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>
43 #include "cgroup-util.h"
44 #include "sd-daemon.h"
47 static char *arg_directory = NULL;
49 static int help(void) {
51 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
52 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
53 " -h --help Show this help\n"
54 " -D --directory=NAME Root directory for the container\n",
55 program_invocation_short_name);
60 static int parse_argv(int argc, char *argv[]) {
62 static const struct option options[] = {
63 { "help", no_argument, NULL, 'h' },
64 { "directory", required_argument, NULL, 'D' },
73 while ((c = getopt_long(argc, argv, "+hD:", options, NULL)) >= 0) {
83 if (!(arg_directory = strdup(optarg))) {
84 log_error("Failed to duplicate root directory.");
94 log_error("Unknown option code %c", c);
102 static int mount_all(const char *dest) {
104 typedef struct MountPoint {
113 static const MountPoint mount_table[] = {
114 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
115 { "/proc/sys", "/proc/sys", "bind", NULL, MS_BIND, true }, /* Bind mount first */
116 { "/proc/sys", "/proc/sys", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, true }, /* Then, make it r/o */
117 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, true },
118 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID, true },
119 { "/dev/pts", "/dev/pts", "bind", NULL, MS_BIND, true },
120 { "tmpfs", "/dev/.run", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
122 { "selinux", "/selinux", "selinuxfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, false },
129 for (k = 0; k < ELEMENTSOF(mount_table); k++) {
133 if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
134 log_error("Out of memory");
142 if ((t = path_is_mount_point(where)) < 0) {
143 log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
152 mkdir_p(where, 0755);
154 if (mount(mount_table[k].what,
157 mount_table[k].flags,
158 mount_table[k].options) < 0 &&
159 mount_table[k].fatal) {
161 log_error("mount(%s) failed: %m", where);
173 static int copy_devnodes(const char *dest, const char *console) {
175 static const char devnodes[] =
190 char *from = NULL, *to = NULL;
197 NULSTR_FOREACH(d, devnodes) {
200 asprintf(&from, "/dev/%s", d);
201 asprintf(&to, "%s/dev/%s", dest, d);
204 log_error("Failed to allocate devnode path");
217 if (stat(from, &st) < 0) {
219 if (errno != ENOENT) {
220 log_error("Failed to stat %s: %m", from);
225 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
227 log_error("%s is not a char or block device, cannot copy.", from);
231 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
233 log_error("mknod(%s) failed: %m", dest);
242 if (stat(console, &st) < 0) {
244 log_error("Failed to stat %s: %m", console);
250 } else if (!S_ISCHR(st.st_mode)) {
252 log_error("/dev/console is not a char device.");
259 if (asprintf(&to, "%s/dev/console", dest) < 0) {
261 log_error("Out of memory");
268 /* We need to bind mount the right tty to /dev/console since
269 * ptys can only exist on pts file systems. To have something
270 * to bind mount things on we create a device node first, that
271 * has the right major/minor (note that the major minor
272 * doesn't actually matter here, since we mount it over
275 if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0)
276 log_error("mknod for /dev/console failed: %m");
278 if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
279 log_error("bind mount for /dev/console failed: %m");
287 if ((k = chmod_and_chown(console, 0600, 0, 0)) < 0) {
288 log_error("Failed to correct access mode for TTY: %s", strerror(-k));
301 static int drop_capabilities(void) {
302 static const unsigned long retain[] = {
312 CAP_NET_BIND_SERVICE,
328 for (l = 0; l <= MAX(63LU, (unsigned long) CAP_LAST_CAP); l ++) {
331 for (i = 0; i < ELEMENTSOF(retain); i++)
335 if (i < ELEMENTSOF(retain))
338 if (prctl(PR_CAPBSET_DROP, l) < 0) {
340 /* If this capability is not known, EINVAL
341 * will be returned, let's ignore this. */
345 log_error("PR_CAPBSET_DROP failed: %m");
353 static int is_os_tree(const char *path) {
356 /* We use /bin/sh as flag file if something is an OS */
358 if (asprintf(&p, "%s/bin/sh", path) < 0)
364 return r < 0 ? 0 : 1;
367 #define BUFFER_SIZE 1024
369 static int process_pty(int master, sigset_t *mask) {
370 char in_buffer[BUFFER_SIZE], out_buffer[BUFFER_SIZE];
371 size_t in_buffer_full = 0, out_buffer_full = 0;
372 struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
373 bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
374 bool stdin_rhup = false, stdout_whup = false, master_rhup = false, master_whup = false;
375 int ep = -1, signal_fd = -1, r;
377 fd_nonblock(STDIN_FILENO, 1);
378 fd_nonblock(STDOUT_FILENO, 1);
379 fd_nonblock(master, 1);
381 if ((signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
382 log_error("signalfd(): %m");
387 if ((ep = epoll_create1(EPOLL_CLOEXEC)) < 0) {
388 log_error("Failed to create epoll: %m");
394 stdin_ev.events = EPOLLIN|EPOLLET;
395 stdin_ev.data.fd = STDIN_FILENO;
398 stdout_ev.events = EPOLLOUT|EPOLLET;
399 stdout_ev.data.fd = STDOUT_FILENO;
402 master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
403 master_ev.data.fd = master;
406 signal_ev.events = EPOLLIN;
407 signal_ev.data.fd = signal_fd;
409 if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0 ||
410 epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0 ||
411 epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
412 epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
413 log_error("Failed to regiser fds in epoll: %m");
419 struct epoll_event ev[16];
423 if ((nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1)) < 0) {
425 if (errno == EINTR || errno == EAGAIN)
428 log_error("epoll_wait(): %m");
435 for (i = 0; i < nfds; i++) {
436 if (ev[i].data.fd == STDIN_FILENO) {
438 if (!stdin_rhup && (ev[i].events & (EPOLLHUP|EPOLLIN)))
439 stdin_readable = true;
441 } else if (ev[i].data.fd == STDOUT_FILENO) {
443 if (ev[i].events & EPOLLHUP) {
444 stdout_writable = false;
448 if (!stdout_whup && (ev[i].events & EPOLLOUT))
449 stdout_writable = true;
451 } else if (ev[i].data.fd == master) {
453 /* We don't connect EPOLLHUP to
454 * master_whup here, since EPOLLHUP
455 * can happen when noone has opened
458 if (!master_rhup && (ev[i].events & (EPOLLHUP|EPOLLIN)))
459 master_readable = true;
461 if (!master_whup && (ev[i].events & EPOLLOUT))
462 master_writable = true;
464 } else if (ev[i].data.fd == signal_fd) {
465 struct signalfd_siginfo sfsi;
468 if ((n = read(signal_fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) {
475 if (errno != EINTR && errno != EAGAIN) {
481 if (sfsi.ssi_signo == SIGWINCH) {
484 /* The window size changed, let's forward that. */
486 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
487 ioctl(master, TIOCSWINSZ, &ws);
496 while ((stdin_readable && in_buffer_full <= 0) ||
497 (master_writable && in_buffer_full > 0) ||
498 (master_readable && out_buffer_full <= 0) ||
499 (stdout_writable && out_buffer_full > 0)) {
501 if (stdin_readable && in_buffer_full < BUFFER_SIZE) {
503 if ((k = read(STDIN_FILENO, in_buffer + in_buffer_full, BUFFER_SIZE - in_buffer_full)) < 0) {
506 stdin_readable = false;
507 else if (errno == EPIPE || errno == ECONNRESET || errno == EIO)
510 log_error("read(): %m");
514 in_buffer_full += (size_t) k;
518 stdin_readable = false;
519 shutdown(STDIN_FILENO, SHUT_RD);
523 if (master_writable && in_buffer_full > 0) {
525 if ((k = write(master, in_buffer, in_buffer_full)) < 0) {
528 master_writable = false;
529 else if (errno == EPIPE || errno == ECONNRESET || errno == EIO) {
531 master_writable = false;
533 log_error("write(): %m");
538 assert(in_buffer_full >= (size_t) k);
539 memmove(in_buffer, in_buffer + k, in_buffer_full - k);
544 if (master_readable && out_buffer_full < BUFFER_SIZE) {
546 if ((k = read(master, out_buffer + out_buffer_full, BUFFER_SIZE - out_buffer_full)) < 0) {
549 master_readable = false;
550 else if (errno == EPIPE || errno == ECONNRESET || errno == EIO)
553 log_error("read(): %m");
557 out_buffer_full += (size_t) k;
561 master_readable = false;
565 if (stdout_writable && out_buffer_full > 0) {
567 if ((k = write(STDOUT_FILENO, out_buffer, out_buffer_full)) < 0) {
570 stdout_writable = false;
571 else if (errno == EPIPE || errno == ECONNRESET || errno == EIO) {
573 stdout_writable = false;
575 log_error("write(): %m");
580 assert(out_buffer_full >= (size_t) k);
581 memmove(out_buffer, out_buffer + k, out_buffer_full - k);
582 out_buffer_full -= k;
587 if (stdin_rhup && in_buffer_full <= 0 && !master_whup) {
589 master_writable = false;
592 if (master_rhup && out_buffer_full <= 0 && !stdout_whup) {
594 stdout_writable = false;
595 shutdown(STDOUT_FILENO, SHUT_WR);
598 } while (!stdout_whup || !master_whup);
602 close_nointr_nofail(ep);
605 close_nointr_nofail(signal_fd);
610 int main(int argc, char *argv[]) {
612 int r = EXIT_FAILURE, k;
613 char *oldcg = NULL, *newcg = NULL;
615 const char *console = NULL;
616 struct termios saved_attr, raw_attr;
618 bool saved_attr_valid = false;
621 log_parse_environment();
624 if ((r = parse_argv(argc, argv)) <= 0)
630 p = path_make_absolute_cwd(arg_directory);
634 arg_directory = get_current_dir_name();
636 if (!arg_directory) {
637 log_error("Failed to determine path");
641 path_kill_slashes(arg_directory);
643 if (geteuid() != 0) {
644 log_error("Need to be root.");
648 if (sd_booted() <= 0) {
649 log_error("Not running on a systemd system.");
653 if (path_equal(arg_directory, "/")) {
654 log_error("Spawning container on root directory not supported.");
658 if (is_os_tree(arg_directory) <= 0) {
659 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
663 if ((k = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &oldcg)) < 0) {
664 log_error("Failed to determine current cgroup: %s", strerror(-k));
668 if (asprintf(&newcg, "%s/nspawn-%lu", oldcg, (unsigned long) getpid()) < 0) {
669 log_error("Failed to allocate cgroup path.");
673 if ((k = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, newcg, 0)) < 0) {
674 log_error("Failed to create cgroup: %s", strerror(-k));
678 if ((master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY)) < 0) {
679 log_error("Failed to acquire pseudo tty: %m");
683 if (!(console = ptsname(master))) {
684 log_error("Failed to determine tty name: %m");
688 log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
690 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
691 ioctl(master, TIOCSWINSZ, &ws);
693 if (unlockpt(master) < 0) {
694 log_error("Failed to unlock tty: %m");
698 if (tcgetattr(STDIN_FILENO, &saved_attr) < 0) {
699 log_error("Failed to get terminal attributes: %m");
703 saved_attr_valid = true;
705 raw_attr = saved_attr;
706 cfmakeraw(&raw_attr);
707 raw_attr.c_lflag &= ~ECHO;
709 if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
710 log_error("Failed to set terminal attributes: %m");
714 assert_se(sigemptyset(&mask) == 0);
715 sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
716 assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
718 if ((pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS, NULL)) < 0) {
719 log_error("clone() failed: %m");
727 const char *envp[] = {
729 "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
734 envp[2] = strv_find_prefix(environ, "TERM=");
736 close_nointr_nofail(master);
738 close_nointr(STDIN_FILENO);
739 close_nointr(STDOUT_FILENO);
740 close_nointr(STDERR_FILENO);
742 close_all_fds(NULL, 0);
744 reset_all_signal_handlers();
746 assert_se(sigemptyset(&mask) == 0);
747 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
752 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
755 if (mount_all(arg_directory) < 0)
758 if (copy_devnodes(arg_directory, console) < 0)
761 if (chdir(arg_directory) < 0) {
762 log_error("chdir(%s) failed: %m", arg_directory);
766 if (open_terminal("dev/console", O_RDWR) != STDIN_FILENO ||
767 dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
768 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
771 if (mount(arg_directory, "/", "bind", MS_BIND|MS_MOVE, NULL) < 0) {
772 log_error("mount(MS_MOVE) failed: %m");
776 if (chroot(".") < 0) {
777 log_error("chroot() failed: %m");
781 if (chdir("/") < 0) {
782 log_error("chdir() failed: %m");
788 if (drop_capabilities() < 0)
791 if ((hn = file_name_from_path(arg_directory)))
792 sethostname(hn, strlen(hn));
795 execvpe(argv[optind], argv + optind, (char**) envp);
798 execle("/bin/bash", "-bash", NULL, (char**) envp);
801 log_error("execv() failed: %m");
807 if (process_pty(master, &mask) < 0)
810 if (saved_attr_valid) {
811 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
812 saved_attr_valid = false;
815 r = wait_for_terminate_and_warn(argc > optind ? argv[optind] : "bash", pid);
821 if (saved_attr_valid)
822 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
825 close_nointr_nofail(master);
828 cg_attach(SYSTEMD_CGROUP_CONTROLLER, oldcg, 0);
831 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);