src/nspawn/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 Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  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   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <signal.h>
  23 #include <sched.h>
  24 #include <unistd.h>
  25 #include <attr/xattr.h>
  26 #include <sys/types.h>
  27 #include <sys/syscall.h>
  28 #include <sys/mount.h>
  29 #include <sys/wait.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <stdio.h>
  33 #include <errno.h>
  34 #include <sys/prctl.h>
  35 #include <sys/capability.h>
  36 #include <getopt.h>
  37 #include <sys/poll.h>
  38 #include <sys/epoll.h>
  39 #include <termios.h>
  40 #include <sys/signalfd.h>
  41 #include <grp.h>
  42 #include <linux/fs.h>
  43 #include <sys/un.h>
  44 #include <sys/socket.h>
  45
  46 #include <systemd/sd-daemon.h>
  47
  48 #include "log.h"
  49 #include "util.h"
  50 #include "mkdir.h"
  51 #include "macro.h"
  52 #include "audit.h"
  53 #include "missing.h"
  54 #include "cgroup-util.h"
  55 #include "strv.h"
  56 #include "path-util.h"
  57 #include "loopback-setup.h"
  58 #include "sd-id128.h"
  59 #include "dev-setup.h"
  60 #include "fdset.h"
  61 #include "build.h"
  62 #include "fileio.h"
  63
  64 #ifndef TTY_GID
  65 #define TTY_GID 5
  66 #endif
  67
  68 typedef enum LinkJournal {
  69         LINK_NO,
  70         LINK_AUTO,
  71         LINK_HOST,
  72         LINK_GUEST
  73 } LinkJournal;
  74
  75 static char *arg_directory = NULL;
  76 static char *arg_user = NULL;
  77 static char **arg_controllers = NULL;
  78 static char *arg_uuid = NULL;
  79 static char *arg_machine = NULL;
  80 static bool arg_private_network = false;
  81 static bool arg_read_only = false;
  82 static bool arg_boot = false;
  83 static LinkJournal arg_link_journal = LINK_AUTO;
  84 static uint64_t arg_retain =
  85         (1ULL << CAP_CHOWN) |
  86         (1ULL << CAP_DAC_OVERRIDE) |
  87         (1ULL << CAP_DAC_READ_SEARCH) |
  88         (1ULL << CAP_FOWNER) |
  89         (1ULL << CAP_FSETID) |
  90         (1ULL << CAP_IPC_OWNER) |
  91         (1ULL << CAP_KILL) |
  92         (1ULL << CAP_LEASE) |
  93         (1ULL << CAP_LINUX_IMMUTABLE) |
  94         (1ULL << CAP_NET_BIND_SERVICE) |
  95         (1ULL << CAP_NET_BROADCAST) |
  96         (1ULL << CAP_NET_RAW) |
  97         (1ULL << CAP_SETGID) |
  98         (1ULL << CAP_SETFCAP) |
  99         (1ULL << CAP_SETPCAP) |
 100         (1ULL << CAP_SETUID) |
 101         (1ULL << CAP_SYS_ADMIN) |
 102         (1ULL << CAP_SYS_CHROOT) |
 103         (1ULL << CAP_SYS_NICE) |
 104         (1ULL << CAP_SYS_PTRACE) |
 105         (1ULL << CAP_SYS_TTY_CONFIG) |
 106         (1ULL << CAP_SYS_RESOURCE) |
 107         (1ULL << CAP_SYS_BOOT) |
 108         (1ULL << CAP_AUDIT_WRITE) |
 109         (1ULL << CAP_AUDIT_CONTROL);
 110 static char **arg_bind = NULL;
 111 static char **arg_bind_ro = NULL;
 112
 113 static int help(void) {
 114
 115         printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
 116                "Spawn a minimal namespace container for debugging, testing and building.\n\n"
 117                "  -h --help                Show this help\n"
 118                "     --version             Print version string\n"
 119                "  -D --directory=NAME      Root directory for the container\n"
 120                "  -b --boot                Boot up full system (i.e. invoke init)\n"
 121                "  -u --user=USER           Run the command under specified user or uid\n"
 122                "  -C --controllers=LIST    Put the container in specified comma-separated\n"
 123                "                           cgroup hierarchies\n"
 124                "     --uuid=UUID           Set a specific machine UUID for the container\n"
 125                "  -M --machine=NAME        Set the machine name for the container\n"
 126                "     --private-network     Disable network in container\n"
 127                "     --read-only           Mount the root directory read-only\n"
 128                "     --capability=CAP      In addition to the default, retain specified\n"
 129                "                           capability\n"
 130                "     --link-journal=MODE   Link up guest journal, one of no, auto, guest, host\n"
 131                "  -j                       Equivalent to --link-journal=host\n"
 132                "     --bind=PATH[:PATH]    Bind mount a file or directory from the host into\n"
 133                "                           the container\n"
 134                "     --bind-ro=PATH[:PATH] Similar, but creates a read-only bind mount\n",
 135                program_invocation_short_name);
 136
 137         return 0;
 138 }
 139
 140 static int parse_argv(int argc, char *argv[]) {
 141
 142         enum {
 143                 ARG_VERSION = 0x100,
 144                 ARG_PRIVATE_NETWORK,
 145                 ARG_UUID,
 146                 ARG_READ_ONLY,
 147                 ARG_CAPABILITY,
 148                 ARG_LINK_JOURNAL,
 149                 ARG_BIND,
 150                 ARG_BIND_RO
 151         };
 152
 153         static const struct option options[] = {
 154                 { "help",            no_argument,       NULL, 'h'                 },
 155                 { "version",         no_argument,       NULL, ARG_VERSION         },
 156                 { "directory",       required_argument, NULL, 'D'                 },
 157                 { "user",            required_argument, NULL, 'u'                 },
 158                 { "controllers",     required_argument, NULL, 'C'                 },
 159                 { "private-network", no_argument,       NULL, ARG_PRIVATE_NETWORK },
 160                 { "boot",            no_argument,       NULL, 'b'                 },
 161                 { "uuid",            required_argument, NULL, ARG_UUID            },
 162                 { "read-only",       no_argument,       NULL, ARG_READ_ONLY       },
 163                 { "capability",      required_argument, NULL, ARG_CAPABILITY      },
 164                 { "link-journal",    required_argument, NULL, ARG_LINK_JOURNAL    },
 165                 { "bind",            required_argument, NULL, ARG_BIND            },
 166                 { "bind-ro",         required_argument, NULL, ARG_BIND_RO         },
 167                 { "machine",         required_argument, NULL, 'M'                 },
 168                 { NULL,              0,                 NULL, 0                   }
 169         };
 170
 171         int c;
 172
 173         assert(argc >= 0);
 174         assert(argv);
 175
 176         while ((c = getopt_long(argc, argv, "+hD:u:C:bj", options, NULL)) >= 0) {
 177
 178                 switch (c) {
 179
 180                 case 'h':
 181                         help();
 182                         return 0;
 183
 184                 case ARG_VERSION:
 185                         puts(PACKAGE_STRING);
 186                         puts(SYSTEMD_FEATURES);
 187                         return 0;
 188
 189                 case 'D':
 190                         free(arg_directory);
 191                         arg_directory = canonicalize_file_name(optarg);
 192                         if (!arg_directory) {
 193                                 log_error("Failed to canonicalize root directory.");
 194                                 return -ENOMEM;
 195                         }
 196
 197                         break;
 198
 199                 case 'u':
 200                         free(arg_user);
 201                         arg_user = strdup(optarg);
 202                         if (!arg_user)
 203                                 return log_oom();
 204
 205                         break;
 206
 207                 case 'C':
 208                         strv_free(arg_controllers);
 209                         arg_controllers = strv_split(optarg, ",");
 210                         if (!arg_controllers)
 211                                 return log_oom();
 212
 213                         cg_shorten_controllers(arg_controllers);
 214                         break;
 215
 216                 case ARG_PRIVATE_NETWORK:
 217                         arg_private_network = true;
 218                         break;
 219
 220                 case 'b':
 221                         arg_boot = true;
 222                         break;
 223
 224                 case ARG_UUID:
 225                         arg_uuid = optarg;
 226                         break;
 227
 228                 case 'M':
 229                         if (!hostname_is_valid(optarg)) {
 230                                 log_error("Invalid machine name: %s", optarg);
 231                                 return -EINVAL;
 232                         }
 233
 234                         free(arg_machine);
 235                         arg_machine = strdup(optarg);
 236                         if (!arg_machine)
 237                                 return log_oom();
 238
 239                         break;
 240
 241                 case ARG_READ_ONLY:
 242                         arg_read_only = true;
 243                         break;
 244
 245                 case ARG_CAPABILITY: {
 246                         char *state, *word;
 247                         size_t length;
 248
 249                         FOREACH_WORD_SEPARATOR(word, length, optarg, ",", state) {
 250                                 cap_value_t cap;
 251                                 char *t;
 252
 253                                 t = strndup(word, length);
 254                                 if (!t)
 255                                         return log_oom();
 256
 257                                 if (cap_from_name(t, &cap) < 0) {
 258                                         log_error("Failed to parse capability %s.", t);
 259                                         free(t);
 260                                         return -EINVAL;
 261                                 }
 262
 263                                 free(t);
 264                                 arg_retain |= 1ULL << (uint64_t) cap;
 265                         }
 266
 267                         break;
 268                 }
 269
 270                 case 'j':
 271                         arg_link_journal = LINK_GUEST;
 272                         break;
 273
 274                 case ARG_LINK_JOURNAL:
 275                         if (streq(optarg, "auto"))
 276                                 arg_link_journal = LINK_AUTO;
 277                         else if (streq(optarg, "no"))
 278                                 arg_link_journal = LINK_NO;
 279                         else if (streq(optarg, "guest"))
 280                                 arg_link_journal = LINK_GUEST;
 281                         else if (streq(optarg, "host"))
 282                                 arg_link_journal = LINK_HOST;
 283                         else {
 284                                 log_error("Failed to parse link journal mode %s", optarg);
 285                                 return -EINVAL;
 286                         }
 287
 288                         break;
 289
 290                 case ARG_BIND:
 291                 case ARG_BIND_RO: {
 292                         _cleanup_free_ char *a = NULL, *b = NULL;
 293                         char *e;
 294                         char ***x;
 295                         int r;
 296
 297                         x = c == ARG_BIND ? &arg_bind : &arg_bind_ro;
 298
 299                         e = strchr(optarg, ':');
 300                         if (e) {
 301                                 a = strndup(optarg, e - optarg);
 302                                 b = strdup(e + 1);
 303                         } else {
 304                                 a = strdup(optarg);
 305                                 b = strdup(optarg);
 306                         }
 307
 308                         if (!a || !b)
 309                                 return log_oom();
 310
 311                         if (!path_is_absolute(a) || !path_is_absolute(b)) {
 312                                 log_error("Invalid bind mount specification: %s", optarg);
 313                                 return -EINVAL;
 314                         }
 315
 316                         r = strv_extend(x, a);
 317                         if (r < 0)
 318                                 return r;
 319
 320                         r = strv_extend(x, b);
 321                         if (r < 0)
 322                                 return r;
 323
 324                         break;
 325                 }
 326
 327                 case '?':
 328                         return -EINVAL;
 329
 330                 default:
 331                         log_error("Unknown option code %c", c);
 332                         return -EINVAL;
 333                 }
 334         }
 335
 336         return 1;
 337 }
 338
 339 static int mount_all(const char *dest) {
 340
 341         typedef struct MountPoint {
 342                 const char *what;
 343                 const char *where;
 344                 const char *type;
 345                 const char *options;
 346                 unsigned long flags;
 347                 bool fatal;
 348         } MountPoint;
 349
 350         static const MountPoint mount_table[] = {
 351                 { "proc",      "/proc",     "proc",  NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 352                 { "/proc/sys", "/proc/sys", NULL,    NULL,       MS_BIND, true                       },   /* Bind mount first */
 353                 { NULL,        "/proc/sys", NULL,    NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 354                 { "sysfs",     "/sys",      "sysfs", NULL,       MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 355                 { "tmpfs",     "/dev",      "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,     true  },
 356                 { "devpts",    "/dev/pts",  "devpts","newinstance,ptmxmode=0666,mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC, true },
 357                 { "tmpfs",     "/dev/shm",  "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true  },
 358                 { "tmpfs",     "/run",      "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true  },
 359 #ifdef HAVE_SELINUX
 360                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND,                      false },  /* Bind mount first */
 361                 { NULL,              "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false },  /* Then, make it r/o */
 362 #endif
 363         };
 364
 365         unsigned k;
 366         int r = 0;
 367
 368         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 369                 _cleanup_free_ char *where = NULL;
 370                 int t;
 371
 372                 where = strjoin(dest, "/", mount_table[k].where, NULL);
 373                 if (!where)
 374                         return log_oom();
 375
 376                 t = path_is_mount_point(where, true);
 377                 if (t < 0) {
 378                         log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
 379
 380                         if (r == 0)
 381                                 r = t;
 382
 383                         continue;
 384                 }
 385
 386                 /* Skip this entry if it is not a remount. */
 387                 if (mount_table[k].what && t > 0)
 388                         continue;
 389
 390                 mkdir_p(where, 0755);
 391
 392                 if (mount(mount_table[k].what,
 393                           where,
 394                           mount_table[k].type,
 395                           mount_table[k].flags,
 396                           mount_table[k].options) < 0 &&
 397                     mount_table[k].fatal) {
 398
 399                         log_error("mount(%s) failed: %m", where);
 400
 401                         if (r == 0)
 402                                 r = -errno;
 403                 }
 404         }
 405
 406         return r;
 407 }
 408
 409 static int mount_binds(const char *dest, char **l, unsigned long flags) {
 410         char **x, **y;
 411
 412         STRV_FOREACH_PAIR(x, y, l) {
 413                 _cleanup_free_ char *where = NULL;
 414
 415                 where = strjoin(dest, "/", *y, NULL);
 416                 if (!where)
 417                         return log_oom();
 418
 419                 mkdir_p_label(where, 0755);
 420
 421                 if (mount(*x, where, "bind", MS_BIND, NULL) < 0) {
 422                         log_error("mount(%s) failed: %m", where);
 423                         return -errno;
 424                 }
 425
 426                 if (flags && mount(NULL, where, NULL, MS_REMOUNT|MS_BIND|flags, NULL) < 0) {
 427                         log_error("mount(%s) failed: %m", where);
 428                         return -errno;
 429                 }
 430         }
 431
 432         return 0;
 433 }
 434
 435 static int setup_timezone(const char *dest) {
 436         _cleanup_free_ char *where = NULL, *p = NULL, *q = NULL, *check = NULL, *what = NULL;
 437         char *z, *y;
 438         int r;
 439
 440         assert(dest);
 441
 442         /* Fix the timezone, if possible */
 443         r = readlink_malloc("/etc/localtime", &p);
 444         if (r < 0) {
 445                 log_warning("/etc/localtime is not a symlink, not updating container timezone.");
 446                 return 0;
 447         }
 448
 449         z = path_startswith(p, "../usr/share/zoneinfo/");
 450         if (!z)
 451                 z = path_startswith(p, "/usr/share/zoneinfo/");
 452         if (!z) {
 453                 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
 454                 return 0;
 455         }
 456
 457         where = strappend(dest, "/etc/localtime");
 458         if (!where)
 459                 return log_oom();
 460
 461         r = readlink_malloc(where, &q);
 462         if (r >= 0) {
 463                 y = path_startswith(q, "../usr/share/zoneinfo/");
 464                 if (!y)
 465                         y = path_startswith(q, "/usr/share/zoneinfo/");
 466
 467
 468                 /* Already pointing to the right place? Then do nothing .. */
 469                 if (y && streq(y, z))
 470                         return 0;
 471         }
 472
 473         check = strjoin(dest, "/usr/share/zoneinfo/", z, NULL);
 474         if (!check)
 475                 return log_oom();
 476
 477         if (access(check, F_OK) < 0) {
 478                 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
 479                 return 0;
 480         }
 481
 482         what = strappend("../usr/share/zoneinfo/", z);
 483         if (!what)
 484                 return log_oom();
 485
 486         unlink(where);
 487         if (symlink(what, where) < 0) {
 488                 log_error("Failed to correct timezone of container: %m");
 489                 return 0;
 490         }
 491
 492         return 0;
 493 }
 494
 495 static int setup_resolv_conf(const char *dest) {
 496         char _cleanup_free_ *where = NULL;
 497         _cleanup_close_ int fd = -1;
 498
 499         assert(dest);
 500
 501         if (arg_private_network)
 502                 return 0;
 503
 504         /* Fix resolv.conf, if possible */
 505         where = strappend(dest, "/etc/resolv.conf");
 506         if (!where)
 507                 return log_oom();
 508
 509         fd = open(where, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, 0644);
 510
 511         /* We don't really care for the results of this really. If it
 512          * fails, it fails, but meh... */
 513         if (mount("/etc/resolv.conf", where, "bind", MS_BIND, NULL) < 0)
 514                 log_warning("Failed to bind mount /etc/resolv.conf: %m");
 515         else
 516                 if (mount("/etc/resolv.conf", where, "bind",
 517                           MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0) {
 518                         log_error("Failed to remount /etc/resolv.conf readonly: %m");
 519                         return -errno;
 520                 }
 521
 522         return 0;
 523 }
 524
 525 static int setup_boot_id(const char *dest) {
 526         _cleanup_free_ char *from = NULL, *to = NULL;
 527         sd_id128_t rnd;
 528         char as_uuid[37];
 529         int r;
 530
 531         assert(dest);
 532
 533         /* Generate a new randomized boot ID, so that each boot-up of
 534          * the container gets a new one */
 535
 536         from = strappend(dest, "/dev/proc-sys-kernel-random-boot-id");
 537         to = strappend(dest, "/proc/sys/kernel/random/boot_id");
 538         if (!from || !to)
 539                 return log_oom();
 540
 541         r = sd_id128_randomize(&rnd);
 542         if (r < 0) {
 543                 log_error("Failed to generate random boot id: %s", strerror(-r));
 544                 return r;
 545         }
 546
 547         snprintf(as_uuid, sizeof(as_uuid),
 548                  "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
 549                  SD_ID128_FORMAT_VAL(rnd));
 550         char_array_0(as_uuid);
 551
 552         r = write_string_file(from, as_uuid);
 553         if (r < 0) {
 554                 log_error("Failed to write boot id: %s", strerror(-r));
 555                 return r;
 556         }
 557
 558         if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
 559                 log_error("Failed to bind mount boot id: %m");
 560                 r = -errno;
 561         } else if (mount(from, to, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL))
 562                 log_warning("Failed to make boot id read-only: %m");
 563
 564         unlink(from);
 565         return r;
 566 }
 567
 568 static int copy_devnodes(const char *dest) {
 569
 570         static const char devnodes[] =
 571                 "null\0"
 572                 "zero\0"
 573                 "full\0"
 574                 "random\0"
 575                 "urandom\0"
 576                 "tty\0";
 577
 578         const char *d;
 579         int r = 0;
 580         _cleanup_umask_ mode_t u;
 581
 582         assert(dest);
 583
 584         u = umask(0000);
 585
 586         NULSTR_FOREACH(d, devnodes) {
 587                 struct stat st;
 588                 _cleanup_free_ char *from = NULL, *to = NULL;
 589
 590                 asprintf(&from, "/dev/%s", d);
 591                 asprintf(&to, "%s/dev/%s", dest, d);
 592
 593                 if (!from || !to) {
 594                         log_oom();
 595
 596                         if (r == 0)
 597                                 r = -ENOMEM;
 598
 599                         break;
 600                 }
 601
 602                 if (stat(from, &st) < 0) {
 603
 604                         if (errno != ENOENT) {
 605                                 log_error("Failed to stat %s: %m", from);
 606                                 if (r == 0)
 607                                         r = -errno;
 608                         }
 609
 610                 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
 611
 612                         log_error("%s is not a char or block device, cannot copy", from);
 613                         if (r == 0)
 614                                 r = -EIO;
 615
 616                 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
 617
 618                         log_error("mknod(%s) failed: %m", dest);
 619                         if (r == 0)
 620                                 r = -errno;
 621                 }
 622         }
 623
 624         return r;
 625 }
 626
 627 static int setup_ptmx(const char *dest) {
 628         _cleanup_free_ char *p = NULL;
 629
 630         p = strappend(dest, "/dev/ptmx");
 631         if (!p)
 632                 return log_oom();
 633
 634         if (symlink("pts/ptmx", p) < 0) {
 635                 log_error("Failed to create /dev/ptmx symlink: %m");
 636                 return -errno;
 637         }
 638
 639         return 0;
 640 }
 641
 642 static int setup_dev_console(const char *dest, const char *console) {
 643         struct stat st;
 644         _cleanup_free_ char *to = NULL;
 645         int r;
 646         _cleanup_umask_ mode_t u;
 647
 648         assert(dest);
 649         assert(console);
 650
 651         u = umask(0000);
 652
 653         if (stat(console, &st) < 0) {
 654                 log_error("Failed to stat %s: %m", console);
 655                 return -errno;
 656
 657         } else if (!S_ISCHR(st.st_mode)) {
 658                 log_error("/dev/console is not a char device");
 659                 return -EIO;
 660         }
 661
 662         r = chmod_and_chown(console, 0600, 0, 0);
 663         if (r < 0) {
 664                 log_error("Failed to correct access mode for TTY: %s", strerror(-r));
 665                 return r;
 666         }
 667
 668         if (asprintf(&to, "%s/dev/console", dest) < 0)
 669                 return log_oom();
 670
 671         /* We need to bind mount the right tty to /dev/console since
 672          * ptys can only exist on pts file systems. To have something
 673          * to bind mount things on we create a device node first, that
 674          * has the right major/minor (note that the major minor
 675          * doesn't actually matter here, since we mount it over
 676          * anyway). */
 677
 678         if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) {
 679                 log_error("mknod() for /dev/console failed: %m");
 680                 return -errno;
 681         }
 682
 683         if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
 684                 log_error("Bind mount for /dev/console failed: %m");
 685                 return -errno;
 686         }
 687
 688         return 0;
 689 }
 690
 691 static int setup_kmsg(const char *dest, int kmsg_socket) {
 692         _cleanup_free_ char *from = NULL, *to = NULL;
 693         int r, fd, k;
 694         _cleanup_umask_ mode_t u;
 695         union {
 696                 struct cmsghdr cmsghdr;
 697                 uint8_t buf[CMSG_SPACE(sizeof(int))];
 698         } control = {};
 699         struct msghdr mh = {
 700                 .msg_control = &control,
 701                 .msg_controllen = sizeof(control),
 702         };
 703         struct cmsghdr *cmsg;
 704
 705         assert(dest);
 706         assert(kmsg_socket >= 0);
 707
 708         u = umask(0000);
 709
 710         /* We create the kmsg FIFO as /dev/kmsg, but immediately
 711          * delete it after bind mounting it to /proc/kmsg. While FIFOs
 712          * on the reading side behave very similar to /proc/kmsg,
 713          * their writing side behaves differently from /dev/kmsg in
 714          * that writing blocks when nothing is reading. In order to
 715          * avoid any problems with containers deadlocking due to this
 716          * we simply make /dev/kmsg unavailable to the container. */
 717         if (asprintf(&from, "%s/dev/kmsg", dest) < 0 ||
 718             asprintf(&to, "%s/proc/kmsg", dest) < 0)
 719                 return log_oom();
 720
 721         if (mkfifo(from, 0600) < 0) {
 722                 log_error("mkfifo() for /dev/kmsg failed: %m");
 723                 return -errno;
 724         }
 725
 726         r = chmod_and_chown(from, 0600, 0, 0);
 727         if (r < 0) {
 728                 log_error("Failed to correct access mode for /dev/kmsg: %s", strerror(-r));
 729                 return r;
 730         }
 731
 732         if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
 733                 log_error("Bind mount for /proc/kmsg failed: %m");
 734                 return -errno;
 735         }
 736
 737         fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
 738         if (fd < 0) {
 739                 log_error("Failed to open fifo: %m");
 740                 return -errno;
 741         }
 742
 743         cmsg = CMSG_FIRSTHDR(&mh);
 744         cmsg->cmsg_level = SOL_SOCKET;
 745         cmsg->cmsg_type = SCM_RIGHTS;
 746         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
 747         memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
 748
 749         mh.msg_controllen = cmsg->cmsg_len;
 750
 751         /* Store away the fd in the socket, so that it stays open as
 752          * long as we run the child */
 753         k = sendmsg(kmsg_socket, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
 754         close_nointr_nofail(fd);
 755
 756         if (k < 0) {
 757                 log_error("Failed to send FIFO fd: %m");
 758                 return -errno;
 759         }
 760
 761         /* And now make the FIFO unavailable as /dev/kmsg... */
 762         unlink(from);
 763         return 0;
 764 }
 765
 766 static int setup_hostname(void) {
 767
 768         if (sethostname(arg_machine, strlen(arg_machine)) < 0)
 769                 return -errno;
 770
 771         return 0;
 772 }
 773
 774 static int setup_journal(const char *directory) {
 775         sd_id128_t machine_id;
 776         _cleanup_free_ char *p = NULL, *b = NULL, *q = NULL, *d = NULL;
 777         char *id;
 778         int r;
 779
 780         if (arg_link_journal == LINK_NO)
 781                 return 0;
 782
 783         p = strappend(directory, "/etc/machine-id");
 784         if (!p)
 785                 return log_oom();
 786
 787         r = read_one_line_file(p, &b);
 788         if (r == -ENOENT && arg_link_journal == LINK_AUTO)
 789                 return 0;
 790         else if (r < 0) {
 791                 log_error("Failed to read machine ID from %s: %s", p, strerror(-r));
 792                 return r;
 793         }
 794
 795         id = strstrip(b);
 796         if (isempty(id) && arg_link_journal == LINK_AUTO)
 797                 return 0;
 798
 799         /* Verify validity */
 800         r = sd_id128_from_string(id, &machine_id);
 801         if (r < 0) {
 802                 log_error("Failed to parse machine ID from %s: %s", p, strerror(-r));
 803                 return r;
 804         }
 805
 806         free(p);
 807         p = strappend("/var/log/journal/", id);
 808         q = strjoin(directory, "/var/log/journal/", id, NULL);
 809         if (!p || !q)
 810                 return log_oom();
 811
 812         if (path_is_mount_point(p, false) > 0) {
 813                 if (arg_link_journal != LINK_AUTO) {
 814                         log_error("%s: already a mount point, refusing to use for journal", p);
 815                         return -EEXIST;
 816                 }
 817
 818                 return 0;
 819         }
 820
 821         if (path_is_mount_point(q, false) > 0) {
 822                 if (arg_link_journal != LINK_AUTO) {
 823                         log_error("%s: already a mount point, refusing to use for journal", q);
 824                         return -EEXIST;
 825                 }
 826
 827                 return 0;
 828         }
 829
 830         r = readlink_and_make_absolute(p, &d);
 831         if (r >= 0) {
 832                 if ((arg_link_journal == LINK_GUEST ||
 833                      arg_link_journal == LINK_AUTO) &&
 834                     path_equal(d, q)) {
 835
 836                         r = mkdir_p(q, 0755);
 837                         if (r < 0)
 838                                 log_warning("failed to create directory %s: %m", q);
 839                         return 0;
 840                 }
 841
 842                 if (unlink(p) < 0) {
 843                         log_error("Failed to remove symlink %s: %m", p);
 844                         return -errno;
 845                 }
 846         } else if (r == -EINVAL) {
 847
 848                 if (arg_link_journal == LINK_GUEST &&
 849                     rmdir(p) < 0) {
 850
 851                         if (errno == ENOTDIR) {
 852                                 log_error("%s already exists and is neither a symlink nor a directory", p);
 853                                 return r;
 854                         } else {
 855                                 log_error("Failed to remove %s: %m", p);
 856                                 return -errno;
 857                         }
 858                 }
 859         } else if (r != -ENOENT) {
 860                 log_error("readlink(%s) failed: %m", p);
 861                 return r;
 862         }
 863
 864         if (arg_link_journal == LINK_GUEST) {
 865
 866                 if (symlink(q, p) < 0) {
 867                         log_error("Failed to symlink %s to %s: %m", q, p);
 868                         return -errno;
 869                 }
 870
 871                 r = mkdir_p(q, 0755);
 872                 if (r < 0)
 873                         log_warning("failed to create directory %s: %m", q);
 874                 return 0;
 875         }
 876
 877         if (arg_link_journal == LINK_HOST) {
 878                 r = mkdir_p(p, 0755);
 879                 if (r < 0) {
 880                         log_error("Failed to create %s: %m", p);
 881                         return r;
 882                 }
 883
 884         } else if (access(p, F_OK) < 0)
 885                 return 0;
 886
 887         if (dir_is_empty(q) == 0) {
 888                 log_error("%s not empty.", q);
 889                 return -ENOTEMPTY;
 890         }
 891
 892         r = mkdir_p(q, 0755);
 893         if (r < 0) {
 894                 log_error("Failed to create %s: %m", q);
 895                 return r;
 896         }
 897
 898         if (mount(p, q, "bind", MS_BIND, NULL) < 0) {
 899                 log_error("Failed to bind mount journal from host into guest: %m");
 900                 return -errno;
 901         }
 902
 903         return 0;
 904 }
 905
 906 static int setup_cgroup(const char *path) {
 907         char **c;
 908         int r;
 909
 910         r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, path, 1);
 911         if (r < 0) {
 912                 log_error("Failed to create cgroup: %s", strerror(-r));
 913                 return r;
 914         }
 915
 916         STRV_FOREACH(c, arg_controllers) {
 917                 r = cg_create_and_attach(*c, path, 1);
 918                 if (r < 0)
 919                         log_warning("Failed to create cgroup in controller %s: %s", *c, strerror(-r));
 920         }
 921
 922         return 0;
 923 }
 924
 925 static int save_attributes(const char *cgroup, pid_t pid, const char *uuid, const char *directory) {
 926         char buf[DECIMAL_STR_MAX(pid_t)], path[PATH_MAX];
 927         int r = 0, k;
 928
 929         assert(cgroup);
 930         assert(pid >= 0);
 931         assert(arg_directory);
 932
 933 #ifdef HAVE_XATTR
 934         assert_se(snprintf(buf, sizeof(buf), "%lu", (unsigned long) pid) < (int) sizeof(buf));
 935
 936         r = snprintf(path, sizeof(path), "/sys/fs/cgroup/systemd/%s", cgroup);
 937         if (r >= (int) sizeof(path)) {
 938                 log_error("cgroup name too long");
 939                 return -EINVAL;
 940         }
 941
 942         r = setxattr(path, "trusted.init_pid", buf, strlen(buf), XATTR_CREATE);
 943         if (r < 0)
 944                 log_warning("Failed to set %s attribute on %s: %m", "trusted.init_pid", path);
 945
 946         if (uuid) {
 947                 k = setxattr(path, "trusted.machine_id", uuid, strlen(uuid), XATTR_CREATE);
 948                 if (k < 0) {
 949                         log_warning("Failed to set %s attribute on %s: %m", "trusted.machine_id", path);
 950                         if (r == 0)
 951                                 r = k;
 952                 }
 953         }
 954
 955         k = setxattr(path, "trusted.root_directory", directory, strlen(directory), XATTR_CREATE);
 956         if (k < 0) {
 957                 log_warning("Failed to set %s attribute on %s: %m", "trusted.machine_id", path);
 958                 if (r == 0)
 959                         r = k;
 960         }
 961 #endif
 962         return r;
 963 }
 964
 965 static int drop_capabilities(void) {
 966         return capability_bounding_set_drop(~arg_retain, false);
 967 }
 968
 969 static int process_pty(int master, pid_t pid, sigset_t *mask) {
 970
 971         char in_buffer[LINE_MAX], out_buffer[LINE_MAX];
 972         size_t in_buffer_full = 0, out_buffer_full = 0;
 973         struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
 974         bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
 975         int ep = -1, signal_fd = -1, r;
 976         bool tried_orderly_shutdown = false;
 977
 978         assert(master >= 0);
 979         assert(pid > 0);
 980         assert(mask);
 981
 982         fd_nonblock(STDIN_FILENO, 1);
 983         fd_nonblock(STDOUT_FILENO, 1);
 984         fd_nonblock(master, 1);
 985
 986         signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC);
 987         if (signal_fd < 0) {
 988                 log_error("signalfd(): %m");
 989                 r = -errno;
 990                 goto finish;
 991         }
 992
 993         ep = epoll_create1(EPOLL_CLOEXEC);
 994         if (ep < 0) {
 995                 log_error("Failed to create epoll: %m");
 996                 r = -errno;
 997                 goto finish;
 998         }
 999
1000         /* We read from STDIN only if this is actually a TTY,
1001          * otherwise we assume non-interactivity. */
1002         if (isatty(STDIN_FILENO)) {
1003                 zero(stdin_ev);
1004                 stdin_ev.events = EPOLLIN|EPOLLET;
1005                 stdin_ev.data.fd = STDIN_FILENO;
1006
1007                 if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0) {
1008                         log_error("Failed to register STDIN in epoll: %m");
1009                         r = -errno;
1010                         goto finish;
1011                 }
1012         }
1013
1014         zero(stdout_ev);
1015         stdout_ev.events = EPOLLOUT|EPOLLET;
1016         stdout_ev.data.fd = STDOUT_FILENO;
1017
1018         zero(master_ev);
1019         master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
1020         master_ev.data.fd = master;
1021
1022         zero(signal_ev);
1023         signal_ev.events = EPOLLIN;
1024         signal_ev.data.fd = signal_fd;
1025
1026         if (epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0) {
1027                 if (errno != EPERM) {
1028                         log_error("Failed to register stdout in epoll: %m");
1029                         r = -errno;
1030                         goto finish;
1031                 }
1032                 /* stdout without epoll support. Likely redirected to regular file. */
1033                 stdout_writable = true;
1034         }
1035
1036         if (epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
1037             epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
1038                 log_error("Failed to register fds in epoll: %m");
1039                 r = -errno;
1040                 goto finish;
1041         }
1042
1043         for (;;) {
1044                 struct epoll_event ev[16];
1045                 ssize_t k;
1046                 int i, nfds;
1047
1048                 nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1);
1049                 if (nfds < 0) {
1050
1051                         if (errno == EINTR || errno == EAGAIN)
1052                                 continue;
1053
1054                         log_error("epoll_wait(): %m");
1055                         r = -errno;
1056                         goto finish;
1057                 }
1058
1059                 assert(nfds >= 1);
1060
1061                 for (i = 0; i < nfds; i++) {
1062                         if (ev[i].data.fd == STDIN_FILENO) {
1063
1064                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
1065                                         stdin_readable = true;
1066
1067                         } else if (ev[i].data.fd == STDOUT_FILENO) {
1068
1069                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
1070                                         stdout_writable = true;
1071
1072                         } else if (ev[i].data.fd == master) {
1073
1074                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
1075                                         master_readable = true;
1076
1077                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
1078                                         master_writable = true;
1079
1080                         } else if (ev[i].data.fd == signal_fd) {
1081                                 struct signalfd_siginfo sfsi;
1082                                 ssize_t n;
1083
1084                                 n = read(signal_fd, &sfsi, sizeof(sfsi));
1085                                 if (n != sizeof(sfsi)) {
1086
1087                                         if (n >= 0) {
1088                                                 log_error("Failed to read from signalfd: invalid block size");
1089                                                 r = -EIO;
1090                                                 goto finish;
1091                                         }
1092
1093                                         if (errno != EINTR && errno != EAGAIN) {
1094                                                 log_error("Failed to read from signalfd: %m");
1095                                                 r = -errno;
1096                                                 goto finish;
1097                                         }
1098                                 } else {
1099
1100                                         if (sfsi.ssi_signo == SIGWINCH) {
1101                                                 struct winsize ws;
1102
1103                                                 /* The window size changed, let's forward that. */
1104                                                 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
1105                                                         ioctl(master, TIOCSWINSZ, &ws);
1106                                         } else if (sfsi.ssi_signo == SIGTERM && arg_boot && !tried_orderly_shutdown) {
1107
1108                                                 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
1109
1110                                                 /* This only works for systemd... */
1111                                                 tried_orderly_shutdown = true;
1112                                                 kill(pid, SIGRTMIN+3);
1113
1114                                         } else {
1115                                                 r = 0;
1116                                                 goto finish;
1117                                         }
1118                                 }
1119                         }
1120                 }
1121
1122                 while ((stdin_readable && in_buffer_full <= 0) ||
1123                        (master_writable && in_buffer_full > 0) ||
1124                        (master_readable && out_buffer_full <= 0) ||
1125                        (stdout_writable && out_buffer_full > 0)) {
1126
1127                         if (stdin_readable && in_buffer_full < LINE_MAX) {
1128
1129                                 k = read(STDIN_FILENO, in_buffer + in_buffer_full, LINE_MAX - in_buffer_full);
1130                                 if (k < 0) {
1131
1132                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
1133                                                 stdin_readable = false;
1134                                         else {
1135                                                 log_error("read(): %m");
1136                                                 r = -errno;
1137                                                 goto finish;
1138                                         }
1139                                 } else
1140                                         in_buffer_full += (size_t) k;
1141                         }
1142
1143                         if (master_writable && in_buffer_full > 0) {
1144
1145                                 k = write(master, in_buffer, in_buffer_full);
1146                                 if (k < 0) {
1147
1148                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
1149                                                 master_writable = false;
1150                                         else {
1151                                                 log_error("write(): %m");
1152                                                 r = -errno;
1153                                                 goto finish;
1154                                         }
1155
1156                                 } else {
1157                                         assert(in_buffer_full >= (size_t) k);
1158                                         memmove(in_buffer, in_buffer + k, in_buffer_full - k);
1159                                         in_buffer_full -= k;
1160                                 }
1161                         }
1162
1163                         if (master_readable && out_buffer_full < LINE_MAX) {
1164
1165                                 k = read(master, out_buffer + out_buffer_full, LINE_MAX - out_buffer_full);
1166                                 if (k < 0) {
1167
1168                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
1169                                                 master_readable = false;
1170                                         else {
1171                                                 log_error("read(): %m");
1172                                                 r = -errno;
1173                                                 goto finish;
1174                                         }
1175                                 }  else
1176                                         out_buffer_full += (size_t) k;
1177                         }
1178
1179                         if (stdout_writable && out_buffer_full > 0) {
1180
1181                                 k = write(STDOUT_FILENO, out_buffer, out_buffer_full);
1182                                 if (k < 0) {
1183
1184                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
1185                                                 stdout_writable = false;
1186                                         else {
1187                                                 log_error("write(): %m");
1188                                                 r = -errno;
1189                                                 goto finish;
1190                                         }
1191
1192                                 } else {
1193                                         assert(out_buffer_full >= (size_t) k);
1194                                         memmove(out_buffer, out_buffer + k, out_buffer_full - k);
1195                                         out_buffer_full -= k;
1196                                 }
1197                         }
1198                 }
1199         }
1200
1201 finish:
1202         if (ep >= 0)
1203                 close_nointr_nofail(ep);
1204
1205         if (signal_fd >= 0)
1206                 close_nointr_nofail(signal_fd);
1207
1208         return r;
1209 }
1210
1211 int main(int argc, char *argv[]) {
1212         pid_t pid = 0;
1213         int r = EXIT_FAILURE, k;
1214         _cleanup_free_ char *machine_root = NULL, *newcg = NULL;
1215         _cleanup_close_ int master = -1;
1216         int n_fd_passed;
1217         const char *console = NULL;
1218         struct termios saved_attr, raw_attr;
1219         sigset_t mask;
1220         bool saved_attr_valid = false;
1221         struct winsize ws;
1222         int kmsg_socket_pair[2] = { -1, -1 };
1223         FDSet *fds = NULL;
1224
1225         log_parse_environment();
1226         log_open();
1227
1228         r = parse_argv(argc, argv);
1229         if (r <= 0)
1230                 goto finish;
1231
1232         if (arg_directory) {
1233                 char *p;
1234
1235                 p = path_make_absolute_cwd(arg_directory);
1236                 free(arg_directory);
1237                 arg_directory = p;
1238         } else
1239                 arg_directory = get_current_dir_name();
1240
1241         if (!arg_directory) {
1242                 log_error("Failed to determine path, please use -D.");
1243                 goto finish;
1244         }
1245
1246         path_kill_slashes(arg_directory);
1247
1248         if (!arg_machine) {
1249                 arg_machine = strdup(path_get_file_name(arg_directory));
1250                 if (!arg_machine) {
1251                         log_oom();
1252                         goto finish;
1253                 }
1254
1255                 hostname_cleanup(arg_machine);
1256                 if (isempty(arg_machine)) {
1257                         log_error("Failed to determine machine name automatically, please use -M.");
1258                         goto finish;
1259                 }
1260         }
1261
1262         if (geteuid() != 0) {
1263                 log_error("Need to be root.");
1264                 goto finish;
1265         }
1266
1267         if (sd_booted() <= 0) {
1268                 log_error("Not running on a systemd system.");
1269                 goto finish;
1270         }
1271
1272         if (path_equal(arg_directory, "/")) {
1273                 log_error("Spawning container on root directory not supported.");
1274                 goto finish;
1275         }
1276
1277         if (path_is_os_tree(arg_directory) <= 0) {
1278                 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
1279                 goto finish;
1280         }
1281
1282         log_close();
1283         n_fd_passed = sd_listen_fds(false);
1284         if (n_fd_passed > 0) {
1285                 k = fdset_new_listen_fds(&fds, false);
1286                 if (k < 0) {
1287                         log_error("Failed to collect file descriptors: %s", strerror(-k));
1288                         goto finish;
1289                 }
1290         }
1291         fdset_close_others(fds);
1292         log_open();
1293
1294         k = cg_get_machine_path(&machine_root);
1295         if (k < 0) {
1296                 log_error("Failed to determine machine cgroup path: %s", strerror(-k));
1297                 goto finish;
1298         }
1299
1300         newcg = strjoin(machine_root, "/", arg_machine, NULL);
1301         if (!newcg) {
1302                 log_error("Failed to allocate cgroup path.");
1303                 goto finish;
1304         }
1305
1306         r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, newcg, false);
1307         if (r <= 0 && r != -ENOENT) {
1308                 log_error("Container already running.");
1309
1310                 free(newcg);
1311                 newcg = NULL;
1312
1313                 goto finish;
1314         }
1315
1316         master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
1317         if (master < 0) {
1318                 log_error("Failed to acquire pseudo tty: %m");
1319                 goto finish;
1320         }
1321
1322         console = ptsname(master);
1323         if (!console) {
1324                 log_error("Failed to determine tty name: %m");
1325                 goto finish;
1326         }
1327
1328         log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
1329
1330         if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
1331                 ioctl(master, TIOCSWINSZ, &ws);
1332
1333         if (unlockpt(master) < 0) {
1334                 log_error("Failed to unlock tty: %m");
1335                 goto finish;
1336         }
1337
1338         if (tcgetattr(STDIN_FILENO, &saved_attr) >= 0) {
1339                 saved_attr_valid = true;
1340
1341                 raw_attr = saved_attr;
1342                 cfmakeraw(&raw_attr);
1343                 raw_attr.c_lflag &= ~ECHO;
1344         }
1345
1346         if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_NONBLOCK|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
1347                 log_error("Failed to create kmsg socket pair.");
1348                 goto finish;
1349         }
1350
1351         assert_se(sigemptyset(&mask) == 0);
1352         sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
1353         assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
1354
1355         for (;;) {
1356                 siginfo_t status;
1357                 int pipefd[2], pipefd2[2];
1358
1359                 if (pipe2(pipefd, O_NONBLOCK|O_CLOEXEC) < 0) {
1360                         log_error("pipe2(): %m");
1361                         goto finish;
1362                 }
1363
1364                 if (pipe2(pipefd2, O_NONBLOCK|O_CLOEXEC) < 0) {
1365                         log_error("pipe2(): %m");
1366                         close_pipe(pipefd);
1367                         goto finish;
1368                 }
1369
1370                 pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS|(arg_private_network ? CLONE_NEWNET : 0), NULL);
1371                 if (pid < 0) {
1372                         if (errno == EINVAL)
1373                                 log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
1374                         else
1375                                 log_error("clone() failed: %m");
1376
1377                         goto finish;
1378                 }
1379
1380                 if (pid == 0) {
1381                         /* child */
1382                         const char *home = NULL;
1383                         uid_t uid = (uid_t) -1;
1384                         gid_t gid = (gid_t) -1;
1385                         unsigned n_env = 2;
1386                         const char *envp[] = {
1387                                 "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
1388                                 "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
1389                                 NULL, /* TERM */
1390                                 NULL, /* HOME */
1391                                 NULL, /* USER */
1392                                 NULL, /* LOGNAME */
1393                                 NULL, /* container_uuid */
1394                                 NULL, /* LISTEN_FDS */
1395                                 NULL, /* LISTEN_PID */
1396                                 NULL
1397                         };
1398
1399                         envp[n_env] = strv_find_prefix(environ, "TERM=");
1400                         if (envp[n_env])
1401                                 n_env ++;
1402
1403                         /* Wait for the parent process to log our PID */
1404                         close_nointr_nofail(pipefd[1]);
1405                         fd_wait_for_event(pipefd[0], POLLHUP, -1);
1406                         close_nointr_nofail(pipefd[0]);
1407
1408                         close_nointr_nofail(master);
1409                         master = -1;
1410
1411                         if (saved_attr_valid) {
1412                                 if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
1413                                         log_error("Failed to set terminal attributes: %m");
1414                                         goto child_fail;
1415                                 }
1416                         }
1417
1418                         close_nointr(STDIN_FILENO);
1419                         close_nointr(STDOUT_FILENO);
1420                         close_nointr(STDERR_FILENO);
1421
1422                         close_nointr_nofail(kmsg_socket_pair[0]);
1423                         kmsg_socket_pair[0] = -1;
1424
1425                         reset_all_signal_handlers();
1426
1427                         assert_se(sigemptyset(&mask) == 0);
1428                         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1429
1430                         k = open_terminal(console, O_RDWR);
1431                         if (k != STDIN_FILENO) {
1432                                 if (k >= 0) {
1433                                         close_nointr_nofail(k);
1434                                         k = -EINVAL;
1435                                 }
1436
1437                                 log_error("Failed to open console: %s", strerror(-k));
1438                                 goto child_fail;
1439                         }
1440
1441                         if (dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
1442                             dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO) {
1443                                 log_error("Failed to duplicate console: %m");
1444                                 goto child_fail;
1445                         }
1446
1447                         if (setsid() < 0) {
1448                                 log_error("setsid() failed: %m");
1449                                 goto child_fail;
1450                         }
1451
1452                         if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0) {
1453                                 log_error("PR_SET_PDEATHSIG failed: %m");
1454                                 goto child_fail;
1455                         }
1456
1457                         if (setup_cgroup(newcg) < 0)
1458                                 goto child_fail;
1459
1460                         close_nointr_nofail(pipefd2[1]);
1461                         close_nointr_nofail(pipefd2[0]);
1462
1463                         /* Mark everything as slave, so that we still
1464                          * receive mounts from the real root, but don't
1465                          * propagate mounts to the real root. */
1466                         if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
1467                                 log_error("MS_SLAVE|MS_REC failed: %m");
1468                                 goto child_fail;
1469                         }
1470
1471                         /* Turn directory into bind mount */
1472                         if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REC, NULL) < 0) {
1473                                 log_error("Failed to make bind mount.");
1474                                 goto child_fail;
1475                         }
1476
1477                         if (arg_read_only)
1478                                 if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY|MS_REC, NULL) < 0) {
1479                                         log_error("Failed to make read-only.");
1480                                         goto child_fail;
1481                                 }
1482
1483                         if (mount_all(arg_directory) < 0)
1484                                 goto child_fail;
1485
1486                         if (copy_devnodes(arg_directory) < 0)
1487                                 goto child_fail;
1488
1489                         if (setup_ptmx(arg_directory) < 0)
1490                                 goto child_fail;
1491
1492                         dev_setup(arg_directory);
1493
1494                         if (setup_dev_console(arg_directory, console) < 0)
1495                                 goto child_fail;
1496
1497                         if (setup_kmsg(arg_directory, kmsg_socket_pair[1]) < 0)
1498                                 goto child_fail;
1499
1500                         close_nointr_nofail(kmsg_socket_pair[1]);
1501                         kmsg_socket_pair[1] = -1;
1502
1503                         if (setup_boot_id(arg_directory) < 0)
1504                                 goto child_fail;
1505
1506                         if (setup_timezone(arg_directory) < 0)
1507                                 goto child_fail;
1508
1509                         if (setup_resolv_conf(arg_directory) < 0)
1510                                 goto child_fail;
1511
1512                         if (setup_journal(arg_directory) < 0)
1513                                 goto child_fail;
1514
1515                         if (mount_binds(arg_directory, arg_bind, 0) < 0)
1516                                 goto child_fail;
1517
1518                         if (mount_binds(arg_directory, arg_bind_ro, MS_RDONLY) < 0)
1519                                 goto child_fail;
1520
1521                         if (chdir(arg_directory) < 0) {
1522                                 log_error("chdir(%s) failed: %m", arg_directory);
1523                                 goto child_fail;
1524                         }
1525
1526                         if (mount(arg_directory, "/", NULL, MS_MOVE, NULL) < 0) {
1527                                 log_error("mount(MS_MOVE) failed: %m");
1528                                 goto child_fail;
1529                         }
1530
1531                         if (chroot(".") < 0) {
1532                                 log_error("chroot() failed: %m");
1533                                 goto child_fail;
1534                         }
1535
1536                         if (chdir("/") < 0) {
1537                                 log_error("chdir() failed: %m");
1538                                 goto child_fail;
1539                         }
1540
1541                         umask(0022);
1542
1543                         loopback_setup();
1544
1545                         if (drop_capabilities() < 0) {
1546                                 log_error("drop_capabilities() failed: %m");
1547                                 goto child_fail;
1548                         }
1549
1550                         if (arg_user) {
1551
1552                                 /* Note that this resolves user names
1553                                  * inside the container, and hence
1554                                  * accesses the NSS modules from the
1555                                  * container and not the host. This is
1556                                  * a bit weird... */
1557
1558                                 if (get_user_creds((const char**)&arg_user, &uid, &gid, &home, NULL) < 0) {
1559                                         log_error("get_user_creds() failed: %m");
1560                                         goto child_fail;
1561                                 }
1562
1563                                 if (mkdir_parents_label(home, 0775) < 0) {
1564                                         log_error("mkdir_parents_label() failed: %m");
1565                                         goto child_fail;
1566                                 }
1567
1568                                 if (mkdir_safe_label(home, 0775, uid, gid) < 0) {
1569                                         log_error("mkdir_safe_label() failed: %m");
1570                                         goto child_fail;
1571                                 }
1572
1573                                 if (initgroups((const char*)arg_user, gid) < 0) {
1574                                         log_error("initgroups() failed: %m");
1575                                         goto child_fail;
1576                                 }
1577
1578                                 if (setresgid(gid, gid, gid) < 0) {
1579                                         log_error("setregid() failed: %m");
1580                                         goto child_fail;
1581                                 }
1582
1583                                 if (setresuid(uid, uid, uid) < 0) {
1584                                         log_error("setreuid() failed: %m");
1585                                         goto child_fail;
1586                                 }
1587                         } else {
1588                                 /* Reset everything fully to 0, just in case */
1589
1590                                 if (setgroups(0, NULL) < 0) {
1591                                         log_error("setgroups() failed: %m");
1592                                         goto child_fail;
1593                                 }
1594
1595                                 if (setresgid(0, 0, 0) < 0) {
1596                                         log_error("setregid() failed: %m");
1597                                         goto child_fail;
1598                                 }
1599
1600                                 if (setresuid(0, 0, 0) < 0) {
1601                                         log_error("setreuid() failed: %m");
1602                                         goto child_fail;
1603                                 }
1604                         }
1605
1606                         if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
1607                             (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
1608                             (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)) {
1609                                 log_oom();
1610                                 goto child_fail;
1611                         }
1612
1613                         if (arg_uuid) {
1614                                 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", arg_uuid) < 0) {
1615                                         log_oom();
1616                                         goto child_fail;
1617                                 }
1618                         }
1619
1620                         if (fdset_size(fds) > 0) {
1621                                 k = fdset_cloexec(fds, false);
1622                                 if (k < 0) {
1623                                         log_error("Failed to unset O_CLOEXEC for file descriptors.");
1624                                         goto child_fail;
1625                                 }
1626
1627                                 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", n_fd_passed) < 0) ||
1628                                     (asprintf((char **)(envp + n_env++), "LISTEN_PID=%lu", (unsigned long) 1) < 0)) {
1629                                         log_oom();
1630                                         goto child_fail;
1631                                 }
1632                         }
1633
1634                         setup_hostname();
1635
1636                         if (arg_boot) {
1637                                 char **a;
1638                                 size_t l;
1639
1640                                 /* Automatically search for the init system */
1641
1642                                 l = 1 + argc - optind;
1643                                 a = newa(char*, l + 1);
1644                                 memcpy(a + 1, argv + optind, l * sizeof(char*));
1645
1646                                 a[0] = (char*) "/usr/lib/systemd/systemd";
1647                                 execve(a[0], a, (char**) envp);
1648
1649                                 a[0] = (char*) "/lib/systemd/systemd";
1650                                 execve(a[0], a, (char**) envp);
1651
1652                                 a[0] = (char*) "/sbin/init";
1653                                 execve(a[0], a, (char**) envp);
1654                         } else if (argc > optind)
1655                                 execvpe(argv[optind], argv + optind, (char**) envp);
1656                         else {
1657                                 chdir(home ? home : "/root");
1658                                 execle("/bin/bash", "-bash", NULL, (char**) envp);
1659                         }
1660
1661                         log_error("execv() failed: %m");
1662
1663                 child_fail:
1664                         _exit(EXIT_FAILURE);
1665                 }
1666
1667                 log_info("Init process in the container running as PID %lu.", (unsigned long) pid);
1668                 close_nointr_nofail(pipefd[0]);
1669                 close_nointr_nofail(pipefd[1]);
1670
1671                 /* Wait for the child process to establish cgroup hierarchy */
1672                 close_nointr_nofail(pipefd2[1]);
1673                 fd_wait_for_event(pipefd2[0], POLLHUP, -1);
1674                 close_nointr_nofail(pipefd2[0]);
1675
1676                 save_attributes(newcg, pid, arg_uuid, arg_directory);
1677
1678                 fdset_free(fds);
1679                 fds = NULL;
1680
1681                 if (process_pty(master, pid, &mask) < 0)
1682                         goto finish;
1683
1684                 if (saved_attr_valid)
1685                         tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
1686
1687                 r = wait_for_terminate(pid, &status);
1688                 if (r < 0) {
1689                         r = EXIT_FAILURE;
1690                         break;
1691                 }
1692
1693                 if (status.si_code == CLD_EXITED) {
1694                         if (status.si_status != 0) {
1695                                 log_error("Container failed with error code %i.", status.si_status);
1696                                 r = status.si_status;
1697                                 break;
1698                         }
1699
1700                         log_debug("Container exited successfully.");
1701                         break;
1702                 } else if (status.si_code == CLD_KILLED &&
1703                            status.si_status == SIGINT) {
1704                         log_info("Container has been shut down.");
1705                         r = 0;
1706                         break;
1707                 } else if (status.si_code == CLD_KILLED &&
1708                            status.si_status == SIGHUP) {
1709                         log_info("Container is being rebooted.");
1710                         continue;
1711                 } else if (status.si_code == CLD_KILLED ||
1712                            status.si_code == CLD_DUMPED) {
1713
1714                         log_error("Container terminated by signal %s.", signal_to_string(status.si_status));
1715                         r = EXIT_FAILURE;
1716                         break;
1717                 } else {
1718                         log_error("Container failed due to unknown reason.");
1719                         r = EXIT_FAILURE;
1720                         break;
1721                 }
1722         }
1723
1724 finish:
1725         if (saved_attr_valid)
1726                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
1727
1728         close_pipe(kmsg_socket_pair);
1729
1730         if (newcg)
1731                 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
1732
1733         free(arg_directory);
1734         free(arg_machine);
1735         strv_free(arg_controllers);
1736
1737         fdset_free(fds);
1738
1739         return r;
1740 }