
AP		 3638
3639 return 0;
3640}
3641
5773024d 3642static int setup_sd_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
9c1e04d0 3643 int r;
9c1e04d0 3644
5773024d 3645 r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
9c1e04d0
AP		 3646 if (r < 0)
3647 return log_error_errno(r, "Failed to allocate notify event source: %m");
3648
5773024d 3649 (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
9c1e04d0
AP		 3650
3651 return 0;
3652}
3653
5d961407
LP		 3654static int merge_settings(Settings *settings, const char *path) {
3655 int rl;
f757855e 3656
5d961407
LP		 3657 assert(settings);
3658 assert(path);
f757855e 3659
5d961407
LP		 3660 /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
3661 * that this steals the fields of the Settings* structure, and hence modifies it. */
f757855e 3662
7732f92b
LP		 3663 if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
3664 settings->start_mode >= 0) {
3665 arg_start_mode = settings->start_mode;
130d3d22 3666 strv_free_and_replace(arg_parameters, settings->parameters);
f757855e
LP		 3667 }
3668
a2f577fc
JL		 3669 if ((arg_settings_mask & SETTING_EPHEMERAL) == 0)
3670 arg_ephemeral = settings->ephemeral;
3671
de40a303
LP		 3672 if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
3673 settings->root) {
3674
3675 if (!arg_settings_trusted)
3676 log_warning("Ignoring root directory setting, file %s is not trusted.", path);
3677 else
3678 free_and_replace(arg_directory, settings->root);
3679 }
3680
b53ede69
PW		 3681 if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
3682 settings->pivot_root_new) {
3683 free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
3684 free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
3685 }
3686
5f932eb9 3687 if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
1cc6c93a
YW		 3688 settings->working_directory)
3689 free_and_replace(arg_chdir, settings->working_directory);
5f932eb9 3690
f757855e 3691 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
130d3d22
YW		 3692 settings->environment)
3693 strv_free_and_replace(arg_setenv, settings->environment);
f757855e 3694
de40a303
LP		 3695 if ((arg_settings_mask & SETTING_USER) == 0) {
3696
3697 if (settings->user)
3698 free_and_replace(arg_user, settings->user);
3699
3700 if (uid_is_valid(settings->uid))
3701 arg_uid = settings->uid;
3702 if (gid_is_valid(settings->gid))
3703 arg_gid = settings->gid;
3704 if (settings->n_supplementary_gids > 0) {
3705 free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
3706 arg_n_supplementary_gids = settings->n_supplementary_gids;
3707 }
3708 }
f757855e
LP		 3709
3710 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
a3fc6b55 3711 uint64_t plus, minus;
f757855e 3712
de40a303
LP		 3713 /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
3714 * Settings structure */
3715
0e265674 3716 plus = settings->capability;
a3fc6b55
LP		 3717 minus = settings->drop_capability;
3718
3719 if ((arg_settings_mask & SETTING_NETWORK) == 0) {
3720 if (settings_private_network(settings))
3721 plus |= UINT64_C(1) << CAP_NET_ADMIN;
3722 else
3723 minus |= UINT64_C(1) << CAP_NET_ADMIN;
3724 }
0e265674
LP		 3725
3726 if (!arg_settings_trusted && plus != 0) {
3727 if (settings->capability != 0)
5d961407 3728 log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
0e265674 3729 } else
520e0d54 3730 arg_caps_retain |= plus;
f757855e 3731
a3fc6b55 3732 arg_caps_retain &= ~minus;
de40a303
LP		 3733
3734 /* Copy the full capabilities over too */
3735 if (capability_quintet_is_set(&settings->full_capabilities)) {
3736 if (!arg_settings_trusted)
5238e957 3737 log_warning("Ignoring capability settings, file %s is not trusted.", path);
de40a303
LP		 3738 else
3739 arg_full_capabilities = settings->full_capabilities;
3740 }
f757855e
LP		 3741 }
3742
3743 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
3744 settings->kill_signal > 0)
3745 arg_kill_signal = settings->kill_signal;
3746
3747 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
3748 settings->personality != PERSONALITY_INVALID)
3749 arg_personality = settings->personality;
3750
3751 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
3752 !sd_id128_is_null(settings->machine_id)) {
3753
3754 if (!arg_settings_trusted)
5d961407 3755 log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
f757855e
LP		 3756 else
3757 arg_uuid = settings->machine_id;
3758 }
3759
3760 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
3761 settings->read_only >= 0)
3762 arg_read_only = settings->read_only;
3763
3764 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
3765 settings->volatile_mode != _VOLATILE_MODE_INVALID)
3766 arg_volatile_mode = settings->volatile_mode;
3767
3768 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
3769 settings->n_custom_mounts > 0) {
3770
3771 if (!arg_settings_trusted)
5d961407 3772 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
f757855e
LP		 3773 else {
3774 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
1cc6c93a 3775 arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
f757855e 3776 arg_n_custom_mounts = settings->n_custom_mounts;
f757855e
LP		 3777 settings->n_custom_mounts = 0;
3778 }
3779 }
3780
3781 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
3782 (settings->private_network >= 0 ||
3783 settings->network_veth >= 0 ||
3784 settings->network_bridge ||
22b28dfd 3785 settings->network_zone ||
f757855e
LP		 3786 settings->network_interfaces ||
3787 settings->network_macvlan ||
f6d6bad1 3788 settings->network_ipvlan ||
de40a303
LP		 3789 settings->network_veth_extra ||
3790 settings->network_namespace_path)) {
f757855e
LP		 3791
3792 if (!arg_settings_trusted)
5d961407 3793 log_warning("Ignoring network settings, file %s is not trusted.", path);
f757855e 3794 else {
f6d6bad1 3795 arg_network_veth = settings_network_veth(settings);
0e265674
LP		 3796 arg_private_network = settings_private_network(settings);
3797
130d3d22
YW		 3798 strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
3799 strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
3800 strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
3801 strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
f6d6bad1 3802
1cc6c93a
YW		 3803 free_and_replace(arg_network_bridge, settings->network_bridge);
3804 free_and_replace(arg_network_zone, settings->network_zone);
de40a303
LP		 3805
3806 free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
f757855e
LP		 3807 }
3808 }
3809
3810 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
3811 settings->expose_ports) {
3812
3813 if (!arg_settings_trusted)
5d961407 3814 log_warning("Ignoring Port= setting, file %s is not trusted.", path);
f757855e
LP		 3815 else {
3816 expose_port_free_all(arg_expose_ports);
1cc6c93a 3817 arg_expose_ports = TAKE_PTR(settings->expose_ports);
f757855e
LP		 3818 }
3819 }
3820
0de7acce
LP		 3821 if ((arg_settings_mask & SETTING_USERNS) == 0 &&
3822 settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
3823
3824 if (!arg_settings_trusted)
5d961407 3825 log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
0de7acce
LP		 3826 else {
3827 arg_userns_mode = settings->userns_mode;
3828 arg_uid_shift = settings->uid_shift;
3829 arg_uid_range = settings->uid_range;
3830 arg_userns_chown = settings->userns_chown;
3831 }
3832 }
3833
9c1e04d0
AP		 3834 if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0)
3835 arg_notify_ready = settings->notify_ready;
3836
960e4569
LP		 3837 if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
3838
de40a303 3839 if (!arg_settings_trusted && !strv_isempty(settings->syscall_whitelist))
5d961407 3840 log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
960e4569 3841 else {
130d3d22
YW		 3842 strv_free_and_replace(arg_syscall_whitelist, settings->syscall_whitelist);
3843 strv_free_and_replace(arg_syscall_blacklist, settings->syscall_blacklist);
960e4569 3844 }
de40a303
LP		 3845
3846#if HAVE_SECCOMP
3847 if (!arg_settings_trusted && settings->seccomp)
3848 log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
3849 else {
3850 seccomp_release(arg_seccomp);
3851 arg_seccomp = TAKE_PTR(settings->seccomp);
3852 }
3853#endif
960e4569
LP		 3854 }
3855
bf428efb
LP		 3856 for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
3857 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
3858 continue;
3859
3860 if (!settings->rlimit[rl])
3861 continue;
3862
3863 if (!arg_settings_trusted) {
5d961407 3864 log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
bf428efb
LP		 3865 continue;
3866 }
3867
3868 free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
3869 }
3870
3a9530e5
LP		 3871 if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
3872 settings->hostname)
3873 free_and_replace(arg_hostname, settings->hostname);
3874
66edd963
LP		 3875 if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
3876 settings->no_new_privileges >= 0)
3877 arg_no_new_privileges = settings->no_new_privileges;
3878
81f345df
LP		 3879 if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
3880 settings->oom_score_adjust_set) {
3881
3882 if (!arg_settings_trusted)
5d961407 3883 log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
81f345df
LP		 3884 else {
3885 arg_oom_score_adjust = settings->oom_score_adjust;
3886 arg_oom_score_adjust_set = true;
3887 }
3888 }
3889
d107bb7d 3890 if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
0985c7c4 3891 settings->cpu_set.set) {
d107bb7d
LP		 3892
3893 if (!arg_settings_trusted)
5d961407 3894 log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
d107bb7d 3895 else {
0985c7c4
ZJS		 3896 cpu_set_reset(&arg_cpu_set);
3897 arg_cpu_set = settings->cpu_set;
3898 settings->cpu_set = (CPUSet) {};
d107bb7d
LP		 3899 }
3900 }
3901
09d423e9
LP		 3902 if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
3903 settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
3904 arg_resolv_conf = settings->resolv_conf;
3905
4e1d6aa9
LP		 3906 if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
3907 settings->link_journal != _LINK_JOURNAL_INVALID) {
3908
3909 if (!arg_settings_trusted)
3910 log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
3911 else {
3912 arg_link_journal = settings->link_journal;
3913 arg_link_journal_try = settings->link_journal_try;
3914 }
3915 }
3916
1688841f
LP		 3917 if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
3918 settings->timezone != _TIMEZONE_MODE_INVALID)
3919 arg_timezone = settings->timezone;
3920
de40a303
LP		 3921 if ((arg_settings_mask & SETTING_SLICE) == 0 &&
3922 settings->slice) {
3923
3924 if (!arg_settings_trusted)
3925 log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
3926 else
3927 free_and_replace(arg_slice, settings->slice);
3928 }
3929
3930 if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
3931 settings->use_cgns >= 0) {
3932
3933 if (!arg_settings_trusted)
3934 log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
3935 else
3936 arg_use_cgns = settings->use_cgns;
3937 }
3938
3939 if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
3940 settings->clone_ns_flags != (unsigned long) -1) {
3941
3942 if (!arg_settings_trusted)
3943 log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
3944 else
3945 arg_clone_ns_flags = settings->clone_ns_flags;
3946 }
3947
3948 if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
3949 settings->console_mode >= 0) {
3950
3951 if (!arg_settings_trusted)
3952 log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
3953 else
3954 arg_console_mode = settings->console_mode;
3955 }
3956
3957 /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
3958 * don't consult arg_settings_mask for them. */
3959
3960 sd_bus_message_unref(arg_property_message);
3961 arg_property_message = TAKE_PTR(settings->properties);
3962
3963 arg_console_width = settings->console_width;
3964 arg_console_height = settings->console_height;
3965
b2645747 3966 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
de40a303
LP		 3967 arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
3968 arg_n_extra_nodes = settings->n_extra_nodes;
3969
f757855e
LP		 3970 return 0;
3971}
3972
5d961407
LP		 3973static int load_settings(void) {
3974 _cleanup_(settings_freep) Settings *settings = NULL;
3975 _cleanup_fclose_ FILE *f = NULL;
3976 _cleanup_free_ char *p = NULL;
3977 const char *fn, *i;
3978 int r;
3979
de40a303
LP		 3980 if (arg_oci_bundle)
3981 return 0;
3982
5d961407
LP		 3983 /* If all settings are masked, there's no point in looking for
3984 * the settings file */
3985 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
3986 return 0;
3987
3988 fn = strjoina(arg_machine, ".nspawn");
3989
3990 /* We first look in the admin's directories in /etc and /run */
3991 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
3992 _cleanup_free_ char *j = NULL;
3993
657ee2d8 3994 j = path_join(i, fn);
5d961407
LP		 3995 if (!j)
3996 return log_oom();
3997
3998 f = fopen(j, "re");
3999 if (f) {
4000 p = TAKE_PTR(j);
4001
4002 /* By default, we trust configuration from /etc and /run */
4003 if (arg_settings_trusted < 0)
4004 arg_settings_trusted = true;
4005
4006 break;
4007 }
4008
4009 if (errno != ENOENT)
4010 return log_error_errno(errno, "Failed to open %s: %m", j);
4011 }
4012
4013 if (!f) {
4014 /* After that, let's look for a file next to the
4015 * actual image we shall boot. */
4016
4017 if (arg_image) {
4018 p = file_in_same_dir(arg_image, fn);
4019 if (!p)
4020 return log_oom();
cd6e3914 4021 } else if (arg_directory && !path_equal(arg_directory, "/")) {
5d961407
LP		 4022 p = file_in_same_dir(arg_directory, fn);
4023 if (!p)
4024 return log_oom();
4025 }
4026
4027 if (p) {
4028 f = fopen(p, "re");
4029 if (!f && errno != ENOENT)
4030 return log_error_errno(errno, "Failed to open %s: %m", p);
4031
4032 /* By default, we do not trust configuration from /var/lib/machines */
4033 if (arg_settings_trusted < 0)
4034 arg_settings_trusted = false;
4035 }
4036 }
4037
4038 if (!f)
4039 return 0;
4040
4041 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4042
4043 r = settings_load(f, p, &settings);
4044 if (r < 0)
4045 return r;
4046
4047 return merge_settings(settings, p);
4048}
4049
de40a303
LP		 4050static int load_oci_bundle(void) {
4051 _cleanup_(settings_freep) Settings *settings = NULL;
4052 int r;
4053
4054 if (!arg_oci_bundle)
4055 return 0;
4056
4057 /* By default let's trust OCI bundles */
4058 if (arg_settings_trusted < 0)
4059 arg_settings_trusted = true;
4060
4061 r = oci_load(NULL, arg_oci_bundle, &settings);
4062 if (r < 0)
4063 return r;
4064
4065 return merge_settings(settings, arg_oci_bundle);
4066}
4067
3acc84eb 4068static int run_container(
2d845785 4069 DissectedImage *dissected_image,
b0067625
ZJS		 4070 bool secondary,
4071 FDSet *fds,
4072 char veth_name[IFNAMSIZ], bool *veth_created,
4073 union in_addr_union *exposed,
3acc84eb 4074 int *master, pid_t *pid, int *ret) {
b0067625
ZJS		 4075
4076 static const struct sigaction sa = {
4077 .sa_handler = nop_signal_handler,
e28c7cd0 4078 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
b0067625
ZJS		 4079 };
4080
8e766630 4081 _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
b0067625
ZJS		 4082 _cleanup_close_ int etc_passwd_lock = -1;
4083 _cleanup_close_pair_ int
4084 kmsg_socket_pair[2] = { -1, -1 },
4085 rtnl_socket_pair[2] = { -1, -1 },
4086 pid_socket_pair[2] = { -1, -1 },
4087 uuid_socket_pair[2] = { -1, -1 },
4088 notify_socket_pair[2] = { -1, -1 },
8199d554 4089 uid_shift_socket_pair[2] = { -1, -1 },
3acc84eb 4090 master_pty_socket_pair[2] = { -1, -1 },
8199d554
LP		 4091 unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
4092
3acc84eb 4093 _cleanup_close_ int notify_socket = -1;
b0067625 4094 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
5773024d 4095 _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
b0067625
ZJS		 4096 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
4097 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4098 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
abdb9b08 4099 _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
b0067625 4100 ContainerStatus container_status = 0;
b0067625
ZJS		 4101 int ifi = 0, r;
4102 ssize_t l;
4103 sigset_t mask_chld;
d7bea6b6 4104 _cleanup_close_ int netns_fd = -1;
b0067625
ZJS		 4105
4106 assert_se(sigemptyset(&mask_chld) == 0);
4107 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4108
4109 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4110 /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
4111 * check with getpwuid() if the specific user already exists. Note that /etc might be
4112 * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
4113 * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
4114 * really just an extra safety net. We kinda assume that the UID range we allocate from is
4115 * really ours. */
4116
4117 etc_passwd_lock = take_etc_passwd_lock(NULL);
4118 if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
4119 return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
4120 }
4121
4122 r = barrier_create(&barrier);
4123 if (r < 0)
4124 return log_error_errno(r, "Cannot initialize IPC barrier: %m");
4125
4126 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
4127 return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4128
4129 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
4130 return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4131
4132 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
4133 return log_error_errno(errno, "Failed to create pid socket pair: %m");
4134
4135 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
4136 return log_error_errno(errno, "Failed to create id socket pair: %m");
4137
4138 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
4139 return log_error_errno(errno, "Failed to create notify socket pair: %m");
4140
3acc84eb
FB		 4141 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
4142 return log_error_errno(errno, "Failed to create console socket pair: %m");
4143
b0067625
ZJS		 4144 if (arg_userns_mode != USER_NAMESPACE_NO)
4145 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
4146 return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4147
8199d554
LP		 4148 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
4149 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
4150 return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
4151
b0067625
ZJS		 4152 /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
4153 * parent's blocking calls and give it a chance to call wait() and terminate. */
4154 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4155 if (r < 0)
4156 return log_error_errno(errno, "Failed to change the signal mask: %m");
4157
4158 r = sigaction(SIGCHLD, &sa, NULL);
4159 if (r < 0)
4160 return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4161
d7bea6b6
DP		 4162 if (arg_network_namespace_path) {
4163 netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
4164 if (netns_fd < 0)
4165 return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
4166
4167 r = fd_is_network_ns(netns_fd);
6619ad88
LP		 4168 if (r == -EUCLEAN)
4169 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);
4170 else if (r < 0)
d7bea6b6 4171 return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
c6147113
LP		 4172 else if (r == 0)
4173 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4174 "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
d7bea6b6
DP		 4175 }
4176
b0067625
ZJS		 4177 *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
4178 if (*pid < 0)
4179 return log_error_errno(errno, "clone() failed%s: %m",
4180 errno == EINVAL ?
4181 ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
4182
4183 if (*pid == 0) {
4184 /* The outer child only has a file system namespace. */
4185 barrier_set_role(&barrier, BARRIER_CHILD);
4186
b0067625
ZJS		 4187 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4188 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4189 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4190 uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
4191 notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
3acc84eb 4192 master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
b0067625 4193 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
8199d554 4194 unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
b0067625
ZJS		 4195
4196 (void) reset_all_signal_handlers();
4197 (void) reset_signal_mask();
4198
4199 r = outer_child(&barrier,
4200 arg_directory,
2d845785 4201 dissected_image,
b0067625
ZJS		 4202 secondary,
4203 pid_socket_pair[1],
4204 uuid_socket_pair[1],
4205 notify_socket_pair[1],
4206 kmsg_socket_pair[1],
4207 rtnl_socket_pair[1],
4208 uid_shift_socket_pair[1],
3acc84eb 4209 master_pty_socket_pair[1],
8199d554 4210 unified_cgroup_hierarchy_socket_pair[1],
d7bea6b6
DP		 4211 fds,
4212 netns_fd);
b0067625
ZJS		 4213 if (r < 0)
4214 _exit(EXIT_FAILURE);
4215
4216 _exit(EXIT_SUCCESS);
4217 }
4218
4219 barrier_set_role(&barrier, BARRIER_PARENT);
4220
e4077ff6 4221 fdset_close(fds);
b0067625
ZJS		 4222
4223 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4224 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4225 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4226 uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
4227 notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
3acc84eb 4228 master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
b0067625 4229 uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
8199d554 4230 unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
b0067625
ZJS		 4231
4232 if (arg_userns_mode != USER_NAMESPACE_NO) {
4233 /* The child just let us know the UID shift it might have read from the image. */
4234 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
4235 if (l < 0)
4236 return log_error_errno(errno, "Failed to read UID shift: %m");
c6147113
LP		 4237 if (l != sizeof arg_uid_shift)
4238 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
b0067625
ZJS		 4239
4240 if (arg_userns_mode == USER_NAMESPACE_PICK) {
4241 /* If we are supposed to pick the UID shift, let's try to use the shift read from the
4242 * image, but if that's already in use, pick a new one, and report back to the child,
4243 * which one we now picked. */
4244
4245 r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
4246 if (r < 0)
4247 return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
4248
4249 l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
4250 if (l < 0)
4251 return log_error_errno(errno, "Failed to send UID shift: %m");
c6147113
LP		 4252 if (l != sizeof arg_uid_shift)
4253 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
b0067625
ZJS		 4254 }
4255 }
4256
8199d554
LP		 4257 if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
4258 /* The child let us know the support cgroup mode it might have read from the image. */
4259 l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
4260 if (l < 0)
4261 return log_error_errno(errno, "Failed to read cgroup mode: %m");
c6147113
LP		 4262 if (l != sizeof(arg_unified_cgroup_hierarchy))
4263 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zu bytes).%s",
4264 l, l == 0 ? " The child is most likely dead." : "");
8199d554
LP		 4265 }
4266
b0067625 4267 /* Wait for the outer child. */
d2e0ac3d
LP		 4268 r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
4269 if (r < 0)
4270 return r;
4271 if (r != EXIT_SUCCESS)
4272 return -EIO;
b0067625
ZJS		 4273
4274 /* And now retrieve the PID of the inner child. */
4275 l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
4276 if (l < 0)
4277 return log_error_errno(errno, "Failed to read inner child PID: %m");
c6147113
LP		 4278 if (l != sizeof *pid)
4279 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
b0067625
ZJS		 4280
4281 /* We also retrieve container UUID in case it was generated by outer child */
4282 l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
4283 if (l < 0)
4284 return log_error_errno(errno, "Failed to read container machine ID: %m");
c6147113
LP		 4285 if (l != sizeof(arg_uuid))
4286 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
b0067625
ZJS		 4287
4288 /* We also retrieve the socket used for notifications generated by outer child */
4289 notify_socket = receive_one_fd(notify_socket_pair[0], 0);
4290 if (notify_socket < 0)
4291 return log_error_errno(notify_socket,
4292 "Failed to receive notification socket from the outer child: %m");
4293
4294 log_debug("Init process invoked as PID "PID_FMT, *pid);
4295
4296 if (arg_userns_mode != USER_NAMESPACE_NO) {
c6147113
LP		 4297 if (!barrier_place_and_sync(&barrier)) /* #1 */
4298 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
b0067625
ZJS		 4299
4300 r = setup_uid_map(*pid);
4301 if (r < 0)
4302 return r;
4303
4304 (void) barrier_place(&barrier); /* #2 */
4305 }
4306
4307 if (arg_private_network) {
75116558
PS		 4308 if (!arg_network_namespace_path) {
4309 /* Wait until the child has unshared its network namespace. */
c6147113
LP		 4310 if (!barrier_place_and_sync(&barrier)) /* #3 */
4311 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
75116558
PS		 4312 }
4313
b0067625
ZJS		 4314 r = move_network_interfaces(*pid, arg_network_interfaces);
4315 if (r < 0)
4316 return r;
4317
4318 if (arg_network_veth) {
4319 r = setup_veth(arg_machine, *pid, veth_name,
4320 arg_network_bridge || arg_network_zone);
4321 if (r < 0)
4322 return r;
4323 else if (r > 0)
4324 ifi = r;
4325
4326 if (arg_network_bridge) {
4327 /* Add the interface to a bridge */
4328 r = setup_bridge(veth_name, arg_network_bridge, false);
4329 if (r < 0)
4330 return r;
4331 if (r > 0)
4332 ifi = r;
4333 } else if (arg_network_zone) {
4334 /* Add the interface to a bridge, possibly creating it */
4335 r = setup_bridge(veth_name, arg_network_zone, true);
4336 if (r < 0)
4337 return r;
4338 if (r > 0)
4339 ifi = r;
4340 }
4341 }
4342
4343 r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
4344 if (r < 0)
4345 return r;
4346
4347 /* We created the primary and extra veth links now; let's remember this, so that we know to
4348 remove them later on. Note that we don't bother with removing veth links that were created
4349 here when their setup failed half-way, because in that case the kernel should be able to
4350 remove them on its own, since they cannot be referenced by anything yet. */
4351 *veth_created = true;
4352
4353 r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
4354 if (r < 0)
4355 return r;
4356
4357 r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
4358 if (r < 0)
4359 return r;
4360 }
4361
abdb9b08
LP		 4362 if (arg_register || !arg_keep_unit) {
4363 r = sd_bus_default_system(&bus);
4364 if (r < 0)
4365 return log_error_errno(r, "Failed to open system bus: %m");
e5a2d8b5
LP		 4366
4367 r = sd_bus_set_close_on_exit(bus, false);
4368 if (r < 0)
4369 return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
abdb9b08
LP		 4370 }
4371
4372 if (!arg_keep_unit) {
4373 /* When a new scope is created for this container, then we'll be registered as its controller, in which
4374 * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
4375 * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
4376
75152a4d
LP		 4377 r = sd_bus_match_signal_async(
4378 bus,
4379 NULL,
4380 "org.freedesktop.systemd1",
4381 NULL,
4382 "org.freedesktop.systemd1.Scope",
4383 "RequestStop",
4384 on_request_stop, NULL, PID_TO_PTR(*pid));
abdb9b08 4385 if (r < 0)
75152a4d 4386 return log_error_errno(r, "Failed to request RequestStop match: %m");
abdb9b08
LP		 4387 }
4388
b0067625
ZJS		 4389 if (arg_register) {
4390 r = register_machine(
abdb9b08 4391 bus,
b0067625
ZJS		 4392 arg_machine,
4393 *pid,
4394 arg_directory,
4395 arg_uuid,
4396 ifi,
4397 arg_slice,
4398 arg_custom_mounts, arg_n_custom_mounts,
4399 arg_kill_signal,
4400 arg_property,
de40a303 4401 arg_property_message,
b0067625
ZJS		 4402 arg_keep_unit,
4403 arg_container_service_name);
4404 if (r < 0)
4405 return r;
abdb9b08 4406
cd2dfc6f
LP		 4407 } else if (!arg_keep_unit) {
4408 r = allocate_scope(
abdb9b08 4409 bus,
cd2dfc6f
LP		 4410 arg_machine,
4411 *pid,
4412 arg_slice,
4413 arg_custom_mounts, arg_n_custom_mounts,
4414 arg_kill_signal,
de40a303
LP		 4415 arg_property,
4416 arg_property_message);
cd2dfc6f
LP		 4417 if (r < 0)
4418 return r;
4419
4420 } else if (arg_slice || arg_property)
4421 log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
b0067625 4422
27da7ef0 4423 r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
b0067625
ZJS		 4424 if (r < 0)
4425 return r;
4426
27da7ef0 4427 r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
720f0a2f
LP		 4428 if (r < 0)
4429 return r;
b0067625 4430
de54e02d 4431 r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
b0067625
ZJS		 4432 if (r < 0)
4433 return r;
4434
4435 /* Notify the child that the parent is ready with all
4436 * its setup (including cgroup-ification), and that
4437 * the child can now hand over control to the code to
4438 * run inside the container. */
75116558 4439 (void) barrier_place(&barrier); /* #4 */
b0067625
ZJS		 4440
4441 /* Block SIGCHLD here, before notifying child.
4442 * process_pty() will handle it with the other signals. */
4443 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4444
4445 /* Reset signal to default */
4446 r = default_signals(SIGCHLD, -1);
4447 if (r < 0)
4448 return log_error_errno(r, "Failed to reset SIGCHLD: %m");
4449
4450 r = sd_event_new(&event);
4451 if (r < 0)
4452 return log_error_errno(r, "Failed to get default event source: %m");
4453
8fd010bb
LP		 4454 (void) sd_event_set_watchdog(event, true);
4455
abdb9b08
LP		 4456 if (bus) {
4457 r = sd_bus_attach_event(bus, event, 0);
4458 if (r < 0)
4459 return log_error_errno(r, "Failed to attach bus to event loop: %m");
4460 }
4461
5773024d 4462 r = setup_sd_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
b0067625
ZJS		 4463 if (r < 0)
4464 return r;
4465
4466 /* Let the child know that we are ready and wait that the child is completely ready now. */
c6147113
LP		 4467 if (!barrier_place_and_sync(&barrier)) /* #5 */
4468 return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
b0067625
ZJS		 4469
4470 /* At this point we have made use of the UID we picked, and thus nss-mymachines
4471 * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
4472 etc_passwd_lock = safe_close(etc_passwd_lock);
4473
04f590a4
LP		 4474 (void) sd_notifyf(false,
4475 "STATUS=Container running.\n"
4476 "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
b0067625 4477 if (!arg_notify_ready)
919f5ae0 4478 (void) sd_notify(false, "READY=1\n");
b0067625
ZJS		 4479
4480 if (arg_kill_signal > 0) {
4481 /* Try to kill the init system on SIGINT or SIGTERM */
919f5ae0
LP		 4482 (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
4483 (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
b0067625
ZJS		 4484 } else {
4485 /* Immediately exit */
919f5ae0
LP		 4486 (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4487 (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
b0067625
ZJS		 4488 }
4489
6916b164 4490 /* Exit when the child exits */
919f5ae0 4491 (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
b0067625
ZJS		 4492
4493 if (arg_expose_ports) {
4494 r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, exposed, &rtnl);
4495 if (r < 0)
4496 return r;
4497
4498 (void) expose_port_execute(rtnl, arg_expose_ports, exposed);
4499 }
4500
4501 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4502
3acc84eb
FB		 4503 if (arg_console_mode != CONSOLE_PIPE) {
4504 _cleanup_close_ int fd = -1;
4505 PTYForwardFlags flags = 0;
de40a303 4506
3acc84eb
FB		 4507 /* Retrieve the master pty allocated by inner child */
4508 fd = receive_one_fd(master_pty_socket_pair[0], 0);
4509 if (fd < 0)
4510 return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
4511
4512 switch (arg_console_mode) {
de40a303 4513
3acc84eb
FB		 4514 case CONSOLE_READ_ONLY:
4515 flags |= PTY_FORWARD_READ_ONLY;
4516
4517 _fallthrough_;
4518
4519 case CONSOLE_INTERACTIVE:
4520 flags |= PTY_FORWARD_IGNORE_VHANGUP;
4521
4522 r = pty_forward_new(event, fd, flags, &forward);
4523 if (r < 0)
4524 return log_error_errno(r, "Failed to create PTY forwarder: %m");
4525
4526 if (arg_console_width != (unsigned) -1 || arg_console_height != (unsigned) -1)
4527 (void) pty_forward_set_width_height(forward,
4528 arg_console_width,
4529 arg_console_height);
4530 break;
4531
4532 default:
4533 assert(arg_console_mode == CONSOLE_PASSIVE);
4534 }
4535
4536 *master = TAKE_FD(fd);
de40a303 4537 }
b0067625
ZJS		 4538
4539 r = sd_event_loop(event);
4540 if (r < 0)
4541 return log_error_errno(r, "Failed to run event loop: %m");
4542
de40a303
LP		 4543 if (forward) {
4544 char last_char = 0;
b0067625 4545
de40a303
LP		 4546 (void) pty_forward_get_last_char(forward, &last_char);
4547 forward = pty_forward_free(forward);
b0067625 4548
de40a303
LP		 4549 if (!arg_quiet && last_char != '\n')
4550 putc('\n', stdout);
4551 }
b0067625
ZJS		 4552
4553 /* Kill if it is not dead yet anyway */
1d78fea2
LP		 4554 if (bus) {
4555 if (arg_register)
4556 terminate_machine(bus, arg_machine);
4557 else if (!arg_keep_unit)
4558 terminate_scope(bus, arg_machine);
4559 }
b0067625
ZJS		 4560
4561 /* Normally redundant, but better safe than sorry */
c67b0082 4562 (void) kill(*pid, SIGKILL);
b0067625
ZJS		 4563
4564 r = wait_for_container(*pid, &container_status);
4565 *pid = 0;
4566
4567 if (r < 0)
4568 /* We failed to wait for the container, or the container exited abnormally. */
4569 return r;
4570 if (r > 0 || container_status == CONTAINER_TERMINATED) {
27e29a1e
ZJS		 4571 /* r > 0 → The container exited with a non-zero status.
4572 * As a special case, we need to replace 133 with a different value,
4573 * because 133 is special-cased in the service file to reboot the container.
4574 * otherwise → The container exited with zero status and a reboot was not requested.
4575 */
2a49b612 4576 if (r == EXIT_FORCE_RESTART)
27e29a1e 4577 r = EXIT_FAILURE; /* replace 133 with the general failure code */
b0067625 4578 *ret = r;
b0067625
ZJS		 4579 return 0; /* finito */
4580 }
4581
4582 /* CONTAINER_REBOOTED, loop again */
4583
4584 if (arg_keep_unit) {
4585 /* Special handling if we are running as a service: instead of simply
4586 * restarting the machine we want to restart the entire service, so let's
4587 * inform systemd about this with the special exit code 133. The service
4588 * file uses RestartForceExitStatus=133 so that this results in a full
4589 * nspawn restart. This is necessary since we might have cgroup parameters
4590 * set we want to have flushed out. */
2a49b612
ZJS		 4591 *ret = EXIT_FORCE_RESTART;
4592 return 0; /* finito */
b0067625
ZJS		 4593 }
4594
4595 expose_port_flush(arg_expose_ports, exposed);
4596
4597 (void) remove_veth_links(veth_name, arg_network_veth_extra);
4598 *veth_created = false;
4599 return 1; /* loop again */
4600}
4601
bf428efb 4602static int initialize_rlimits(void) {
bf428efb
LP		 4603 /* The default resource limits the kernel passes to PID 1, as per kernel 4.16. Let's pass our container payload
4604 * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
4605 * container execution environments. */
4606
4607 static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
4608 [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
4609 [RLIMIT_CORE] = { 0, RLIM_INFINITY },
4610 [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
4611 [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
4612 [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
4613 [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
4614 [RLIMIT_MEMLOCK] = { 65536, 65536 },
4615 [RLIMIT_MSGQUEUE] = { 819200, 819200 },
4616 [RLIMIT_NICE] = { 0, 0 },
4617 [RLIMIT_NOFILE] = { 1024, 4096 },
4618 [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
4619 [RLIMIT_RTPRIO] = { 0, 0 },
4620 [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
4621 [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
4622
4623 /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
4624 * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
4625 * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
4626 * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
4627 * that PID 1 changes a number of other resource limits during early initialization which is why we
4628 * don't read the other limits from PID 1 but prefer the static table above. */
4629 };
4630
4631 int rl;
4632
4633 for (rl = 0; rl < _RLIMIT_MAX; rl++) {
bf428efb
LP		 4634 /* Let's only fill in what the user hasn't explicitly configured anyway */
4635 if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
4636 const struct rlimit *v;
4637 struct rlimit buffer;
4638
4639 if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
4640 /* For these two let's read the limits off PID 1. See above for an explanation. */
4641
4642 if (prlimit(1, rl, NULL, &buffer) < 0)
4643 return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
4644
4645 v = &buffer;
4646 } else
4647 v = kernel_defaults + rl;
4648
4649 arg_rlimit[rl] = newdup(struct rlimit, v, 1);
4650 if (!arg_rlimit[rl])
4651 return log_oom();
4652 }
4653
4654 if (DEBUG_LOGGING) {
4655 _cleanup_free_ char *k = NULL;
4656
4657 (void) rlimit_format(arg_rlimit[rl], &k);
4658 log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
4659 }
4660 }
4661
4662 return 0;
4663}
4664
44dbef90 4665static int run(int argc, char *argv[]) {
7bf011e3
LP		 4666 bool secondary = false, remove_directory = false, remove_image = false,
4667 veth_created = false, remove_tmprootdir = false;
2d845785 4668 _cleanup_close_ int master = -1;
03cfe0d5 4669 _cleanup_fdset_free_ FDSet *fds = NULL;
2d845785 4670 int r, n_fd_passed, ret = EXIT_SUCCESS;
5aa3eba5 4671 char veth_name[IFNAMSIZ] = "";
03cfe0d5 4672 union in_addr_union exposed = {};
8e766630 4673 _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
c67b0082 4674 char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
2d845785 4675 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
18b5886e
LP		 4676 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4677 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
7bf011e3 4678 pid_t pid = 0;
03cfe0d5
LP		 4679
4680 log_parse_environment();
4681 log_open();
415fc41c 4682
03cfe0d5
LP		 4683 r = parse_argv(argc, argv);
4684 if (r <= 0)
4685 goto finish;
4686
fba868fa
LP		 4687 r = must_be_root();
4688 if (r < 0)
03cfe0d5 4689 goto finish;
fba868fa 4690
bf428efb
LP		 4691 r = initialize_rlimits();
4692 if (r < 0)
4693 goto finish;
4694
de40a303
LP		 4695 r = load_oci_bundle();
4696 if (r < 0)
4697 goto finish;
4698
f757855e
LP		 4699 r = determine_names();
4700 if (r < 0)
4701 goto finish;
4702
4703 r = load_settings();
4704 if (r < 0)
4705 goto finish;
4706
5eee8290
LP		 4707 r = cg_unified_flush();
4708 if (r < 0) {
4709 log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
4710 goto finish;
4711 }
4712
f757855e
LP		 4713 r = verify_arguments();
4714 if (r < 0)
4715 goto finish;
03cfe0d5 4716
8199d554
LP		 4717 r = detect_unified_cgroup_hierarchy_from_environment();
4718 if (r < 0)
4719 goto finish;
4720
2949ff26
LP		 4721 /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
4722 * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
4723 * so just turning this off here means we only turn it off in nspawn itself, not any children. */
4724 (void) ignore_signals(SIGPIPE, -1);
4725
03cfe0d5
LP		 4726 n_fd_passed = sd_listen_fds(false);
4727 if (n_fd_passed > 0) {
4728 r = fdset_new_listen_fds(&fds, false);
4729 if (r < 0) {
4730 log_error_errno(r, "Failed to collect file descriptors: %m");
4731 goto finish;
4732 }
4733 }
4734
83e803a9
ZJS		 4735 /* The "default" umask. This is appropriate for most file and directory
4736 * operations performed by nspawn, and is the umask that will be used for
4737 * the child. Functions like copy_devnodes() change the umask temporarily. */
4738 umask(0022);
4739
03cfe0d5
LP		 4740 if (arg_directory) {
4741 assert(!arg_image);
4742
b35ca61a
LP		 4743 /* Safety precaution: let's not allow running images from the live host OS image, as long as
4744 * /var from the host will propagate into container dynamically (because bad things happen if
4745 * two systems write to the same /var). Let's allow it for the special cases where /var is
4746 * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
4747 if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
4748 log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
03cfe0d5
LP		 4749 r = -EINVAL;
4750 goto finish;
4751 }
4752
4753 if (arg_ephemeral) {
4754 _cleanup_free_ char *np = NULL;
4755
8d4aa2bb 4756 r = chase_symlinks_and_update(&arg_directory, 0);
3f342ec4
LP		 4757 if (r < 0)
4758 goto finish;
4759
7bf011e3
LP		 4760 /* If the specified path is a mount point we generate the new snapshot immediately
4761 * inside it under a random name. However if the specified is not a mount point we
4762 * create the new snapshot in the parent directory, just next to it. */
e1873695 4763 r = path_is_mount_point(arg_directory, NULL, 0);
03cfe0d5
LP		 4764 if (r < 0) {
4765 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4766 goto finish;
4767 }
4768 if (r > 0)
770b5ce4 4769 r = tempfn_random_child(arg_directory, "machine.", &np);
03cfe0d5 4770 else
770b5ce4 4771 r = tempfn_random(arg_directory, "machine.", &np);
03cfe0d5 4772 if (r < 0) {
0f3be6ca 4773 log_error_errno(r, "Failed to generate name for directory snapshot: %m");
03cfe0d5
LP		 4774 goto finish;
4775 }
4776
6992459c
LP		 4777 /* We take an exclusive lock on this image, since it's our private, ephemeral copy
4778 * only owned by us and noone else. */
4779 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
03cfe0d5
LP		 4780 if (r < 0) {
4781 log_error_errno(r, "Failed to lock %s: %m", np);
4782 goto finish;
4783 }
4784
7bf011e3
LP		 4785 {
4786 BLOCK_SIGNALS(SIGINT);
4787 r = btrfs_subvol_snapshot(arg_directory, np,
4788 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4789 BTRFS_SNAPSHOT_FALLBACK_COPY |
4790 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4791 BTRFS_SNAPSHOT_RECURSIVE |
4792 BTRFS_SNAPSHOT_QUOTA |
4793 BTRFS_SNAPSHOT_SIGINT);
4794 }
4795 if (r == -EINTR) {
4796 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
4797 goto finish;
4798 }
03cfe0d5
LP		 4799 if (r < 0) {
4800 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4801 goto finish;
ec16945e
LP		 4802 }
4803
1cc6c93a 4804 free_and_replace(arg_directory, np);
17cbb288 4805 remove_directory = true;
30535c16 4806 } else {
cb638b5e 4807 r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
8d4aa2bb
LP		 4808 if (r < 0)
4809 goto finish;
4810
30535c16
LP		 4811 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4812 if (r == -EBUSY) {
4813 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4814 goto finish;
4815 }
4816 if (r < 0) {
4817 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
476b8254 4818 goto finish;
30535c16
LP		 4819 }
4820
4821 if (arg_template) {
8d4aa2bb 4822 r = chase_symlinks_and_update(&arg_template, 0);
3f342ec4
LP		 4823 if (r < 0)
4824 goto finish;
4825
7bf011e3
LP		 4826 {
4827 BLOCK_SIGNALS(SIGINT);
4828 r = btrfs_subvol_snapshot(arg_template, arg_directory,
4829 (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
4830 BTRFS_SNAPSHOT_FALLBACK_COPY |
4831 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
4832 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
4833 BTRFS_SNAPSHOT_RECURSIVE |
4834 BTRFS_SNAPSHOT_QUOTA |
4835 BTRFS_SNAPSHOT_SIGINT);
4836 }
ff6c6cc1
LP		 4837 if (r == -EEXIST)
4838 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4839 "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
7bf011e3
LP		 4840 else if (r == -EINTR) {
4841 log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
4842 goto finish;
4843 } else if (r < 0) {
83521414 4844 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
30535c16 4845 goto finish;
ff6c6cc1
LP		 4846 } else
4847 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
4848 "Populated %s from template %s.", arg_directory, arg_template);
30535c16 4849 }
ec16945e
LP		 4850 }
4851
7732f92b 4852 if (arg_start_mode == START_BOOT) {
a5201ed6 4853 const char *p;
c9fe05e0 4854
a5201ed6
LP		 4855 if (arg_pivot_root_new)
4856 p = prefix_roota(arg_directory, arg_pivot_root_new);
4857 else
4858 p = arg_directory;
c9fe05e0
AR		 4859
4860 if (path_is_os_tree(p) <= 0) {
4861 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
ec16945e 4862 r = -EINVAL;
1b9e5b12
LP		 4863 goto finish;
4864 }
4865 } else {
c9fe05e0
AR		 4866 const char *p, *q;
4867
a5201ed6
LP		 4868 if (arg_pivot_root_new)
4869 p = prefix_roota(arg_directory, arg_pivot_root_new);
4870 else
4871 p = arg_directory;
c9fe05e0
AR		 4872
4873 q = strjoina(p, "/usr/");
1b9e5b12 4874
c9fe05e0
AR		 4875 if (laccess(q, F_OK) < 0) {
4876 log_error("Directory %s doesn't look like it has an OS tree. Refusing.", p);
ec16945e 4877 r = -EINVAL;
1b9e5b12 4878 goto finish;
1b9e5b12
LP		 4879 }
4880 }
ec16945e 4881
6b9132a9 4882 } else {
ec16945e
LP		 4883 assert(arg_image);
4884 assert(!arg_template);
4885
8d4aa2bb 4886 r = chase_symlinks_and_update(&arg_image, 0);
3f342ec4
LP		 4887 if (r < 0)
4888 goto finish;
4889
0f3be6ca
LP		 4890 if (arg_ephemeral) {
4891 _cleanup_free_ char *np = NULL;
4892
4893 r = tempfn_random(arg_image, "machine.", &np);
4894 if (r < 0) {
4895 log_error_errno(r, "Failed to generate name for image snapshot: %m");
4896 goto finish;
4897 }
4898
6992459c
LP		 4899 /* Always take an exclusive lock on our own ephemeral copy. */
4900 r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
0f3be6ca
LP		 4901 if (r < 0) {
4902 r = log_error_errno(r, "Failed to create image lock: %m");
4903 goto finish;
4904 }
4905
7bf011e3
LP		 4906 {
4907 BLOCK_SIGNALS(SIGINT);
4908 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);
4909 }
4910 if (r == -EINTR) {
4911 log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
4912 goto finish;
4913 }
0f3be6ca
LP		 4914 if (r < 0) {
4915 r = log_error_errno(r, "Failed to copy image file: %m");
4916 goto finish;
4917 }
4918
1cc6c93a 4919 free_and_replace(arg_image, np);
0f3be6ca
LP		 4920 remove_image = true;
4921 } else {
4922 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4923 if (r == -EBUSY) {
4924 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4925 goto finish;
4926 }
4927 if (r < 0) {
4928 r = log_error_errno(r, "Failed to create image lock: %m");
4929 goto finish;
4930 }
4623e8e6 4931
78ebe980
LP		 4932 if (!arg_root_hash) {
4933 r = root_hash_load(arg_image, &arg_root_hash, &arg_root_hash_size);
4934 if (r < 0) {
4935 log_error_errno(r, "Failed to load root hash file for %s: %m", arg_image);
4936 goto finish;
4937 }
4938 }
30535c16
LP		 4939 }
4940
c67b0082 4941 if (!mkdtemp(tmprootdir)) {
0f3be6ca 4942 r = log_error_errno(errno, "Failed to create temporary directory: %m");
6b9132a9 4943 goto finish;
1b9e5b12 4944 }
6b9132a9 4945
c67b0082
LP		 4946 remove_tmprootdir = true;
4947
4948 arg_directory = strdup(tmprootdir);
1b9e5b12
LP		 4949 if (!arg_directory) {
4950 r = log_oom();
4951 goto finish;
6b9132a9 4952 }
88213476 4953
2d845785
LP		 4954 r = loop_device_make_by_path(arg_image, arg_read_only ? O_RDONLY : O_RDWR, &loop);
4955 if (r < 0) {
4956 log_error_errno(r, "Failed to set up loopback block device: %m");
842f3b0f
LP		 4957 goto finish;
4958 }
1b9e5b12 4959
4526113f 4960 r = dissect_image_and_warn(
e0f9e7bd 4961 loop->fd,
4526113f 4962 arg_image,
e0f9e7bd
LP		 4963 arg_root_hash, arg_root_hash_size,
4964 DISSECT_IMAGE_REQUIRE_ROOT,
4965 &dissected_image);
2d845785 4966 if (r == -ENOPKG) {
4526113f 4967 /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
2d845785
LP		 4968 log_notice("Note that the disk image needs to\n"
4969 " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
4970 " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
4971 " c) or follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n"
4972 " d) or contain a file system without a partition table\n"
4973 "in order to be bootable with systemd-nspawn.");
1b9e5b12 4974 goto finish;
2d845785 4975 }
4526113f 4976 if (r < 0)
842f3b0f 4977 goto finish;
1b9e5b12 4978
4623e8e6
LP		 4979 if (!arg_root_hash && dissected_image->can_verity)
4980 log_notice("Note: image %s contains verity information, but no root hash specified! Proceeding without integrity checking.", arg_image);
4981
4982 r = dissected_image_decrypt_interactively(dissected_image, NULL, arg_root_hash, arg_root_hash_size, 0, &decrypted_image);
1b9e5b12
LP		 4983 if (r < 0)
4984 goto finish;
0f3be6ca
LP		 4985
4986 /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
4987 if (remove_image && unlink(arg_image) >= 0)
4988 remove_image = false;
842f3b0f 4989 }
842f3b0f 4990
86c0dd4a 4991 r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
5a8af538
LP		 4992 if (r < 0)
4993 goto finish;
4994
de40a303
LP		 4995 if (arg_console_mode < 0)
4996 arg_console_mode =
4997 isatty(STDIN_FILENO) > 0 &&
4998 isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
9c857b9d 4999
de40a303
LP		 5000 if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
5001 arg_quiet = true;
a258bf26 5002
9c857b9d
LP		 5003 if (!arg_quiet)
5004 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
5005 arg_machine, arg_image ?: arg_directory);
5006
72c0a2c2 5007 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
a258bf26 5008
66edd963 5009 if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
03cfe0d5
LP		 5010 r = log_error_errno(errno, "Failed to become subreaper: %m");
5011 goto finish;
5012 }
5013
d87be9b0 5014 for (;;) {
3acc84eb 5015 r = run_container(dissected_image,
44dbef90
LP		 5016 secondary,
5017 fds,
5018 veth_name, &veth_created,
3acc84eb 5019 &exposed, &master,
44dbef90 5020 &pid, &ret);
b0067625 5021 if (r <= 0)
d87be9b0 5022 break;
d87be9b0 5023 }
88213476
LP		 5024
5025finish:
04f590a4
LP		 5026 (void) sd_notify(false,
5027 r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
5028 "STOPPING=1\nSTATUS=Terminating...");
af4ec430 5029
9444b1f2 5030 if (pid > 0)
c67b0082 5031 (void) kill(pid, SIGKILL);
88213476 5032
503546da 5033 /* Try to flush whatever is still queued in the pty */
6a0f896b 5034 if (master >= 0) {
1c876927 5035 (void) copy_bytes(master, STDOUT_FILENO, (uint64_t) -1, 0);
6a0f896b
LP		 5036 master = safe_close(master);
5037 }
5038
5039 if (pid > 0)
5040 (void) wait_for_terminate(pid, NULL);
503546da 5041
50ebcf6c
LP		 5042 pager_close();
5043
17cbb288 5044 if (remove_directory && arg_directory) {
ec16945e
LP		 5045 int k;
5046
17cbb288 5047 k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
ec16945e 5048 if (k < 0)
17cbb288 5049 log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
ec16945e
LP		 5050 }
5051
0f3be6ca
LP		 5052 if (remove_image && arg_image) {
5053 if (unlink(arg_image) < 0)
5054 log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
5055 }
5056
c67b0082
LP		 5057 if (remove_tmprootdir) {
5058 if (rmdir(tmprootdir) < 0)
5059 log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
5060 }
5061
785890ac
LP		 5062 if (arg_machine) {
5063 const char *p;
5064
63c372cb 5065 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
c6878637 5066 (void) rm_rf(p, REMOVE_ROOT);
785890ac
LP		 5067 }
5068
7a8f6325 5069 expose_port_flush(arg_expose_ports, &exposed);
7513c5b8
LP		 5070
5071 if (veth_created)
5072 (void) remove_veth_links(veth_name, arg_network_veth_extra);
22b28dfd 5073 (void) remove_bridge(arg_network_zone);
f757855e 5074
f757855e
LP		 5075 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
5076 expose_port_free_all(arg_expose_ports);
bf428efb 5077 rlimit_free_all(arg_rlimit);
b2645747 5078 device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
6d0b55c2 5079
44dbef90
LP		 5080 if (r < 0)
5081 return r;
5082
5083 return ret;
88213476 5084}
44dbef90
LP		 5085
5086DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.