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 "audit.h"
  47 #include "missing.h"
  48 #include "cgroup-util.h"
  49 #include "strv.h"
  50 #include "loopback-setup.h"
  51
  52 static char *arg_directory = NULL;
  53 static char *arg_user = NULL;
  54 static bool arg_private_network = false;
  55
  56 static int help(void) {
  57
  58         printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
  59                "Spawn a minimal namespace container for debugging, testing and building.\n\n"
  60                "  -h --help            Show this help\n"
  61                "  -D --directory=NAME  Root directory for the container\n"
  62                "  -u --user=USER       Run the command under specified user or uid\n"
  63                "     --private-network Disable network in container\n",
  64                program_invocation_short_name);
  65
  66         return 0;
  67 }
  68
  69 static int parse_argv(int argc, char *argv[]) {
  70
  71         enum {
  72                 ARG_PRIVATE_NETWORK = 0x100
  73         };
  74
  75         static const struct option options[] = {
  76                 { "help",            no_argument,       NULL, 'h'                 },
  77                 { "directory",       required_argument, NULL, 'D'                 },
  78                 { "user",            required_argument, NULL, 'u'                 },
  79                 { "private-network", no_argument,       NULL, ARG_PRIVATE_NETWORK },
  80                 { NULL,              0,                 NULL, 0                   }
  81         };
  82
  83         int c;
  84
  85         assert(argc >= 0);
  86         assert(argv);
  87
  88         while ((c = getopt_long(argc, argv, "+hD:u:", options, NULL)) >= 0) {
  89
  90                 switch (c) {
  91
  92                 case 'h':
  93                         help();
  94                         return 0;
  95
  96                 case 'D':
  97                         free(arg_directory);
  98                         if (!(arg_directory = strdup(optarg))) {
  99                                 log_error("Failed to duplicate root directory.");
 100                                 return -ENOMEM;
 101                         }
 102
 103                         break;
 104
 105                 case 'u':
 106                         free(arg_user);
 107                         if (!(arg_user = strdup(optarg))) {
 108                                 log_error("Failed to duplicate user name.");
 109                                 return -ENOMEM;
 110                         }
 111
 112                         break;
 113
 114                 case ARG_PRIVATE_NETWORK:
 115                         arg_private_network = true;
 116                         break;
 117
 118                 case '?':
 119                         return -EINVAL;
 120
 121                 default:
 122                         log_error("Unknown option code %c", c);
 123                         return -EINVAL;
 124                 }
 125         }
 126
 127         return 1;
 128 }
 129
 130 static int mount_all(const char *dest) {
 131
 132         typedef struct MountPoint {
 133                 const char *what;
 134                 const char *where;
 135                 const char *type;
 136                 const char *options;
 137                 unsigned long flags;
 138                 bool fatal;
 139         } MountPoint;
 140
 141         static const MountPoint mount_table[] = {
 142                 { "proc",      "/proc",     "proc",  NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 143                 { "/proc/sys", "/proc/sys", "bind",  NULL,       MS_BIND, true                       },   /* Bind mount first */
 144                 { "/proc/sys", "/proc/sys", "bind",  NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 145                 { "/sys",      "/sys",      "bind",  NULL,       MS_BIND,                      true  },   /* Bind mount first */
 146                 { "/sys",      "/sys",      "bind",  NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 147                 { "tmpfs",     "/dev",      "tmpfs", "mode=755", MS_NOSUID,                    true  },
 148                 { "/dev/pts",  "/dev/pts",  "bind",  NULL,       MS_BIND,                      true  },
 149                 { "tmpfs",     "/run",      "tmpfs", "mode=755", MS_NOSUID|MS_NODEV,           true  },
 150 #ifdef HAVE_SELINUX
 151                 { "/sys/fs/selinux", "/sys/fs/selinux", "bind", NULL, MS_BIND,                      false },  /* Bind mount first */
 152                 { "/sys/fs/selinux", "/sys/fs/selinux", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false },  /* Then, make it r/o */
 153 #endif
 154         };
 155
 156         unsigned k;
 157         int r = 0;
 158         char *where;
 159
 160         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 161                 int t;
 162
 163                 if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
 164                         log_error("Out of memory");
 165
 166                         if (r == 0)
 167                                 r = -ENOMEM;
 168
 169                         break;
 170                 }
 171
 172                 if ((t = path_is_mount_point(where, false)) < 0) {
 173                         log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
 174                         free(where);
 175
 176                         if (r == 0)
 177                                 r = t;
 178
 179                         continue;
 180                 }
 181
 182                 mkdir_p(where, 0755);
 183
 184                 if (mount(mount_table[k].what,
 185                           where,
 186                           mount_table[k].type,
 187                           mount_table[k].flags,
 188                           mount_table[k].options) < 0 &&
 189                     mount_table[k].fatal) {
 190
 191                         log_error("mount(%s) failed: %m", where);
 192
 193                         if (r == 0)
 194                                 r = -errno;
 195                 }
 196
 197                 free(where);
 198         }
 199
 200         /* Fix the timezone, if possible */
 201         if (asprintf(&where, "%s/etc/localtime", dest) >= 0) {
 202
 203                 if (mount("/etc/localtime", where, "bind", MS_BIND, NULL) >= 0)
 204                         mount("/etc/localtime", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 205
 206                 free(where);
 207         }
 208
 209         if (asprintf(&where, "%s/etc/timezone", dest) >= 0) {
 210
 211                 if (mount("/etc/timezone", where, "bind", MS_BIND, NULL) >= 0)
 212                         mount("/etc/timezone", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 213
 214                 free(where);
 215         }
 216
 217         return r;
 218 }
 219
 220 static int copy_devnodes(const char *dest, const char *console) {
 221
 222         static const char devnodes[] =
 223                 "null\0"
 224                 "zero\0"
 225                 "full\0"
 226                 "random\0"
 227                 "urandom\0"
 228                 "tty\0"
 229                 "ptmx\0"
 230                 "kmsg\0"
 231                 "rtc0\0";
 232
 233         const char *d;
 234         int r = 0, k;
 235         mode_t u;
 236         struct stat st;
 237         char *from = NULL, *to = NULL;
 238
 239         assert(dest);
 240         assert(console);
 241
 242         u = umask(0000);
 243
 244         NULSTR_FOREACH(d, devnodes) {
 245                 from = to = NULL;
 246
 247                 asprintf(&from, "/dev/%s", d);
 248                 asprintf(&to, "%s/dev/%s", dest, d);
 249
 250                 if (!from || !to) {
 251                         log_error("Failed to allocate devnode path");
 252
 253                         free(from);
 254                         free(to);
 255
 256                         from = to = NULL;
 257
 258                         if (r == 0)
 259                                 r = -ENOMEM;
 260
 261                         break;
 262                 }
 263
 264                 if (stat(from, &st) < 0) {
 265
 266                         if (errno != ENOENT) {
 267                                 log_error("Failed to stat %s: %m", from);
 268                                 if (r == 0)
 269                                         r = -errno;
 270                         }
 271
 272                 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
 273
 274                         log_error("%s is not a char or block device, cannot copy.", from);
 275                         if (r == 0)
 276                                 r = -EIO;
 277
 278                 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
 279
 280                         log_error("mknod(%s) failed: %m", dest);
 281                         if (r == 0)
 282                                 r = -errno;
 283                 }
 284
 285                 free(from);
 286                 free(to);
 287         }
 288
 289         if (stat(console, &st) < 0) {
 290
 291                 log_error("Failed to stat %s: %m", console);
 292                 if (r == 0)
 293                         r = -errno;
 294
 295                 goto finish;
 296
 297         } else if (!S_ISCHR(st.st_mode)) {
 298
 299                 log_error("/dev/console is not a char device.");
 300                 if (r == 0)
 301                         r = -EIO;
 302
 303                 goto finish;
 304         }
 305
 306         if (asprintf(&to, "%s/dev/console", dest) < 0) {
 307
 308                 log_error("Out of memory");
 309                 if (r == 0)
 310                         r = -ENOMEM;
 311
 312                  goto finish;
 313         }
 314
 315         /* We need to bind mount the right tty to /dev/console since
 316          * ptys can only exist on pts file systems. To have something
 317          * to bind mount things on we create a device node first, that
 318          * has the right major/minor (note that the major minor
 319          * doesn't actually matter here, since we mount it over
 320          * anyway). */
 321
 322         if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0)
 323                 log_error("mknod for /dev/console failed: %m");
 324
 325         if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
 326                 log_error("bind mount for /dev/console failed: %m");
 327
 328                 if (r == 0)
 329                         r = -errno;
 330         }
 331
 332         free(to);
 333
 334         if ((k = chmod_and_chown(console, 0600, 0, 0)) < 0) {
 335                 log_error("Failed to correct access mode for TTY: %s", strerror(-k));
 336
 337                 if (r == 0)
 338                         r = k;
 339         }
 340
 341 finish:
 342         umask(u);
 343
 344         return r;
 345 }
 346
 347 static int drop_capabilities(void) {
 348         static const unsigned long retain[] = {
 349                 CAP_CHOWN,
 350                 CAP_DAC_OVERRIDE,
 351                 CAP_DAC_READ_SEARCH,
 352                 CAP_FOWNER,
 353                 CAP_FSETID,
 354                 CAP_IPC_OWNER,
 355                 CAP_KILL,
 356                 CAP_LEASE,
 357                 CAP_LINUX_IMMUTABLE,
 358                 CAP_NET_BIND_SERVICE,
 359                 CAP_NET_BROADCAST,
 360                 CAP_NET_RAW,
 361                 CAP_SETGID,
 362                 CAP_SETFCAP,
 363                 CAP_SETPCAP,
 364                 CAP_SETUID,
 365                 CAP_SYS_ADMIN,
 366                 CAP_SYS_CHROOT,
 367                 CAP_SYS_NICE,
 368                 CAP_SYS_PTRACE,
 369                 CAP_SYS_TTY_CONFIG
 370         };
 371
 372         unsigned long l;
 373
 374         for (l = 0; l <= cap_last_cap(); l++) {
 375                 unsigned i;
 376
 377                 for (i = 0; i < ELEMENTSOF(retain); i++)
 378                         if (retain[i] == l)
 379                                 break;
 380
 381                 if (i < ELEMENTSOF(retain))
 382                         continue;
 383
 384                 if (prctl(PR_CAPBSET_DROP, l) < 0) {
 385                         log_error("PR_CAPBSET_DROP failed: %m");
 386                         return -errno;
 387                 }
 388         }
 389
 390         return 0;
 391 }
 392
 393 static int is_os_tree(const char *path) {
 394         int r;
 395         char *p;
 396         /* We use /bin/sh as flag file if something is an OS */
 397
 398         if (asprintf(&p, "%s/bin/sh", path) < 0)
 399                 return -ENOMEM;
 400
 401         r = access(p, F_OK);
 402         free(p);
 403
 404         return r < 0 ? 0 : 1;
 405 }
 406
 407 static int process_pty(int master, sigset_t *mask) {
 408
 409         char in_buffer[LINE_MAX], out_buffer[LINE_MAX];
 410         size_t in_buffer_full = 0, out_buffer_full = 0;
 411         struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
 412         bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
 413         int ep = -1, signal_fd = -1, r;
 414
 415         fd_nonblock(STDIN_FILENO, 1);
 416         fd_nonblock(STDOUT_FILENO, 1);
 417         fd_nonblock(master, 1);
 418
 419         if ((signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
 420                 log_error("signalfd(): %m");
 421                 r = -errno;
 422                 goto finish;
 423         }
 424
 425         if ((ep = epoll_create1(EPOLL_CLOEXEC)) < 0) {
 426                 log_error("Failed to create epoll: %m");
 427                 r = -errno;
 428                 goto finish;
 429         }
 430
 431         zero(stdin_ev);
 432         stdin_ev.events = EPOLLIN|EPOLLET;
 433         stdin_ev.data.fd = STDIN_FILENO;
 434
 435         zero(stdout_ev);
 436         stdout_ev.events = EPOLLOUT|EPOLLET;
 437         stdout_ev.data.fd = STDOUT_FILENO;
 438
 439         zero(master_ev);
 440         master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
 441         master_ev.data.fd = master;
 442
 443         zero(signal_ev);
 444         signal_ev.events = EPOLLIN;
 445         signal_ev.data.fd = signal_fd;
 446
 447         if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0 ||
 448             epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0 ||
 449             epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
 450             epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
 451                 log_error("Failed to regiser fds in epoll: %m");
 452                 r = -errno;
 453                 goto finish;
 454         }
 455
 456         for (;;) {
 457                 struct epoll_event ev[16];
 458                 ssize_t k;
 459                 int i, nfds;
 460
 461                 if ((nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1)) < 0) {
 462
 463                         if (errno == EINTR || errno == EAGAIN)
 464                                 continue;
 465
 466                         log_error("epoll_wait(): %m");
 467                         r = -errno;
 468                         goto finish;
 469                 }
 470
 471                 assert(nfds >= 1);
 472
 473                 for (i = 0; i < nfds; i++) {
 474                         if (ev[i].data.fd == STDIN_FILENO) {
 475
 476                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
 477                                         stdin_readable = true;
 478
 479                         } else if (ev[i].data.fd == STDOUT_FILENO) {
 480
 481                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
 482                                         stdout_writable = true;
 483
 484                         } else if (ev[i].data.fd == master) {
 485
 486                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
 487                                         master_readable = true;
 488
 489                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
 490                                         master_writable = true;
 491
 492                         } else if (ev[i].data.fd == signal_fd) {
 493                                 struct signalfd_siginfo sfsi;
 494                                 ssize_t n;
 495
 496                                 if ((n = read(signal_fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) {
 497
 498                                         if (n >= 0) {
 499                                                 log_error("Failed to read from signalfd: invalid block size");
 500                                                 r = -EIO;
 501                                                 goto finish;
 502                                         }
 503
 504                                         if (errno != EINTR && errno != EAGAIN) {
 505                                                 log_error("Failed to read from signalfd: %m");
 506                                                 r = -errno;
 507                                                 goto finish;
 508                                         }
 509                                 } else {
 510
 511                                         if (sfsi.ssi_signo == SIGWINCH) {
 512                                                 struct winsize ws;
 513
 514                                                 /* The window size changed, let's forward that. */
 515                                                 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
 516                                                         ioctl(master, TIOCSWINSZ, &ws);
 517                                         } else {
 518                                                 r = 0;
 519                                                 goto finish;
 520                                         }
 521                                 }
 522                         }
 523                 }
 524
 525                 while ((stdin_readable && in_buffer_full <= 0) ||
 526                        (master_writable && in_buffer_full > 0) ||
 527                        (master_readable && out_buffer_full <= 0) ||
 528                        (stdout_writable && out_buffer_full > 0)) {
 529
 530                         if (stdin_readable && in_buffer_full < LINE_MAX) {
 531
 532                                 if ((k = read(STDIN_FILENO, in_buffer + in_buffer_full, LINE_MAX - in_buffer_full)) < 0) {
 533
 534                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 535                                                 stdin_readable = false;
 536                                         else {
 537                                                 log_error("read(): %m");
 538                                                 r = -errno;
 539                                                 goto finish;
 540                                         }
 541                                 } else
 542                                         in_buffer_full += (size_t) k;
 543                         }
 544
 545                         if (master_writable && in_buffer_full > 0) {
 546
 547                                 if ((k = write(master, in_buffer, in_buffer_full)) < 0) {
 548
 549                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 550                                                 master_writable = false;
 551                                         else {
 552                                                 log_error("write(): %m");
 553                                                 r = -errno;
 554                                                 goto finish;
 555                                         }
 556
 557                                 } else {
 558                                         assert(in_buffer_full >= (size_t) k);
 559                                         memmove(in_buffer, in_buffer + k, in_buffer_full - k);
 560                                         in_buffer_full -= k;
 561                                 }
 562                         }
 563
 564                         if (master_readable && out_buffer_full < LINE_MAX) {
 565
 566                                 if ((k = read(master, out_buffer + out_buffer_full, LINE_MAX - out_buffer_full)) < 0) {
 567
 568                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 569                                                 master_readable = false;
 570                                         else {
 571                                                 log_error("read(): %m");
 572                                                 r = -errno;
 573                                                 goto finish;
 574                                         }
 575                                 }  else
 576                                         out_buffer_full += (size_t) k;
 577                         }
 578
 579                         if (stdout_writable && out_buffer_full > 0) {
 580
 581                                 if ((k = write(STDOUT_FILENO, out_buffer, out_buffer_full)) < 0) {
 582
 583                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 584                                                 stdout_writable = false;
 585                                         else {
 586                                                 log_error("write(): %m");
 587                                                 r = -errno;
 588                                                 goto finish;
 589                                         }
 590
 591                                 } else {
 592                                         assert(out_buffer_full >= (size_t) k);
 593                                         memmove(out_buffer, out_buffer + k, out_buffer_full - k);
 594                                         out_buffer_full -= k;
 595                                 }
 596                         }
 597                 }
 598         }
 599
 600 finish:
 601         if (ep >= 0)
 602                 close_nointr_nofail(ep);
 603
 604         if (signal_fd >= 0)
 605                 close_nointr_nofail(signal_fd);
 606
 607         return r;
 608 }
 609
 610 int main(int argc, char *argv[]) {
 611         pid_t pid = 0;
 612         int r = EXIT_FAILURE, k;
 613         char *oldcg = NULL, *newcg = NULL;
 614         int master = -1;
 615         const char *console = NULL;
 616         struct termios saved_attr, raw_attr;
 617         sigset_t mask;
 618         bool saved_attr_valid = false;
 619         struct winsize ws;
 620
 621         log_parse_environment();
 622         log_open();
 623
 624         if ((r = parse_argv(argc, argv)) <= 0)
 625                 goto finish;
 626
 627         if (arg_directory) {
 628                 char *p;
 629
 630                 p = path_make_absolute_cwd(arg_directory);
 631                 free(arg_directory);
 632                 arg_directory = p;
 633         } else
 634                 arg_directory = get_current_dir_name();
 635
 636         if (!arg_directory) {
 637                 log_error("Failed to determine path");
 638                 goto finish;
 639         }
 640
 641         path_kill_slashes(arg_directory);
 642
 643         if (geteuid() != 0) {
 644                 log_error("Need to be root.");
 645                 goto finish;
 646         }
 647
 648         if (sd_booted() <= 0) {
 649                 log_error("Not running on a systemd system.");
 650                 goto finish;
 651         }
 652
 653         if (path_equal(arg_directory, "/")) {
 654                 log_error("Spawning container on root directory not supported.");
 655                 goto finish;
 656         }
 657
 658         if (is_os_tree(arg_directory) <= 0) {
 659                 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
 660                 goto finish;
 661         }
 662
 663         if ((k = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &oldcg)) < 0) {
 664                 log_error("Failed to determine current cgroup: %s", strerror(-k));
 665                 goto finish;
 666         }
 667
 668         if (asprintf(&newcg, "%s/nspawn-%lu", oldcg, (unsigned long) getpid()) < 0) {
 669                 log_error("Failed to allocate cgroup path.");
 670                 goto finish;
 671         }
 672
 673         if ((k = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, newcg, 0)) < 0)  {
 674                 log_error("Failed to create cgroup: %s", strerror(-k));
 675                 goto finish;
 676         }
 677
 678         if ((master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY)) < 0) {
 679                 log_error("Failed to acquire pseudo tty: %m");
 680                 goto finish;
 681         }
 682
 683         if (!(console = ptsname(master))) {
 684                 log_error("Failed to determine tty name: %m");
 685                 goto finish;
 686         }
 687
 688         log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
 689
 690         if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
 691                 ioctl(master, TIOCSWINSZ, &ws);
 692
 693         if (unlockpt(master) < 0) {
 694                 log_error("Failed to unlock tty: %m");
 695                 goto finish;
 696         }
 697
 698         if (tcgetattr(STDIN_FILENO, &saved_attr) < 0) {
 699                 log_error("Failed to get terminal attributes: %m");
 700                 goto finish;
 701         }
 702
 703         saved_attr_valid = true;
 704
 705         raw_attr = saved_attr;
 706         cfmakeraw(&raw_attr);
 707         raw_attr.c_lflag &= ~ECHO;
 708
 709         if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
 710                 log_error("Failed to set terminal attributes: %m");
 711                 goto finish;
 712         }
 713
 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);
 717
 718         pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS|(arg_private_network ? CLONE_NEWNET : 0), NULL);
 719         if (pid < 0) {
 720                 if (errno == EINVAL)
 721                         log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
 722                 else
 723                         log_error("clone() failed: %m");
 724
 725                 goto finish;
 726         }
 727
 728         if (pid == 0) {
 729                 /* child */
 730
 731                 const char *hn;
 732                 const char *home = NULL;
 733                 uid_t uid = (uid_t) -1;
 734                 gid_t gid = (gid_t) -1;
 735                 const char *envp[] = {
 736                         "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
 737                         "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
 738                         NULL, /* TERM */
 739                         NULL, /* HOME */
 740                         NULL, /* USER */
 741                         NULL, /* LOGNAME */
 742                         NULL
 743                 };
 744
 745                 envp[2] = strv_find_prefix(environ, "TERM=");
 746
 747                 close_nointr_nofail(master);
 748
 749                 close_nointr(STDIN_FILENO);
 750                 close_nointr(STDOUT_FILENO);
 751                 close_nointr(STDERR_FILENO);
 752
 753                 close_all_fds(NULL, 0);
 754
 755                 reset_all_signal_handlers();
 756
 757                 assert_se(sigemptyset(&mask) == 0);
 758                 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 759
 760                 if (setsid() < 0)
 761                         goto child_fail;
 762
 763                 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
 764                         goto child_fail;
 765
 766                 /* Mark / as private, in case somebody marked it shared */
 767                 if (mount(NULL, "/", NULL, MS_PRIVATE|MS_REC, NULL) < 0)
 768                         goto child_fail;
 769
 770                 if (mount_all(arg_directory) < 0)
 771                         goto child_fail;
 772
 773                 if (copy_devnodes(arg_directory, console) < 0)
 774                         goto child_fail;
 775
 776                 if (chdir(arg_directory) < 0) {
 777                         log_error("chdir(%s) failed: %m", arg_directory);
 778                         goto child_fail;
 779                 }
 780
 781                 if (open_terminal("dev/console", O_RDWR) != STDIN_FILENO ||
 782                     dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
 783                     dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
 784                         goto child_fail;
 785
 786                 if (mount(arg_directory, "/", "bind", MS_BIND|MS_MOVE, NULL) < 0) {
 787                         log_error("mount(MS_MOVE) failed: %m");
 788                         goto child_fail;
 789                 }
 790
 791                 if (chroot(".") < 0) {
 792                         log_error("chroot() failed: %m");
 793                         goto child_fail;
 794                 }
 795
 796                 if (chdir("/") < 0) {
 797                         log_error("chdir() failed: %m");
 798                         goto child_fail;
 799                 }
 800
 801                 umask(0022);
 802
 803                 loopback_setup();
 804
 805                 if (drop_capabilities() < 0)
 806                         goto child_fail;
 807
 808                 if (arg_user) {
 809
 810                         if (get_user_creds((const char**)&arg_user, &uid, &gid, &home) < 0) {
 811                                 log_error("get_user_creds() failed: %m");
 812                                 goto child_fail;
 813                         }
 814
 815                         if (mkdir_parents(home, 0775) < 0) {
 816                                 log_error("mkdir_parents() failed: %m");
 817                                 goto child_fail;
 818                         }
 819
 820                         if (safe_mkdir(home, 0775, uid, gid) < 0) {
 821                                 log_error("safe_mkdir() failed: %m");
 822                                 goto child_fail;
 823                         }
 824
 825                         if (initgroups((const char*)arg_user, gid) < 0) {
 826                                 log_error("initgroups() failed: %m");
 827                                 goto child_fail;
 828                         }
 829
 830                         if (setresgid(gid, gid, gid) < 0) {
 831                                 log_error("setregid() failed: %m");
 832                                 goto child_fail;
 833                         }
 834
 835                         if (setresuid(uid, uid, uid) < 0) {
 836                                 log_error("setreuid() failed: %m");
 837                                 goto child_fail;
 838                         }
 839                 }
 840
 841                 if ((asprintf((char**)(envp + 3), "HOME=%s", home? home: "/root") < 0) ||
 842                     (asprintf((char**)(envp + 4), "USER=%s", arg_user? arg_user : "root") < 0) ||
 843                     (asprintf((char**)(envp + 5), "LOGNAME=%s", arg_user? arg_user : "root") < 0)) {
 844                     log_error("Out of memory");
 845                     goto child_fail;
 846                 }
 847
 848                 if ((hn = file_name_from_path(arg_directory)))
 849                         sethostname(hn, strlen(hn));
 850
 851                 if (argc > optind)
 852                         execvpe(argv[optind], argv + optind, (char**) envp);
 853                 else {
 854                         chdir(home ? home : "/root");
 855                         execle("/bin/bash", "-bash", NULL, (char**) envp);
 856                 }
 857
 858                 log_error("execv() failed: %m");
 859
 860         child_fail:
 861                 _exit(EXIT_FAILURE);
 862         }
 863
 864         if (process_pty(master, &mask) < 0)
 865                 goto finish;
 866
 867         if (saved_attr_valid) {
 868                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
 869                 saved_attr_valid = false;
 870         }
 871
 872         r = wait_for_terminate_and_warn(argc > optind ? argv[optind] : "bash", pid);
 873
 874         if (r < 0)
 875                 r = EXIT_FAILURE;
 876
 877 finish:
 878         if (saved_attr_valid)
 879                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
 880
 881         if (master >= 0)
 882                 close_nointr_nofail(master);
 883
 884         if (oldcg)
 885                 cg_attach(SYSTEMD_CGROUP_CONTROLLER, oldcg, 0);
 886
 887         if (newcg)
 888                 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
 889
 890         free(arg_directory);
 891         free(oldcg);
 892         free(newcg);
 893
 894         return r;
 895 }