src/nspawn.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   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.
  12
  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.
  17
  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/>.
  20 ***/
  21
  22 #include <signal.h>
  23 #include <sched.h>
  24 #include <unistd.h>
  25 #include <sys/types.h>
  26 #include <sys/syscall.h>
  27 #include <sys/mount.h>
  28 #include <sys/wait.h>
  29 #include <stdlib.h>
  30 #include <string.h>
  31 #include <stdio.h>
  32 #include <errno.h>
  33 #include <sys/prctl.h>
  34 #include <sys/capability.h>
  35 #include <getopt.h>
  36 #include <sys/epoll.h>
  37 #include <termios.h>
  38 #include <sys/signalfd.h>
  39 #include <grp.h>
  40 #include <linux/fs.h>
  41
  42 #include <systemd/sd-daemon.h>
  43
  44 #include "log.h"
  45 #include "util.h"
  46 #include "missing.h"
  47 #include "cgroup-util.h"
  48 #include "strv.h"
  49 #include "loopback-setup.h"
  50
  51 static char *arg_directory = NULL;
  52 static char *arg_user = NULL;
  53 static bool arg_private_network = false;
  54
  55 static int help(void) {
  56
  57         printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
  58                "Spawn a minimal namespace container for debugging, testing and building.\n\n"
  59                "  -h --help            Show this help\n"
  60                "  -D --directory=NAME  Root directory for the container\n"
  61                "  -u --user=USER       Run the command under specified user or uid\n"
  62                "     --private-network Disable network in container\n",
  63                program_invocation_short_name);
  64
  65         return 0;
  66 }
  67
  68 static int parse_argv(int argc, char *argv[]) {
  69
  70         enum {
  71                 ARG_PRIVATE_NETWORK = 0x100
  72         };
  73
  74         static const struct option options[] = {
  75                 { "help",            no_argument,       NULL, 'h'                 },
  76                 { "directory",       required_argument, NULL, 'D'                 },
  77                 { "user",            required_argument, NULL, 'u'                 },
  78                 { "private-network", no_argument,       NULL, ARG_PRIVATE_NETWORK },
  79                 { NULL,              0,                 NULL, 0                   }
  80         };
  81
  82         int c;
  83
  84         assert(argc >= 0);
  85         assert(argv);
  86
  87         while ((c = getopt_long(argc, argv, "+hD:u:", options, NULL)) >= 0) {
  88
  89                 switch (c) {
  90
  91                 case 'h':
  92                         help();
  93                         return 0;
  94
  95                 case 'D':
  96                         free(arg_directory);
  97                         if (!(arg_directory = strdup(optarg))) {
  98                                 log_error("Failed to duplicate root directory.");
  99                                 return -ENOMEM;
 100                         }
 101
 102                         break;
 103
 104                 case 'u':
 105                         free(arg_user);
 106                         if (!(arg_user = strdup(optarg))) {
 107                                 log_error("Failed to duplicate user name.");
 108                                 return -ENOMEM;
 109                         }
 110
 111                         break;
 112
 113                 case ARG_PRIVATE_NETWORK:
 114                         arg_private_network = true;
 115                         break;
 116
 117                 case '?':
 118                         return -EINVAL;
 119
 120                 default:
 121                         log_error("Unknown option code %c", c);
 122                         return -EINVAL;
 123                 }
 124         }
 125
 126         return 1;
 127 }
 128
 129 static int mount_all(const char *dest) {
 130
 131         typedef struct MountPoint {
 132                 const char *what;
 133                 const char *where;
 134                 const char *type;
 135                 const char *options;
 136                 unsigned long flags;
 137                 bool fatal;
 138         } MountPoint;
 139
 140         static const MountPoint mount_table[] = {
 141                 { "proc",      "/proc",     "proc",  NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 142                 { "/proc/sys", "/proc/sys", "bind",  NULL,       MS_BIND, true                       },   /* Bind mount first */
 143                 { "/proc/sys", "/proc/sys", "bind",  NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 144                 { "/sys",      "/sys",      "bind",  NULL,       MS_BIND,                      true  },   /* Bind mount first */
 145                 { "/sys",      "/sys",      "bind",  NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 146                 { "tmpfs",     "/dev",      "tmpfs", "mode=755", MS_NOSUID,                    true  },
 147                 { "/dev/pts",  "/dev/pts",  "bind",  NULL,       MS_BIND,                      true  },
 148                 { "tmpfs",     "/run",      "tmpfs", "mode=755", MS_NOSUID|MS_NODEV,           true  },
 149 #ifdef HAVE_SELINUX
 150                 { "/sys/fs/selinux", "/sys/fs/selinux", "bind", NULL, MS_BIND,                      false },  /* Bind mount first */
 151                 { "/sys/fs/selinux", "/sys/fs/selinux", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false },  /* Then, make it r/o */
 152 #endif
 153         };
 154
 155         unsigned k;
 156         int r = 0;
 157         char *where;
 158
 159         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 160                 int t;
 161
 162                 if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
 163                         log_error("Out of memory");
 164
 165                         if (r == 0)
 166                                 r = -ENOMEM;
 167
 168                         break;
 169                 }
 170
 171                 if ((t = path_is_mount_point(where, false)) < 0) {
 172                         log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
 173                         free(where);
 174
 175                         if (r == 0)
 176                                 r = t;
 177
 178                         continue;
 179                 }
 180
 181                 mkdir_p(where, 0755);
 182
 183                 if (mount(mount_table[k].what,
 184                           where,
 185                           mount_table[k].type,
 186                           mount_table[k].flags,
 187                           mount_table[k].options) < 0 &&
 188                     mount_table[k].fatal) {
 189
 190                         log_error("mount(%s) failed: %m", where);
 191
 192                         if (r == 0)
 193                                 r = -errno;
 194                 }
 195
 196                 free(where);
 197         }
 198
 199         /* Fix the timezone, if possible */
 200         if (asprintf(&where, "%s/etc/localtime", dest) >= 0) {
 201
 202                 if (mount("/etc/localtime", where, "bind", MS_BIND, NULL) >= 0)
 203                         mount("/etc/localtime", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 204
 205                 free(where);
 206         }
 207
 208         if (asprintf(&where, "%s/etc/timezone", dest) >= 0) {
 209
 210                 if (mount("/etc/timezone", where, "bind", MS_BIND, NULL) >= 0)
 211                         mount("/etc/timezone", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 212
 213                 free(where);
 214         }
 215
 216         return r;
 217 }
 218
 219 static int copy_devnodes(const char *dest, const char *console) {
 220
 221         static const char devnodes[] =
 222                 "null\0"
 223                 "zero\0"
 224                 "full\0"
 225                 "random\0"
 226                 "urandom\0"
 227                 "tty\0"
 228                 "ptmx\0"
 229                 "kmsg\0"
 230                 "rtc0\0";
 231
 232         const char *d;
 233         int r = 0, k;
 234         mode_t u;
 235         struct stat st;
 236         char *from = NULL, *to = NULL;
 237
 238         assert(dest);
 239         assert(console);
 240
 241         u = umask(0000);
 242
 243         NULSTR_FOREACH(d, devnodes) {
 244                 from = to = NULL;
 245
 246                 asprintf(&from, "/dev/%s", d);
 247                 asprintf(&to, "%s/dev/%s", dest, d);
 248
 249                 if (!from || !to) {
 250                         log_error("Failed to allocate devnode path");
 251
 252                         free(from);
 253                         free(to);
 254
 255                         from = to = NULL;
 256
 257                         if (r == 0)
 258                                 r = -ENOMEM;
 259
 260                         break;
 261                 }
 262
 263                 if (stat(from, &st) < 0) {
 264
 265                         if (errno != ENOENT) {
 266                                 log_error("Failed to stat %s: %m", from);
 267                                 if (r == 0)
 268                                         r = -errno;
 269                         }
 270
 271                 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
 272
 273                         log_error("%s is not a char or block device, cannot copy.", from);
 274                         if (r == 0)
 275                                 r = -EIO;
 276
 277                 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
 278
 279                         log_error("mknod(%s) failed: %m", dest);
 280                         if (r == 0)
 281                                 r = -errno;
 282                 }
 283
 284                 free(from);
 285                 free(to);
 286         }
 287
 288         if (stat(console, &st) < 0) {
 289
 290                 log_error("Failed to stat %s: %m", console);
 291                 if (r == 0)
 292                         r = -errno;
 293
 294                 goto finish;
 295
 296         } else if (!S_ISCHR(st.st_mode)) {
 297
 298                 log_error("/dev/console is not a char device.");
 299                 if (r == 0)
 300                         r = -EIO;
 301
 302                 goto finish;
 303         }
 304
 305         if (asprintf(&to, "%s/dev/console", dest) < 0) {
 306
 307                 log_error("Out of memory");
 308                 if (r == 0)
 309                         r = -ENOMEM;
 310
 311                  goto finish;
 312         }
 313
 314         /* We need to bind mount the right tty to /dev/console since
 315          * ptys can only exist on pts file systems. To have something
 316          * to bind mount things on we create a device node first, that
 317          * has the right major/minor (note that the major minor
 318          * doesn't actually matter here, since we mount it over
 319          * anyway). */
 320
 321         if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0)
 322                 log_error("mknod for /dev/console failed: %m");
 323
 324         if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
 325                 log_error("bind mount for /dev/console failed: %m");
 326
 327                 if (r == 0)
 328                         r = -errno;
 329         }
 330
 331         free(to);
 332
 333         if ((k = chmod_and_chown(console, 0600, 0, 0)) < 0) {
 334                 log_error("Failed to correct access mode for TTY: %s", strerror(-k));
 335
 336                 if (r == 0)
 337                         r = k;
 338         }
 339
 340 finish:
 341         umask(u);
 342
 343         return r;
 344 }
 345
 346 static int drop_capabilities(void) {
 347         static const unsigned long retain[] = {
 348                 CAP_CHOWN,
 349                 CAP_DAC_OVERRIDE,
 350                 CAP_DAC_READ_SEARCH,
 351                 CAP_FOWNER,
 352                 CAP_FSETID,
 353                 CAP_IPC_OWNER,
 354                 CAP_KILL,
 355                 CAP_LEASE,
 356                 CAP_LINUX_IMMUTABLE,
 357                 CAP_NET_BIND_SERVICE,
 358                 CAP_NET_BROADCAST,
 359                 CAP_NET_RAW,
 360                 CAP_SETGID,
 361                 CAP_SETFCAP,
 362                 CAP_SETPCAP,
 363                 CAP_SETUID,
 364                 CAP_SYS_ADMIN,
 365                 CAP_SYS_CHROOT,
 366                 CAP_SYS_NICE,
 367                 CAP_SYS_PTRACE,
 368                 CAP_SYS_TTY_CONFIG
 369         };
 370
 371         unsigned long l;
 372
 373         for (l = 0; l <= cap_last_cap(); l++) {
 374                 unsigned i;
 375
 376                 for (i = 0; i < ELEMENTSOF(retain); i++)
 377                         if (retain[i] == l)
 378                                 break;
 379
 380                 if (i < ELEMENTSOF(retain))
 381                         continue;
 382
 383                 if (prctl(PR_CAPBSET_DROP, l) < 0) {
 384                         log_error("PR_CAPBSET_DROP failed: %m");
 385                         return -errno;
 386                 }
 387         }
 388
 389         return 0;
 390 }
 391
 392 static int is_os_tree(const char *path) {
 393         int r;
 394         char *p;
 395         /* We use /bin/sh as flag file if something is an OS */
 396
 397         if (asprintf(&p, "%s/bin/sh", path) < 0)
 398                 return -ENOMEM;
 399
 400         r = access(p, F_OK);
 401         free(p);
 402
 403         return r < 0 ? 0 : 1;
 404 }
 405
 406 static int process_pty(int master, sigset_t *mask) {
 407
 408         char in_buffer[LINE_MAX], out_buffer[LINE_MAX];
 409         size_t in_buffer_full = 0, out_buffer_full = 0;
 410         struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
 411         bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
 412         int ep = -1, signal_fd = -1, r;
 413
 414         fd_nonblock(STDIN_FILENO, 1);
 415         fd_nonblock(STDOUT_FILENO, 1);
 416         fd_nonblock(master, 1);
 417
 418         if ((signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
 419                 log_error("signalfd(): %m");
 420                 r = -errno;
 421                 goto finish;
 422         }
 423
 424         if ((ep = epoll_create1(EPOLL_CLOEXEC)) < 0) {
 425                 log_error("Failed to create epoll: %m");
 426                 r = -errno;
 427                 goto finish;
 428         }
 429
 430         zero(stdin_ev);
 431         stdin_ev.events = EPOLLIN|EPOLLET;
 432         stdin_ev.data.fd = STDIN_FILENO;
 433
 434         zero(stdout_ev);
 435         stdout_ev.events = EPOLLOUT|EPOLLET;
 436         stdout_ev.data.fd = STDOUT_FILENO;
 437
 438         zero(master_ev);
 439         master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
 440         master_ev.data.fd = master;
 441
 442         zero(signal_ev);
 443         signal_ev.events = EPOLLIN;
 444         signal_ev.data.fd = signal_fd;
 445
 446         if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0 ||
 447             epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0 ||
 448             epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
 449             epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
 450                 log_error("Failed to regiser fds in epoll: %m");
 451                 r = -errno;
 452                 goto finish;
 453         }
 454
 455         for (;;) {
 456                 struct epoll_event ev[16];
 457                 ssize_t k;
 458                 int i, nfds;
 459
 460                 if ((nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1)) < 0) {
 461
 462                         if (errno == EINTR || errno == EAGAIN)
 463                                 continue;
 464
 465                         log_error("epoll_wait(): %m");
 466                         r = -errno;
 467                         goto finish;
 468                 }
 469
 470                 assert(nfds >= 1);
 471
 472                 for (i = 0; i < nfds; i++) {
 473                         if (ev[i].data.fd == STDIN_FILENO) {
 474
 475                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
 476                                         stdin_readable = true;
 477
 478                         } else if (ev[i].data.fd == STDOUT_FILENO) {
 479
 480                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
 481                                         stdout_writable = true;
 482
 483                         } else if (ev[i].data.fd == master) {
 484
 485                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
 486                                         master_readable = true;
 487
 488                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
 489                                         master_writable = true;
 490
 491                         } else if (ev[i].data.fd == signal_fd) {
 492                                 struct signalfd_siginfo sfsi;
 493                                 ssize_t n;
 494
 495                                 if ((n = read(signal_fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) {
 496
 497                                         if (n >= 0) {
 498                                                 log_error("Failed to read from signalfd: invalid block size");
 499                                                 r = -EIO;
 500                                                 goto finish;
 501                                         }
 502
 503                                         if (errno != EINTR && errno != EAGAIN) {
 504                                                 log_error("Failed to read from signalfd: %m");
 505                                                 r = -errno;
 506                                                 goto finish;
 507                                         }
 508                                 } else {
 509
 510                                         if (sfsi.ssi_signo == SIGWINCH) {
 511                                                 struct winsize ws;
 512
 513                                                 /* The window size changed, let's forward that. */
 514                                                 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
 515                                                         ioctl(master, TIOCSWINSZ, &ws);
 516                                         } else {
 517                                                 r = 0;
 518                                                 goto finish;
 519                                         }
 520                                 }
 521                         }
 522                 }
 523
 524                 while ((stdin_readable && in_buffer_full <= 0) ||
 525                        (master_writable && in_buffer_full > 0) ||
 526                        (master_readable && out_buffer_full <= 0) ||
 527                        (stdout_writable && out_buffer_full > 0)) {
 528
 529                         if (stdin_readable && in_buffer_full < LINE_MAX) {
 530
 531                                 if ((k = read(STDIN_FILENO, in_buffer + in_buffer_full, LINE_MAX - in_buffer_full)) < 0) {
 532
 533                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 534                                                 stdin_readable = false;
 535                                         else {
 536                                                 log_error("read(): %m");
 537                                                 r = -errno;
 538                                                 goto finish;
 539                                         }
 540                                 } else
 541                                         in_buffer_full += (size_t) k;
 542                         }
 543
 544                         if (master_writable && in_buffer_full > 0) {
 545
 546                                 if ((k = write(master, in_buffer, in_buffer_full)) < 0) {
 547
 548                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 549                                                 master_writable = false;
 550                                         else {
 551                                                 log_error("write(): %m");
 552                                                 r = -errno;
 553                                                 goto finish;
 554                                         }
 555
 556                                 } else {
 557                                         assert(in_buffer_full >= (size_t) k);
 558                                         memmove(in_buffer, in_buffer + k, in_buffer_full - k);
 559                                         in_buffer_full -= k;
 560                                 }
 561                         }
 562
 563                         if (master_readable && out_buffer_full < LINE_MAX) {
 564
 565                                 if ((k = read(master, out_buffer + out_buffer_full, LINE_MAX - out_buffer_full)) < 0) {
 566
 567                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 568                                                 master_readable = false;
 569                                         else {
 570                                                 log_error("read(): %m");
 571                                                 r = -errno;
 572                                                 goto finish;
 573                                         }
 574                                 }  else
 575                                         out_buffer_full += (size_t) k;
 576                         }
 577
 578                         if (stdout_writable && out_buffer_full > 0) {
 579
 580                                 if ((k = write(STDOUT_FILENO, out_buffer, out_buffer_full)) < 0) {
 581
 582                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 583                                                 stdout_writable = false;
 584                                         else {
 585                                                 log_error("write(): %m");
 586                                                 r = -errno;
 587                                                 goto finish;
 588                                         }
 589
 590                                 } else {
 591                                         assert(out_buffer_full >= (size_t) k);
 592                                         memmove(out_buffer, out_buffer + k, out_buffer_full - k);
 593                                         out_buffer_full -= k;
 594                                 }
 595                         }
 596                 }
 597         }
 598
 599 finish:
 600         if (ep >= 0)
 601                 close_nointr_nofail(ep);
 602
 603         if (signal_fd >= 0)
 604                 close_nointr_nofail(signal_fd);
 605
 606         return r;
 607 }
 608
 609 int main(int argc, char *argv[]) {
 610         pid_t pid = 0;
 611         int r = EXIT_FAILURE, k;
 612         char *oldcg = NULL, *newcg = NULL;
 613         int master = -1;
 614         const char *console = NULL;
 615         struct termios saved_attr, raw_attr;
 616         sigset_t mask;
 617         bool saved_attr_valid = false;
 618         struct winsize ws;
 619
 620         log_parse_environment();
 621         log_open();
 622
 623         if ((r = parse_argv(argc, argv)) <= 0)
 624                 goto finish;
 625
 626         if (arg_directory) {
 627                 char *p;
 628
 629                 p = path_make_absolute_cwd(arg_directory);
 630                 free(arg_directory);
 631                 arg_directory = p;
 632         } else
 633                 arg_directory = get_current_dir_name();
 634
 635         if (!arg_directory) {
 636                 log_error("Failed to determine path");
 637                 goto finish;
 638         }
 639
 640         path_kill_slashes(arg_directory);
 641
 642         if (geteuid() != 0) {
 643                 log_error("Need to be root.");
 644                 goto finish;
 645         }
 646
 647         if (sd_booted() <= 0) {
 648                 log_error("Not running on a systemd system.");
 649                 goto finish;
 650         }
 651
 652         if (path_equal(arg_directory, "/")) {
 653                 log_error("Spawning container on root directory not supported.");
 654                 goto finish;
 655         }
 656
 657         if (is_os_tree(arg_directory) <= 0) {
 658                 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
 659                 goto finish;
 660         }
 661
 662         if ((k = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &oldcg)) < 0) {
 663                 log_error("Failed to determine current cgroup: %s", strerror(-k));
 664                 goto finish;
 665         }
 666
 667         if (asprintf(&newcg, "%s/nspawn-%lu", oldcg, (unsigned long) getpid()) < 0) {
 668                 log_error("Failed to allocate cgroup path.");
 669                 goto finish;
 670         }
 671
 672         if ((k = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, newcg, 0)) < 0)  {
 673                 log_error("Failed to create cgroup: %s", strerror(-k));
 674                 goto finish;
 675         }
 676
 677         if ((master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY)) < 0) {
 678                 log_error("Failed to acquire pseudo tty: %m");
 679                 goto finish;
 680         }
 681
 682         if (!(console = ptsname(master))) {
 683                 log_error("Failed to determine tty name: %m");
 684                 goto finish;
 685         }
 686
 687         log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
 688
 689         if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
 690                 ioctl(master, TIOCSWINSZ, &ws);
 691
 692         if (unlockpt(master) < 0) {
 693                 log_error("Failed to unlock tty: %m");
 694                 goto finish;
 695         }
 696
 697         if (tcgetattr(STDIN_FILENO, &saved_attr) < 0) {
 698                 log_error("Failed to get terminal attributes: %m");
 699                 goto finish;
 700         }
 701
 702         saved_attr_valid = true;
 703
 704         raw_attr = saved_attr;
 705         cfmakeraw(&raw_attr);
 706         raw_attr.c_lflag &= ~ECHO;
 707
 708         if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
 709                 log_error("Failed to set terminal attributes: %m");
 710                 goto finish;
 711         }
 712
 713         assert_se(sigemptyset(&mask) == 0);
 714         sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
 715         assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
 716
 717         pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS|(arg_private_network ? CLONE_NEWNET : 0), NULL);
 718         if (pid < 0) {
 719                 if (errno == EINVAL)
 720                         log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
 721                 else
 722                         log_error("clone() failed: %m");
 723
 724                 goto finish;
 725         }
 726
 727         if (pid == 0) {
 728                 /* child */
 729
 730                 const char *hn;
 731                 const char *home = NULL;
 732                 uid_t uid = (uid_t) -1;
 733                 gid_t gid = (gid_t) -1;
 734                 const char *envp[] = {
 735                         "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
 736                         "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
 737                         NULL, /* TERM */
 738                         NULL, /* HOME */
 739                         NULL, /* USER */
 740                         NULL, /* LOGNAME */
 741                         NULL
 742                 };
 743
 744                 envp[2] = strv_find_prefix(environ, "TERM=");
 745
 746                 close_nointr_nofail(master);
 747
 748                 close_nointr(STDIN_FILENO);
 749                 close_nointr(STDOUT_FILENO);
 750                 close_nointr(STDERR_FILENO);
 751
 752                 close_all_fds(NULL, 0);
 753
 754                 reset_all_signal_handlers();
 755
 756                 assert_se(sigemptyset(&mask) == 0);
 757                 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 758
 759                 if (setsid() < 0)
 760                         goto child_fail;
 761
 762                 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
 763                         goto child_fail;
 764
 765                 /* Mark / as private, in case somebody marked it shared */
 766                 if (mount(NULL, "/", NULL, MS_PRIVATE|MS_REC, NULL) < 0)
 767                         goto child_fail;
 768
 769                 if (mount_all(arg_directory) < 0)
 770                         goto child_fail;
 771
 772                 if (copy_devnodes(arg_directory, console) < 0)
 773                         goto child_fail;
 774
 775                 if (chdir(arg_directory) < 0) {
 776                         log_error("chdir(%s) failed: %m", arg_directory);
 777                         goto child_fail;
 778                 }
 779
 780                 if (open_terminal("dev/console", O_RDWR) != STDIN_FILENO ||
 781                     dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
 782                     dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
 783                         goto child_fail;
 784
 785                 if (mount(arg_directory, "/", "bind", MS_BIND|MS_MOVE, NULL) < 0) {
 786                         log_error("mount(MS_MOVE) failed: %m");
 787                         goto child_fail;
 788                 }
 789
 790                 if (chroot(".") < 0) {
 791                         log_error("chroot() failed: %m");
 792                         goto child_fail;
 793                 }
 794
 795                 if (chdir("/") < 0) {
 796                         log_error("chdir() failed: %m");
 797                         goto child_fail;
 798                 }
 799
 800                 umask(0022);
 801
 802                 loopback_setup();
 803
 804                 if (drop_capabilities() < 0)
 805                         goto child_fail;
 806
 807                 if (arg_user) {
 808
 809                         if (get_user_creds((const char**)&arg_user, &uid, &gid, &home) < 0) {
 810                                 log_error("get_user_creds() failed: %m");
 811                                 goto child_fail;
 812                         }
 813
 814                         if (mkdir_parents(home, 0775) < 0) {
 815                                 log_error("mkdir_parents() failed: %m");
 816                                 goto child_fail;
 817                         }
 818
 819                         if (safe_mkdir(home, 0775, uid, gid) < 0) {
 820                                 log_error("safe_mkdir() failed: %m");
 821                                 goto child_fail;
 822                         }
 823
 824                         if (initgroups((const char*)arg_user, gid) < 0) {
 825                                 log_error("initgroups() failed: %m");
 826                                 goto child_fail;
 827                         }
 828
 829                         if (setresgid(gid, gid, gid) < 0) {
 830                                 log_error("setregid() failed: %m");
 831                                 goto child_fail;
 832                         }
 833
 834                         if (setresuid(uid, uid, uid) < 0) {
 835                                 log_error("setreuid() failed: %m");
 836                                 goto child_fail;
 837                         }
 838                 }
 839
 840                 if ((asprintf((char**)(envp + 3), "HOME=%s", home? home: "/root") < 0) ||
 841                     (asprintf((char**)(envp + 4), "USER=%s", arg_user? arg_user : "root") < 0) ||
 842                     (asprintf((char**)(envp + 5), "LOGNAME=%s", arg_user? arg_user : "root") < 0)) {
 843                     log_error("Out of memory");
 844                     goto child_fail;
 845                 }
 846
 847                 if ((hn = file_name_from_path(arg_directory)))
 848                         sethostname(hn, strlen(hn));
 849
 850                 if (argc > optind)
 851                         execvpe(argv[optind], argv + optind, (char**) envp);
 852                 else {
 853                         chdir(home ? home : "/root");
 854                         execle("/bin/bash", "-bash", NULL, (char**) envp);
 855                 }
 856
 857                 log_error("execv() failed: %m");
 858
 859         child_fail:
 860                 _exit(EXIT_FAILURE);
 861         }
 862
 863         if (process_pty(master, &mask) < 0)
 864                 goto finish;
 865
 866         if (saved_attr_valid) {
 867                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
 868                 saved_attr_valid = false;
 869         }
 870
 871         r = wait_for_terminate_and_warn(argc > optind ? argv[optind] : "bash", pid);
 872
 873         if (r < 0)
 874                 r = EXIT_FAILURE;
 875
 876 finish:
 877         if (saved_attr_valid)
 878                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
 879
 880         if (master >= 0)
 881                 close_nointr_nofail(master);
 882
 883         if (oldcg)
 884                 cg_attach(SYSTEMD_CGROUP_CONTROLLER, oldcg, 0);
 885
 886         if (newcg)
 887                 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
 888
 889         free(arg_directory);
 890         free(oldcg);
 891         free(newcg);
 892
 893         return r;
 894 }