1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <sys/types.h>
26 #include <sys/syscall.h>
27 #include <sys/mount.h>
33 #include <sys/prctl.h>
36 #include <sys/signalfd.h>
40 #include <sys/socket.h>
41 #include <linux/netlink.h>
43 #include <linux/veth.h>
44 #include <sys/personality.h>
45 #include <linux/loop.h>
50 #include <selinux/selinux.h>
58 #include <blkid/blkid.h>
61 #include "sd-daemon.h"
71 #include "cgroup-util.h"
73 #include "path-util.h"
74 #include "loopback-setup.h"
75 #include "dev-setup.h"
80 #include "bus-error.h"
82 #include "bus-kernel.h"
85 #include "rtnl-util.h"
86 #include "udev-util.h"
87 #include "blkid-util.h"
89 #include "siphash24.h"
91 #include "base-filesystem.h"
93 #include "event-util.h"
94 #include "capability.h"
96 #include "btrfs-util.h"
97 #include "machine-image.h"
99 #include "in-addr-util.h"
101 #include "local-addresses.h"
104 #include "seccomp-util.h"
107 typedef struct ExposePort {
110 uint16_t container_port;
111 LIST_FIELDS(struct ExposePort, ports);
114 typedef enum ContainerStatus {
115 CONTAINER_TERMINATED,
119 typedef enum LinkJournal {
126 typedef enum Volatile {
132 static char *arg_directory = NULL;
133 static char *arg_template = NULL;
134 static char *arg_user = NULL;
135 static sd_id128_t arg_uuid = {};
136 static char *arg_machine = NULL;
137 static const char *arg_selinux_context = NULL;
138 static const char *arg_selinux_apifs_context = NULL;
139 static const char *arg_slice = NULL;
140 static bool arg_private_network = false;
141 static bool arg_read_only = false;
142 static bool arg_boot = false;
143 static bool arg_ephemeral = false;
144 static LinkJournal arg_link_journal = LINK_AUTO;
145 static bool arg_link_journal_try = false;
146 static uint64_t arg_retain =
147 (1ULL << CAP_CHOWN) |
148 (1ULL << CAP_DAC_OVERRIDE) |
149 (1ULL << CAP_DAC_READ_SEARCH) |
150 (1ULL << CAP_FOWNER) |
151 (1ULL << CAP_FSETID) |
152 (1ULL << CAP_IPC_OWNER) |
154 (1ULL << CAP_LEASE) |
155 (1ULL << CAP_LINUX_IMMUTABLE) |
156 (1ULL << CAP_NET_BIND_SERVICE) |
157 (1ULL << CAP_NET_BROADCAST) |
158 (1ULL << CAP_NET_RAW) |
159 (1ULL << CAP_SETGID) |
160 (1ULL << CAP_SETFCAP) |
161 (1ULL << CAP_SETPCAP) |
162 (1ULL << CAP_SETUID) |
163 (1ULL << CAP_SYS_ADMIN) |
164 (1ULL << CAP_SYS_CHROOT) |
165 (1ULL << CAP_SYS_NICE) |
166 (1ULL << CAP_SYS_PTRACE) |
167 (1ULL << CAP_SYS_TTY_CONFIG) |
168 (1ULL << CAP_SYS_RESOURCE) |
169 (1ULL << CAP_SYS_BOOT) |
170 (1ULL << CAP_AUDIT_WRITE) |
171 (1ULL << CAP_AUDIT_CONTROL) |
173 static char **arg_bind = NULL;
174 static char **arg_bind_ro = NULL;
175 static char **arg_tmpfs = NULL;
176 static char **arg_setenv = NULL;
177 static bool arg_quiet = false;
178 static bool arg_share_system = false;
179 static bool arg_register = true;
180 static bool arg_keep_unit = false;
181 static char **arg_network_interfaces = NULL;
182 static char **arg_network_macvlan = NULL;
183 static char **arg_network_ipvlan = NULL;
184 static bool arg_network_veth = false;
185 static const char *arg_network_bridge = NULL;
186 static unsigned long arg_personality = 0xffffffffLU;
187 static char *arg_image = NULL;
188 static Volatile arg_volatile = VOLATILE_NO;
189 static ExposePort *arg_expose_ports = NULL;
190 static char **arg_property = NULL;
191 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
192 static bool arg_userns = false;
194 static void help(void) {
195 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
196 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
197 " -h --help Show this help\n"
198 " --version Print version string\n"
199 " -q --quiet Do not show status information\n"
200 " -D --directory=PATH Root directory for the container\n"
201 " --template=PATH Initialize root directory from template directory,\n"
203 " -x --ephemeral Run container with snapshot of root directory, and\n"
204 " remove it after exit\n"
205 " -i --image=PATH File system device or disk image for the container\n"
206 " -b --boot Boot up full system (i.e. invoke init)\n"
207 " -u --user=USER Run the command under specified user or uid\n"
208 " -M --machine=NAME Set the machine name for the container\n"
209 " --uuid=UUID Set a specific machine UUID for the container\n"
210 " -S --slice=SLICE Place the container in the specified slice\n"
211 " --property=NAME=VALUE Set scope unit property\n"
212 " --private-network Disable network in container\n"
213 " --network-interface=INTERFACE\n"
214 " Assign an existing network interface to the\n"
216 " --network-macvlan=INTERFACE\n"
217 " Create a macvlan network interface based on an\n"
218 " existing network interface to the container\n"
219 " --network-ipvlan=INTERFACE\n"
220 " Create a ipvlan network interface based on an\n"
221 " existing network interface to the container\n"
222 " -n --network-veth Add a virtual ethernet connection between host\n"
224 " --network-bridge=INTERFACE\n"
225 " Add a virtual ethernet connection between host\n"
226 " and container and add it to an existing bridge on\n"
228 " --private-users[=UIDBASE[:NUIDS]]\n"
229 " Run within user namespace\n"
230 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
231 " Expose a container IP port on the host\n"
232 " -Z --selinux-context=SECLABEL\n"
233 " Set the SELinux security context to be used by\n"
234 " processes in the container\n"
235 " -L --selinux-apifs-context=SECLABEL\n"
236 " Set the SELinux security context to be used by\n"
237 " API/tmpfs file systems in the container\n"
238 " --capability=CAP In addition to the default, retain specified\n"
240 " --drop-capability=CAP Drop the specified capability from the default set\n"
241 " --link-journal=MODE Link up guest journal, one of no, auto, guest, host,\n"
242 " try-guest, try-host\n"
243 " -j Equivalent to --link-journal=try-guest\n"
244 " --read-only Mount the root directory read-only\n"
245 " --bind=PATH[:PATH] Bind mount a file or directory from the host into\n"
247 " --bind-ro=PATH[:PATH] Similar, but creates a read-only bind mount\n"
248 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
249 " --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
250 " --share-system Share system namespaces with host\n"
251 " --register=BOOLEAN Register container as machine\n"
252 " --keep-unit Do not register a scope for the machine, reuse\n"
253 " the service unit nspawn is running in\n"
254 " --volatile[=MODE] Run the system in volatile mode\n"
255 , program_invocation_short_name);
258 static int set_sanitized_path(char **b, const char *path) {
264 p = canonicalize_file_name(path);
269 p = path_make_absolute_cwd(path);
275 *b = path_kill_slashes(p);
279 static int parse_argv(int argc, char *argv[]) {
296 ARG_NETWORK_INTERFACE,
307 static const struct option options[] = {
308 { "help", no_argument, NULL, 'h' },
309 { "version", no_argument, NULL, ARG_VERSION },
310 { "directory", required_argument, NULL, 'D' },
311 { "template", required_argument, NULL, ARG_TEMPLATE },
312 { "ephemeral", no_argument, NULL, 'x' },
313 { "user", required_argument, NULL, 'u' },
314 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
315 { "boot", no_argument, NULL, 'b' },
316 { "uuid", required_argument, NULL, ARG_UUID },
317 { "read-only", no_argument, NULL, ARG_READ_ONLY },
318 { "capability", required_argument, NULL, ARG_CAPABILITY },
319 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
320 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
321 { "bind", required_argument, NULL, ARG_BIND },
322 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
323 { "tmpfs", required_argument, NULL, ARG_TMPFS },
324 { "machine", required_argument, NULL, 'M' },
325 { "slice", required_argument, NULL, 'S' },
326 { "setenv", required_argument, NULL, ARG_SETENV },
327 { "selinux-context", required_argument, NULL, 'Z' },
328 { "selinux-apifs-context", required_argument, NULL, 'L' },
329 { "quiet", no_argument, NULL, 'q' },
330 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM },
331 { "register", required_argument, NULL, ARG_REGISTER },
332 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
333 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
334 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
335 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
336 { "network-veth", no_argument, NULL, 'n' },
337 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
338 { "personality", required_argument, NULL, ARG_PERSONALITY },
339 { "image", required_argument, NULL, 'i' },
340 { "volatile", optional_argument, NULL, ARG_VOLATILE },
341 { "port", required_argument, NULL, 'p' },
342 { "property", required_argument, NULL, ARG_PROPERTY },
343 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
348 uint64_t plus = 0, minus = 0;
353 while ((c = getopt_long(argc, argv, "+hD:u:bL:M:jS:Z:qi:xp:n", options, NULL)) >= 0)
362 puts(PACKAGE_STRING);
363 puts(SYSTEMD_FEATURES);
367 r = set_sanitized_path(&arg_directory, optarg);
369 return log_error_errno(r, "Invalid root directory: %m");
374 r = set_sanitized_path(&arg_template, optarg);
376 return log_error_errno(r, "Invalid template directory: %m");
381 r = set_sanitized_path(&arg_image, optarg);
383 return log_error_errno(r, "Invalid image path: %m");
388 arg_ephemeral = true;
393 arg_user = strdup(optarg);
399 case ARG_NETWORK_BRIDGE:
400 arg_network_bridge = optarg;
405 arg_network_veth = true;
406 arg_private_network = true;
409 case ARG_NETWORK_INTERFACE:
410 if (strv_extend(&arg_network_interfaces, optarg) < 0)
413 arg_private_network = true;
416 case ARG_NETWORK_MACVLAN:
417 if (strv_extend(&arg_network_macvlan, optarg) < 0)
420 arg_private_network = true;
423 case ARG_NETWORK_IPVLAN:
424 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
429 case ARG_PRIVATE_NETWORK:
430 arg_private_network = true;
438 r = sd_id128_from_string(optarg, &arg_uuid);
440 log_error("Invalid UUID: %s", optarg);
450 if (isempty(optarg)) {
454 if (!machine_name_is_valid(optarg)) {
455 log_error("Invalid machine name: %s", optarg);
459 r = free_and_strdup(&arg_machine, optarg);
467 arg_selinux_context = optarg;
471 arg_selinux_apifs_context = optarg;
475 arg_read_only = true;
479 case ARG_DROP_CAPABILITY: {
480 const char *state, *word;
483 FOREACH_WORD_SEPARATOR(word, length, optarg, ",", state) {
484 _cleanup_free_ char *t;
486 t = strndup(word, length);
490 if (streq(t, "all")) {
491 if (c == ARG_CAPABILITY)
492 plus = (uint64_t) -1;
494 minus = (uint64_t) -1;
498 cap = capability_from_name(t);
500 log_error("Failed to parse capability %s.", t);
504 if (c == ARG_CAPABILITY)
505 plus |= 1ULL << (uint64_t) cap;
507 minus |= 1ULL << (uint64_t) cap;
515 arg_link_journal = LINK_GUEST;
516 arg_link_journal_try = true;
519 case ARG_LINK_JOURNAL:
520 if (streq(optarg, "auto")) {
521 arg_link_journal = LINK_AUTO;
522 arg_link_journal_try = false;
523 } else if (streq(optarg, "no")) {
524 arg_link_journal = LINK_NO;
525 arg_link_journal_try = false;
526 } else if (streq(optarg, "guest")) {
527 arg_link_journal = LINK_GUEST;
528 arg_link_journal_try = false;
529 } else if (streq(optarg, "host")) {
530 arg_link_journal = LINK_HOST;
531 arg_link_journal_try = false;
532 } else if (streq(optarg, "try-guest")) {
533 arg_link_journal = LINK_GUEST;
534 arg_link_journal_try = true;
535 } else if (streq(optarg, "try-host")) {
536 arg_link_journal = LINK_HOST;
537 arg_link_journal_try = true;
539 log_error("Failed to parse link journal mode %s", optarg);
547 _cleanup_free_ char *a = NULL, *b = NULL;
551 x = c == ARG_BIND ? &arg_bind : &arg_bind_ro;
553 e = strchr(optarg, ':');
555 a = strndup(optarg, e - optarg);
565 if (!path_is_absolute(a) || !path_is_absolute(b)) {
566 log_error("Invalid bind mount specification: %s", optarg);
570 r = strv_extend(x, a);
574 r = strv_extend(x, b);
582 _cleanup_free_ char *a = NULL, *b = NULL;
585 e = strchr(optarg, ':');
587 a = strndup(optarg, e - optarg);
591 b = strdup("mode=0755");
597 if (!path_is_absolute(a)) {
598 log_error("Invalid tmpfs specification: %s", optarg);
602 r = strv_push(&arg_tmpfs, a);
608 r = strv_push(&arg_tmpfs, b);
620 if (!env_assignment_is_valid(optarg)) {
621 log_error("Environment variable assignment '%s' is not valid.", optarg);
625 n = strv_env_set(arg_setenv, optarg);
629 strv_free(arg_setenv);
638 case ARG_SHARE_SYSTEM:
639 arg_share_system = true;
643 r = parse_boolean(optarg);
645 log_error("Failed to parse --register= argument: %s", optarg);
653 arg_keep_unit = true;
656 case ARG_PERSONALITY:
658 arg_personality = personality_from_string(optarg);
659 if (arg_personality == 0xffffffffLU) {
660 log_error("Unknown or unsupported personality '%s'.", optarg);
669 arg_volatile = VOLATILE_YES;
671 r = parse_boolean(optarg);
673 if (streq(optarg, "state"))
674 arg_volatile = VOLATILE_STATE;
676 log_error("Failed to parse --volatile= argument: %s", optarg);
680 arg_volatile = r ? VOLATILE_YES : VOLATILE_NO;
686 const char *split, *e;
687 uint16_t container_port, host_port;
691 if ((e = startswith(optarg, "tcp:")))
692 protocol = IPPROTO_TCP;
693 else if ((e = startswith(optarg, "udp:")))
694 protocol = IPPROTO_UDP;
697 protocol = IPPROTO_TCP;
700 split = strchr(e, ':');
702 char v[split - e + 1];
704 memcpy(v, e, split - e);
707 r = safe_atou16(v, &host_port);
708 if (r < 0 || host_port <= 0) {
709 log_error("Failed to parse host port: %s", optarg);
713 r = safe_atou16(split + 1, &container_port);
715 r = safe_atou16(e, &container_port);
716 host_port = container_port;
719 if (r < 0 || container_port <= 0) {
720 log_error("Failed to parse host port: %s", optarg);
724 LIST_FOREACH(ports, p, arg_expose_ports) {
725 if (p->protocol == protocol && p->host_port == host_port) {
726 log_error("Duplicate port specification: %s", optarg);
731 p = new(ExposePort, 1);
735 p->protocol = protocol;
736 p->host_port = host_port;
737 p->container_port = container_port;
739 LIST_PREPEND(ports, arg_expose_ports, p);
745 if (strv_extend(&arg_property, optarg) < 0)
750 case ARG_PRIVATE_USERS:
752 _cleanup_free_ char *buffer = NULL;
753 const char *range, *shift;
755 range = strchr(optarg, ':');
757 buffer = strndup(optarg, range - optarg);
763 if (safe_atou32(range, &arg_uid_range) < 0 || arg_uid_range <= 0) {
764 log_error("Failed to parse UID range: %s", range);
770 if (parse_uid(shift, &arg_uid_shift) < 0) {
771 log_error("Failed to parse UID: %s", optarg);
783 assert_not_reached("Unhandled option");
786 if (arg_share_system)
787 arg_register = false;
789 if (arg_boot && arg_share_system) {
790 log_error("--boot and --share-system may not be combined.");
794 if (arg_keep_unit && cg_pid_get_owner_uid(0, NULL) >= 0) {
795 log_error("--keep-unit may not be used when invoked from a user session.");
799 if (arg_directory && arg_image) {
800 log_error("--directory= and --image= may not be combined.");
804 if (arg_template && arg_image) {
805 log_error("--template= and --image= may not be combined.");
809 if (arg_template && !(arg_directory || arg_machine)) {
810 log_error("--template= needs --directory= or --machine=.");
814 if (arg_ephemeral && arg_template) {
815 log_error("--ephemeral and --template= may not be combined.");
819 if (arg_ephemeral && arg_image) {
820 log_error("--ephemeral and --image= may not be combined.");
824 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
825 log_error("--ephemeral and --link-journal= may not be combined.");
829 if (arg_volatile != VOLATILE_NO && arg_read_only) {
830 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
834 if (arg_expose_ports && !arg_private_network) {
835 log_error("Cannot use --port= without private networking.");
839 arg_retain = (arg_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
844 static int mount_all(const char *dest) {
846 typedef struct MountPoint {
855 static const MountPoint mount_table[] = {
856 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
857 { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND, true }, /* Bind mount first */
858 { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, true }, /* Then, make it r/o */
859 { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
860 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME, true },
861 { "devpts", "/dev/pts", "devpts","newinstance,ptmxmode=0666,mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC, true },
862 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true },
863 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true },
864 { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME, true },
866 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND, false }, /* Bind mount first */
867 { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false }, /* Then, make it r/o */
874 for (k = 0; k < ELEMENTSOF(mount_table); k++) {
875 _cleanup_free_ char *where = NULL, *options = NULL;
879 where = strjoin(dest, "/", mount_table[k].where, NULL);
883 t = path_is_mount_point(where, true);
885 log_error_errno(t, "Failed to detect whether %s is a mount point: %m", where);
893 /* Skip this entry if it is not a remount. */
894 if (mount_table[k].what && t > 0)
897 t = mkdir_p(where, 0755);
899 if (mount_table[k].fatal) {
900 log_error_errno(t, "Failed to create directory %s: %m", where);
905 log_warning_errno(t, "Failed to create directory %s: %m", where);
911 if (arg_selinux_apifs_context &&
912 (streq_ptr(mount_table[k].what, "tmpfs") || streq_ptr(mount_table[k].what, "devpts"))) {
913 options = strjoin(mount_table[k].options, ",context=\"", arg_selinux_apifs_context, "\"", NULL);
920 o = mount_table[k].options;
922 if (arg_userns && arg_uid_shift != UID_INVALID && streq_ptr(mount_table[k].type, "tmpfs")) {
923 char *uid_options = NULL;
926 asprintf(&uid_options, "%s,uid=" UID_FMT ",gid=" UID_FMT, o, arg_uid_shift, arg_uid_shift);
928 asprintf(&uid_options, "uid=" UID_FMT ",gid=" UID_FMT, arg_uid_shift, arg_uid_shift);
933 o = options = uid_options;
936 if (mount(mount_table[k].what,
939 mount_table[k].flags,
942 if (mount_table[k].fatal) {
943 log_error_errno(errno, "mount(%s) failed: %m", where);
948 log_warning_errno(errno, "mount(%s) failed: %m", where);
955 static int mount_binds(const char *dest, char **l, bool ro) {
958 STRV_FOREACH_PAIR(x, y, l) {
959 _cleanup_free_ char *where = NULL;
960 struct stat source_st, dest_st;
963 if (stat(*x, &source_st) < 0)
964 return log_error_errno(errno, "Failed to stat %s: %m", *x);
966 where = strappend(dest, *y);
970 r = stat(where, &dest_st);
972 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
973 log_error("Cannot bind mount directory %s on file %s.", *x, where);
976 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
977 log_error("Cannot bind mount file %s on directory %s.", *x, where);
980 } else if (errno == ENOENT) {
981 r = mkdir_parents_label(where, 0755);
983 return log_error_errno(r, "Failed to bind mount %s: %m", *x);
985 log_error_errno(errno, "Failed to bind mount %s: %m", *x);
989 /* Create the mount point. Any non-directory file can be
990 * mounted on any non-directory file (regular, fifo, socket,
993 if (S_ISDIR(source_st.st_mode)) {
994 r = mkdir_label(where, 0755);
995 if (r < 0 && errno != EEXIST)
996 return log_error_errno(r, "Failed to create mount point %s: %m", where);
1000 return log_error_errno(r, "Failed to create mount point %s: %m", where);
1003 if (mount(*x, where, "bind", MS_BIND, NULL) < 0)
1004 return log_error_errno(errno, "mount(%s) failed: %m", where);
1007 r = bind_remount_recursive(where, true);
1009 return log_error_errno(r, "Read-Only bind mount failed: %m");
1016 static int mount_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, bool read_only) {
1020 to = strjoina(dest, "/sys/fs/cgroup/", hierarchy);
1022 r = path_is_mount_point(to, false);
1024 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
1030 /* The superblock mount options of the mount point need to be
1031 * identical to the hosts', and hence writable... */
1032 if (mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller) < 0)
1033 return log_error_errno(errno, "Failed to mount to %s: %m", to);
1035 /* ... hence let's only make the bind mount read-only, not the
1038 if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
1039 return log_error_errno(errno, "Failed to remount %s read-only: %m", to);
1044 static int mount_cgroup(const char *dest) {
1045 _cleanup_set_free_free_ Set *controllers = NULL;
1046 _cleanup_free_ char *own_cgroup_path = NULL;
1047 const char *cgroup_root, *systemd_root, *systemd_own;
1050 controllers = set_new(&string_hash_ops);
1054 r = cg_kernel_controllers(controllers);
1056 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1058 r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1060 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1062 cgroup_root = strjoina(dest, "/sys/fs/cgroup");
1063 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, "mode=755") < 0)
1064 return log_error_errno(errno, "Failed to mount tmpfs to /sys/fs/cgroup: %m");
1067 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
1069 controller = set_steal_first(controllers);
1073 origin = strappend("/sys/fs/cgroup/", controller);
1077 r = readlink_malloc(origin, &combined);
1079 /* Not a symbolic link, but directly a single cgroup hierarchy */
1081 r = mount_cgroup_hierarchy(dest, controller, controller, true);
1086 return log_error_errno(r, "Failed to read link %s: %m", origin);
1088 _cleanup_free_ char *target = NULL;
1090 target = strjoin(dest, "/sys/fs/cgroup/", controller, NULL);
1094 /* A symbolic link, a combination of controllers in one hierarchy */
1096 if (!filename_is_valid(combined)) {
1097 log_warning("Ignoring invalid combined hierarchy %s.", combined);
1101 r = mount_cgroup_hierarchy(dest, combined, combined, true);
1105 if (symlink(combined, target) < 0)
1106 return log_error_errno(errno, "Failed to create symlink for combined hierarchy: %m");
1110 r = mount_cgroup_hierarchy(dest, "name=systemd,xattr", "systemd", false);
1114 /* Make our own cgroup a (writable) bind mount */
1115 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1116 if (mount(systemd_own, systemd_own, NULL, MS_BIND, NULL) < 0)
1117 return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path);
1119 /* And then remount the systemd cgroup root read-only */
1120 systemd_root = strjoina(dest, "/sys/fs/cgroup/systemd");
1121 if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
1122 return log_error_errno(errno, "Failed to mount cgroup root read-only: %m");
1124 if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
1125 return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
1130 static int mount_tmpfs(const char *dest) {
1133 STRV_FOREACH_PAIR(i, o, arg_tmpfs) {
1134 _cleanup_free_ char *where = NULL;
1137 where = strappend(dest, *i);
1141 r = mkdir_label(where, 0755);
1142 if (r < 0 && r != -EEXIST)
1143 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
1145 if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, *o) < 0)
1146 return log_error_errno(errno, "tmpfs mount to %s failed: %m", where);
1152 static int setup_timezone(const char *dest) {
1153 _cleanup_free_ char *where = NULL, *p = NULL, *q = NULL, *check = NULL, *what = NULL;
1159 /* Fix the timezone, if possible */
1160 r = readlink_malloc("/etc/localtime", &p);
1162 log_warning("/etc/localtime is not a symlink, not updating container timezone.");
1166 z = path_startswith(p, "../usr/share/zoneinfo/");
1168 z = path_startswith(p, "/usr/share/zoneinfo/");
1170 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1174 where = strappend(dest, "/etc/localtime");
1178 r = readlink_malloc(where, &q);
1180 y = path_startswith(q, "../usr/share/zoneinfo/");
1182 y = path_startswith(q, "/usr/share/zoneinfo/");
1184 /* Already pointing to the right place? Then do nothing .. */
1185 if (y && streq(y, z))
1189 check = strjoin(dest, "/usr/share/zoneinfo/", z, NULL);
1193 if (access(check, F_OK) < 0) {
1194 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1198 what = strappend("../usr/share/zoneinfo/", z);
1202 r = mkdir_parents(where, 0755);
1204 log_error_errno(r, "Failed to create directory for timezone info %s in container: %m", where);
1210 if (r < 0 && errno != ENOENT) {
1211 log_error_errno(errno, "Failed to remove existing timezone info %s in container: %m", where);
1216 if (symlink(what, where) < 0) {
1217 log_error_errno(errno, "Failed to correct timezone of container: %m");
1224 static int setup_resolv_conf(const char *dest) {
1225 _cleanup_free_ char *where = NULL;
1230 if (arg_private_network)
1233 /* Fix resolv.conf, if possible */
1234 where = strappend(dest, "/etc/resolv.conf");
1238 /* We don't really care for the results of this really. If it
1239 * fails, it fails, but meh... */
1240 r = mkdir_parents(where, 0755);
1242 log_warning_errno(r, "Failed to create parent directory for resolv.conf %s: %m", where);
1247 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0);
1249 log_warning_errno(r, "Failed to copy /etc/resolv.conf to %s: %m", where);
1257 static int setup_volatile_state(const char *directory) {
1263 if (arg_volatile != VOLATILE_STATE)
1266 /* --volatile=state means we simply overmount /var
1267 with a tmpfs, and the rest read-only. */
1269 r = bind_remount_recursive(directory, true);
1271 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1273 p = strjoina(directory, "/var");
1275 if (r < 0 && errno != EEXIST)
1276 return log_error_errno(errno, "Failed to create %s: %m", directory);
1278 if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, "mode=755") < 0)
1279 return log_error_errno(errno, "Failed to mount tmpfs to /var: %m");
1284 static int setup_volatile(const char *directory) {
1285 bool tmpfs_mounted = false, bind_mounted = false;
1286 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1292 if (arg_volatile != VOLATILE_YES)
1295 /* --volatile=yes means we mount a tmpfs to the root dir, and
1296 the original /usr to use inside it, and that read-only. */
1298 if (!mkdtemp(template))
1299 return log_error_errno(errno, "Failed to create temporary directory: %m");
1301 if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, "mode=755") < 0) {
1302 log_error_errno(errno, "Failed to mount tmpfs for root directory: %m");
1307 tmpfs_mounted = true;
1309 f = strjoina(directory, "/usr");
1310 t = strjoina(template, "/usr");
1313 if (r < 0 && errno != EEXIST) {
1314 log_error_errno(errno, "Failed to create %s: %m", t);
1319 if (mount(f, t, "bind", MS_BIND|MS_REC, NULL) < 0) {
1320 log_error_errno(errno, "Failed to create /usr bind mount: %m");
1325 bind_mounted = true;
1327 r = bind_remount_recursive(t, true);
1329 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1333 if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) {
1334 log_error_errno(errno, "Failed to move root mount: %m");
1352 static char* id128_format_as_uuid(sd_id128_t id, char s[37]) {
1355 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
1356 SD_ID128_FORMAT_VAL(id));
1361 static int setup_boot_id(const char *dest) {
1362 _cleanup_free_ char *from = NULL, *to = NULL;
1363 sd_id128_t rnd = {};
1369 if (arg_share_system)
1372 /* Generate a new randomized boot ID, so that each boot-up of
1373 * the container gets a new one */
1375 from = strappend(dest, "/dev/proc-sys-kernel-random-boot-id");
1376 to = strappend(dest, "/proc/sys/kernel/random/boot_id");
1380 r = sd_id128_randomize(&rnd);
1382 return log_error_errno(r, "Failed to generate random boot id: %m");
1384 id128_format_as_uuid(rnd, as_uuid);
1386 r = write_string_file(from, as_uuid);
1388 return log_error_errno(r, "Failed to write boot id: %m");
1390 if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
1391 log_error_errno(errno, "Failed to bind mount boot id: %m");
1393 } else if (mount(from, to, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL))
1394 log_warning_errno(errno, "Failed to make boot id read-only: %m");
1400 static int copy_devnodes(const char *dest) {
1402 static const char devnodes[] =
1413 _cleanup_umask_ mode_t u;
1419 NULSTR_FOREACH(d, devnodes) {
1420 _cleanup_free_ char *from = NULL, *to = NULL;
1423 from = strappend("/dev/", d);
1424 to = strjoin(dest, "/dev/", d, NULL);
1428 if (stat(from, &st) < 0) {
1430 if (errno != ENOENT)
1431 return log_error_errno(errno, "Failed to stat %s: %m", from);
1433 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1435 log_error("%s is not a char or block device, cannot copy", from);
1439 r = mkdir_parents(to, 0775);
1441 log_error_errno(r, "Failed to create parent directory of %s: %m", to);
1445 if (mknod(to, st.st_mode, st.st_rdev) < 0)
1446 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1453 static int setup_ptmx(const char *dest) {
1454 _cleanup_free_ char *p = NULL;
1456 p = strappend(dest, "/dev/ptmx");
1460 if (symlink("pts/ptmx", p) < 0)
1461 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1466 static int setup_dev_console(const char *dest, const char *console) {
1467 _cleanup_umask_ mode_t u;
1477 if (stat("/dev/null", &st) < 0)
1478 return log_error_errno(errno, "Failed to stat /dev/null: %m");
1480 r = chmod_and_chown(console, 0600, 0, 0);
1482 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1484 /* We need to bind mount the right tty to /dev/console since
1485 * ptys can only exist on pts file systems. To have something
1486 * to bind mount things on we create a device node first, and
1487 * use /dev/null for that since we the cgroups device policy
1488 * allows us to create that freely, while we cannot create
1489 * /dev/console. (Note that the major minor doesn't actually
1490 * matter here, since we mount it over anyway). */
1492 to = strjoina(dest, "/dev/console");
1493 if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0)
1494 return log_error_errno(errno, "mknod() for /dev/console failed: %m");
1496 if (mount(console, to, "bind", MS_BIND, NULL) < 0)
1497 return log_error_errno(errno, "Bind mount for /dev/console failed: %m");
1502 static int setup_kmsg(const char *dest, int kmsg_socket) {
1503 _cleanup_free_ char *from = NULL, *to = NULL;
1504 _cleanup_umask_ mode_t u;
1507 struct cmsghdr cmsghdr;
1508 uint8_t buf[CMSG_SPACE(sizeof(int))];
1510 struct msghdr mh = {
1511 .msg_control = &control,
1512 .msg_controllen = sizeof(control),
1514 struct cmsghdr *cmsg;
1517 assert(kmsg_socket >= 0);
1521 /* We create the kmsg FIFO as /dev/kmsg, but immediately
1522 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1523 * on the reading side behave very similar to /proc/kmsg,
1524 * their writing side behaves differently from /dev/kmsg in
1525 * that writing blocks when nothing is reading. In order to
1526 * avoid any problems with containers deadlocking due to this
1527 * we simply make /dev/kmsg unavailable to the container. */
1528 if (asprintf(&from, "%s/dev/kmsg", dest) < 0 ||
1529 asprintf(&to, "%s/proc/kmsg", dest) < 0)
1532 if (mkfifo(from, 0600) < 0)
1533 return log_error_errno(errno, "mkfifo() for /dev/kmsg failed: %m");
1535 r = chmod_and_chown(from, 0600, 0, 0);
1537 return log_error_errno(r, "Failed to correct access mode for /dev/kmsg: %m");
1539 if (mount(from, to, "bind", MS_BIND, NULL) < 0)
1540 return log_error_errno(errno, "Bind mount for /proc/kmsg failed: %m");
1542 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
1544 return log_error_errno(errno, "Failed to open fifo: %m");
1546 cmsg = CMSG_FIRSTHDR(&mh);
1547 cmsg->cmsg_level = SOL_SOCKET;
1548 cmsg->cmsg_type = SCM_RIGHTS;
1549 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1550 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1552 mh.msg_controllen = cmsg->cmsg_len;
1554 /* Store away the fd in the socket, so that it stays open as
1555 * long as we run the child */
1556 k = sendmsg(kmsg_socket, &mh, MSG_NOSIGNAL);
1560 return log_error_errno(errno, "Failed to send FIFO fd: %m");
1562 /* And now make the FIFO unavailable as /dev/kmsg... */
1567 static int send_rtnl(int send_fd) {
1569 struct cmsghdr cmsghdr;
1570 uint8_t buf[CMSG_SPACE(sizeof(int))];
1572 struct msghdr mh = {
1573 .msg_control = &control,
1574 .msg_controllen = sizeof(control),
1576 struct cmsghdr *cmsg;
1577 _cleanup_close_ int fd = -1;
1580 assert(send_fd >= 0);
1582 if (!arg_expose_ports)
1585 fd = socket(PF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_ROUTE);
1587 return log_error_errno(errno, "failed to allocate container netlink: %m");
1589 cmsg = CMSG_FIRSTHDR(&mh);
1590 cmsg->cmsg_level = SOL_SOCKET;
1591 cmsg->cmsg_type = SCM_RIGHTS;
1592 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1593 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1595 mh.msg_controllen = cmsg->cmsg_len;
1597 /* Store away the fd in the socket, so that it stays open as
1598 * long as we run the child */
1599 k = sendmsg(send_fd, &mh, MSG_NOSIGNAL);
1601 return log_error_errno(errno, "Failed to send netlink fd: %m");
1606 static int flush_ports(union in_addr_union *exposed) {
1608 int r, af = AF_INET;
1612 if (!arg_expose_ports)
1615 if (in_addr_is_null(af, exposed))
1618 log_debug("Lost IP address.");
1620 LIST_FOREACH(ports, p, arg_expose_ports) {
1621 r = fw_add_local_dnat(false,
1632 log_warning_errno(r, "Failed to modify firewall: %m");
1635 *exposed = IN_ADDR_NULL;
1639 static int expose_ports(sd_rtnl *rtnl, union in_addr_union *exposed) {
1640 _cleanup_free_ struct local_address *addresses = NULL;
1641 _cleanup_free_ char *pretty = NULL;
1642 union in_addr_union new_exposed;
1645 int af = AF_INET, r;
1649 /* Invoked each time an address is added or removed inside the
1652 if (!arg_expose_ports)
1655 r = local_addresses(rtnl, 0, af, &addresses);
1657 return log_error_errno(r, "Failed to enumerate local addresses: %m");
1660 addresses[0].family == af &&
1661 addresses[0].scope < RT_SCOPE_LINK;
1664 return flush_ports(exposed);
1666 new_exposed = addresses[0].address;
1667 if (in_addr_equal(af, exposed, &new_exposed))
1670 in_addr_to_string(af, &new_exposed, &pretty);
1671 log_debug("New container IP is %s.", strna(pretty));
1673 LIST_FOREACH(ports, p, arg_expose_ports) {
1675 r = fw_add_local_dnat(true,
1684 in_addr_is_null(af, exposed) ? NULL : exposed);
1686 log_warning_errno(r, "Failed to modify firewall: %m");
1689 *exposed = new_exposed;
1693 static int on_address_change(sd_rtnl *rtnl, sd_rtnl_message *m, void *userdata) {
1694 union in_addr_union *exposed = userdata;
1700 expose_ports(rtnl, exposed);
1704 static int watch_rtnl(sd_event *event, int recv_fd, union in_addr_union *exposed, sd_rtnl **ret) {
1706 struct cmsghdr cmsghdr;
1707 uint8_t buf[CMSG_SPACE(sizeof(int))];
1709 struct msghdr mh = {
1710 .msg_control = &control,
1711 .msg_controllen = sizeof(control),
1713 struct cmsghdr *cmsg;
1714 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
1719 assert(recv_fd >= 0);
1722 if (!arg_expose_ports)
1725 k = recvmsg(recv_fd, &mh, MSG_NOSIGNAL);
1727 return log_error_errno(errno, "Failed to recv netlink fd: %m");
1729 cmsg = CMSG_FIRSTHDR(&mh);
1730 assert(cmsg->cmsg_level == SOL_SOCKET);
1731 assert(cmsg->cmsg_type == SCM_RIGHTS);
1732 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
1733 memcpy(&fd, CMSG_DATA(cmsg), sizeof(int));
1735 r = sd_rtnl_open_fd(&rtnl, fd, 1, RTNLGRP_IPV4_IFADDR);
1738 return log_error_errno(r, "Failed to create rtnl object: %m");
1741 r = sd_rtnl_add_match(rtnl, RTM_NEWADDR, on_address_change, exposed);
1743 return log_error_errno(r, "Failed to subscribe to RTM_NEWADDR messages: %m");
1745 r = sd_rtnl_add_match(rtnl, RTM_DELADDR, on_address_change, exposed);
1747 return log_error_errno(r, "Failed to subscribe to RTM_DELADDR messages: %m");
1749 r = sd_rtnl_attach_event(rtnl, event, 0);
1751 return log_error_errno(r, "Failed to add to even loop: %m");
1759 static int setup_hostname(void) {
1761 if (arg_share_system)
1764 if (sethostname_idempotent(arg_machine) < 0)
1770 static int setup_journal(const char *directory) {
1771 sd_id128_t machine_id, this_id;
1772 _cleanup_free_ char *p = NULL, *b = NULL, *q = NULL, *d = NULL;
1776 /* Don't link journals in ephemeral mode */
1780 p = strappend(directory, "/etc/machine-id");
1784 r = read_one_line_file(p, &b);
1785 if (r == -ENOENT && arg_link_journal == LINK_AUTO)
1788 return log_error_errno(r, "Failed to read machine ID from %s: %m", p);
1791 if (isempty(id) && arg_link_journal == LINK_AUTO)
1794 /* Verify validity */
1795 r = sd_id128_from_string(id, &machine_id);
1797 return log_error_errno(r, "Failed to parse machine ID from %s: %m", p);
1799 r = sd_id128_get_machine(&this_id);
1801 return log_error_errno(r, "Failed to retrieve machine ID: %m");
1803 if (sd_id128_equal(machine_id, this_id)) {
1804 log_full(arg_link_journal == LINK_AUTO ? LOG_WARNING : LOG_ERR,
1805 "Host and machine ids are equal (%s): refusing to link journals", id);
1806 if (arg_link_journal == LINK_AUTO)
1811 if (arg_link_journal == LINK_NO)
1815 p = strappend("/var/log/journal/", id);
1816 q = strjoin(directory, "/var/log/journal/", id, NULL);
1820 if (path_is_mount_point(p, false) > 0) {
1821 if (arg_link_journal != LINK_AUTO) {
1822 log_error("%s: already a mount point, refusing to use for journal", p);
1829 if (path_is_mount_point(q, false) > 0) {
1830 if (arg_link_journal != LINK_AUTO) {
1831 log_error("%s: already a mount point, refusing to use for journal", q);
1838 r = readlink_and_make_absolute(p, &d);
1840 if ((arg_link_journal == LINK_GUEST ||
1841 arg_link_journal == LINK_AUTO) &&
1844 r = mkdir_p(q, 0755);
1846 log_warning_errno(errno, "Failed to create directory %s: %m", q);
1851 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
1852 } else if (r == -EINVAL) {
1854 if (arg_link_journal == LINK_GUEST &&
1857 if (errno == ENOTDIR) {
1858 log_error("%s already exists and is neither a symlink nor a directory", p);
1861 log_error_errno(errno, "Failed to remove %s: %m", p);
1865 } else if (r != -ENOENT) {
1866 log_error_errno(errno, "readlink(%s) failed: %m", p);
1870 if (arg_link_journal == LINK_GUEST) {
1872 if (symlink(q, p) < 0) {
1873 if (arg_link_journal_try) {
1874 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
1877 log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
1882 r = mkdir_p(q, 0755);
1884 log_warning_errno(errno, "Failed to create directory %s: %m", q);
1888 if (arg_link_journal == LINK_HOST) {
1889 /* don't create parents here -- if the host doesn't have
1890 * permanent journal set up, don't force it here */
1893 if (arg_link_journal_try) {
1894 log_debug_errno(errno, "Failed to create %s, skipping journal setup: %m", p);
1897 log_error_errno(errno, "Failed to create %s: %m", p);
1902 } else if (access(p, F_OK) < 0)
1905 if (dir_is_empty(q) == 0)
1906 log_warning("%s is not empty, proceeding anyway.", q);
1908 r = mkdir_p(q, 0755);
1910 log_error_errno(errno, "Failed to create %s: %m", q);
1914 if (mount(p, q, "bind", MS_BIND, NULL) < 0)
1915 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
1920 static int drop_capabilities(void) {
1921 return capability_bounding_set_drop(~arg_retain, false);
1924 static int register_machine(pid_t pid, int local_ifindex) {
1925 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1926 _cleanup_bus_close_unref_ sd_bus *bus = NULL;
1932 r = sd_bus_default_system(&bus);
1934 return log_error_errno(r, "Failed to open system bus: %m");
1936 if (arg_keep_unit) {
1937 r = sd_bus_call_method(
1939 "org.freedesktop.machine1",
1940 "/org/freedesktop/machine1",
1941 "org.freedesktop.machine1.Manager",
1942 "RegisterMachineWithNetwork",
1947 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
1951 strempty(arg_directory),
1952 local_ifindex > 0 ? 1 : 0, local_ifindex);
1954 _cleanup_bus_message_unref_ sd_bus_message *m = NULL;
1957 r = sd_bus_message_new_method_call(
1960 "org.freedesktop.machine1",
1961 "/org/freedesktop/machine1",
1962 "org.freedesktop.machine1.Manager",
1963 "CreateMachineWithNetwork");
1965 return bus_log_create_error(r);
1967 r = sd_bus_message_append(
1971 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
1975 strempty(arg_directory),
1976 local_ifindex > 0 ? 1 : 0, local_ifindex);
1978 return bus_log_create_error(r);
1980 r = sd_bus_message_open_container(m, 'a', "(sv)");
1982 return bus_log_create_error(r);
1984 if (!isempty(arg_slice)) {
1985 r = sd_bus_message_append(m, "(sv)", "Slice", "s", arg_slice);
1987 return bus_log_create_error(r);
1990 r = sd_bus_message_append(m, "(sv)", "DevicePolicy", "s", "strict");
1992 return bus_log_create_error(r);
1994 r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 9,
1995 /* Allow the container to
1996 * access and create the API
1997 * device nodes, so that
1998 * PrivateDevices= in the
1999 * container can work
2004 "/dev/random", "rwm",
2005 "/dev/urandom", "rwm",
2007 "/dev/net/tun", "rwm",
2008 /* Allow the container
2009 * access to ptys. However,
2011 * container to ever create
2012 * these device nodes. */
2013 "/dev/pts/ptmx", "rw",
2016 return log_error_errno(r, "Failed to add device whitelist: %m");
2018 STRV_FOREACH(i, arg_property) {
2019 r = sd_bus_message_open_container(m, 'r', "sv");
2021 return bus_log_create_error(r);
2023 r = bus_append_unit_property_assignment(m, *i);
2027 r = sd_bus_message_close_container(m);
2029 return bus_log_create_error(r);
2032 r = sd_bus_message_close_container(m);
2034 return bus_log_create_error(r);
2036 r = sd_bus_call(bus, m, 0, &error, NULL);
2040 log_error("Failed to register machine: %s", bus_error_message(&error, r));
2047 static int terminate_machine(pid_t pid) {
2048 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
2049 _cleanup_bus_message_unref_ sd_bus_message *reply = NULL;
2050 _cleanup_bus_close_unref_ sd_bus *bus = NULL;
2057 r = sd_bus_default_system(&bus);
2059 return log_error_errno(r, "Failed to open system bus: %m");
2061 r = sd_bus_call_method(
2063 "org.freedesktop.machine1",
2064 "/org/freedesktop/machine1",
2065 "org.freedesktop.machine1.Manager",
2072 /* Note that the machine might already have been
2073 * cleaned up automatically, hence don't consider it a
2074 * failure if we cannot get the machine object. */
2075 log_debug("Failed to get machine: %s", bus_error_message(&error, r));
2079 r = sd_bus_message_read(reply, "o", &path);
2081 return bus_log_parse_error(r);
2083 r = sd_bus_call_method(
2085 "org.freedesktop.machine1",
2087 "org.freedesktop.machine1.Machine",
2093 log_debug("Failed to terminate machine: %s", bus_error_message(&error, r));
2100 static int reset_audit_loginuid(void) {
2101 _cleanup_free_ char *p = NULL;
2104 if (arg_share_system)
2107 r = read_one_line_file("/proc/self/loginuid", &p);
2111 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2113 /* Already reset? */
2114 if (streq(p, "4294967295"))
2117 r = write_string_file("/proc/self/loginuid", "4294967295");
2119 log_error("Failed to reset audit login UID. This probably means that your kernel is too\n"
2120 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2121 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2122 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2123 "using systemd-nspawn. Sleeping for 5s... (%s)\n", strerror(-r));
2131 #define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
2132 #define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
2133 #define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
2135 static int generate_mac(struct ether_addr *mac, sd_id128_t hash_key, uint64_t idx) {
2141 l = strlen(arg_machine);
2142 sz = sizeof(sd_id128_t) + l;
2148 /* fetch some persistent data unique to the host */
2149 r = sd_id128_get_machine((sd_id128_t*) v);
2153 /* combine with some data unique (on this host) to this
2154 * container instance */
2155 i = mempcpy(v + sizeof(sd_id128_t), arg_machine, l);
2158 memcpy(i, &idx, sizeof(idx));
2161 /* Let's hash the host machine ID plus the container name. We
2162 * use a fixed, but originally randomly created hash key here. */
2163 siphash24(result, v, sz, hash_key.bytes);
2165 assert_cc(ETH_ALEN <= sizeof(result));
2166 memcpy(mac->ether_addr_octet, result, ETH_ALEN);
2168 /* see eth_random_addr in the kernel */
2169 mac->ether_addr_octet[0] &= 0xfe; /* clear multicast bit */
2170 mac->ether_addr_octet[0] |= 0x02; /* set local assignment bit (IEEE802) */
2175 static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ], int *ifi) {
2176 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
2177 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
2178 struct ether_addr mac_host, mac_container;
2181 if (!arg_private_network)
2184 if (!arg_network_veth)
2187 /* Use two different interface name prefixes depending whether
2188 * we are in bridge mode or not. */
2189 snprintf(iface_name, IFNAMSIZ - 1, "%s-%s",
2190 arg_network_bridge ? "vb" : "ve", arg_machine);
2192 r = generate_mac(&mac_container, CONTAINER_HASH_KEY, 0);
2194 return log_error_errno(r, "Failed to generate predictable MAC address for container side: %m");
2196 r = generate_mac(&mac_host, HOST_HASH_KEY, 0);
2198 return log_error_errno(r, "Failed to generate predictable MAC address for host side: %m");
2200 r = sd_rtnl_open(&rtnl, 0);
2202 return log_error_errno(r, "Failed to connect to netlink: %m");
2204 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2206 return log_error_errno(r, "Failed to allocate netlink message: %m");
2208 r = sd_rtnl_message_append_string(m, IFLA_IFNAME, iface_name);
2210 return log_error_errno(r, "Failed to add netlink interface name: %m");
2212 r = sd_rtnl_message_append_ether_addr(m, IFLA_ADDRESS, &mac_host);
2214 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2216 r = sd_rtnl_message_open_container(m, IFLA_LINKINFO);
2218 return log_error_errno(r, "Failed to open netlink container: %m");
2220 r = sd_rtnl_message_open_container_union(m, IFLA_INFO_DATA, "veth");
2222 return log_error_errno(r, "Failed to open netlink container: %m");
2224 r = sd_rtnl_message_open_container(m, VETH_INFO_PEER);
2226 return log_error_errno(r, "Failed to open netlink container: %m");
2228 r = sd_rtnl_message_append_string(m, IFLA_IFNAME, "host0");
2230 return log_error_errno(r, "Failed to add netlink interface name: %m");
2232 r = sd_rtnl_message_append_ether_addr(m, IFLA_ADDRESS, &mac_container);
2234 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2236 r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
2238 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2240 r = sd_rtnl_message_close_container(m);
2242 return log_error_errno(r, "Failed to close netlink container: %m");
2244 r = sd_rtnl_message_close_container(m);
2246 return log_error_errno(r, "Failed to close netlink container: %m");
2248 r = sd_rtnl_message_close_container(m);
2250 return log_error_errno(r, "Failed to close netlink container: %m");
2252 r = sd_rtnl_call(rtnl, m, 0, NULL);
2254 return log_error_errno(r, "Failed to add new veth interfaces: %m");
2256 i = (int) if_nametoindex(iface_name);
2258 return log_error_errno(errno, "Failed to resolve interface %s: %m", iface_name);
2265 static int setup_bridge(const char veth_name[], int *ifi) {
2266 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
2267 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
2270 if (!arg_private_network)
2273 if (!arg_network_veth)
2276 if (!arg_network_bridge)
2279 bridge = (int) if_nametoindex(arg_network_bridge);
2281 return log_error_errno(errno, "Failed to resolve interface %s: %m", arg_network_bridge);
2285 r = sd_rtnl_open(&rtnl, 0);
2287 return log_error_errno(r, "Failed to connect to netlink: %m");
2289 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, 0);
2291 return log_error_errno(r, "Failed to allocate netlink message: %m");
2293 r = sd_rtnl_message_link_set_flags(m, IFF_UP, IFF_UP);
2295 return log_error_errno(r, "Failed to set IFF_UP flag: %m");
2297 r = sd_rtnl_message_append_string(m, IFLA_IFNAME, veth_name);
2299 return log_error_errno(r, "Failed to add netlink interface name field: %m");
2301 r = sd_rtnl_message_append_u32(m, IFLA_MASTER, bridge);
2303 return log_error_errno(r, "Failed to add netlink master field: %m");
2305 r = sd_rtnl_call(rtnl, m, 0, NULL);
2307 return log_error_errno(r, "Failed to add veth interface to bridge: %m");
2312 static int parse_interface(struct udev *udev, const char *name) {
2313 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
2314 char ifi_str[2 + DECIMAL_STR_MAX(int)];
2317 ifi = (int) if_nametoindex(name);
2319 return log_error_errno(errno, "Failed to resolve interface %s: %m", name);
2321 sprintf(ifi_str, "n%i", ifi);
2322 d = udev_device_new_from_device_id(udev, ifi_str);
2324 return log_error_errno(errno, "Failed to get udev device for interface %s: %m", name);
2326 if (udev_device_get_is_initialized(d) <= 0) {
2327 log_error("Network interface %s is not initialized yet.", name);
2334 static int move_network_interfaces(pid_t pid) {
2335 _cleanup_udev_unref_ struct udev *udev = NULL;
2336 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
2340 if (!arg_private_network)
2343 if (strv_isempty(arg_network_interfaces))
2346 r = sd_rtnl_open(&rtnl, 0);
2348 return log_error_errno(r, "Failed to connect to netlink: %m");
2352 log_error("Failed to connect to udev.");
2356 STRV_FOREACH(i, arg_network_interfaces) {
2357 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
2360 ifi = parse_interface(udev, *i);
2364 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, ifi);
2366 return log_error_errno(r, "Failed to allocate netlink message: %m");
2368 r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
2370 return log_error_errno(r, "Failed to append namespace PID to netlink message: %m");
2372 r = sd_rtnl_call(rtnl, m, 0, NULL);
2374 return log_error_errno(r, "Failed to move interface %s to namespace: %m", *i);
2380 static int setup_macvlan(pid_t pid) {
2381 _cleanup_udev_unref_ struct udev *udev = NULL;
2382 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
2387 if (!arg_private_network)
2390 if (strv_isempty(arg_network_macvlan))
2393 r = sd_rtnl_open(&rtnl, 0);
2395 return log_error_errno(r, "Failed to connect to netlink: %m");
2399 log_error("Failed to connect to udev.");
2403 STRV_FOREACH(i, arg_network_macvlan) {
2404 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
2405 _cleanup_free_ char *n = NULL;
2406 struct ether_addr mac;
2409 ifi = parse_interface(udev, *i);
2413 r = generate_mac(&mac, MACVLAN_HASH_KEY, idx++);
2415 return log_error_errno(r, "Failed to create MACVLAN MAC address: %m");
2417 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2419 return log_error_errno(r, "Failed to allocate netlink message: %m");
2421 r = sd_rtnl_message_append_u32(m, IFLA_LINK, ifi);
2423 return log_error_errno(r, "Failed to add netlink interface index: %m");
2425 n = strappend("mv-", *i);
2429 strshorten(n, IFNAMSIZ-1);
2431 r = sd_rtnl_message_append_string(m, IFLA_IFNAME, n);
2433 return log_error_errno(r, "Failed to add netlink interface name: %m");
2435 r = sd_rtnl_message_append_ether_addr(m, IFLA_ADDRESS, &mac);
2437 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2439 r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
2441 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2443 r = sd_rtnl_message_open_container(m, IFLA_LINKINFO);
2445 return log_error_errno(r, "Failed to open netlink container: %m");
2447 r = sd_rtnl_message_open_container_union(m, IFLA_INFO_DATA, "macvlan");
2449 return log_error_errno(r, "Failed to open netlink container: %m");
2451 r = sd_rtnl_message_append_u32(m, IFLA_MACVLAN_MODE, MACVLAN_MODE_BRIDGE);
2453 return log_error_errno(r, "Failed to append macvlan mode: %m");
2455 r = sd_rtnl_message_close_container(m);
2457 return log_error_errno(r, "Failed to close netlink container: %m");
2459 r = sd_rtnl_message_close_container(m);
2461 return log_error_errno(r, "Failed to close netlink container: %m");
2463 r = sd_rtnl_call(rtnl, m, 0, NULL);
2465 return log_error_errno(r, "Failed to add new macvlan interfaces: %m");
2471 static int setup_ipvlan(pid_t pid) {
2472 _cleanup_udev_unref_ struct udev *udev = NULL;
2473 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
2477 if (!arg_private_network)
2480 if (strv_isempty(arg_network_ipvlan))
2483 r = sd_rtnl_open(&rtnl, 0);
2485 return log_error_errno(r, "Failed to connect to netlink: %m");
2489 log_error("Failed to connect to udev.");
2493 STRV_FOREACH(i, arg_network_ipvlan) {
2494 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
2495 _cleanup_free_ char *n = NULL;
2498 ifi = parse_interface(udev, *i);
2502 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2504 return log_error_errno(r, "Failed to allocate netlink message: %m");
2506 r = sd_rtnl_message_append_u32(m, IFLA_LINK, ifi);
2508 return log_error_errno(r, "Failed to add netlink interface index: %m");
2510 n = strappend("iv-", *i);
2514 strshorten(n, IFNAMSIZ-1);
2516 r = sd_rtnl_message_append_string(m, IFLA_IFNAME, n);
2518 return log_error_errno(r, "Failed to add netlink interface name: %m");
2520 r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
2522 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2524 r = sd_rtnl_message_open_container(m, IFLA_LINKINFO);
2526 return log_error_errno(r, "Failed to open netlink container: %m");
2528 r = sd_rtnl_message_open_container_union(m, IFLA_INFO_DATA, "ipvlan");
2530 return log_error_errno(r, "Failed to open netlink container: %m");
2532 r = sd_rtnl_message_append_u16(m, IFLA_IPVLAN_MODE, IPVLAN_MODE_L2);
2534 return log_error_errno(r, "Failed to add ipvlan mode: %m");
2536 r = sd_rtnl_message_close_container(m);
2538 return log_error_errno(r, "Failed to close netlink container: %m");
2540 r = sd_rtnl_message_close_container(m);
2542 return log_error_errno(r, "Failed to close netlink container: %m");
2544 r = sd_rtnl_call(rtnl, m, 0, NULL);
2546 return log_error_errno(r, "Failed to add new ipvlan interfaces: %m");
2552 static int setup_seccomp(void) {
2555 static const int blacklist[] = {
2556 SCMP_SYS(kexec_load),
2557 SCMP_SYS(open_by_handle_at),
2564 static const int kmod_blacklist[] = {
2565 SCMP_SYS(init_module),
2566 SCMP_SYS(finit_module),
2567 SCMP_SYS(delete_module),
2570 scmp_filter_ctx seccomp;
2574 seccomp = seccomp_init(SCMP_ACT_ALLOW);
2578 r = seccomp_add_secondary_archs(seccomp);
2580 log_error_errno(r, "Failed to add secondary archs to seccomp filter: %m");
2584 for (i = 0; i < ELEMENTSOF(blacklist); i++) {
2585 r = seccomp_rule_add(seccomp, SCMP_ACT_ERRNO(EPERM), blacklist[i], 0);
2587 continue; /* unknown syscall */
2589 log_error_errno(r, "Failed to block syscall: %m");
2594 /* If the CAP_SYS_MODULE capability is not requested then
2595 * we'll block the kmod syscalls too */
2596 if (!(arg_retain & (1ULL << CAP_SYS_MODULE))) {
2597 for (i = 0; i < ELEMENTSOF(kmod_blacklist); i++) {
2598 r = seccomp_rule_add(seccomp, SCMP_ACT_ERRNO(EPERM), kmod_blacklist[i], 0);
2600 continue; /* unknown syscall */
2602 log_error_errno(r, "Failed to block syscall: %m");
2609 Audit is broken in containers, much of the userspace audit
2610 hookup will fail if running inside a container. We don't
2611 care and just turn off creation of audit sockets.
2613 This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail
2614 with EAFNOSUPPORT which audit userspace uses as indication
2615 that audit is disabled in the kernel.
2618 r = seccomp_rule_add(
2620 SCMP_ACT_ERRNO(EAFNOSUPPORT),
2623 SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
2624 SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
2626 log_error_errno(r, "Failed to add audit seccomp rule: %m");
2630 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
2632 log_error_errno(r, "Failed to unset NO_NEW_PRIVS: %m");
2636 r = seccomp_load(seccomp);
2638 log_error_errno(r, "Failed to install seccomp audit filter: %m");
2641 seccomp_release(seccomp);
2649 static int setup_propagate(const char *root) {
2652 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2653 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2654 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2655 (void) mkdir_p(p, 0600);
2657 q = strjoina(root, "/run/systemd/nspawn/incoming");
2658 mkdir_parents(q, 0755);
2661 if (mount(p, q, NULL, MS_BIND, NULL) < 0)
2662 return log_error_errno(errno, "Failed to install propagation bind mount.");
2664 if (mount(NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0)
2665 return log_error_errno(errno, "Failed to make propagation mount read-only");
2670 static int setup_image(char **device_path, int *loop_nr) {
2671 struct loop_info64 info = {
2672 .lo_flags = LO_FLAGS_AUTOCLEAR|LO_FLAGS_PARTSCAN
2674 _cleanup_close_ int fd = -1, control = -1, loop = -1;
2675 _cleanup_free_ char* loopdev = NULL;
2679 assert(device_path);
2683 fd = open(arg_image, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
2685 return log_error_errno(errno, "Failed to open %s: %m", arg_image);
2687 if (fstat(fd, &st) < 0)
2688 return log_error_errno(errno, "Failed to stat %s: %m", arg_image);
2690 if (S_ISBLK(st.st_mode)) {
2693 p = strdup(arg_image);
2707 if (!S_ISREG(st.st_mode)) {
2708 log_error_errno(errno, "%s is not a regular file or block device: %m", arg_image);
2712 control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
2714 return log_error_errno(errno, "Failed to open /dev/loop-control: %m");
2716 nr = ioctl(control, LOOP_CTL_GET_FREE);
2718 return log_error_errno(errno, "Failed to allocate loop device: %m");
2720 if (asprintf(&loopdev, "/dev/loop%i", nr) < 0)
2723 loop = open(loopdev, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
2725 return log_error_errno(errno, "Failed to open loop device %s: %m", loopdev);
2727 if (ioctl(loop, LOOP_SET_FD, fd) < 0)
2728 return log_error_errno(errno, "Failed to set loopback file descriptor on %s: %m", loopdev);
2731 info.lo_flags |= LO_FLAGS_READ_ONLY;
2733 if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0)
2734 return log_error_errno(errno, "Failed to set loopback settings on %s: %m", loopdev);
2736 *device_path = loopdev;
2747 #define PARTITION_TABLE_BLURB \
2748 "Note that the disk image needs to either contain only a single MBR partition of\n" \
2749 "type 0x83 that is marked bootable, or a sinlge GPT partition of type" \
2750 "0FC63DAF-8483-4772-8E79-3D69D8477DE4 or follow\n" \
2751 " http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n" \
2752 "to be bootable with systemd-nspawn."
2754 static int dissect_image(
2756 char **root_device, bool *root_device_rw,
2757 char **home_device, bool *home_device_rw,
2758 char **srv_device, bool *srv_device_rw,
2762 int home_nr = -1, srv_nr = -1;
2763 #ifdef GPT_ROOT_NATIVE
2766 #ifdef GPT_ROOT_SECONDARY
2767 int secondary_root_nr = -1;
2769 _cleanup_free_ char *home = NULL, *root = NULL, *secondary_root = NULL, *srv = NULL, *generic = NULL;
2770 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
2771 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
2772 _cleanup_blkid_free_probe_ blkid_probe b = NULL;
2773 _cleanup_udev_unref_ struct udev *udev = NULL;
2774 struct udev_list_entry *first, *item;
2775 bool home_rw = true, root_rw = true, secondary_root_rw = true, srv_rw = true, generic_rw = true;
2776 bool is_gpt, is_mbr, multiple_generic = false;
2777 const char *pttype = NULL;
2784 assert(root_device);
2785 assert(home_device);
2790 b = blkid_new_probe();
2795 r = blkid_probe_set_device(b, fd, 0, 0);
2800 log_error_errno(errno, "Failed to set device on blkid probe: %m");
2804 blkid_probe_enable_partitions(b, 1);
2805 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
2808 r = blkid_do_safeprobe(b);
2809 if (r == -2 || r == 1) {
2810 log_error("Failed to identify any partition table on\n"
2812 PARTITION_TABLE_BLURB, arg_image);
2814 } else if (r != 0) {
2817 log_error_errno(errno, "Failed to probe: %m");
2821 blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
2823 is_gpt = streq_ptr(pttype, "gpt");
2824 is_mbr = streq_ptr(pttype, "dos");
2826 if (!is_gpt && !is_mbr) {
2827 log_error("No GPT or MBR partition table discovered on\n"
2829 PARTITION_TABLE_BLURB, arg_image);
2834 pl = blkid_probe_get_partitions(b);
2839 log_error("Failed to list partitions of %s", arg_image);
2847 if (fstat(fd, &st) < 0)
2848 return log_error_errno(errno, "Failed to stat block device: %m");
2850 d = udev_device_new_from_devnum(udev, 'b', st.st_rdev);
2858 log_error("Kernel partitions never appeared.");
2862 e = udev_enumerate_new(udev);
2866 r = udev_enumerate_add_match_parent(e, d);
2870 r = udev_enumerate_scan_devices(e);
2872 return log_error_errno(r, "Failed to scan for partition devices of %s: %m", arg_image);
2874 /* Count the partitions enumerated by the kernel */
2876 first = udev_enumerate_get_list_entry(e);
2877 udev_list_entry_foreach(item, first)
2880 /* Count the partitions enumerated by blkid */
2881 m = blkid_partlist_numof_partitions(pl);
2885 log_error("blkid and kernel partition list do not match.");
2891 /* The kernel has probed fewer partitions than
2892 * blkid? Maybe the kernel prober is still
2893 * running or it got EBUSY because udev
2894 * already opened the device. Let's reprobe
2895 * the device, which is a synchronous call
2896 * that waits until probing is complete. */
2898 for (j = 0; j < 20; j++) {
2900 r = ioctl(fd, BLKRRPART, 0);
2903 if (r >= 0 || r != -EBUSY)
2906 /* If something else has the device
2907 * open, such as an udev rule, the
2908 * ioctl will return EBUSY. Since
2909 * there's no way to wait until it
2910 * isn't busy anymore, let's just wait
2911 * a bit, and try again.
2913 * This is really something they
2914 * should fix in the kernel! */
2916 usleep(50 * USEC_PER_MSEC);
2920 return log_error_errno(r, "Failed to reread partition table: %m");
2923 e = udev_enumerate_unref(e);
2926 first = udev_enumerate_get_list_entry(e);
2927 udev_list_entry_foreach(item, first) {
2928 _cleanup_udev_device_unref_ struct udev_device *q;
2930 unsigned long long flags;
2936 q = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
2941 log_error_errno(errno, "Failed to get partition device of %s: %m", arg_image);
2945 qn = udev_device_get_devnum(q);
2949 if (st.st_rdev == qn)
2952 node = udev_device_get_devnode(q);
2956 pp = blkid_partlist_devno_to_partition(pl, qn);
2960 flags = blkid_partition_get_flags(pp);
2962 nr = blkid_partition_get_partno(pp);
2970 if (flags & GPT_FLAG_NO_AUTO)
2973 stype = blkid_partition_get_type_string(pp);
2977 if (sd_id128_from_string(stype, &type_id) < 0)
2980 if (sd_id128_equal(type_id, GPT_HOME)) {
2982 if (home && nr >= home_nr)
2986 home_rw = !(flags & GPT_FLAG_READ_ONLY);
2988 r = free_and_strdup(&home, node);
2992 } else if (sd_id128_equal(type_id, GPT_SRV)) {
2994 if (srv && nr >= srv_nr)
2998 srv_rw = !(flags & GPT_FLAG_READ_ONLY);
3000 r = free_and_strdup(&srv, node);
3004 #ifdef GPT_ROOT_NATIVE
3005 else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
3007 if (root && nr >= root_nr)
3011 root_rw = !(flags & GPT_FLAG_READ_ONLY);
3013 r = free_and_strdup(&root, node);
3018 #ifdef GPT_ROOT_SECONDARY
3019 else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
3021 if (secondary_root && nr >= secondary_root_nr)
3024 secondary_root_nr = nr;
3025 secondary_root_rw = !(flags & GPT_FLAG_READ_ONLY);
3027 r = free_and_strdup(&secondary_root, node);
3032 else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
3035 multiple_generic = true;
3037 generic_rw = !(flags & GPT_FLAG_READ_ONLY);
3039 r = free_and_strdup(&generic, node);
3045 } else if (is_mbr) {
3048 if (flags != 0x80) /* Bootable flag */
3051 type = blkid_partition_get_type(pp);
3052 if (type != 0x83) /* Linux partition */
3056 multiple_generic = true;
3060 r = free_and_strdup(&root, node);
3068 *root_device = root;
3071 *root_device_rw = root_rw;
3073 } else if (secondary_root) {
3074 *root_device = secondary_root;
3075 secondary_root = NULL;
3077 *root_device_rw = secondary_root_rw;
3079 } else if (generic) {
3081 /* There were no partitions with precise meanings
3082 * around, but we found generic partitions. In this
3083 * case, if there's only one, we can go ahead and boot
3084 * it, otherwise we bail out, because we really cannot
3085 * make any sense of it. */
3087 if (multiple_generic) {
3088 log_error("Identified multiple bootable Linux partitions on\n"
3090 PARTITION_TABLE_BLURB, arg_image);
3094 *root_device = generic;
3097 *root_device_rw = generic_rw;
3100 log_error("Failed to identify root partition in disk image\n"
3102 PARTITION_TABLE_BLURB, arg_image);
3107 *home_device = home;
3110 *home_device_rw = home_rw;
3117 *srv_device_rw = srv_rw;
3122 log_error("--image= is not supported, compiled without blkid support.");
3127 static int mount_device(const char *what, const char *where, const char *directory, bool rw) {
3129 _cleanup_blkid_free_probe_ blkid_probe b = NULL;
3130 const char *fstype, *p;
3140 p = strjoina(where, directory);
3145 b = blkid_new_probe_from_filename(what);
3149 log_error_errno(errno, "Failed to allocate prober for %s: %m", what);
3153 blkid_probe_enable_superblocks(b, 1);
3154 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
3157 r = blkid_do_safeprobe(b);
3158 if (r == -1 || r == 1) {
3159 log_error("Cannot determine file system type of %s", what);
3161 } else if (r != 0) {
3164 log_error_errno(errno, "Failed to probe %s: %m", what);
3169 if (blkid_probe_lookup_value(b, "TYPE", &fstype, NULL) < 0) {
3172 log_error("Failed to determine file system type of %s", what);
3176 if (streq(fstype, "crypto_LUKS")) {
3177 log_error("nspawn currently does not support LUKS disk images.");
3181 if (mount(what, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), NULL) < 0)
3182 return log_error_errno(errno, "Failed to mount %s: %m", what);
3186 log_error("--image= is not supported, compiled without blkid support.");
3191 static int mount_devices(
3193 const char *root_device, bool root_device_rw,
3194 const char *home_device, bool home_device_rw,
3195 const char *srv_device, bool srv_device_rw) {
3201 r = mount_device(root_device, arg_directory, NULL, root_device_rw);
3203 return log_error_errno(r, "Failed to mount root directory: %m");
3207 r = mount_device(home_device, arg_directory, "/home", home_device_rw);
3209 return log_error_errno(r, "Failed to mount home directory: %m");
3213 r = mount_device(srv_device, arg_directory, "/srv", srv_device_rw);
3215 return log_error_errno(r, "Failed to mount server data directory: %m");
3221 static void loop_remove(int nr, int *image_fd) {
3222 _cleanup_close_ int control = -1;
3228 if (image_fd && *image_fd >= 0) {
3229 r = ioctl(*image_fd, LOOP_CLR_FD);
3231 log_debug_errno(errno, "Failed to close loop image: %m");
3232 *image_fd = safe_close(*image_fd);
3235 control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
3237 log_warning_errno(errno, "Failed to open /dev/loop-control: %m");
3241 r = ioctl(control, LOOP_CTL_REMOVE, nr);
3243 log_debug_errno(errno, "Failed to remove loop %d: %m", nr);
3246 static int spawn_getent(const char *database, const char *key, pid_t *rpid) {
3254 if (pipe2(pipe_fds, O_CLOEXEC) < 0)
3255 return log_error_errno(errno, "Failed to allocate pipe: %m");
3259 return log_error_errno(errno, "Failed to fork getent child: %m");
3260 else if (pid == 0) {
3262 char *empty_env = NULL;
3264 if (dup3(pipe_fds[1], STDOUT_FILENO, 0) < 0)
3265 _exit(EXIT_FAILURE);
3267 if (pipe_fds[0] > 2)
3268 safe_close(pipe_fds[0]);
3269 if (pipe_fds[1] > 2)
3270 safe_close(pipe_fds[1]);
3272 nullfd = open("/dev/null", O_RDWR);
3274 _exit(EXIT_FAILURE);
3276 if (dup3(nullfd, STDIN_FILENO, 0) < 0)
3277 _exit(EXIT_FAILURE);
3279 if (dup3(nullfd, STDERR_FILENO, 0) < 0)
3280 _exit(EXIT_FAILURE);
3285 reset_all_signal_handlers();
3286 close_all_fds(NULL, 0);
3288 execle("/usr/bin/getent", "getent", database, key, NULL, &empty_env);
3289 execle("/bin/getent", "getent", database, key, NULL, &empty_env);
3290 _exit(EXIT_FAILURE);
3293 pipe_fds[1] = safe_close(pipe_fds[1]);
3300 static int change_uid_gid(char **_home) {
3301 char line[LINE_MAX], *x, *u, *g, *h;
3302 const char *word, *state;
3303 _cleanup_free_ uid_t *uids = NULL;
3304 _cleanup_free_ char *home = NULL;
3305 _cleanup_fclose_ FILE *f = NULL;
3306 _cleanup_close_ int fd = -1;
3307 unsigned n_uids = 0;
3316 if (!arg_user || streq(arg_user, "root") || streq(arg_user, "0")) {
3317 /* Reset everything fully to 0, just in case */
3319 if (setgroups(0, NULL) < 0)
3320 return log_error_errno(errno, "setgroups() failed: %m");
3322 if (setresgid(0, 0, 0) < 0)
3323 return log_error_errno(errno, "setregid() failed: %m");
3325 if (setresuid(0, 0, 0) < 0)
3326 return log_error_errno(errno, "setreuid() failed: %m");
3332 /* First, get user credentials */
3333 fd = spawn_getent("passwd", arg_user, &pid);
3337 f = fdopen(fd, "r");
3342 if (!fgets(line, sizeof(line), f)) {
3345 log_error("Failed to resolve user %s.", arg_user);
3349 log_error_errno(errno, "Failed to read from getent: %m");
3355 wait_for_terminate_and_warn("getent passwd", pid, true);
3357 x = strchr(line, ':');
3359 log_error("/etc/passwd entry has invalid user field.");
3363 u = strchr(x+1, ':');
3365 log_error("/etc/passwd entry has invalid password field.");
3372 log_error("/etc/passwd entry has invalid UID field.");
3380 log_error("/etc/passwd entry has invalid GID field.");
3385 h = strchr(x+1, ':');
3387 log_error("/etc/passwd entry has invalid GECOS field.");
3394 log_error("/etc/passwd entry has invalid home directory field.");
3400 r = parse_uid(u, &uid);
3402 log_error("Failed to parse UID of user.");
3406 r = parse_gid(g, &gid);
3408 log_error("Failed to parse GID of user.");
3416 /* Second, get group memberships */
3417 fd = spawn_getent("initgroups", arg_user, &pid);
3422 f = fdopen(fd, "r");
3427 if (!fgets(line, sizeof(line), f)) {
3429 log_error("Failed to resolve user %s.", arg_user);
3433 log_error_errno(errno, "Failed to read from getent: %m");
3439 wait_for_terminate_and_warn("getent initgroups", pid, true);
3441 /* Skip over the username and subsequent separator whitespace */
3443 x += strcspn(x, WHITESPACE);
3444 x += strspn(x, WHITESPACE);
3446 FOREACH_WORD(word, l, x, state) {
3452 if (!GREEDY_REALLOC(uids, sz, n_uids+1))
3455 r = parse_uid(c, &uids[n_uids++]);
3457 log_error("Failed to parse group data from getent.");
3462 r = mkdir_parents(home, 0775);
3464 return log_error_errno(r, "Failed to make home root directory: %m");
3466 r = mkdir_safe(home, 0755, uid, gid);
3467 if (r < 0 && r != -EEXIST)
3468 return log_error_errno(r, "Failed to make home directory: %m");
3470 fchown(STDIN_FILENO, uid, gid);
3471 fchown(STDOUT_FILENO, uid, gid);
3472 fchown(STDERR_FILENO, uid, gid);
3474 if (setgroups(n_uids, uids) < 0)
3475 return log_error_errno(errno, "Failed to set auxiliary groups: %m");
3477 if (setresgid(gid, gid, gid) < 0)
3478 return log_error_errno(errno, "setregid() failed: %m");
3480 if (setresuid(uid, uid, uid) < 0)
3481 return log_error_errno(errno, "setreuid() failed: %m");
3493 * < 0 : wait_for_terminate() failed to get the state of the
3494 * container, the container was terminated by a signal, or
3495 * failed for an unknown reason. No change is made to the
3496 * container argument.
3497 * > 0 : The program executed in the container terminated with an
3498 * error. The exit code of the program executed in the
3499 * container is returned. The container argument has been set
3500 * to CONTAINER_TERMINATED.
3501 * 0 : The container is being rebooted, has been shut down or exited
3502 * successfully. The container argument has been set to either
3503 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
3505 * That is, success is indicated by a return value of zero, and an
3506 * error is indicated by a non-zero value.
3508 static int wait_for_container(pid_t pid, ContainerStatus *container) {
3512 r = wait_for_terminate(pid, &status);
3514 return log_warning_errno(r, "Failed to wait for container: %m");
3516 switch (status.si_code) {
3519 if (status.si_status == 0) {
3520 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
3523 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
3525 *container = CONTAINER_TERMINATED;
3526 return status.si_status;
3529 if (status.si_status == SIGINT) {
3531 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
3532 *container = CONTAINER_TERMINATED;
3535 } else if (status.si_status == SIGHUP) {
3537 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
3538 *container = CONTAINER_REBOOTED;
3542 /* CLD_KILLED fallthrough */
3545 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
3549 log_error("Container %s failed due to unknown reason.", arg_machine);
3556 static void nop_handler(int sig) {}
3558 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
3561 pid = PTR_TO_UINT32(userdata);
3563 if (kill(pid, SIGRTMIN+3) >= 0) {
3564 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
3565 sd_event_source_set_userdata(s, NULL);
3570 sd_event_exit(sd_event_source_get_event(s), 0);
3574 static int determine_names(void) {
3577 if (!arg_image && !arg_directory) {
3579 _cleanup_(image_unrefp) Image *i = NULL;
3581 r = image_find(arg_machine, &i);
3583 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
3585 log_error("No image for machine '%s': %m", arg_machine);
3589 if (i->type == IMAGE_RAW)
3590 r = set_sanitized_path(&arg_image, i->path);
3592 r = set_sanitized_path(&arg_directory, i->path);
3594 return log_error_errno(r, "Invalid image directory: %m");
3596 arg_read_only = arg_read_only || i->read_only;
3598 arg_directory = get_current_dir_name();
3600 if (!arg_directory && !arg_machine) {
3601 log_error("Failed to determine path, please use -D or -i.");
3607 if (arg_directory && path_equal(arg_directory, "/"))
3608 arg_machine = gethostname_malloc();
3610 arg_machine = strdup(basename(arg_image ?: arg_directory));
3615 hostname_cleanup(arg_machine, false);
3616 if (!machine_name_is_valid(arg_machine)) {
3617 log_error("Failed to determine machine name automatically, please use -M.");
3621 if (arg_ephemeral) {
3624 /* Add a random suffix when this is an
3625 * ephemeral machine, so that we can run many
3626 * instances at once without manually having
3627 * to specify -M each time. */
3629 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
3640 static int determine_uid_shift(void) {
3646 if (arg_uid_shift == UID_INVALID) {
3649 r = stat(arg_directory, &st);
3651 return log_error_errno(errno, "Failed to determine UID base of %s: %m", arg_directory);
3653 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
3655 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
3656 log_error("UID and GID base of %s don't match.", arg_directory);
3660 arg_uid_range = UINT32_C(0x10000);
3663 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
3664 log_error("UID base too high for UID range.");
3668 log_info("Using user namespaces with base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3672 int main(int argc, char *argv[]) {
3674 _cleanup_free_ char *device_path = NULL, *root_device = NULL, *home_device = NULL, *srv_device = NULL, *console = NULL;
3675 bool root_device_rw = true, home_device_rw = true, srv_device_rw = true;
3676 _cleanup_close_ int master = -1, image_fd = -1;
3677 _cleanup_fdset_free_ FDSet *fds = NULL;
3678 int r, n_fd_passed, loop_nr = -1;
3679 char veth_name[IFNAMSIZ];
3680 bool secondary = false, remove_subvol = false;
3681 sigset_t mask, mask_chld;
3683 int ret = EXIT_SUCCESS;
3684 union in_addr_union exposed = {};
3685 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
3688 log_parse_environment();
3691 r = parse_argv(argc, argv);
3695 r = determine_names();
3699 if (geteuid() != 0) {
3700 log_error("Need to be root.");
3705 if (sd_booted() <= 0) {
3706 log_error("Not running on a systemd system.");
3712 n_fd_passed = sd_listen_fds(false);
3713 if (n_fd_passed > 0) {
3714 r = fdset_new_listen_fds(&fds, false);
3716 log_error_errno(r, "Failed to collect file descriptors: %m");
3720 fdset_close_others(fds);
3723 if (arg_directory) {
3726 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
3727 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
3732 if (arg_ephemeral) {
3735 /* If the specified path is a mount point we
3736 * generate the new snapshot immediately
3737 * inside it under a random name. However if
3738 * the specified is not a mount point we
3739 * create the new snapshot in the parent
3740 * directory, just next to it. */
3741 r = path_is_mount_point(arg_directory, false);
3743 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
3747 r = tempfn_random_child(arg_directory, &np);
3749 r = tempfn_random(arg_directory, &np);
3751 log_error_errno(r, "Failed to generate name for snapshot: %m");
3755 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3757 log_error_errno(r, "Failed to lock %s: %m", np);
3761 r = btrfs_subvol_snapshot(arg_directory, np, arg_read_only, true);
3764 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
3768 free(arg_directory);
3771 remove_subvol = true;
3774 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3776 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
3780 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
3785 r = btrfs_subvol_snapshot(arg_template, arg_directory, arg_read_only, true);
3788 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
3790 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
3794 log_info("Populated %s from template %s.", arg_directory, arg_template);
3800 if (path_is_os_tree(arg_directory) <= 0) {
3801 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
3808 p = strjoina(arg_directory,
3809 argc > optind && path_is_absolute(argv[optind]) ? argv[optind] : "/usr/bin/");
3810 if (access(p, F_OK) < 0) {
3811 log_error("Directory %s lacks the binary to execute or doesn't look like a binary tree. Refusing.", arg_directory);
3818 char template[] = "/tmp/nspawn-root-XXXXXX";
3821 assert(!arg_template);
3823 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
3825 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
3829 r = log_error_errno(r, "Failed to create image lock: %m");
3833 if (!mkdtemp(template)) {
3834 log_error_errno(errno, "Failed to create temporary directory: %m");
3839 arg_directory = strdup(template);
3840 if (!arg_directory) {
3845 image_fd = setup_image(&device_path, &loop_nr);
3851 r = dissect_image(image_fd,
3852 &root_device, &root_device_rw,
3853 &home_device, &home_device_rw,
3854 &srv_device, &srv_device_rw,
3860 r = determine_uid_shift();
3864 interactive = isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0;
3866 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
3868 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
3872 r = ptsname_malloc(master, &console);
3874 r = log_error_errno(r, "Failed to determine tty name: %m");
3878 if (unlockpt(master) < 0) {
3879 r = log_error_errno(errno, "Failed to unlock tty: %m");
3884 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
3885 arg_machine, arg_image ?: arg_directory);
3887 assert_se(sigemptyset(&mask) == 0);
3888 sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
3889 assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
3891 assert_se(sigemptyset(&mask_chld) == 0);
3892 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
3895 _cleanup_close_pair_ int kmsg_socket_pair[2] = { -1, -1 }, rtnl_socket_pair[2] = { -1, -1 };
3896 ContainerStatus container_status;
3897 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
3898 struct sigaction sa = {
3899 .sa_handler = nop_handler,
3900 .sa_flags = SA_NOCLDSTOP,
3903 r = barrier_create(&barrier);
3905 log_error_errno(r, "Cannot initialize IPC barrier: %m");
3909 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
3910 r = log_error_errno(errno, "Failed to create kmsg socket pair: %m");
3914 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0) {
3915 r = log_error_errno(errno, "Failed to create rtnl socket pair: %m");
3919 /* Child can be killed before execv(), so handle SIGCHLD
3920 * in order to interrupt parent's blocking calls and
3921 * give it a chance to call wait() and terminate. */
3922 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
3924 r = log_error_errno(errno, "Failed to change the signal mask: %m");
3928 r = sigaction(SIGCHLD, &sa, NULL);
3930 r = log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
3934 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3935 (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS)|
3936 (arg_private_network ? CLONE_NEWNET : 0), NULL);
3938 if (errno == EINVAL)
3939 r = log_error_errno(errno, "clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
3941 r = log_error_errno(errno, "clone() failed: %m");
3948 _cleanup_free_ char *home = NULL;
3950 const char *envp[] = {
3951 "PATH=" DEFAULT_PATH_SPLIT_USR,
3952 "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
3957 NULL, /* container_uuid */
3958 NULL, /* LISTEN_FDS */
3959 NULL, /* LISTEN_PID */
3964 barrier_set_role(&barrier, BARRIER_CHILD);
3966 envp[n_env] = strv_find_prefix(environ, "TERM=");
3970 master = safe_close(master);
3972 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
3973 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
3975 reset_all_signal_handlers();
3976 reset_signal_mask();
3979 close_nointr(STDIN_FILENO);
3980 close_nointr(STDOUT_FILENO);
3981 close_nointr(STDERR_FILENO);
3983 r = open_terminal(console, O_RDWR);
3984 if (r != STDIN_FILENO) {
3990 log_error_errno(r, "Failed to open console: %m");
3991 _exit(EXIT_FAILURE);
3994 if (dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
3995 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO) {
3996 log_error_errno(errno, "Failed to duplicate console: %m");
3997 _exit(EXIT_FAILURE);
4002 log_error_errno(errno, "setsid() failed: %m");
4003 _exit(EXIT_FAILURE);
4006 if (reset_audit_loginuid() < 0)
4007 _exit(EXIT_FAILURE);
4009 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0) {
4010 log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
4011 _exit(EXIT_FAILURE);
4014 if (arg_private_network)
4017 /* Mark everything as slave, so that we still
4018 * receive mounts from the real root, but don't
4019 * propagate mounts to the real root. */
4020 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
4021 log_error_errno(errno, "MS_SLAVE|MS_REC failed: %m");
4022 _exit(EXIT_FAILURE);
4025 if (mount_devices(arg_directory,
4026 root_device, root_device_rw,
4027 home_device, home_device_rw,
4028 srv_device, srv_device_rw) < 0)
4029 _exit(EXIT_FAILURE);
4031 /* Turn directory into bind mount */
4032 if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REC, NULL) < 0) {
4033 log_error_errno(errno, "Failed to make bind mount: %m");
4034 _exit(EXIT_FAILURE);
4037 r = setup_volatile(arg_directory);
4039 _exit(EXIT_FAILURE);
4041 if (setup_volatile_state(arg_directory) < 0)
4042 _exit(EXIT_FAILURE);
4044 r = base_filesystem_create(arg_directory);
4046 _exit(EXIT_FAILURE);
4048 if (arg_read_only) {
4049 r = bind_remount_recursive(arg_directory, true);
4051 log_error_errno(r, "Failed to make tree read-only: %m");
4052 _exit(EXIT_FAILURE);
4056 if (mount_all(arg_directory) < 0)
4057 _exit(EXIT_FAILURE);
4059 if (copy_devnodes(arg_directory) < 0)
4060 _exit(EXIT_FAILURE);
4062 if (setup_ptmx(arg_directory) < 0)
4063 _exit(EXIT_FAILURE);
4065 dev_setup(arg_directory);
4067 if (setup_propagate(arg_directory) < 0)
4068 _exit(EXIT_FAILURE);
4070 if (setup_seccomp() < 0)
4071 _exit(EXIT_FAILURE);
4073 if (setup_dev_console(arg_directory, console) < 0)
4074 _exit(EXIT_FAILURE);
4076 if (setup_kmsg(arg_directory, kmsg_socket_pair[1]) < 0)
4077 _exit(EXIT_FAILURE);
4078 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4080 if (send_rtnl(rtnl_socket_pair[1]) < 0)
4081 _exit(EXIT_FAILURE);
4082 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4084 /* Tell the parent that we are ready, and that
4085 * it can cgroupify us to that we lack access
4086 * to certain devices and resources. */
4087 (void) barrier_place(&barrier); /* #1 */
4089 if (setup_boot_id(arg_directory) < 0)
4090 _exit(EXIT_FAILURE);
4092 if (setup_timezone(arg_directory) < 0)
4093 _exit(EXIT_FAILURE);
4095 if (setup_resolv_conf(arg_directory) < 0)
4096 _exit(EXIT_FAILURE);
4098 if (setup_journal(arg_directory) < 0)
4099 _exit(EXIT_FAILURE);
4101 if (mount_binds(arg_directory, arg_bind, false) < 0)
4102 _exit(EXIT_FAILURE);
4104 if (mount_binds(arg_directory, arg_bind_ro, true) < 0)
4105 _exit(EXIT_FAILURE);
4107 if (mount_tmpfs(arg_directory) < 0)
4108 _exit(EXIT_FAILURE);
4110 /* Wait until we are cgroup-ified, so that we
4111 * can mount the right cgroup path writable */
4112 (void) barrier_place_and_sync(&barrier); /* #2 */
4114 if (mount_cgroup(arg_directory) < 0)
4115 _exit(EXIT_FAILURE);
4117 if (chdir(arg_directory) < 0) {
4118 log_error_errno(errno, "chdir(%s) failed: %m", arg_directory);
4119 _exit(EXIT_FAILURE);
4122 if (mount(arg_directory, "/", NULL, MS_MOVE, NULL) < 0) {
4123 log_error_errno(errno, "mount(MS_MOVE) failed: %m");
4124 _exit(EXIT_FAILURE);
4127 if (chroot(".") < 0) {
4128 log_error_errno(errno, "chroot() failed: %m");
4129 _exit(EXIT_FAILURE);
4132 if (chdir("/") < 0) {
4133 log_error_errno(errno, "chdir() failed: %m");
4134 _exit(EXIT_FAILURE);
4138 if (unshare(CLONE_NEWUSER) < 0) {
4139 log_error_errno(errno, "unshare(CLONE_NEWUSER) failed: %m");
4140 _exit(EXIT_FAILURE);
4143 /* Tell the parent, that it now can
4144 * write the UID map. */
4145 (void) barrier_place(&barrier); /* #3 */
4147 /* Wait until the parent wrote the UID
4149 (void) barrier_place_and_sync(&barrier); /* #4 */
4154 if (drop_capabilities() < 0) {
4155 log_error_errno(errno, "drop_capabilities() failed: %m");
4156 _exit(EXIT_FAILURE);
4161 if (arg_personality != 0xffffffffLU) {
4162 if (personality(arg_personality) < 0) {
4163 log_error_errno(errno, "personality() failed: %m");
4164 _exit(EXIT_FAILURE);
4166 } else if (secondary) {
4167 if (personality(PER_LINUX32) < 0) {
4168 log_error_errno(errno, "personality() failed: %m");
4169 _exit(EXIT_FAILURE);
4174 if (arg_selinux_context)
4175 if (setexeccon((security_context_t) arg_selinux_context) < 0) {
4176 log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
4177 _exit(EXIT_FAILURE);
4181 r = change_uid_gid(&home);
4183 _exit(EXIT_FAILURE);
4185 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
4186 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
4187 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)) {
4189 _exit(EXIT_FAILURE);
4192 if (!sd_id128_equal(arg_uuid, SD_ID128_NULL)) {
4195 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_format_as_uuid(arg_uuid, as_uuid)) < 0) {
4197 _exit(EXIT_FAILURE);
4201 if (fdset_size(fds) > 0) {
4202 r = fdset_cloexec(fds, false);
4204 log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
4205 _exit(EXIT_FAILURE);
4208 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", n_fd_passed) < 0) ||
4209 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0)) {
4211 _exit(EXIT_FAILURE);
4215 if (!strv_isempty(arg_setenv)) {
4218 n = strv_env_merge(2, envp, arg_setenv);
4221 _exit(EXIT_FAILURE);
4226 env_use = (char**) envp;
4228 /* Let the parent know that we are ready and
4229 * wait until the parent is ready with the
4231 (void) barrier_place_and_sync(&barrier); /* #5 */
4237 /* Automatically search for the init system */
4239 l = 1 + argc - optind;
4240 a = newa(char*, l + 1);
4241 memcpy(a + 1, argv + optind, l * sizeof(char*));
4243 a[0] = (char*) "/usr/lib/systemd/systemd";
4244 execve(a[0], a, env_use);
4246 a[0] = (char*) "/lib/systemd/systemd";
4247 execve(a[0], a, env_use);
4249 a[0] = (char*) "/sbin/init";
4250 execve(a[0], a, env_use);
4251 } else if (argc > optind)
4252 execvpe(argv[optind], argv + optind, env_use);
4254 chdir(home ? home : "/root");
4255 execle("/bin/bash", "-bash", NULL, env_use);
4256 execle("/bin/sh", "-sh", NULL, env_use);
4259 log_error_errno(errno, "execv() failed: %m");
4260 _exit(EXIT_FAILURE);
4263 barrier_set_role(&barrier, BARRIER_PARENT);
4267 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4268 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4270 (void) barrier_place(&barrier); /* #1 */
4272 /* Wait for the most basic Child-setup to be done,
4273 * before we add hardware to it, and place it in a
4275 if (barrier_sync(&barrier)) { /* #1 */
4278 r = move_network_interfaces(pid);
4282 r = setup_veth(pid, veth_name, &ifi);
4286 r = setup_bridge(veth_name, &ifi);
4290 r = setup_macvlan(pid);
4294 r = setup_ipvlan(pid);
4298 r = register_machine(pid, ifi);
4302 /* Notify the child that the parent is ready with all
4303 * its setup, and that the child can now hand over
4304 * control to the code to run inside the container. */
4305 (void) barrier_place(&barrier); /* #2 */
4308 char uid_map[strlen("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
4310 (void) barrier_place_and_sync(&barrier); /* #3 */
4312 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
4313 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
4314 r = write_string_file(uid_map, line);
4316 log_error_errno(r, "Failed to write UID map: %m");
4320 /* We always assign the same UID and GID ranges */
4321 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
4322 r = write_string_file(uid_map, line);
4324 log_error_errno(r, "Failed to write GID map: %m");
4328 (void) barrier_place(&barrier); /* #4 */
4331 /* Block SIGCHLD here, before notifying child.
4332 * process_pty() will handle it with the other signals. */
4333 r = sigprocmask(SIG_BLOCK, &mask_chld, NULL);
4337 /* Reset signal to default */
4338 r = default_signals(SIGCHLD, -1);
4342 /* Let the child know that we are ready and wait that the child is completely ready now. */
4343 if (barrier_place_and_sync(&barrier)) { /* #5 */
4344 _cleanup_event_unref_ sd_event *event = NULL;
4345 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4346 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
4351 "STATUS=Container running.\n"
4352 "X_NSPAWN_LEADER_PID=" PID_FMT, pid);
4354 r = sd_event_new(&event);
4356 log_error_errno(r, "Failed to get default event source: %m");
4361 /* Try to kill the init system on SIGINT or SIGTERM */
4362 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, UINT32_TO_PTR(pid));
4363 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, UINT32_TO_PTR(pid));
4365 /* Immediately exit */
4366 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4367 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4370 /* simply exit on sigchld */
4371 sd_event_add_signal(event, NULL, SIGCHLD, NULL, NULL);
4373 if (arg_expose_ports) {
4374 r = watch_rtnl(event, rtnl_socket_pair[0], &exposed, &rtnl);
4378 (void) expose_ports(rtnl, &exposed);
4381 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4383 r = pty_forward_new(event, master, true, !interactive, &forward);
4385 log_error_errno(r, "Failed to create PTY forwarder: %m");
4389 r = sd_event_loop(event);
4391 log_error_errno(r, "Failed to run event loop: %m");
4395 pty_forward_get_last_char(forward, &last_char);
4397 forward = pty_forward_free(forward);
4399 if (!arg_quiet && last_char != '\n')
4402 /* Kill if it is not dead yet anyway */
4403 terminate_machine(pid);
4407 /* Normally redundant, but better safe than sorry */
4410 r = wait_for_container(pid, &container_status);
4414 /* We failed to wait for the container, or the
4415 * container exited abnormally */
4417 else if (r > 0 || container_status == CONTAINER_TERMINATED){
4418 /* The container exited with a non-zero
4419 * status, or with zero status and no reboot
4425 /* CONTAINER_REBOOTED, loop again */
4427 if (arg_keep_unit) {
4428 /* Special handling if we are running as a
4429 * service: instead of simply restarting the
4430 * machine we want to restart the entire
4431 * service, so let's inform systemd about this
4432 * with the special exit code 133. The service
4433 * file uses RestartForceExitStatus=133 so
4434 * that this results in a full nspawn
4435 * restart. This is necessary since we might
4436 * have cgroup parameters set we want to have
4443 flush_ports(&exposed);
4449 "STATUS=Terminating...");
4451 loop_remove(loop_nr, &image_fd);
4456 if (remove_subvol && arg_directory) {
4459 k = btrfs_subvol_remove(arg_directory);
4461 log_warning_errno(k, "Cannot remove subvolume '%s', ignoring: %m", arg_directory);
4467 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
4468 (void) rm_rf(p, false, true, false);
4471 free(arg_directory);
4476 strv_free(arg_setenv);
4477 strv_free(arg_network_interfaces);
4478 strv_free(arg_network_macvlan);
4479 strv_free(arg_network_ipvlan);
4480 strv_free(arg_bind);
4481 strv_free(arg_bind_ro);
4482 strv_free(arg_tmpfs);
4484 flush_ports(&exposed);
4486 while (arg_expose_ports) {
4487 ExposePort *p = arg_expose_ports;
4488 LIST_REMOVE(ports, arg_expose_ports, p);
4492 return r < 0 ? EXIT_FAILURE : ret;