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>
34 #include <sys/capability.h>
36 #include <sys/epoll.h>
38 #include <sys/signalfd.h>
42 #include <sys/socket.h>
44 #include <systemd/sd-daemon.h>
52 #include "cgroup-util.h"
54 #include "path-util.h"
55 #include "loopback-setup.h"
57 #include "dev-setup.h"
59 typedef enum LinkJournal {
66 static char *arg_directory = NULL;
67 static char *arg_user = NULL;
68 static char **arg_controllers = NULL;
69 static char *arg_uuid = NULL;
70 static bool arg_private_network = false;
71 static bool arg_read_only = false;
72 static bool arg_boot = false;
73 static LinkJournal arg_link_journal = LINK_AUTO;
74 static uint64_t arg_retain =
76 (1ULL << CAP_DAC_OVERRIDE) |
77 (1ULL << CAP_DAC_READ_SEARCH) |
78 (1ULL << CAP_FOWNER) |
79 (1ULL << CAP_FSETID) |
80 (1ULL << CAP_IPC_OWNER) |
83 (1ULL << CAP_LINUX_IMMUTABLE) |
84 (1ULL << CAP_NET_BIND_SERVICE) |
85 (1ULL << CAP_NET_BROADCAST) |
86 (1ULL << CAP_NET_RAW) |
87 (1ULL << CAP_SETGID) |
88 (1ULL << CAP_SETFCAP) |
89 (1ULL << CAP_SETPCAP) |
90 (1ULL << CAP_SETUID) |
91 (1ULL << CAP_SYS_ADMIN) |
92 (1ULL << CAP_SYS_CHROOT) |
93 (1ULL << CAP_SYS_NICE) |
94 (1ULL << CAP_SYS_PTRACE) |
95 (1ULL << CAP_SYS_TTY_CONFIG) |
96 (1ULL << CAP_SYS_RESOURCE) |
97 (1ULL << CAP_SYS_BOOT);
99 static int help(void) {
101 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
102 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
103 " -h --help Show this help\n"
104 " -D --directory=NAME Root directory for the container\n"
105 " -b --boot Boot up full system (i.e. invoke init)\n"
106 " -u --user=USER Run the command under specified user or uid\n"
107 " -C --controllers=LIST Put the container in specified comma-separated cgroup hierarchies\n"
108 " --uuid=UUID Set a specific machine UUID for the container\n"
109 " --private-network Disable network in container\n"
110 " --read-only Mount the root directory read-only\n"
111 " --capability=CAP In addition to the default, retain specified capability\n"
112 " --link-journal=MODE Link up guest journal, one of no, auto, guest, host\n"
113 " -j Equivalent to --link-journal=host\n",
114 program_invocation_short_name);
119 static int parse_argv(int argc, char *argv[]) {
122 ARG_PRIVATE_NETWORK = 0x100,
129 static const struct option options[] = {
130 { "help", no_argument, NULL, 'h' },
131 { "directory", required_argument, NULL, 'D' },
132 { "user", required_argument, NULL, 'u' },
133 { "controllers", required_argument, NULL, 'C' },
134 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
135 { "boot", no_argument, NULL, 'b' },
136 { "uuid", required_argument, NULL, ARG_UUID },
137 { "read-only", no_argument, NULL, ARG_READ_ONLY },
138 { "capability", required_argument, NULL, ARG_CAPABILITY },
139 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
148 while ((c = getopt_long(argc, argv, "+hD:u:C:bj", options, NULL)) >= 0) {
158 arg_directory = canonicalize_file_name(optarg);
159 if (!arg_directory) {
160 log_error("Failed to canonicalize root directory.");
168 if (!(arg_user = strdup(optarg))) {
169 log_error("Failed to duplicate user name.");
176 strv_free(arg_controllers);
177 arg_controllers = strv_split(optarg, ",");
178 if (!arg_controllers) {
179 log_error("Failed to split controllers list.");
182 strv_uniq(arg_controllers);
186 case ARG_PRIVATE_NETWORK:
187 arg_private_network = true;
199 arg_read_only = true;
202 case ARG_CAPABILITY: {
206 FOREACH_WORD_SEPARATOR(word, length, optarg, ",", state) {
210 t = strndup(word, length);
214 if (cap_from_name(t, &cap) < 0) {
215 log_error("Failed to parse capability %s.", t);
221 arg_retain |= 1ULL << (uint64_t) cap;
228 arg_link_journal = LINK_GUEST;
231 case ARG_LINK_JOURNAL:
232 if (streq(optarg, "auto"))
233 arg_link_journal = LINK_AUTO;
234 else if (streq(optarg, "no"))
235 arg_link_journal = LINK_NO;
236 else if (streq(optarg, "guest"))
237 arg_link_journal = LINK_GUEST;
238 else if (streq(optarg, "host"))
239 arg_link_journal = LINK_HOST;
241 log_error("Failed to parse link journal mode %s", optarg);
251 log_error("Unknown option code %c", c);
259 static int mount_all(const char *dest) {
261 typedef struct MountPoint {
270 static const MountPoint mount_table[] = {
271 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
272 { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND, true }, /* Bind mount first */
273 { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, true }, /* Then, make it r/o */
274 { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
275 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME, true },
276 { "/dev/pts", "/dev/pts", NULL, NULL, MS_BIND, true },
277 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true },
278 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true },
280 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND, false }, /* Bind mount first */
281 { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false }, /* Then, make it r/o */
288 for (k = 0; k < ELEMENTSOF(mount_table); k++) {
289 char _cleanup_free_ *where = NULL;
292 if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
301 t = path_is_mount_point(where, true);
303 log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
311 /* Skip this entry if it is not a remount. */
312 if (mount_table[k].what && t > 0)
315 mkdir_p_label(where, 0755);
317 if (mount(mount_table[k].what,
320 mount_table[k].flags,
321 mount_table[k].options) < 0 &&
322 mount_table[k].fatal) {
324 log_error("mount(%s) failed: %m", where);
334 static int setup_timezone(const char *dest) {
335 _cleanup_free_ char *where = NULL, *p = NULL, *q = NULL, *check = NULL, *what = NULL;
341 /* Fix the timezone, if possible */
342 r = readlink_malloc("/etc/localtime", &p);
344 log_warning("/etc/localtime is not a symlink, not updating container timezone.");
348 z = path_startswith(p, "../usr/share/zoneinfo/");
350 z = path_startswith(p, "/usr/share/zoneinfo/");
352 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
356 where = strappend(dest, "/etc/localtime");
360 r = readlink_malloc(where, &q);
362 y = path_startswith(q, "../usr/share/zoneinfo/");
364 y = path_startswith(q, "/usr/share/zoneinfo/");
367 /* Already pointing to the right place? Then do nothing .. */
368 if (y && streq(y, z))
372 check = strjoin(dest, "/usr/share/zoneinfo/", z, NULL);
376 if (access(check, F_OK) < 0) {
377 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
381 what = strappend("../usr/share/zoneinfo/", z);
386 if (symlink(what, where) < 0) {
387 log_error("Failed to correct timezone of container: %m");
394 static int setup_resolv_conf(const char *dest) {
399 if (arg_private_network)
402 /* Fix resolv.conf, if possible */
403 where = strappend(dest, "/etc/resolv.conf");
407 /* We don't really care for the results of this really. If it
408 * fails, it fails, but meh... */
409 if (mount("/etc/resolv.conf", where, "bind", MS_BIND, NULL) >= 0)
410 mount("/etc/resolv.conf", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
417 static int setup_boot_id(const char *dest) {
418 char _cleanup_free_ *from = NULL, *to = NULL;
425 /* Generate a new randomized boot ID, so that each boot-up of
426 * the container gets a new one */
428 from = strappend(dest, "/dev/proc-sys-kernel-random-boot-id");
429 to = strappend(dest, "/proc/sys/kernel/random/boot_id");
433 r = sd_id128_randomize(&rnd);
435 log_error("Failed to generate random boot id: %s", strerror(-r));
439 snprintf(as_uuid, sizeof(as_uuid),
440 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
441 SD_ID128_FORMAT_VAL(rnd));
442 char_array_0(as_uuid);
444 r = write_one_line_file(from, as_uuid);
446 log_error("Failed to write boot id: %s", strerror(-r));
450 if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
451 log_error("Failed to bind mount boot id: %m");
454 mount(from, to, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
460 static int copy_devnodes(const char *dest) {
462 static const char devnodes[] =
473 mode_t _cleanup_umask_ u;
479 NULSTR_FOREACH(d, devnodes) {
481 char _cleanup_free_ *from = NULL, *to = NULL;
483 asprintf(&from, "/dev/%s", d);
484 asprintf(&to, "%s/dev/%s", dest, d);
495 if (stat(from, &st) < 0) {
497 if (errno != ENOENT) {
498 log_error("Failed to stat %s: %m", from);
503 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
505 log_error("%s is not a char or block device, cannot copy", from);
509 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
511 log_error("mknod(%s) failed: %m", dest);
520 static int setup_dev_console(const char *dest, const char *console) {
522 char _cleanup_free_ *to = NULL;
524 mode_t _cleanup_umask_ u;
531 if (stat(console, &st) < 0) {
532 log_error("Failed to stat %s: %m", console);
535 } else if (!S_ISCHR(st.st_mode)) {
536 log_error("/dev/console is not a char device");
540 r = chmod_and_chown(console, 0600, 0, 0);
542 log_error("Failed to correct access mode for TTY: %s", strerror(-r));
546 if (asprintf(&to, "%s/dev/console", dest) < 0)
549 /* We need to bind mount the right tty to /dev/console since
550 * ptys can only exist on pts file systems. To have something
551 * to bind mount things on we create a device node first, that
552 * has the right major/minor (note that the major minor
553 * doesn't actually matter here, since we mount it over
556 if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) {
557 log_error("mknod() for /dev/console failed: %m");
561 if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
562 log_error("Bind mount for /dev/console failed: %m");
569 static int setup_kmsg(const char *dest, int kmsg_socket) {
570 char _cleanup_free_ *from = NULL, *to = NULL;
572 mode_t _cleanup_umask_ u;
574 struct cmsghdr cmsghdr;
575 uint8_t buf[CMSG_SPACE(sizeof(int))];
578 struct cmsghdr *cmsg;
581 assert(kmsg_socket >= 0);
585 /* We create the kmsg FIFO as /dev/kmsg, but immediately
586 * delete it after bind mounting it to /proc/kmsg. While FIFOs
587 * on the reading side behave very similar to /proc/kmsg,
588 * their writing side behaves differently from /dev/kmsg in
589 * that writing blocks when nothing is reading. In order to
590 * avoid any problems with containers deadlocking due to this
591 * we simply make /dev/kmsg unavailable to the container. */
592 if (asprintf(&from, "%s/dev/kmsg", dest) < 0 ||
593 asprintf(&to, "%s/proc/kmsg", dest) < 0)
596 if (mkfifo(from, 0600) < 0) {
597 log_error("mkfifo() for /dev/kmsg failed: %m");
601 r = chmod_and_chown(from, 0600, 0, 0);
603 log_error("Failed to correct access mode for /dev/kmsg: %s", strerror(-r));
607 if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
608 log_error("Bind mount for /proc/kmsg failed: %m");
612 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
614 log_error("Failed to open fifo: %m");
621 mh.msg_control = &control;
622 mh.msg_controllen = sizeof(control);
624 cmsg = CMSG_FIRSTHDR(&mh);
625 cmsg->cmsg_level = SOL_SOCKET;
626 cmsg->cmsg_type = SCM_RIGHTS;
627 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
628 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
630 mh.msg_controllen = cmsg->cmsg_len;
632 /* Store away the fd in the socket, so that it stays open as
633 * long as we run the child */
634 k = sendmsg(kmsg_socket, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
635 close_nointr_nofail(fd);
638 log_error("Failed to send FIFO fd: %m");
642 /* And now make the FIFO unavailable as /dev/kmsg... */
647 static int setup_hostname(void) {
651 hn = path_get_file_name(arg_directory);
657 hostname_cleanup(hn);
660 if (sethostname(hn, strlen(hn)) < 0)
669 static int setup_journal(const char *directory) {
670 sd_id128_t machine_id;
671 char _cleanup_free_ *p = NULL, *b = NULL, *q = NULL, *d = NULL;
675 if (arg_link_journal == LINK_NO)
678 p = strappend(directory, "/etc/machine-id");
682 r = read_one_line_file(p, &b);
683 if (r == -ENOENT && arg_link_journal == LINK_AUTO)
686 log_error("Failed to read machine ID from %s: %s", p, strerror(-r));
691 if (isempty(id) && arg_link_journal == LINK_AUTO)
694 /* Verify validity */
695 r = sd_id128_from_string(id, &machine_id);
697 log_error("Failed to parse machine ID from %s: %s", p, strerror(-r));
702 p = strappend("/var/log/journal/", id);
703 q = strjoin(directory, "/var/log/journal/", id, NULL);
707 if (path_is_mount_point(p, false) > 0) {
708 if (arg_link_journal != LINK_AUTO) {
709 log_error("%s: already a mount point, refusing to use for journal", p);
716 if (path_is_mount_point(q, false) > 0) {
717 if (arg_link_journal != LINK_AUTO) {
718 log_error("%s: already a mount point, refusing to use for journal", q);
725 r = readlink_and_make_absolute(p, &d);
727 if ((arg_link_journal == LINK_GUEST ||
728 arg_link_journal == LINK_AUTO) &&
731 r = mkdir_p(q, 0755);
733 log_warning("failed to create directory %s: %m", q);
738 log_error("Failed to remove symlink %s: %m", p);
741 } else if (r == -EINVAL) {
743 if (arg_link_journal == LINK_GUEST &&
746 if (errno == ENOTDIR) {
747 log_error("%s already exists and is neither a symlink nor a directory", p);
750 log_error("Failed to remove %s: %m", p);
754 } else if (r != -ENOENT) {
755 log_error("readlink(%s) failed: %m", p);
759 if (arg_link_journal == LINK_GUEST) {
761 if (symlink(q, p) < 0) {
762 log_error("Failed to symlink %s to %s: %m", q, p);
766 r = mkdir_p(q, 0755);
768 log_warning("failed to create directory %s: %m", q);
772 if (arg_link_journal == LINK_HOST) {
773 r = mkdir_p(p, 0755);
775 log_error("Failed to create %s: %m", p);
779 } else if (access(p, F_OK) < 0)
782 if (dir_is_empty(q) == 0) {
783 log_error("%s not empty.", q);
787 r = mkdir_p(q, 0755);
789 log_error("Failed to create %s: %m", q);
793 if (mount(p, q, "bind", MS_BIND, NULL) < 0) {
794 log_error("Failed to bind mount journal from host into guest: %m");
801 static int drop_capabilities(void) {
802 return capability_bounding_set_drop(~arg_retain, false);
805 static int is_os_tree(const char *path) {
808 /* We use /bin/sh as flag file if something is an OS */
810 if (asprintf(&p, "%s/bin/sh", path) < 0)
816 return r < 0 ? 0 : 1;
819 static int process_pty(int master, sigset_t *mask) {
821 char in_buffer[LINE_MAX], out_buffer[LINE_MAX];
822 size_t in_buffer_full = 0, out_buffer_full = 0;
823 struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
824 bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
825 int ep = -1, signal_fd = -1, r;
827 fd_nonblock(STDIN_FILENO, 1);
828 fd_nonblock(STDOUT_FILENO, 1);
829 fd_nonblock(master, 1);
831 signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC);
833 log_error("signalfd(): %m");
838 ep = epoll_create1(EPOLL_CLOEXEC);
840 log_error("Failed to create epoll: %m");
846 stdin_ev.events = EPOLLIN|EPOLLET;
847 stdin_ev.data.fd = STDIN_FILENO;
850 stdout_ev.events = EPOLLOUT|EPOLLET;
851 stdout_ev.data.fd = STDOUT_FILENO;
854 master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
855 master_ev.data.fd = master;
858 signal_ev.events = EPOLLIN;
859 signal_ev.data.fd = signal_fd;
861 if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0 ||
862 epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0 ||
863 epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
864 epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
865 log_error("Failed to regiser fds in epoll: %m");
871 struct epoll_event ev[16];
875 nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1);
878 if (errno == EINTR || errno == EAGAIN)
881 log_error("epoll_wait(): %m");
888 for (i = 0; i < nfds; i++) {
889 if (ev[i].data.fd == STDIN_FILENO) {
891 if (ev[i].events & (EPOLLIN|EPOLLHUP))
892 stdin_readable = true;
894 } else if (ev[i].data.fd == STDOUT_FILENO) {
896 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
897 stdout_writable = true;
899 } else if (ev[i].data.fd == master) {
901 if (ev[i].events & (EPOLLIN|EPOLLHUP))
902 master_readable = true;
904 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
905 master_writable = true;
907 } else if (ev[i].data.fd == signal_fd) {
908 struct signalfd_siginfo sfsi;
911 n = read(signal_fd, &sfsi, sizeof(sfsi));
912 if (n != sizeof(sfsi)) {
915 log_error("Failed to read from signalfd: invalid block size");
920 if (errno != EINTR && errno != EAGAIN) {
921 log_error("Failed to read from signalfd: %m");
927 if (sfsi.ssi_signo == SIGWINCH) {
930 /* The window size changed, let's forward that. */
931 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
932 ioctl(master, TIOCSWINSZ, &ws);
941 while ((stdin_readable && in_buffer_full <= 0) ||
942 (master_writable && in_buffer_full > 0) ||
943 (master_readable && out_buffer_full <= 0) ||
944 (stdout_writable && out_buffer_full > 0)) {
946 if (stdin_readable && in_buffer_full < LINE_MAX) {
948 k = read(STDIN_FILENO, in_buffer + in_buffer_full, LINE_MAX - in_buffer_full);
951 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
952 stdin_readable = false;
954 log_error("read(): %m");
959 in_buffer_full += (size_t) k;
962 if (master_writable && in_buffer_full > 0) {
964 k = write(master, in_buffer, in_buffer_full);
967 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
968 master_writable = false;
970 log_error("write(): %m");
976 assert(in_buffer_full >= (size_t) k);
977 memmove(in_buffer, in_buffer + k, in_buffer_full - k);
982 if (master_readable && out_buffer_full < LINE_MAX) {
984 k = read(master, out_buffer + out_buffer_full, LINE_MAX - out_buffer_full);
987 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
988 master_readable = false;
990 log_error("read(): %m");
995 out_buffer_full += (size_t) k;
998 if (stdout_writable && out_buffer_full > 0) {
1000 k = write(STDOUT_FILENO, out_buffer, out_buffer_full);
1003 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
1004 stdout_writable = false;
1006 log_error("write(): %m");
1012 assert(out_buffer_full >= (size_t) k);
1013 memmove(out_buffer, out_buffer + k, out_buffer_full - k);
1014 out_buffer_full -= k;
1022 close_nointr_nofail(ep);
1025 close_nointr_nofail(signal_fd);
1030 int main(int argc, char *argv[]) {
1032 int r = EXIT_FAILURE, k;
1033 char *oldcg = NULL, *newcg = NULL;
1034 char **controller = NULL;
1036 const char *console = NULL;
1037 struct termios saved_attr, raw_attr;
1039 bool saved_attr_valid = false;
1041 int kmsg_socket_pair[2] = { -1, -1 };
1043 log_parse_environment();
1046 r = parse_argv(argc, argv);
1050 if (arg_directory) {
1053 p = path_make_absolute_cwd(arg_directory);
1054 free(arg_directory);
1057 arg_directory = get_current_dir_name();
1059 if (!arg_directory) {
1060 log_error("Failed to determine path");
1064 path_kill_slashes(arg_directory);
1066 if (geteuid() != 0) {
1067 log_error("Need to be root.");
1071 if (sd_booted() <= 0) {
1072 log_error("Not running on a systemd system.");
1076 if (path_equal(arg_directory, "/")) {
1077 log_error("Spawning container on root directory not supported.");
1081 if (is_os_tree(arg_directory) <= 0) {
1082 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
1086 k = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &oldcg);
1088 log_error("Failed to determine current cgroup: %s", strerror(-k));
1092 if (asprintf(&newcg, "%s/nspawn-%lu", oldcg, (unsigned long) getpid()) < 0) {
1093 log_error("Failed to allocate cgroup path.");
1097 k = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, newcg, 0);
1099 log_error("Failed to create cgroup: %s", strerror(-k));
1103 STRV_FOREACH(controller, arg_controllers) {
1104 k = cg_create_and_attach(*controller, newcg, 0);
1106 log_warning("Failed to create cgroup in controller %s: %s", *controller, strerror(-k));
1109 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
1111 log_error("Failed to acquire pseudo tty: %m");
1115 console = ptsname(master);
1117 log_error("Failed to determine tty name: %m");
1121 log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
1123 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
1124 ioctl(master, TIOCSWINSZ, &ws);
1126 if (unlockpt(master) < 0) {
1127 log_error("Failed to unlock tty: %m");
1131 if (tcgetattr(STDIN_FILENO, &saved_attr) < 0) {
1132 log_error("Failed to get terminal attributes: %m");
1136 saved_attr_valid = true;
1138 raw_attr = saved_attr;
1139 cfmakeraw(&raw_attr);
1140 raw_attr.c_lflag &= ~ECHO;
1142 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_NONBLOCK|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
1143 log_error("Failed to create kmsg socket pair");
1147 assert_se(sigemptyset(&mask) == 0);
1148 sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
1149 assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
1154 if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
1155 log_error("Failed to set terminal attributes: %m");
1159 pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS|(arg_private_network ? CLONE_NEWNET : 0), NULL);
1161 if (errno == EINVAL)
1162 log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
1164 log_error("clone() failed: %m");
1172 const char *home = NULL;
1173 uid_t uid = (uid_t) -1;
1174 gid_t gid = (gid_t) -1;
1175 const char *envp[] = {
1176 "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
1177 "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
1182 NULL, /* container_uuid */
1186 envp[2] = strv_find_prefix(environ, "TERM=");
1188 close_nointr_nofail(master);
1190 close_nointr(STDIN_FILENO);
1191 close_nointr(STDOUT_FILENO);
1192 close_nointr(STDERR_FILENO);
1194 close_all_fds(&kmsg_socket_pair[1], 1);
1196 reset_all_signal_handlers();
1198 assert_se(sigemptyset(&mask) == 0);
1199 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1201 if (open_terminal(console, O_RDWR) != STDIN_FILENO ||
1202 dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
1203 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
1207 log_error("setsid() failed: %m");
1211 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0) {
1212 log_error("PR_SET_PDEATHSIG failed: %m");
1216 /* Mark everything as slave, so that we still
1217 * receive mounts from the real root, but don't
1218 * propagate mounts to the real root. */
1219 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
1220 log_error("MS_SLAVE|MS_REC failed: %m");
1224 /* Turn directory into bind mount */
1225 if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REC, NULL) < 0) {
1226 log_error("Failed to make bind mount.");
1231 if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY|MS_REC, NULL) < 0) {
1232 log_error("Failed to make read-only.");
1236 if (mount_all(arg_directory) < 0)
1239 if (copy_devnodes(arg_directory) < 0)
1242 dev_setup(arg_directory);
1244 if (setup_dev_console(arg_directory, console) < 0)
1247 if (setup_kmsg(arg_directory, kmsg_socket_pair[1]) < 0)
1250 close_nointr_nofail(kmsg_socket_pair[1]);
1252 if (setup_boot_id(arg_directory) < 0)
1255 if (setup_timezone(arg_directory) < 0)
1258 if (setup_resolv_conf(arg_directory) < 0)
1261 if (setup_journal(arg_directory) < 0)
1264 if (chdir(arg_directory) < 0) {
1265 log_error("chdir(%s) failed: %m", arg_directory);
1269 if (mount(arg_directory, "/", NULL, MS_MOVE, NULL) < 0) {
1270 log_error("mount(MS_MOVE) failed: %m");
1274 if (chroot(".") < 0) {
1275 log_error("chroot() failed: %m");
1279 if (chdir("/") < 0) {
1280 log_error("chdir() failed: %m");
1288 if (drop_capabilities() < 0) {
1289 log_error("drop_capabilities() failed: %m");
1295 /* Note that this resolves user names
1296 * inside the container, and hence
1297 * accesses the NSS modules from the
1298 * container and not the host. This is
1301 if (get_user_creds((const char**)&arg_user, &uid, &gid, &home, NULL) < 0) {
1302 log_error("get_user_creds() failed: %m");
1306 if (mkdir_parents_label(home, 0775) < 0) {
1307 log_error("mkdir_parents_label() failed: %m");
1311 if (mkdir_safe_label(home, 0775, uid, gid) < 0) {
1312 log_error("mkdir_safe_label() failed: %m");
1316 if (initgroups((const char*)arg_user, gid) < 0) {
1317 log_error("initgroups() failed: %m");
1321 if (setresgid(gid, gid, gid) < 0) {
1322 log_error("setregid() failed: %m");
1326 if (setresuid(uid, uid, uid) < 0) {
1327 log_error("setreuid() failed: %m");
1332 if ((asprintf((char**)(envp + 3), "HOME=%s", home ? home: "/root") < 0) ||
1333 (asprintf((char**)(envp + 4), "USER=%s", arg_user ? arg_user : "root") < 0) ||
1334 (asprintf((char**)(envp + 5), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)) {
1340 if (asprintf((char**)(envp + 6), "container_uuid=%s", arg_uuid) < 0) {
1352 /* Automatically search for the init system */
1354 l = 1 + argc - optind;
1355 a = newa(char*, l + 1);
1356 memcpy(a + 1, argv + optind, l * sizeof(char*));
1358 a[0] = (char*) "/usr/lib/systemd/systemd";
1359 execve(a[0], a, (char**) envp);
1361 a[0] = (char*) "/lib/systemd/systemd";
1362 execve(a[0], a, (char**) envp);
1364 a[0] = (char*) "/sbin/init";
1365 execve(a[0], a, (char**) envp);
1366 } else if (argc > optind)
1367 execvpe(argv[optind], argv + optind, (char**) envp);
1369 chdir(home ? home : "/root");
1370 execle("/bin/bash", "-bash", NULL, (char**) envp);
1373 log_error("execv() failed: %m");
1376 _exit(EXIT_FAILURE);
1379 if (process_pty(master, &mask) < 0)
1383 if (saved_attr_valid)
1384 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
1386 r = wait_for_terminate(pid, &status);
1392 if (status.si_code == CLD_EXITED) {
1393 if (status.si_status != 0) {
1394 log_error("Container failed with error code %i.", status.si_status);
1395 r = status.si_status;
1399 log_debug("Container exited successfully.");
1401 } else if (status.si_code == CLD_KILLED &&
1402 status.si_status == SIGINT) {
1403 log_info("Container has been shut down.");
1406 } else if (status.si_code == CLD_KILLED &&
1407 status.si_status == SIGHUP) {
1408 log_info("Container is being rebooted.");
1410 } else if (status.si_code == CLD_KILLED ||
1411 status.si_code == CLD_DUMPED) {
1413 log_error("Container terminated by signal %s.", signal_to_string(status.si_status));
1417 log_error("Container failed due to unknown reason.");
1424 if (saved_attr_valid)
1425 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
1428 close_nointr_nofail(master);
1430 close_pipe(kmsg_socket_pair);
1433 cg_attach(SYSTEMD_CGROUP_CONTROLLER, oldcg, 0);
1436 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
1438 free(arg_directory);
1439 strv_free(arg_controllers);