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 General Public License as published by
10 the Free Software Foundation; either version 2 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 General Public License for more details.
18 You should have received a copy of the GNU 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>
44 #include "cgroup-util.h"
45 #include "sd-daemon.h"
48 static char *arg_directory = NULL;
49 static char *arg_user = NULL;
51 static int help(void) {
53 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
54 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
55 " -h --help Show this help\n"
56 " -D --directory=NAME Root directory for the container\n"
57 " -u --user=USER Run the command under specified user or uid\n",
58 program_invocation_short_name);
63 static int parse_argv(int argc, char *argv[]) {
65 static const struct option options[] = {
66 { "help", no_argument, NULL, 'h' },
67 { "directory", required_argument, NULL, 'D' },
68 { "user", optional_argument, NULL, 'u' },
77 while ((c = getopt_long(argc, argv, "+hD:u:", options, NULL)) >= 0) {
87 if (!(arg_directory = strdup(optarg))) {
88 log_error("Failed to duplicate root directory.");
96 if (!(arg_user = strdup(optarg))) {
97 log_error("Failed to duplicate user name.");
107 log_error("Unknown option code %c", c);
115 static int mount_all(const char *dest) {
117 typedef struct MountPoint {
126 static const MountPoint mount_table[] = {
127 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
128 { "/proc/sys", "/proc/sys", "bind", NULL, MS_BIND, true }, /* Bind mount first */
129 { "/proc/sys", "/proc/sys", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, true }, /* Then, make it r/o */
130 { "/sys", "/sys", "bind", NULL, MS_BIND, true }, /* Bind mount first */
131 { "/sys", "/sys", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, true }, /* Then, make it r/o */
132 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID, true },
133 { "/dev/pts", "/dev/pts", "bind", NULL, MS_BIND, true },
134 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV, true },
136 { "/selinux", "/selinux", "bind", NULL, MS_BIND, false }, /* Bind mount first */
137 { "/selinux", "/selinux", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false }, /* Then, make it r/o */
145 for (k = 0; k < ELEMENTSOF(mount_table); k++) {
148 if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
149 log_error("Out of memory");
157 if ((t = path_is_mount_point(where)) < 0) {
158 log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
167 mkdir_p(where, 0755);
169 if (mount(mount_table[k].what,
172 mount_table[k].flags,
173 mount_table[k].options) < 0 &&
174 mount_table[k].fatal) {
176 log_error("mount(%s) failed: %m", where);
185 /* Fix the timezone, if possible */
186 if (asprintf(&where, "%s/%s", dest, "/etc/localtime") >= 0) {
187 mount("/etc/localtime", where, "bind", MS_BIND, NULL);
188 mount("/etc/localtime", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
195 static int copy_devnodes(const char *dest, const char *console) {
197 static const char devnodes[] =
212 char *from = NULL, *to = NULL;
219 NULSTR_FOREACH(d, devnodes) {
222 asprintf(&from, "/dev/%s", d);
223 asprintf(&to, "%s/dev/%s", dest, d);
226 log_error("Failed to allocate devnode path");
239 if (stat(from, &st) < 0) {
241 if (errno != ENOENT) {
242 log_error("Failed to stat %s: %m", from);
247 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
249 log_error("%s is not a char or block device, cannot copy.", from);
253 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
255 log_error("mknod(%s) failed: %m", dest);
264 if (stat(console, &st) < 0) {
266 log_error("Failed to stat %s: %m", console);
272 } else if (!S_ISCHR(st.st_mode)) {
274 log_error("/dev/console is not a char device.");
281 if (asprintf(&to, "%s/dev/console", dest) < 0) {
283 log_error("Out of memory");
290 /* We need to bind mount the right tty to /dev/console since
291 * ptys can only exist on pts file systems. To have something
292 * to bind mount things on we create a device node first, that
293 * has the right major/minor (note that the major minor
294 * doesn't actually matter here, since we mount it over
297 if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0)
298 log_error("mknod for /dev/console failed: %m");
300 if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
301 log_error("bind mount for /dev/console failed: %m");
309 if ((k = chmod_and_chown(console, 0600, 0, 0)) < 0) {
310 log_error("Failed to correct access mode for TTY: %s", strerror(-k));
323 static int drop_capabilities(void) {
324 static const unsigned long retain[] = {
334 CAP_NET_BIND_SERVICE,
350 for (l = 0; l <= MAX(63LU, (unsigned long) CAP_LAST_CAP); l++) {
353 for (i = 0; i < ELEMENTSOF(retain); i++)
357 if (i < ELEMENTSOF(retain))
360 if (prctl(PR_CAPBSET_DROP, l) < 0) {
362 /* If this capability is not known, EINVAL
363 * will be returned, let's ignore this. */
367 log_error("PR_CAPBSET_DROP failed: %m");
375 static int is_os_tree(const char *path) {
378 /* We use /bin/sh as flag file if something is an OS */
380 if (asprintf(&p, "%s/bin/sh", path) < 0)
386 return r < 0 ? 0 : 1;
389 #define BUFFER_SIZE 1024
391 static int process_pty(int master, sigset_t *mask) {
393 char in_buffer[BUFFER_SIZE], out_buffer[BUFFER_SIZE];
394 size_t in_buffer_full = 0, out_buffer_full = 0;
395 struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
396 bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
397 int ep = -1, signal_fd = -1, r;
399 fd_nonblock(STDIN_FILENO, 1);
400 fd_nonblock(STDOUT_FILENO, 1);
401 fd_nonblock(master, 1);
403 if ((signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
404 log_error("signalfd(): %m");
409 if ((ep = epoll_create1(EPOLL_CLOEXEC)) < 0) {
410 log_error("Failed to create epoll: %m");
416 stdin_ev.events = EPOLLIN|EPOLLET;
417 stdin_ev.data.fd = STDIN_FILENO;
420 stdout_ev.events = EPOLLOUT|EPOLLET;
421 stdout_ev.data.fd = STDOUT_FILENO;
424 master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
425 master_ev.data.fd = master;
428 signal_ev.events = EPOLLIN;
429 signal_ev.data.fd = signal_fd;
431 if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0 ||
432 epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0 ||
433 epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
434 epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
435 log_error("Failed to regiser fds in epoll: %m");
441 struct epoll_event ev[16];
445 if ((nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1)) < 0) {
447 if (errno == EINTR || errno == EAGAIN)
450 log_error("epoll_wait(): %m");
457 for (i = 0; i < nfds; i++) {
458 if (ev[i].data.fd == STDIN_FILENO) {
460 if (ev[i].events & (EPOLLIN|EPOLLHUP))
461 stdin_readable = true;
463 } else if (ev[i].data.fd == STDOUT_FILENO) {
465 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
466 stdout_writable = true;
468 } else if (ev[i].data.fd == master) {
470 if (ev[i].events & (EPOLLIN|EPOLLHUP))
471 master_readable = true;
473 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
474 master_writable = true;
476 } else if (ev[i].data.fd == signal_fd) {
477 struct signalfd_siginfo sfsi;
480 if ((n = read(signal_fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) {
483 log_error("Failed to read from signalfd: invalid block size");
488 if (errno != EINTR && errno != EAGAIN) {
489 log_error("Failed to read from signalfd: %m");
495 if (sfsi.ssi_signo == SIGWINCH) {
498 /* The window size changed, let's forward that. */
499 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
500 ioctl(master, TIOCSWINSZ, &ws);
509 while ((stdin_readable && in_buffer_full <= 0) ||
510 (master_writable && in_buffer_full > 0) ||
511 (master_readable && out_buffer_full <= 0) ||
512 (stdout_writable && out_buffer_full > 0)) {
514 if (stdin_readable && in_buffer_full < BUFFER_SIZE) {
516 if ((k = read(STDIN_FILENO, in_buffer + in_buffer_full, BUFFER_SIZE - in_buffer_full)) < 0) {
518 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
519 stdin_readable = false;
521 log_error("read(): %m");
526 in_buffer_full += (size_t) k;
529 if (master_writable && in_buffer_full > 0) {
531 if ((k = write(master, in_buffer, in_buffer_full)) < 0) {
533 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
534 master_writable = false;
536 log_error("write(): %m");
542 assert(in_buffer_full >= (size_t) k);
543 memmove(in_buffer, in_buffer + k, in_buffer_full - k);
548 if (master_readable && out_buffer_full < BUFFER_SIZE) {
550 if ((k = read(master, out_buffer + out_buffer_full, BUFFER_SIZE - out_buffer_full)) < 0) {
552 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
553 master_readable = false;
555 log_error("read(): %m");
560 out_buffer_full += (size_t) k;
563 if (stdout_writable && out_buffer_full > 0) {
565 if ((k = write(STDOUT_FILENO, out_buffer, out_buffer_full)) < 0) {
567 if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
568 stdout_writable = false;
570 log_error("write(): %m");
576 assert(out_buffer_full >= (size_t) k);
577 memmove(out_buffer, out_buffer + k, out_buffer_full - k);
578 out_buffer_full -= k;
586 close_nointr_nofail(ep);
589 close_nointr_nofail(signal_fd);
594 int main(int argc, char *argv[]) {
596 int r = EXIT_FAILURE, k;
597 char *oldcg = NULL, *newcg = NULL;
599 const char *console = NULL;
600 struct termios saved_attr, raw_attr;
602 bool saved_attr_valid = false;
605 log_parse_environment();
608 if ((r = parse_argv(argc, argv)) <= 0)
614 p = path_make_absolute_cwd(arg_directory);
618 arg_directory = get_current_dir_name();
620 if (!arg_directory) {
621 log_error("Failed to determine path");
625 path_kill_slashes(arg_directory);
627 if (geteuid() != 0) {
628 log_error("Need to be root.");
632 if (sd_booted() <= 0) {
633 log_error("Not running on a systemd system.");
637 if (path_equal(arg_directory, "/")) {
638 log_error("Spawning container on root directory not supported.");
642 if (is_os_tree(arg_directory) <= 0) {
643 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
647 if ((k = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &oldcg)) < 0) {
648 log_error("Failed to determine current cgroup: %s", strerror(-k));
652 if (asprintf(&newcg, "%s/nspawn-%lu", oldcg, (unsigned long) getpid()) < 0) {
653 log_error("Failed to allocate cgroup path.");
657 if ((k = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, newcg, 0)) < 0) {
658 log_error("Failed to create cgroup: %s", strerror(-k));
662 if ((master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY)) < 0) {
663 log_error("Failed to acquire pseudo tty: %m");
667 if (!(console = ptsname(master))) {
668 log_error("Failed to determine tty name: %m");
672 log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
674 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
675 ioctl(master, TIOCSWINSZ, &ws);
677 if (unlockpt(master) < 0) {
678 log_error("Failed to unlock tty: %m");
682 if (tcgetattr(STDIN_FILENO, &saved_attr) < 0) {
683 log_error("Failed to get terminal attributes: %m");
687 saved_attr_valid = true;
689 raw_attr = saved_attr;
690 cfmakeraw(&raw_attr);
691 raw_attr.c_lflag &= ~ECHO;
693 if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
694 log_error("Failed to set terminal attributes: %m");
698 assert_se(sigemptyset(&mask) == 0);
699 sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
700 assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
702 if ((pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS, NULL)) < 0) {
703 log_error("clone() failed: %m");
711 const char *home = NULL;
712 uid_t uid = (uid_t) -1;
713 gid_t gid = (gid_t) -1;
714 const char *envp[] = {
715 "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
723 envp[1] = strv_find_prefix(environ, "TERM=");
725 close_nointr_nofail(master);
727 close_nointr(STDIN_FILENO);
728 close_nointr(STDOUT_FILENO);
729 close_nointr(STDERR_FILENO);
731 close_all_fds(NULL, 0);
733 reset_all_signal_handlers();
735 assert_se(sigemptyset(&mask) == 0);
736 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
741 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
744 /* Mark / as private, in case somebody marked it shared */
745 if (mount(NULL, "/", NULL, MS_PRIVATE|MS_REC, NULL) < 0)
748 if (mount_all(arg_directory) < 0)
751 if (copy_devnodes(arg_directory, console) < 0)
754 if (chdir(arg_directory) < 0) {
755 log_error("chdir(%s) failed: %m", arg_directory);
759 if (open_terminal("dev/console", O_RDWR) != STDIN_FILENO ||
760 dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
761 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
764 if (mount(arg_directory, "/", "bind", MS_BIND|MS_MOVE, NULL) < 0) {
765 log_error("mount(MS_MOVE) failed: %m");
769 if (chroot(".") < 0) {
770 log_error("chroot() failed: %m");
774 if (chdir("/") < 0) {
775 log_error("chdir() failed: %m");
781 if (drop_capabilities() < 0)
786 if (get_user_creds((const char**)&arg_user, &uid, &gid, &home) < 0) {
787 log_error("get_user_creds() failed: %m");
791 if (mkdir_parents(home, 0775) < 0) {
792 log_error("mkdir_parents() failed: %m");
796 if (safe_mkdir(home, 0775, uid, gid) < 0) {
797 log_error("safe_mkdir() failed: %m");
801 if (initgroups((const char*)arg_user, gid) < 0) {
802 log_error("initgroups() failed: %m");
806 if (setresgid(gid, gid, gid) < 0) {
807 log_error("setregid() failed: %m");
811 if (setresuid(uid, uid, uid) < 0) {
812 log_error("setreuid() failed: %m");
817 if ((asprintf((char**)(envp + 2), "HOME=%s", home? home: "/root") < 0) ||
818 (asprintf((char**)(envp + 3), "USER=%s", arg_user? arg_user : "root") < 0) ||
819 (asprintf((char**)(envp + 4), "LOGNAME=%s", arg_user? arg_user : "root") < 0)) {
820 log_error("Out of memory");
824 if ((hn = file_name_from_path(arg_directory)))
825 sethostname(hn, strlen(hn));
828 execvpe(argv[optind], argv + optind, (char**) envp);
830 chdir(home ? home : "/root");
831 execle("/bin/bash", "-bash", NULL, (char**) envp);
834 log_error("execv() failed: %m");
840 if (process_pty(master, &mask) < 0)
843 if (saved_attr_valid) {
844 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
845 saved_attr_valid = false;
848 r = wait_for_terminate_and_warn(argc > optind ? argv[optind] : "bash", pid);
854 if (saved_attr_valid)
855 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
858 close_nointr_nofail(master);
861 cg_attach(SYSTEMD_CGROUP_CONTROLLER, oldcg, 0);
864 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);