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 <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 <termios.h>
  37 #include <sys/signalfd.h>
  38 #include <grp.h>
  39 #include <linux/fs.h>
  40 #include <sys/un.h>
  41 #include <sys/socket.h>
  42 #include <linux/netlink.h>
  43 #include <sys/eventfd.h>
  44 #include <net/if.h>
  45 #include <linux/veth.h>
  46 #include <sys/personality.h>
  47 #include <linux/loop.h>
  48
  49 #ifdef HAVE_SELINUX
  50 #include <selinux/selinux.h>
  51 #endif
  52
  53 #ifdef HAVE_SECCOMP
  54 #include <seccomp.h>
  55 #endif
  56
  57 #ifdef HAVE_BLKID
  58 #include <blkid/blkid.h>
  59 #endif
  60
  61 #include "sd-daemon.h"
  62 #include "sd-bus.h"
  63 #include "sd-id128.h"
  64 #include "sd-rtnl.h"
  65 #include "log.h"
  66 #include "util.h"
  67 #include "mkdir.h"
  68 #include "macro.h"
  69 #include "audit.h"
  70 #include "missing.h"
  71 #include "cgroup-util.h"
  72 #include "strv.h"
  73 #include "path-util.h"
  74 #include "loopback-setup.h"
  75 #include "dev-setup.h"
  76 #include "fdset.h"
  77 #include "build.h"
  78 #include "fileio.h"
  79 #include "bus-util.h"
  80 #include "bus-error.h"
  81 #include "ptyfwd.h"
  82 #include "bus-kernel.h"
  83 #include "env-util.h"
  84 #include "def.h"
  85 #include "rtnl-util.h"
  86 #include "udev-util.h"
  87 #include "blkid-util.h"
  88 #include "gpt.h"
  89 #include "siphash24.h"
  90
  91 #ifdef HAVE_SECCOMP
  92 #include "seccomp-util.h"
  93 #endif
  94
  95 typedef enum LinkJournal {
  96         LINK_NO,
  97         LINK_AUTO,
  98         LINK_HOST,
  99         LINK_GUEST
 100 } LinkJournal;
 101
 102 static char *arg_directory = NULL;
 103 static char *arg_user = NULL;
 104 static sd_id128_t arg_uuid = {};
 105 static char *arg_machine = NULL;
 106 static const char *arg_selinux_context = NULL;
 107 static const char *arg_selinux_apifs_context = NULL;
 108 static const char *arg_slice = NULL;
 109 static bool arg_private_network = false;
 110 static bool arg_read_only = false;
 111 static bool arg_boot = false;
 112 static LinkJournal arg_link_journal = LINK_AUTO;
 113 static uint64_t arg_retain =
 114         (1ULL << CAP_CHOWN) |
 115         (1ULL << CAP_DAC_OVERRIDE) |
 116         (1ULL << CAP_DAC_READ_SEARCH) |
 117         (1ULL << CAP_FOWNER) |
 118         (1ULL << CAP_FSETID) |
 119         (1ULL << CAP_IPC_OWNER) |
 120         (1ULL << CAP_KILL) |
 121         (1ULL << CAP_LEASE) |
 122         (1ULL << CAP_LINUX_IMMUTABLE) |
 123         (1ULL << CAP_NET_BIND_SERVICE) |
 124         (1ULL << CAP_NET_BROADCAST) |
 125         (1ULL << CAP_NET_RAW) |
 126         (1ULL << CAP_SETGID) |
 127         (1ULL << CAP_SETFCAP) |
 128         (1ULL << CAP_SETPCAP) |
 129         (1ULL << CAP_SETUID) |
 130         (1ULL << CAP_SYS_ADMIN) |
 131         (1ULL << CAP_SYS_CHROOT) |
 132         (1ULL << CAP_SYS_NICE) |
 133         (1ULL << CAP_SYS_PTRACE) |
 134         (1ULL << CAP_SYS_TTY_CONFIG) |
 135         (1ULL << CAP_SYS_RESOURCE) |
 136         (1ULL << CAP_SYS_BOOT) |
 137         (1ULL << CAP_AUDIT_WRITE) |
 138         (1ULL << CAP_AUDIT_CONTROL) |
 139         (1ULL << CAP_MKNOD);
 140 static char **arg_bind = NULL;
 141 static char **arg_bind_ro = NULL;
 142 static char **arg_setenv = NULL;
 143 static bool arg_quiet = false;
 144 static bool arg_share_system = false;
 145 static bool arg_register = true;
 146 static bool arg_keep_unit = false;
 147 static char **arg_network_interfaces = NULL;
 148 static char **arg_network_macvlan = NULL;
 149 static bool arg_network_veth = false;
 150 static const char *arg_network_bridge = NULL;
 151 static unsigned long arg_personality = 0xffffffffLU;
 152 static const char *arg_image = NULL;
 153
 154 static int help(void) {
 155
 156         printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
 157                "Spawn a minimal namespace container for debugging, testing and building.\n\n"
 158                "  -h --help                 Show this help\n"
 159                "     --version              Print version string\n"
 160                "  -q --quiet                Do not show status information\n"
 161                "  -D --directory=PATH       Root directory for the container\n"
 162                "  -i --image=PATH           File system device or image for the container\n"
 163                "  -b --boot                 Boot up full system (i.e. invoke init)\n"
 164                "  -u --user=USER            Run the command under specified user or uid\n"
 165                "  -M --machine=NAME         Set the machine name for the container\n"
 166                "     --uuid=UUID            Set a specific machine UUID for the container\n"
 167                "  -S --slice=SLICE          Place the container in the specified slice\n"
 168                "     --private-network      Disable network in container\n"
 169                "     --network-interface=INTERFACE\n"
 170                "                            Assign an existing network interface to the\n"
 171                "                            container\n"
 172                "     --network-macvlan=INTERFACE\n"
 173                "                            Create a macvlan network interface based on an\n"
 174                "                            existing network interface to the container\n"
 175                "     --network-veth         Add a virtual ethernet connection between host\n"
 176                "                            and container\n"
 177                "     --network-bridge=INTERFACE\n"
 178                "                            Add a virtual ethernet connection between host\n"
 179                "                            and container and add it to an existing bridge on\n"
 180                "                            the host\n"
 181                "  -Z --selinux-context=SECLABEL\n"
 182                "                            Set the SELinux security context to be used by\n"
 183                "                            processes in the container\n"
 184                "  -L --selinux-apifs-context=SECLABEL\n"
 185                "                            Set the SELinux security context to be used by\n"
 186                "                            API/tmpfs file systems in the container\n"
 187                "     --capability=CAP       In addition to the default, retain specified\n"
 188                "                            capability\n"
 189                "     --drop-capability=CAP  Drop the specified capability from the default set\n"
 190                "     --link-journal=MODE    Link up guest journal, one of no, auto, guest, host\n"
 191                "  -j                        Equivalent to --link-journal=host\n"
 192                "     --read-only            Mount the root directory read-only\n"
 193                "     --bind=PATH[:PATH]     Bind mount a file or directory from the host into\n"
 194                "                            the container\n"
 195                "     --bind-ro=PATH[:PATH]  Similar, but creates a read-only bind mount\n"
 196                "     --setenv=NAME=VALUE    Pass an environment variable to PID 1\n"
 197                "     --share-system         Share system namespaces with host\n"
 198                "     --register=BOOLEAN     Register container as machine\n"
 199                "     --keep-unit            Do not register a scope for the machine, reuse\n"
 200                "                            the service unit nspawn is running in\n",
 201                program_invocation_short_name);
 202
 203         return 0;
 204 }
 205
 206 static int parse_argv(int argc, char *argv[]) {
 207
 208         enum {
 209                 ARG_VERSION = 0x100,
 210                 ARG_PRIVATE_NETWORK,
 211                 ARG_UUID,
 212                 ARG_READ_ONLY,
 213                 ARG_CAPABILITY,
 214                 ARG_DROP_CAPABILITY,
 215                 ARG_LINK_JOURNAL,
 216                 ARG_BIND,
 217                 ARG_BIND_RO,
 218                 ARG_SETENV,
 219                 ARG_SHARE_SYSTEM,
 220                 ARG_REGISTER,
 221                 ARG_KEEP_UNIT,
 222                 ARG_NETWORK_INTERFACE,
 223                 ARG_NETWORK_MACVLAN,
 224                 ARG_NETWORK_VETH,
 225                 ARG_NETWORK_BRIDGE,
 226                 ARG_PERSONALITY,
 227         };
 228
 229         static const struct option options[] = {
 230                 { "help",                  no_argument,       NULL, 'h'                   },
 231                 { "version",               no_argument,       NULL, ARG_VERSION           },
 232                 { "directory",             required_argument, NULL, 'D'                   },
 233                 { "user",                  required_argument, NULL, 'u'                   },
 234                 { "private-network",       no_argument,       NULL, ARG_PRIVATE_NETWORK   },
 235                 { "boot",                  no_argument,       NULL, 'b'                   },
 236                 { "uuid",                  required_argument, NULL, ARG_UUID              },
 237                 { "read-only",             no_argument,       NULL, ARG_READ_ONLY         },
 238                 { "capability",            required_argument, NULL, ARG_CAPABILITY        },
 239                 { "drop-capability",       required_argument, NULL, ARG_DROP_CAPABILITY   },
 240                 { "link-journal",          required_argument, NULL, ARG_LINK_JOURNAL      },
 241                 { "bind",                  required_argument, NULL, ARG_BIND              },
 242                 { "bind-ro",               required_argument, NULL, ARG_BIND_RO           },
 243                 { "machine",               required_argument, NULL, 'M'                   },
 244                 { "slice",                 required_argument, NULL, 'S'                   },
 245                 { "setenv",                required_argument, NULL, ARG_SETENV            },
 246                 { "selinux-context",       required_argument, NULL, 'Z'                   },
 247                 { "selinux-apifs-context", required_argument, NULL, 'L'                   },
 248                 { "quiet",                 no_argument,       NULL, 'q'                   },
 249                 { "share-system",          no_argument,       NULL, ARG_SHARE_SYSTEM      },
 250                 { "register",              required_argument, NULL, ARG_REGISTER          },
 251                 { "keep-unit",             no_argument,       NULL, ARG_KEEP_UNIT         },
 252                 { "network-interface",     required_argument, NULL, ARG_NETWORK_INTERFACE },
 253                 { "network-macvlan",       required_argument, NULL, ARG_NETWORK_MACVLAN   },
 254                 { "network-veth",          no_argument,       NULL, ARG_NETWORK_VETH      },
 255                 { "network-bridge",        required_argument, NULL, ARG_NETWORK_BRIDGE    },
 256                 { "personality",           required_argument, NULL, ARG_PERSONALITY       },
 257                 { "image",                 required_argument, NULL, 'i'                   },
 258                 {}
 259         };
 260
 261         int c, r;
 262         uint64_t plus = 0, minus = 0;
 263
 264         assert(argc >= 0);
 265         assert(argv);
 266
 267         while ((c = getopt_long(argc, argv, "+hD:u:bL:M:jS:Z:qi:", options, NULL)) >= 0) {
 268
 269                 switch (c) {
 270
 271                 case 'h':
 272                         return help();
 273
 274                 case ARG_VERSION:
 275                         puts(PACKAGE_STRING);
 276                         puts(SYSTEMD_FEATURES);
 277                         return 0;
 278
 279                 case 'D':
 280                         free(arg_directory);
 281                         arg_directory = canonicalize_file_name(optarg);
 282                         if (!arg_directory) {
 283                                 log_error("Invalid root directory: %m");
 284                                 return -ENOMEM;
 285                         }
 286
 287                         break;
 288
 289                 case 'i':
 290                         arg_image = optarg;
 291                         break;
 292
 293                 case 'u':
 294                         free(arg_user);
 295                         arg_user = strdup(optarg);
 296                         if (!arg_user)
 297                                 return log_oom();
 298
 299                         break;
 300
 301                 case ARG_NETWORK_BRIDGE:
 302                         arg_network_bridge = optarg;
 303
 304                         /* fall through */
 305
 306                 case ARG_NETWORK_VETH:
 307                         arg_network_veth = true;
 308                         arg_private_network = true;
 309                         break;
 310
 311                 case ARG_NETWORK_INTERFACE:
 312                         if (strv_extend(&arg_network_interfaces, optarg) < 0)
 313                                 return log_oom();
 314
 315                         arg_private_network = true;
 316                         break;
 317
 318                 case ARG_NETWORK_MACVLAN:
 319                         if (strv_extend(&arg_network_macvlan, optarg) < 0)
 320                                 return log_oom();
 321
 322                         /* fall through */
 323
 324                 case ARG_PRIVATE_NETWORK:
 325                         arg_private_network = true;
 326                         break;
 327
 328                 case 'b':
 329                         arg_boot = true;
 330                         break;
 331
 332                 case ARG_UUID:
 333                         r = sd_id128_from_string(optarg, &arg_uuid);
 334                         if (r < 0) {
 335                                 log_error("Invalid UUID: %s", optarg);
 336                                 return r;
 337                         }
 338                         break;
 339
 340                 case 'S':
 341                         arg_slice = optarg;
 342                         break;
 343
 344                 case 'M':
 345                         if (isempty(optarg)) {
 346                                 free(arg_machine);
 347                                 arg_machine = NULL;
 348                         } else {
 349
 350                                 if (!hostname_is_valid(optarg)) {
 351                                         log_error("Invalid machine name: %s", optarg);
 352                                         return -EINVAL;
 353                                 }
 354
 355                                 free(arg_machine);
 356                                 arg_machine = strdup(optarg);
 357                                 if (!arg_machine)
 358                                         return log_oom();
 359
 360                                 break;
 361                         }
 362
 363                 case 'Z':
 364                         arg_selinux_context = optarg;
 365                         break;
 366
 367                 case 'L':
 368                         arg_selinux_apifs_context = optarg;
 369                         break;
 370
 371                 case ARG_READ_ONLY:
 372                         arg_read_only = true;
 373                         break;
 374
 375                 case ARG_CAPABILITY:
 376                 case ARG_DROP_CAPABILITY: {
 377                         char *state, *word;
 378                         size_t length;
 379
 380                         FOREACH_WORD_SEPARATOR(word, length, optarg, ",", state) {
 381                                 _cleanup_free_ char *t;
 382                                 cap_value_t cap;
 383
 384                                 t = strndup(word, length);
 385                                 if (!t)
 386                                         return log_oom();
 387
 388                                 if (streq(t, "all")) {
 389                                         if (c == ARG_CAPABILITY)
 390                                                 plus = (uint64_t) -1;
 391                                         else
 392                                                 minus = (uint64_t) -1;
 393                                 } else {
 394                                         if (cap_from_name(t, &cap) < 0) {
 395                                                 log_error("Failed to parse capability %s.", t);
 396                                                 return -EINVAL;
 397                                         }
 398
 399                                         if (c == ARG_CAPABILITY)
 400                                                 plus |= 1ULL << (uint64_t) cap;
 401                                         else
 402                                                 minus |= 1ULL << (uint64_t) cap;
 403                                 }
 404                         }
 405
 406                         break;
 407                 }
 408
 409                 case 'j':
 410                         arg_link_journal = LINK_GUEST;
 411                         break;
 412
 413                 case ARG_LINK_JOURNAL:
 414                         if (streq(optarg, "auto"))
 415                                 arg_link_journal = LINK_AUTO;
 416                         else if (streq(optarg, "no"))
 417                                 arg_link_journal = LINK_NO;
 418                         else if (streq(optarg, "guest"))
 419                                 arg_link_journal = LINK_GUEST;
 420                         else if (streq(optarg, "host"))
 421                                 arg_link_journal = LINK_HOST;
 422                         else {
 423                                 log_error("Failed to parse link journal mode %s", optarg);
 424                                 return -EINVAL;
 425                         }
 426
 427                         break;
 428
 429                 case ARG_BIND:
 430                 case ARG_BIND_RO: {
 431                         _cleanup_free_ char *a = NULL, *b = NULL;
 432                         char *e;
 433                         char ***x;
 434
 435                         x = c == ARG_BIND ? &arg_bind : &arg_bind_ro;
 436
 437                         e = strchr(optarg, ':');
 438                         if (e) {
 439                                 a = strndup(optarg, e - optarg);
 440                                 b = strdup(e + 1);
 441                         } else {
 442                                 a = strdup(optarg);
 443                                 b = strdup(optarg);
 444                         }
 445
 446                         if (!a || !b)
 447                                 return log_oom();
 448
 449                         if (!path_is_absolute(a) || !path_is_absolute(b)) {
 450                                 log_error("Invalid bind mount specification: %s", optarg);
 451                                 return -EINVAL;
 452                         }
 453
 454                         r = strv_extend(x, a);
 455                         if (r < 0)
 456                                 return log_oom();
 457
 458                         r = strv_extend(x, b);
 459                         if (r < 0)
 460                                 return log_oom();
 461
 462                         break;
 463                 }
 464
 465                 case ARG_SETENV: {
 466                         char **n;
 467
 468                         if (!env_assignment_is_valid(optarg)) {
 469                                 log_error("Environment variable assignment '%s' is not valid.", optarg);
 470                                 return -EINVAL;
 471                         }
 472
 473                         n = strv_env_set(arg_setenv, optarg);
 474                         if (!n)
 475                                 return log_oom();
 476
 477                         strv_free(arg_setenv);
 478                         arg_setenv = n;
 479                         break;
 480                 }
 481
 482                 case 'q':
 483                         arg_quiet = true;
 484                         break;
 485
 486                 case ARG_SHARE_SYSTEM:
 487                         arg_share_system = true;
 488                         break;
 489
 490                 case ARG_REGISTER:
 491                         r = parse_boolean(optarg);
 492                         if (r < 0) {
 493                                 log_error("Failed to parse --register= argument: %s", optarg);
 494                                 return r;
 495                         }
 496
 497                         arg_register = r;
 498                         break;
 499
 500                 case ARG_KEEP_UNIT:
 501                         arg_keep_unit = true;
 502                         break;
 503
 504                 case ARG_PERSONALITY:
 505
 506                         arg_personality = personality_from_string(optarg);
 507                         if (arg_personality == 0xffffffffLU) {
 508                                 log_error("Unknown or unsupported personality '%s'.", optarg);
 509                                 return -EINVAL;
 510                         }
 511
 512                         break;
 513
 514                 case '?':
 515                         return -EINVAL;
 516
 517                 default:
 518                         assert_not_reached("Unhandled option");
 519                 }
 520         }
 521
 522         if (arg_share_system)
 523                 arg_register = false;
 524
 525         if (arg_boot && arg_share_system) {
 526                 log_error("--boot and --share-system may not be combined.");
 527                 return -EINVAL;
 528         }
 529
 530         if (arg_keep_unit && cg_pid_get_owner_uid(0, NULL) >= 0) {
 531                 log_error("--keep-unit may not be used when invoked from a user session.");
 532                 return -EINVAL;
 533         }
 534
 535         if (arg_directory && arg_image) {
 536                 log_error("--directory= and --image= may not be combined.");
 537                 return -EINVAL;
 538         }
 539
 540         arg_retain = (arg_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
 541
 542         return 1;
 543 }
 544
 545 static int mount_all(const char *dest) {
 546
 547         typedef struct MountPoint {
 548                 const char *what;
 549                 const char *where;
 550                 const char *type;
 551                 const char *options;
 552                 unsigned long flags;
 553                 bool fatal;
 554         } MountPoint;
 555
 556         static const MountPoint mount_table[] = {
 557                 { "proc",      "/proc",     "proc",  NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 558                 { "/proc/sys", "/proc/sys", NULL,    NULL,       MS_BIND, true                       },   /* Bind mount first */
 559                 { NULL,        "/proc/sys", NULL,    NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 560                 { "sysfs",     "/sys",      "sysfs", NULL,       MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 561                 { "tmpfs",     "/dev",      "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,     true  },
 562                 { "devpts",    "/dev/pts",  "devpts","newinstance,ptmxmode=0666,mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC, true },
 563                 { "tmpfs",     "/dev/shm",  "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true  },
 564                 { "tmpfs",     "/run",      "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true  },
 565 #ifdef HAVE_SELINUX
 566                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND,                      false },  /* Bind mount first */
 567                 { NULL,              "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false },  /* Then, make it r/o */
 568 #endif
 569         };
 570
 571         unsigned k;
 572         int r = 0;
 573
 574         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 575                 _cleanup_free_ char *where = NULL;
 576 #ifdef HAVE_SELINUX
 577                 _cleanup_free_ char *options = NULL;
 578 #endif
 579                 const char *o;
 580                 int t;
 581
 582                 where = strjoin(dest, "/", mount_table[k].where, NULL);
 583                 if (!where)
 584                         return log_oom();
 585
 586                 t = path_is_mount_point(where, true);
 587                 if (t < 0) {
 588                         log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
 589
 590                         if (r == 0)
 591                                 r = t;
 592
 593                         continue;
 594                 }
 595
 596                 /* Skip this entry if it is not a remount. */
 597                 if (mount_table[k].what && t > 0)
 598                         continue;
 599
 600                 mkdir_p(where, 0755);
 601
 602 #ifdef HAVE_SELINUX
 603                 if (arg_selinux_apifs_context &&
 604                     (streq_ptr(mount_table[k].what, "tmpfs") || streq_ptr(mount_table[k].what, "devpts"))) {
 605                         options = strjoin(mount_table[k].options, ",context=\"", arg_selinux_apifs_context, "\"", NULL);
 606                         if (!options)
 607                                 return log_oom();
 608
 609                         o = options;
 610                 } else
 611 #endif
 612                         o = mount_table[k].options;
 613
 614
 615                 if (mount(mount_table[k].what,
 616                           where,
 617                           mount_table[k].type,
 618                           mount_table[k].flags,
 619                           o) < 0 &&
 620                     mount_table[k].fatal) {
 621
 622                         log_error("mount(%s) failed: %m", where);
 623
 624                         if (r == 0)
 625                                 r = -errno;
 626                 }
 627         }
 628
 629         return r;
 630 }
 631
 632 static int mount_binds(const char *dest, char **l, unsigned long flags) {
 633         char **x, **y;
 634
 635         STRV_FOREACH_PAIR(x, y, l) {
 636                 char *where;
 637                 struct stat source_st, dest_st;
 638                 int r;
 639
 640                 if (stat(*x, &source_st) < 0) {
 641                         log_error("Failed to stat %s: %m", *x);
 642                         return -errno;
 643                 }
 644
 645                 where = strappenda(dest, *y);
 646                 r = stat(where, &dest_st);
 647                 if (r == 0) {
 648                         if ((source_st.st_mode & S_IFMT) != (dest_st.st_mode & S_IFMT)) {
 649                                 log_error("The file types of %s and %s do not match. Refusing bind mount",
 650                                                 *x, where);
 651                                 return -EINVAL;
 652                         }
 653                 } else if (errno == ENOENT) {
 654                         r = mkdir_parents_label(where, 0755);
 655                         if (r < 0) {
 656                                 log_error("Failed to bind mount %s: %s", *x, strerror(-r));
 657                                 return r;
 658                         }
 659                 } else {
 660                         log_error("Failed to bind mount %s: %s", *x, strerror(errno));
 661                         return -errno;
 662                 }
 663                 /* Create the mount point, but be conservative -- refuse to create block
 664                 * and char devices. */
 665                 if (S_ISDIR(source_st.st_mode))
 666                         mkdir_label(where, 0755);
 667                 else if (S_ISFIFO(source_st.st_mode))
 668                         mkfifo(where, 0644);
 669                 else if (S_ISSOCK(source_st.st_mode))
 670                         mknod(where, 0644 | S_IFSOCK, 0);
 671                 else if (S_ISREG(source_st.st_mode))
 672                         touch(where);
 673                 else {
 674                         log_error("Refusing to create mountpoint for file: %s", *x);
 675                         return -ENOTSUP;
 676                 }
 677
 678                 if (mount(*x, where, "bind", MS_BIND, NULL) < 0) {
 679                         log_error("mount(%s) failed: %m", where);
 680                         return -errno;
 681                 }
 682
 683                 if (flags && mount(NULL, where, NULL, MS_REMOUNT|MS_BIND|flags, NULL) < 0) {
 684                         log_error("mount(%s) failed: %m", where);
 685                         return -errno;
 686                 }
 687         }
 688
 689         return 0;
 690 }
 691
 692 static int setup_timezone(const char *dest) {
 693         _cleanup_free_ char *where = NULL, *p = NULL, *q = NULL, *check = NULL, *what = NULL;
 694         char *z, *y;
 695         int r;
 696
 697         assert(dest);
 698
 699         /* Fix the timezone, if possible */
 700         r = readlink_malloc("/etc/localtime", &p);
 701         if (r < 0) {
 702                 log_warning("/etc/localtime is not a symlink, not updating container timezone.");
 703                 return 0;
 704         }
 705
 706         z = path_startswith(p, "../usr/share/zoneinfo/");
 707         if (!z)
 708                 z = path_startswith(p, "/usr/share/zoneinfo/");
 709         if (!z) {
 710                 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
 711                 return 0;
 712         }
 713
 714         where = strappend(dest, "/etc/localtime");
 715         if (!where)
 716                 return log_oom();
 717
 718         r = readlink_malloc(where, &q);
 719         if (r >= 0) {
 720                 y = path_startswith(q, "../usr/share/zoneinfo/");
 721                 if (!y)
 722                         y = path_startswith(q, "/usr/share/zoneinfo/");
 723
 724
 725                 /* Already pointing to the right place? Then do nothing .. */
 726                 if (y && streq(y, z))
 727                         return 0;
 728         }
 729
 730         check = strjoin(dest, "/usr/share/zoneinfo/", z, NULL);
 731         if (!check)
 732                 return log_oom();
 733
 734         if (access(check, F_OK) < 0) {
 735                 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
 736                 return 0;
 737         }
 738
 739         what = strappend("../usr/share/zoneinfo/", z);
 740         if (!what)
 741                 return log_oom();
 742
 743         unlink(where);
 744         if (symlink(what, where) < 0) {
 745                 log_error("Failed to correct timezone of container: %m");
 746                 return 0;
 747         }
 748
 749         return 0;
 750 }
 751
 752 static int setup_resolv_conf(const char *dest) {
 753         char _cleanup_free_ *where = NULL;
 754
 755         assert(dest);
 756
 757         if (arg_private_network)
 758                 return 0;
 759
 760         /* Fix resolv.conf, if possible */
 761         where = strappend(dest, "/etc/resolv.conf");
 762         if (!where)
 763                 return log_oom();
 764
 765         /* We don't really care for the results of this really. If it
 766          * fails, it fails, but meh... */
 767         copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW);
 768
 769         return 0;
 770 }
 771
 772 static int setup_boot_id(const char *dest) {
 773         _cleanup_free_ char *from = NULL, *to = NULL;
 774         sd_id128_t rnd = {};
 775         char as_uuid[37];
 776         int r;
 777
 778         assert(dest);
 779
 780         if (arg_share_system)
 781                 return 0;
 782
 783         /* Generate a new randomized boot ID, so that each boot-up of
 784          * the container gets a new one */
 785
 786         from = strappend(dest, "/dev/proc-sys-kernel-random-boot-id");
 787         to = strappend(dest, "/proc/sys/kernel/random/boot_id");
 788         if (!from || !to)
 789                 return log_oom();
 790
 791         r = sd_id128_randomize(&rnd);
 792         if (r < 0) {
 793                 log_error("Failed to generate random boot id: %s", strerror(-r));
 794                 return r;
 795         }
 796
 797         snprintf(as_uuid, sizeof(as_uuid),
 798                  "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
 799                  SD_ID128_FORMAT_VAL(rnd));
 800         char_array_0(as_uuid);
 801
 802         r = write_string_file(from, as_uuid);
 803         if (r < 0) {
 804                 log_error("Failed to write boot id: %s", strerror(-r));
 805                 return r;
 806         }
 807
 808         if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
 809                 log_error("Failed to bind mount boot id: %m");
 810                 r = -errno;
 811         } else if (mount(from, to, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL))
 812                 log_warning("Failed to make boot id read-only: %m");
 813
 814         unlink(from);
 815         return r;
 816 }
 817
 818 static int copy_devnodes(const char *dest) {
 819
 820         static const char devnodes[] =
 821                 "null\0"
 822                 "zero\0"
 823                 "full\0"
 824                 "random\0"
 825                 "urandom\0"
 826                 "tty\0";
 827
 828         const char *d;
 829         int r = 0;
 830         _cleanup_umask_ mode_t u;
 831
 832         assert(dest);
 833
 834         u = umask(0000);
 835
 836         NULSTR_FOREACH(d, devnodes) {
 837                 _cleanup_free_ char *from = NULL, *to = NULL;
 838                 struct stat st;
 839
 840                 from = strappend("/dev/", d);
 841                 to = strjoin(dest, "/dev/", d, NULL);
 842                 if (!from || !to)
 843                         return log_oom();
 844
 845                 if (stat(from, &st) < 0) {
 846
 847                         if (errno != ENOENT) {
 848                                 log_error("Failed to stat %s: %m", from);
 849                                 return -errno;
 850                         }
 851
 852                 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
 853
 854                         log_error("%s is not a char or block device, cannot copy", from);
 855                         return -EIO;
 856
 857                 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
 858
 859                         log_error("mknod(%s) failed: %m", dest);
 860                         return  -errno;
 861                 }
 862         }
 863
 864         return r;
 865 }
 866
 867 static int setup_ptmx(const char *dest) {
 868         _cleanup_free_ char *p = NULL;
 869
 870         p = strappend(dest, "/dev/ptmx");
 871         if (!p)
 872                 return log_oom();
 873
 874         if (symlink("pts/ptmx", p) < 0) {
 875                 log_error("Failed to create /dev/ptmx symlink: %m");
 876                 return -errno;
 877         }
 878
 879         return 0;
 880 }
 881
 882 static int setup_dev_console(const char *dest, const char *console) {
 883         _cleanup_umask_ mode_t u;
 884         const char *to;
 885         struct stat st;
 886         int r;
 887
 888         assert(dest);
 889         assert(console);
 890
 891         u = umask(0000);
 892
 893         if (stat("/dev/null", &st) < 0) {
 894                 log_error("Failed to stat /dev/null: %m");
 895                 return -errno;
 896         }
 897
 898         r = chmod_and_chown(console, 0600, 0, 0);
 899         if (r < 0) {
 900                 log_error("Failed to correct access mode for TTY: %s", strerror(-r));
 901                 return r;
 902         }
 903
 904         /* We need to bind mount the right tty to /dev/console since
 905          * ptys can only exist on pts file systems. To have something
 906          * to bind mount things on we create a device node first, and
 907          * use /dev/null for that since we the cgroups device policy
 908          * allows us to create that freely, while we cannot create
 909          * /dev/console. (Note that the major minor doesn't actually
 910          * matter here, since we mount it over anyway). */
 911
 912         to = strappenda(dest, "/dev/console");
 913         if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) {
 914                 log_error("mknod() for /dev/console failed: %m");
 915                 return -errno;
 916         }
 917
 918         if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
 919                 log_error("Bind mount for /dev/console failed: %m");
 920                 return -errno;
 921         }
 922
 923         return 0;
 924 }
 925
 926 static int setup_kmsg(const char *dest, int kmsg_socket) {
 927         _cleanup_free_ char *from = NULL, *to = NULL;
 928         int r, fd, k;
 929         _cleanup_umask_ mode_t u;
 930         union {
 931                 struct cmsghdr cmsghdr;
 932                 uint8_t buf[CMSG_SPACE(sizeof(int))];
 933         } control = {};
 934         struct msghdr mh = {
 935                 .msg_control = &control,
 936                 .msg_controllen = sizeof(control),
 937         };
 938         struct cmsghdr *cmsg;
 939
 940         assert(dest);
 941         assert(kmsg_socket >= 0);
 942
 943         u = umask(0000);
 944
 945         /* We create the kmsg FIFO as /dev/kmsg, but immediately
 946          * delete it after bind mounting it to /proc/kmsg. While FIFOs
 947          * on the reading side behave very similar to /proc/kmsg,
 948          * their writing side behaves differently from /dev/kmsg in
 949          * that writing blocks when nothing is reading. In order to
 950          * avoid any problems with containers deadlocking due to this
 951          * we simply make /dev/kmsg unavailable to the container. */
 952         if (asprintf(&from, "%s/dev/kmsg", dest) < 0 ||
 953             asprintf(&to, "%s/proc/kmsg", dest) < 0)
 954                 return log_oom();
 955
 956         if (mkfifo(from, 0600) < 0) {
 957                 log_error("mkfifo() for /dev/kmsg failed: %m");
 958                 return -errno;
 959         }
 960
 961         r = chmod_and_chown(from, 0600, 0, 0);
 962         if (r < 0) {
 963                 log_error("Failed to correct access mode for /dev/kmsg: %s", strerror(-r));
 964                 return r;
 965         }
 966
 967         if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
 968                 log_error("Bind mount for /proc/kmsg failed: %m");
 969                 return -errno;
 970         }
 971
 972         fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
 973         if (fd < 0) {
 974                 log_error("Failed to open fifo: %m");
 975                 return -errno;
 976         }
 977
 978         cmsg = CMSG_FIRSTHDR(&mh);
 979         cmsg->cmsg_level = SOL_SOCKET;
 980         cmsg->cmsg_type = SCM_RIGHTS;
 981         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
 982         memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
 983
 984         mh.msg_controllen = cmsg->cmsg_len;
 985
 986         /* Store away the fd in the socket, so that it stays open as
 987          * long as we run the child */
 988         k = sendmsg(kmsg_socket, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
 989         close_nointr_nofail(fd);
 990
 991         if (k < 0) {
 992                 log_error("Failed to send FIFO fd: %m");
 993                 return -errno;
 994         }
 995
 996         /* And now make the FIFO unavailable as /dev/kmsg... */
 997         unlink(from);
 998         return 0;
 999 }
1000
1001 static int setup_hostname(void) {
1002
1003         if (arg_share_system)
1004                 return 0;
1005
1006         if (sethostname(arg_machine, strlen(arg_machine)) < 0)
1007                 return -errno;
1008
1009         return 0;
1010 }
1011
1012 static int setup_journal(const char *directory) {
1013         sd_id128_t machine_id, this_id;
1014         _cleanup_free_ char *p = NULL, *b = NULL, *q = NULL, *d = NULL;
1015         char *id;
1016         int r;
1017
1018         p = strappend(directory, "/etc/machine-id");
1019         if (!p)
1020                 return log_oom();
1021
1022         r = read_one_line_file(p, &b);
1023         if (r == -ENOENT && arg_link_journal == LINK_AUTO)
1024                 return 0;
1025         else if (r < 0) {
1026                 log_error("Failed to read machine ID from %s: %s", p, strerror(-r));
1027                 return r;
1028         }
1029
1030         id = strstrip(b);
1031         if (isempty(id) && arg_link_journal == LINK_AUTO)
1032                 return 0;
1033
1034         /* Verify validity */
1035         r = sd_id128_from_string(id, &machine_id);
1036         if (r < 0) {
1037                 log_error("Failed to parse machine ID from %s: %s", p, strerror(-r));
1038                 return r;
1039         }
1040
1041         r = sd_id128_get_machine(&this_id);
1042         if (r < 0) {
1043                 log_error("Failed to retrieve machine ID: %s", strerror(-r));
1044                 return r;
1045         }
1046
1047         if (sd_id128_equal(machine_id, this_id)) {
1048                 log_full(arg_link_journal == LINK_AUTO ? LOG_WARNING : LOG_ERR,
1049                          "Host and machine ids are equal (%s): refusing to link journals", id);
1050                 if (arg_link_journal == LINK_AUTO)
1051                         return 0;
1052                 return
1053                         -EEXIST;
1054         }
1055
1056         if (arg_link_journal == LINK_NO)
1057                 return 0;
1058
1059         free(p);
1060         p = strappend("/var/log/journal/", id);
1061         q = strjoin(directory, "/var/log/journal/", id, NULL);
1062         if (!p || !q)
1063                 return log_oom();
1064
1065         if (path_is_mount_point(p, false) > 0) {
1066                 if (arg_link_journal != LINK_AUTO) {
1067                         log_error("%s: already a mount point, refusing to use for journal", p);
1068                         return -EEXIST;
1069                 }
1070
1071                 return 0;
1072         }
1073
1074         if (path_is_mount_point(q, false) > 0) {
1075                 if (arg_link_journal != LINK_AUTO) {
1076                         log_error("%s: already a mount point, refusing to use for journal", q);
1077                         return -EEXIST;
1078                 }
1079
1080                 return 0;
1081         }
1082
1083         r = readlink_and_make_absolute(p, &d);
1084         if (r >= 0) {
1085                 if ((arg_link_journal == LINK_GUEST ||
1086                      arg_link_journal == LINK_AUTO) &&
1087                     path_equal(d, q)) {
1088
1089                         r = mkdir_p(q, 0755);
1090                         if (r < 0)
1091                                 log_warning("failed to create directory %s: %m", q);
1092                         return 0;
1093                 }
1094
1095                 if (unlink(p) < 0) {
1096                         log_error("Failed to remove symlink %s: %m", p);
1097                         return -errno;
1098                 }
1099         } else if (r == -EINVAL) {
1100
1101                 if (arg_link_journal == LINK_GUEST &&
1102                     rmdir(p) < 0) {
1103
1104                         if (errno == ENOTDIR) {
1105                                 log_error("%s already exists and is neither a symlink nor a directory", p);
1106                                 return r;
1107                         } else {
1108                                 log_error("Failed to remove %s: %m", p);
1109                                 return -errno;
1110                         }
1111                 }
1112         } else if (r != -ENOENT) {
1113                 log_error("readlink(%s) failed: %m", p);
1114                 return r;
1115         }
1116
1117         if (arg_link_journal == LINK_GUEST) {
1118
1119                 if (symlink(q, p) < 0) {
1120                         log_error("Failed to symlink %s to %s: %m", q, p);
1121                         return -errno;
1122                 }
1123
1124                 r = mkdir_p(q, 0755);
1125                 if (r < 0)
1126                         log_warning("failed to create directory %s: %m", q);
1127                 return 0;
1128         }
1129
1130         if (arg_link_journal == LINK_HOST) {
1131                 r = mkdir_p(p, 0755);
1132                 if (r < 0) {
1133                         log_error("Failed to create %s: %m", p);
1134                         return r;
1135                 }
1136
1137         } else if (access(p, F_OK) < 0)
1138                 return 0;
1139
1140         if (dir_is_empty(q) == 0) {
1141                 log_error("%s not empty.", q);
1142                 return -ENOTEMPTY;
1143         }
1144
1145         r = mkdir_p(q, 0755);
1146         if (r < 0) {
1147                 log_error("Failed to create %s: %m", q);
1148                 return r;
1149         }
1150
1151         if (mount(p, q, "bind", MS_BIND, NULL) < 0) {
1152                 log_error("Failed to bind mount journal from host into guest: %m");
1153                 return -errno;
1154         }
1155
1156         return 0;
1157 }
1158
1159 static int setup_kdbus(const char *dest, const char *path) {
1160         const char *p;
1161
1162         if (!path)
1163                 return 0;
1164
1165         p = strappenda(dest, "/dev/kdbus");
1166         if (mkdir(p, 0755) < 0) {
1167                 log_error("Failed to create kdbus path: %m");
1168                 return  -errno;
1169         }
1170
1171         if (mount(path, p, "bind", MS_BIND, NULL) < 0) {
1172                 log_error("Failed to mount kdbus domain path: %m");
1173                 return -errno;
1174         }
1175
1176         return 0;
1177 }
1178
1179 static int drop_capabilities(void) {
1180         return capability_bounding_set_drop(~arg_retain, false);
1181 }
1182
1183 static int register_machine(pid_t pid) {
1184         _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1185         _cleanup_bus_unref_ sd_bus *bus = NULL;
1186         int r;
1187
1188         if (!arg_register)
1189                 return 0;
1190
1191         r = sd_bus_default_system(&bus);
1192         if (r < 0) {
1193                 log_error("Failed to open system bus: %s", strerror(-r));
1194                 return r;
1195         }
1196
1197         if (arg_keep_unit) {
1198                 r = sd_bus_call_method(
1199                                 bus,
1200                                 "org.freedesktop.machine1",
1201                                 "/org/freedesktop/machine1",
1202                                 "org.freedesktop.machine1.Manager",
1203                                 "RegisterMachine",
1204                                 &error,
1205                                 NULL,
1206                                 "sayssus",
1207                                 arg_machine,
1208                                 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
1209                                 "nspawn",
1210                                 "container",
1211                                 (uint32_t) pid,
1212                                 strempty(arg_directory));
1213         } else {
1214                 _cleanup_bus_message_unref_ sd_bus_message *m = NULL;
1215
1216                 r = sd_bus_message_new_method_call(
1217                                 bus,
1218                                 &m,
1219                                 "org.freedesktop.machine1",
1220                                 "/org/freedesktop/machine1",
1221                                 "org.freedesktop.machine1.Manager",
1222                                 "CreateMachine");
1223                 if (r < 0) {
1224                         log_error("Failed to create message: %s", strerror(-r));
1225                         return r;
1226                 }
1227
1228                 r = sd_bus_message_append(
1229                                 m,
1230                                 "sayssus",
1231                                 arg_machine,
1232                                 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
1233                                 "nspawn",
1234                                 "container",
1235                                 (uint32_t) pid,
1236                                 strempty(arg_directory));
1237                 if (r < 0) {
1238                         log_error("Failed to append message arguments: %s", strerror(-r));
1239                         return r;
1240                 }
1241
1242                 r = sd_bus_message_open_container(m, 'a', "(sv)");
1243                 if (r < 0) {
1244                         log_error("Failed to open container: %s", strerror(-r));
1245                         return r;
1246                 }
1247
1248                 if (!isempty(arg_slice)) {
1249                         r = sd_bus_message_append(m, "(sv)", "Slice", "s", arg_slice);
1250                         if (r < 0) {
1251                                 log_error("Failed to append slice: %s", strerror(-r));
1252                                 return r;
1253                         }
1254                 }
1255
1256                 r = sd_bus_message_append(m, "(sv)", "DevicePolicy", "s", "strict");
1257                 if (r < 0) {
1258                         log_error("Failed to add device policy: %s", strerror(-r));
1259                         return r;
1260                 }
1261
1262                 r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 10,
1263                                           /* Allow the container to
1264                                            * access and create the API
1265                                            * device nodes, so that
1266                                            * PrivateDevices= in the
1267                                            * container can work
1268                                            * fine */
1269                                           "/dev/null", "rwm",
1270                                           "/dev/zero", "rwm",
1271                                           "/dev/full", "rwm",
1272                                           "/dev/random", "rwm",
1273                                           "/dev/urandom", "rwm",
1274                                           "/dev/tty", "rwm",
1275                                           /* Allow the container
1276                                            * access to ptys. However,
1277                                            * do not permit the
1278                                            * container to ever create
1279                                            * these device nodes. */
1280                                           "/dev/pts/ptmx", "rw",
1281                                           "char-pts", "rw",
1282                                           /* Allow the container
1283                                            * access to all kdbus
1284                                            * devices. Again, the
1285                                            * container cannot create
1286                                            * these nodes, only use
1287                                            * them. We use a pretty
1288                                            * open match here, so that
1289                                            * the kernel API can still
1290                                            * change. */
1291                                           "char-kdbus", "rw",
1292                                           "char-kdbus/*", "rw");
1293                 if (r < 0) {
1294                         log_error("Failed to add device whitelist: %s", strerror(-r));
1295                         return r;
1296                 }
1297
1298                 r = sd_bus_message_close_container(m);
1299                 if (r < 0) {
1300                         log_error("Failed to close container: %s", strerror(-r));
1301                         return r;
1302                 }
1303
1304                 r = sd_bus_call(bus, m, 0, &error, NULL);
1305         }
1306
1307         if (r < 0) {
1308                 log_error("Failed to register machine: %s", bus_error_message(&error, r));
1309                 return r;
1310         }
1311
1312         return 0;
1313 }
1314
1315 static int terminate_machine(pid_t pid) {
1316         _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1317         _cleanup_bus_message_unref_ sd_bus_message *reply = NULL;
1318         _cleanup_bus_unref_ sd_bus *bus = NULL;
1319         const char *path;
1320         int r;
1321
1322         if (!arg_register)
1323                 return 0;
1324
1325         r = sd_bus_default_system(&bus);
1326         if (r < 0) {
1327                 log_error("Failed to open system bus: %s", strerror(-r));
1328                 return r;
1329         }
1330
1331         r = sd_bus_call_method(
1332                         bus,
1333                         "org.freedesktop.machine1",
1334                         "/org/freedesktop/machine1",
1335                         "org.freedesktop.machine1.Manager",
1336                         "GetMachineByPID",
1337                         &error,
1338                         &reply,
1339                         "u",
1340                         (uint32_t) pid);
1341         if (r < 0) {
1342                 /* Note that the machine might already have been
1343                  * cleaned up automatically, hence don't consider it a
1344                  * failure if we cannot get the machine object. */
1345                 log_debug("Failed to get machine: %s", bus_error_message(&error, r));
1346                 return 0;
1347         }
1348
1349         r = sd_bus_message_read(reply, "o", &path);
1350         if (r < 0)
1351                 return bus_log_parse_error(r);
1352
1353         r = sd_bus_call_method(
1354                         bus,
1355                         "org.freedesktop.machine1",
1356                         path,
1357                         "org.freedesktop.machine1.Machine",
1358                         "Terminate",
1359                         &error,
1360                         NULL,
1361                         NULL);
1362         if (r < 0) {
1363                 log_debug("Failed to terminate machine: %s", bus_error_message(&error, r));
1364                 return 0;
1365         }
1366
1367         return 0;
1368 }
1369
1370 static int reset_audit_loginuid(void) {
1371         _cleanup_free_ char *p = NULL;
1372         int r;
1373
1374         if (arg_share_system)
1375                 return 0;
1376
1377         r = read_one_line_file("/proc/self/loginuid", &p);
1378         if (r == -ENOENT)
1379                 return 0;
1380         if (r < 0) {
1381                 log_error("Failed to read /proc/self/loginuid: %s", strerror(-r));
1382                 return r;
1383         }
1384
1385         /* Already reset? */
1386         if (streq(p, "4294967295"))
1387                 return 0;
1388
1389         r = write_string_file("/proc/self/loginuid", "4294967295");
1390         if (r < 0) {
1391                 log_error("Failed to reset audit login UID. This probably means that your kernel is too\n"
1392                           "old and you have audit enabled. Note that the auditing subsystem is known to\n"
1393                           "be incompatible with containers on old kernels. Please make sure to upgrade\n"
1394                           "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
1395                           "using systemd-nspawn. Sleeping for 5s... (%s)\n", strerror(-r));
1396
1397                 sleep(5);
1398         }
1399
1400         return 0;
1401 }
1402
1403 #define HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
1404
1405 static int get_mac(struct ether_addr *mac) {
1406         int r;
1407
1408         uint8_t result[8];
1409         size_t l, sz;
1410         uint8_t *v;
1411
1412         l = strlen(arg_machine);
1413         sz = sizeof(sd_id128_t) + l;
1414         v = alloca(sz);
1415
1416         /* fetch some persistent data unique to the host */
1417         r = sd_id128_get_machine((sd_id128_t*) v);
1418         if (r < 0)
1419                 return r;
1420
1421         /* combine with some data unique (on this host) to this
1422          * container instance */
1423         memcpy(v + sizeof(sd_id128_t), arg_machine, l);
1424
1425         /* Let's hash the host machine ID plus the container name. We
1426          * use a fixed, but originally randomly created hash key here. */
1427         siphash24(result, v, sz, HASH_KEY.bytes);
1428
1429         assert_cc(ETH_ALEN <= sizeof(result));
1430         memcpy(mac->ether_addr_octet, result, ETH_ALEN);
1431
1432         /* see eth_random_addr in the kernel */
1433         mac->ether_addr_octet[0] &= 0xfe;        /* clear multicast bit */
1434         mac->ether_addr_octet[0] |= 0x02;        /* set local assignment bit (IEEE802) */
1435
1436         return 0;
1437 }
1438
1439 static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ]) {
1440         _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
1441         _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
1442         struct ether_addr mac;
1443         int r;
1444
1445         if (!arg_private_network)
1446                 return 0;
1447
1448         if (!arg_network_veth)
1449                 return 0;
1450
1451         /* Use two different interface name prefixes depending whether
1452          * we are in bridge mode or not. */
1453         if (arg_network_bridge)
1454                 memcpy(iface_name, "vb-", 3);
1455         else
1456                 memcpy(iface_name, "ve-", 3);
1457         strncpy(iface_name+3, arg_machine, IFNAMSIZ - 3);
1458
1459         r = get_mac(&mac);
1460         if (r < 0) {
1461                 log_error("Failed to generate predictable MAC address for host0");
1462                 return r;
1463         }
1464
1465         r = sd_rtnl_open(&rtnl, 0);
1466         if (r < 0) {
1467                 log_error("Failed to connect to netlink: %s", strerror(-r));
1468                 return r;
1469         }
1470
1471         r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
1472         if (r < 0) {
1473                 log_error("Failed to allocate netlink message: %s", strerror(-r));
1474                 return r;
1475         }
1476
1477         r = sd_rtnl_message_append_string(m, IFLA_IFNAME, iface_name);
1478         if (r < 0) {
1479                 log_error("Failed to add netlink interface name: %s", strerror(-r));
1480                 return r;
1481         }
1482
1483         r = sd_rtnl_message_open_container(m, IFLA_LINKINFO);
1484         if (r < 0) {
1485                 log_error("Failed to open netlink container: %s", strerror(-r));
1486                 return r;
1487         }
1488
1489         r = sd_rtnl_message_append_string(m, IFLA_INFO_KIND, "veth");
1490         if (r < 0) {
1491                 log_error("Failed to append netlink kind: %s", strerror(-r));
1492                 return r;
1493         }
1494
1495         r = sd_rtnl_message_open_container(m, IFLA_INFO_DATA);
1496         if (r < 0) {
1497                 log_error("Failed to open netlink container: %s", strerror(-r));
1498                 return r;
1499         }
1500
1501         r = sd_rtnl_message_open_container(m, VETH_INFO_PEER);
1502         if (r < 0) {
1503                 log_error("Failed to open netlink container: %s", strerror(-r));
1504                 return r;
1505         }
1506
1507         r = sd_rtnl_message_append_string(m, IFLA_IFNAME, "host0");
1508         if (r < 0) {
1509                 log_error("Failed to add netlink interface name: %s", strerror(-r));
1510                 return r;
1511         }
1512
1513         r = sd_rtnl_message_append_ether_addr(m, IFLA_ADDRESS, &mac);
1514         if (r < 0) {
1515                 log_error("Failed to add netlink MAC address: %s", strerror(-r));
1516                 return r;
1517         }
1518
1519         r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
1520         if (r < 0) {
1521                 log_error("Failed to add netlink namespace field: %s", strerror(-r));
1522                 return r;
1523         }
1524
1525         r = sd_rtnl_message_close_container(m);
1526         if (r < 0) {
1527                 log_error("Failed to close netlink container: %s", strerror(-r));
1528                 return r;
1529         }
1530
1531         r = sd_rtnl_message_close_container(m);
1532         if (r < 0) {
1533                 log_error("Failed to close netlink container: %s", strerror(-r));
1534                 return r;
1535         }
1536
1537         r = sd_rtnl_message_close_container(m);
1538         if (r < 0) {
1539                 log_error("Failed to close netlink container: %s", strerror(-r));
1540                 return r;
1541         }
1542
1543         r = sd_rtnl_call(rtnl, m, 0, NULL);
1544         if (r < 0) {
1545                 log_error("Failed to add new veth interfaces: %s", strerror(-r));
1546                 return r;
1547         }
1548
1549         return 0;
1550 }
1551
1552 static int setup_bridge(const char veth_name[]) {
1553         _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
1554         _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
1555         int r, bridge;
1556
1557         if (!arg_private_network)
1558                 return 0;
1559
1560         if (!arg_network_veth)
1561                 return 0;
1562
1563         if (!arg_network_bridge)
1564                 return 0;
1565
1566         bridge = (int) if_nametoindex(arg_network_bridge);
1567         if (bridge <= 0) {
1568                 log_error("Failed to resolve interface %s: %m", arg_network_bridge);
1569                 return -errno;
1570         }
1571
1572         r = sd_rtnl_open(&rtnl, 0);
1573         if (r < 0) {
1574                 log_error("Failed to connect to netlink: %s", strerror(-r));
1575                 return r;
1576         }
1577
1578         r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, 0);
1579         if (r < 0) {
1580                 log_error("Failed to allocate netlink message: %s", strerror(-r));
1581                 return r;
1582         }
1583
1584         r = sd_rtnl_message_append_string(m, IFLA_IFNAME, veth_name);
1585         if (r < 0) {
1586                 log_error("Failed to add netlink interface name field: %s", strerror(-r));
1587                 return r;
1588         }
1589
1590         r = sd_rtnl_message_append_u32(m, IFLA_MASTER, bridge);
1591         if (r < 0) {
1592                 log_error("Failed to add netlink master field: %s", strerror(-r));
1593                 return r;
1594         }
1595
1596         r = sd_rtnl_call(rtnl, m, 0, NULL);
1597         if (r < 0) {
1598                 log_error("Failed to add veth interface to bridge: %s", strerror(-r));
1599                 return r;
1600         }
1601
1602         return 0;
1603 }
1604
1605 static int parse_interface(struct udev *udev, const char *name) {
1606         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
1607         char ifi_str[2 + DECIMAL_STR_MAX(int)];
1608         int ifi;
1609
1610         ifi = (int) if_nametoindex(name);
1611         if (ifi <= 0) {
1612                 log_error("Failed to resolve interface %s: %m", name);
1613                 return -errno;
1614         }
1615
1616         sprintf(ifi_str, "n%i", ifi);
1617         d = udev_device_new_from_device_id(udev, ifi_str);
1618         if (!d) {
1619                 log_error("Failed to get udev device for interface %s: %m", name);
1620                 return -errno;
1621         }
1622
1623         if (udev_device_get_is_initialized(d) <= 0) {
1624                 log_error("Network interface %s is not initialized yet.", name);
1625                 return -EBUSY;
1626         }
1627
1628         return ifi;
1629 }
1630
1631 static int move_network_interfaces(pid_t pid) {
1632         _cleanup_udev_unref_ struct udev *udev = NULL;
1633         _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
1634         char **i;
1635         int r;
1636
1637         if (!arg_private_network)
1638                 return 0;
1639
1640         if (strv_isempty(arg_network_interfaces))
1641                 return 0;
1642
1643         r = sd_rtnl_open(&rtnl, 0);
1644         if (r < 0) {
1645                 log_error("Failed to connect to netlink: %s", strerror(-r));
1646                 return r;
1647         }
1648
1649         udev = udev_new();
1650         if (!udev) {
1651                 log_error("Failed to connect to udev.");
1652                 return -ENOMEM;
1653         }
1654
1655         STRV_FOREACH(i, arg_network_interfaces) {
1656                 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
1657                 int ifi;
1658
1659                 ifi = parse_interface(udev, *i);
1660                 if (ifi < 0)
1661                         return ifi;
1662
1663                 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, ifi);
1664                 if (r < 0) {
1665                         log_error("Failed to allocate netlink message: %s", strerror(-r));
1666                         return r;
1667                 }
1668
1669                 r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
1670                 if (r < 0) {
1671                         log_error("Failed to append namespace PID to netlink message: %s", strerror(-r));
1672                         return r;
1673                 }
1674
1675                 r = sd_rtnl_call(rtnl, m, 0, NULL);
1676                 if (r < 0) {
1677                         log_error("Failed to move interface %s to namespace: %s", *i, strerror(-r));
1678                         return r;
1679                 }
1680         }
1681
1682         return 0;
1683 }
1684
1685 static int setup_macvlan(pid_t pid) {
1686         _cleanup_udev_unref_ struct udev *udev = NULL;
1687         _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
1688         char **i;
1689         int r;
1690
1691         if (!arg_private_network)
1692                 return 0;
1693
1694         if (strv_isempty(arg_network_macvlan))
1695                 return 0;
1696
1697         r = sd_rtnl_open(&rtnl, 0);
1698         if (r < 0) {
1699                 log_error("Failed to connect to netlink: %s", strerror(-r));
1700                 return r;
1701         }
1702
1703         udev = udev_new();
1704         if (!udev) {
1705                 log_error("Failed to connect to udev.");
1706                 return -ENOMEM;
1707         }
1708
1709         STRV_FOREACH(i, arg_network_macvlan) {
1710                 _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL;
1711                 _cleanup_free_ char *n = NULL;
1712                 int ifi;
1713
1714                 ifi = parse_interface(udev, *i);
1715                 if (ifi < 0)
1716                         return ifi;
1717
1718                 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
1719                 if (r < 0) {
1720                         log_error("Failed to allocate netlink message: %s", strerror(-r));
1721                         return r;
1722                 }
1723
1724                 r = sd_rtnl_message_append_u32(m, IFLA_LINK, ifi);
1725                 if (r < 0) {
1726                         log_error("Failed to add netlink interface index: %s", strerror(-r));
1727                         return r;
1728                 }
1729
1730                 n = strappend("mv-", *i);
1731                 if (!n)
1732                         return log_oom();
1733
1734                 strshorten(n, IFNAMSIZ-1);
1735
1736                 r = sd_rtnl_message_append_string(m, IFLA_IFNAME, n);
1737                 if (r < 0) {
1738                         log_error("Failed to add netlink interface name: %s", strerror(-r));
1739                         return r;
1740                 }
1741
1742                 r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid);
1743                 if (r < 0) {
1744                         log_error("Failed to add netlink namespace field: %s", strerror(-r));
1745                         return r;
1746                 }
1747
1748                 r = sd_rtnl_message_open_container(m, IFLA_LINKINFO);
1749                 if (r < 0) {
1750                         log_error("Failed to open netlink container: %s", strerror(-r));
1751                         return r;
1752                 }
1753
1754                 r = sd_rtnl_message_append_string(m, IFLA_INFO_KIND, "macvlan");
1755                 if (r < 0) {
1756                         log_error("Failed to append netlink kind: %s", strerror(-r));
1757                         return r;
1758                 }
1759
1760                 r = sd_rtnl_message_open_container(m, IFLA_INFO_DATA);
1761                 if (r < 0) {
1762                         log_error("Failed to open netlink container: %s", strerror(-r));
1763                         return r;
1764                 }
1765
1766                 r = sd_rtnl_message_append_u32(m, IFLA_MACVLAN_MODE, MACVLAN_MODE_BRIDGE);
1767                 if (r < 0) {
1768                         log_error("Failed to append macvlan mode: %s", strerror(-r));
1769                         return r;
1770                 }
1771
1772                 r = sd_rtnl_message_close_container(m);
1773                 if (r < 0) {
1774                         log_error("Failed to close netlink container: %s", strerror(-r));
1775                         return r;
1776                 }
1777
1778                 r = sd_rtnl_message_close_container(m);
1779                 if (r < 0) {
1780                         log_error("Failed to close netlink container: %s", strerror(-r));
1781                         return r;
1782                 }
1783
1784                 r = sd_rtnl_call(rtnl, m, 0, NULL);
1785                 if (r < 0) {
1786                         log_error("Failed to add new macvlan interfaces: %s", strerror(-r));
1787                         return r;
1788                 }
1789         }
1790
1791         return 0;
1792 }
1793
1794 static int audit_still_doesnt_work_in_containers(void) {
1795
1796 #ifdef HAVE_SECCOMP
1797         scmp_filter_ctx seccomp;
1798         int r;
1799
1800         /*
1801            Audit is broken in containers, much of the userspace audit
1802            hookup will fail if running inside a container. We don't
1803            care and just turn off creation of audit sockets.
1804
1805            This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail
1806            with EAFNOSUPPORT which audit userspace uses as indication
1807            that audit is disabled in the kernel.
1808          */
1809
1810         seccomp = seccomp_init(SCMP_ACT_ALLOW);
1811         if (!seccomp)
1812                 return log_oom();
1813
1814         r = seccomp_add_secondary_archs(seccomp);
1815         if (r < 0) {
1816                 log_error("Failed to add secondary archs to seccomp filter: %s", strerror(-r));
1817                 goto finish;
1818         }
1819
1820         r = seccomp_rule_add(
1821                         seccomp,
1822                         SCMP_ACT_ERRNO(EAFNOSUPPORT),
1823                         SCMP_SYS(socket),
1824                         2,
1825                         SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
1826                         SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
1827         if (r < 0) {
1828                 log_error("Failed to add audit seccomp rule: %s", strerror(-r));
1829                 goto finish;
1830         }
1831
1832         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1833         if (r < 0) {
1834                 log_error("Failed to unset NO_NEW_PRIVS: %s", strerror(-r));
1835                 goto finish;
1836         }
1837
1838         r = seccomp_load(seccomp);
1839         if (r < 0)
1840                 log_error("Failed to install seccomp audit filter: %s", strerror(-r));
1841
1842 finish:
1843         seccomp_release(seccomp);
1844         return r;
1845 #else
1846         return 0;
1847 #endif
1848
1849 }
1850
1851 static int setup_image(char **device_path, int *loop_nr) {
1852         struct loop_info64 info = {
1853                 .lo_flags = LO_FLAGS_AUTOCLEAR|LO_FLAGS_PARTSCAN
1854         };
1855         _cleanup_close_ int fd = -1, control = -1, loop = -1;
1856         _cleanup_free_ char* loopdev = NULL;
1857         struct stat st;
1858         int r, nr;
1859
1860         assert(device_path);
1861         assert(loop_nr);
1862
1863         fd = open(arg_image, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
1864         if (fd < 0) {
1865                 log_error("Failed to open %s: %m", arg_image);
1866                 return -errno;
1867         }
1868
1869         if (fstat(fd, &st) < 0) {
1870                 log_error("Failed to stat %s: %m", arg_image);
1871                 return -errno;
1872         }
1873
1874         if (S_ISBLK(st.st_mode)) {
1875                 char *p;
1876
1877                 p = strdup(arg_image);
1878                 if (!p)
1879                         return log_oom();
1880
1881                 *device_path = p;
1882
1883                 *loop_nr = -1;
1884
1885                 r = fd;
1886                 fd = -1;
1887
1888                 return r;
1889         }
1890
1891         if (!S_ISREG(st.st_mode)) {
1892                 log_error("%s is not a regular file or block device: %m", arg_image);
1893                 return -EINVAL;
1894         }
1895
1896         control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
1897         if (control < 0) {
1898                 log_error("Failed to open /dev/loop-control: %m");
1899                 return -errno;
1900         }
1901
1902         nr = ioctl(control, LOOP_CTL_GET_FREE);
1903         if (nr < 0) {
1904                 log_error("Failed to allocate loop device: %m");
1905                 return -errno;
1906         }
1907
1908         if (asprintf(&loopdev, "/dev/loop%i", nr) < 0)
1909                 return log_oom();
1910
1911         loop = open(loopdev, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
1912         if (loop < 0) {
1913                 log_error("Failed to open loop device %s: %m", loopdev);
1914                 return -errno;
1915         }
1916
1917         if (ioctl(loop, LOOP_SET_FD, fd) < 0) {
1918                 log_error("Failed to set loopback file descriptor on %s: %m", loopdev);
1919                 return -errno;
1920         }
1921
1922         if (arg_read_only)
1923                 info.lo_flags |= LO_FLAGS_READ_ONLY;
1924
1925         if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0) {
1926                 log_error("Failed to set loopback settings on %s: %m", loopdev);
1927                 return -errno;
1928         }
1929
1930         *device_path = loopdev;
1931         loopdev = NULL;
1932
1933         *loop_nr = nr;
1934
1935         r = loop;
1936         loop = -1;
1937
1938         return r;
1939 }
1940
1941 static int dissect_image(
1942                 int fd,
1943                 char **root_device, bool *root_device_rw,
1944                 char **home_device, bool *home_device_rw,
1945                 char **srv_device, bool *srv_device_rw,
1946                 bool *secondary) {
1947
1948 #ifdef HAVE_BLKID
1949         int home_nr = -1, root_nr = -1, secondary_root_nr = -1, srv_nr = -1;
1950         _cleanup_free_ char *home = NULL, *root = NULL, *secondary_root = NULL, *srv = NULL;
1951         _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
1952         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
1953         _cleanup_blkid_free_probe_ blkid_probe b = NULL;
1954         _cleanup_udev_unref_ struct udev *udev = NULL;
1955         struct udev_list_entry *first, *item;
1956         bool home_rw = true, root_rw = true, secondary_root_rw = true, srv_rw = true;
1957         const char *pttype = NULL;
1958         blkid_partlist pl;
1959         struct stat st;
1960         int r;
1961
1962         assert(fd >= 0);
1963         assert(root_device);
1964         assert(home_device);
1965         assert(srv_device);
1966         assert(secondary);
1967
1968         b = blkid_new_probe();
1969         if (!b)
1970                 return log_oom();
1971
1972         errno = 0;
1973         r = blkid_probe_set_device(b, fd, 0, 0);
1974         if (r != 0) {
1975                 if (errno == 0)
1976                         return log_oom();
1977
1978                 log_error("Failed to set device on blkid probe: %m");
1979                 return -errno;
1980         }
1981
1982         blkid_probe_enable_partitions(b, 1);
1983         blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
1984
1985         errno = 0;
1986         r = blkid_do_safeprobe(b);
1987         if (r == -2 || r == 1) {
1988                 log_error("Failed to identify any partition table on %s.\n"
1989                           "Note that the disk image needs to follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/ to be supported by systemd-nspawn.", arg_image);
1990                 return -EINVAL;
1991         } else if (r != 0) {
1992                 if (errno == 0)
1993                         errno = EIO;
1994                 log_error("Failed to probe: %m");
1995                 return -errno;
1996         }
1997
1998         blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
1999         if (!streq_ptr(pttype, "gpt")) {
2000                 log_error("Image %s does not carry a GUID Partition Table.\n"
2001                           "Note that the disk image needs to follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/ to be supported by systemd-nspawn.", arg_image);
2002                 return -EINVAL;
2003         }
2004
2005         errno = 0;
2006         pl = blkid_probe_get_partitions(b);
2007         if (!pl) {
2008                 if (errno == 0)
2009                         return log_oom();
2010
2011                 log_error("Failed to list partitions of %s", arg_image);
2012                 return -errno;
2013         }
2014
2015         udev = udev_new();
2016         if (!udev)
2017                 return log_oom();
2018
2019         if (fstat(fd, &st) < 0) {
2020                 log_error("Failed to stat block device: %m");
2021                 return -errno;
2022         }
2023
2024         d = udev_device_new_from_devnum(udev, 'b', st.st_rdev);
2025         if (!d)
2026                 return log_oom();
2027
2028         e = udev_enumerate_new(udev);
2029         if (!e)
2030                 return log_oom();
2031
2032         r = udev_enumerate_add_match_parent(e, d);
2033         if (r < 0)
2034                 return log_oom();
2035
2036         r = udev_enumerate_scan_devices(e);
2037         if (r < 0) {
2038                 log_error("Failed to scan for partition devices of %s: %s", arg_image, strerror(-r));
2039                 return r;
2040         }
2041
2042         first = udev_enumerate_get_list_entry(e);
2043         udev_list_entry_foreach(item, first) {
2044                 _cleanup_udev_device_unref_ struct udev_device *q;
2045                 const char *stype, *node;
2046                 unsigned long long flags;
2047                 sd_id128_t type_id;
2048                 blkid_partition pp;
2049                 dev_t qn;
2050                 int nr;
2051
2052                 errno = 0;
2053                 q = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
2054                 if (!q) {
2055                         if (!errno)
2056                                 errno = ENOMEM;
2057
2058                         log_error("Failed to get partition device of %s: %m", arg_image);
2059                         return -errno;
2060                 }
2061
2062                 qn = udev_device_get_devnum(q);
2063                 if (major(qn) == 0)
2064                         continue;
2065
2066                 if (st.st_rdev == qn)
2067                         continue;
2068
2069                 node = udev_device_get_devnode(q);
2070                 if (!node)
2071                         continue;
2072
2073                 pp = blkid_partlist_devno_to_partition(pl, qn);
2074                 if (!pp)
2075                         continue;
2076
2077                 flags = blkid_partition_get_flags(pp);
2078                 if (flags & GPT_FLAG_NO_AUTO)
2079                         continue;
2080
2081                 nr = blkid_partition_get_partno(pp);
2082                 if (nr < 0)
2083                         continue;
2084
2085                 stype = blkid_partition_get_type_string(pp);
2086                 if (!stype)
2087                         continue;
2088
2089                 if (sd_id128_from_string(stype, &type_id) < 0)
2090                         continue;
2091
2092                 if (sd_id128_equal(type_id, GPT_HOME)) {
2093
2094                         if (home && nr >= home_nr)
2095                                 continue;
2096
2097                         home_nr = nr;
2098                         home_rw = !(flags & GPT_FLAG_READ_ONLY);
2099
2100                         free(home);
2101                         home = strdup(node);
2102                         if (!home)
2103                                 return log_oom();
2104                 } else if (sd_id128_equal(type_id, GPT_SRV)) {
2105
2106                         if (srv && nr >= srv_nr)
2107                                 continue;
2108
2109                         srv_nr = nr;
2110                         srv_rw = !(flags & GPT_FLAG_READ_ONLY);
2111
2112                         free(srv);
2113                         srv = strdup(node);
2114                         if (!srv)
2115                                 return log_oom();
2116                 }
2117 #ifdef GPT_ROOT_NATIVE
2118                 else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
2119
2120                         if (root && nr >= root_nr)
2121                                 continue;
2122
2123                         root_nr = nr;
2124                         root_rw = !(flags & GPT_FLAG_READ_ONLY);
2125
2126                         free(root);
2127                         root = strdup(node);
2128                         if (!root)
2129                                 return log_oom();
2130                 }
2131 #endif
2132 #ifdef GPT_ROOT_SECONDARY
2133                 else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
2134
2135                         if (secondary_root && nr >= secondary_root_nr)
2136                                 continue;
2137
2138                         secondary_root_nr = nr;
2139                         secondary_root_rw = !(flags & GPT_FLAG_READ_ONLY);
2140
2141
2142                         free(secondary_root);
2143                         secondary_root = strdup(node);
2144                         if (!secondary_root)
2145                                 return log_oom();
2146                 }
2147 #endif
2148         }
2149
2150         if (!root && !secondary_root) {
2151                 log_error("Failed to identify root partition in disk image %s.\n"
2152                           "Note that the disk image needs to follow http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/ to be supported by systemd-nspawn.", arg_image);
2153                 return -EINVAL;
2154         }
2155
2156         if (root) {
2157                 *root_device = root;
2158                 root = NULL;
2159
2160                 *root_device_rw = root_rw;
2161                 *secondary = false;
2162         } else if (secondary_root) {
2163                 *root_device = secondary_root;
2164                 secondary_root = NULL;
2165
2166                 *root_device_rw = secondary_root_rw;
2167                 *secondary = true;
2168         }
2169
2170         if (home) {
2171                 *home_device = home;
2172                 home = NULL;
2173
2174                 *home_device_rw = home_rw;
2175         }
2176
2177         if (srv) {
2178                 *srv_device = srv;
2179                 srv = NULL;
2180
2181                 *srv_device_rw = srv_rw;
2182         }
2183
2184         return 0;
2185 #else
2186         log_error("--image= is not supported, compiled without blkid support.");
2187         return -ENOTSUP;
2188 #endif
2189 }
2190
2191 static int mount_device(const char *what, const char *where, const char *directory, bool rw) {
2192 #ifdef HAVE_BLKID
2193         _cleanup_blkid_free_probe_ blkid_probe b = NULL;
2194         const char *fstype, *p;
2195         int r;
2196
2197         assert(what);
2198         assert(where);
2199
2200         if (arg_read_only)
2201                 rw = false;
2202
2203         if (directory)
2204                 p = strappenda(where, directory);
2205         else
2206                 p = where;
2207
2208         errno = 0;
2209         b = blkid_new_probe_from_filename(what);
2210         if (!b) {
2211                 if (errno == 0)
2212                         return log_oom();
2213                 log_error("Failed to allocate prober for %s: %m", what);
2214                 return -errno;
2215         }
2216
2217         blkid_probe_enable_superblocks(b, 1);
2218         blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
2219
2220         errno = 0;
2221         r = blkid_do_safeprobe(b);
2222         if (r == -1 || r == 1) {
2223                 log_error("Cannot determine file system type of %s", what);
2224                 return -EINVAL;
2225         } else if (r != 0) {
2226                 if (errno == 0)
2227                         errno = EIO;
2228                 log_error("Failed to probe %s: %m", what);
2229                 return -errno;
2230         }
2231
2232         errno = 0;
2233         if (blkid_probe_lookup_value(b, "TYPE", &fstype, NULL) < 0) {
2234                 if (errno == 0)
2235                         errno = EINVAL;
2236                 log_error("Failed to determine file system type of %s", what);
2237                 return -errno;
2238         }
2239
2240         if (streq(fstype, "crypto_LUKS")) {
2241                 log_error("nspawn currently does not support LUKS disk images.");
2242                 return -ENOTSUP;
2243         }
2244
2245         if (mount(what, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), NULL) < 0) {
2246                 log_error("Failed to mount %s: %m", what);
2247                 return -errno;
2248         }
2249
2250         return 0;
2251 #else
2252         log_error("--image= is not supported, compiled without blkid support.");
2253         return -ENOTSUP;
2254 #endif
2255 }
2256
2257 static int mount_devices(
2258                 const char *where,
2259                 const char *root_device, bool root_device_rw,
2260                 const char *home_device, bool home_device_rw,
2261                 const char *srv_device, bool srv_device_rw) {
2262         int r;
2263
2264         assert(where);
2265
2266         if (root_device) {
2267                 r = mount_device(root_device, arg_directory, NULL, root_device_rw);
2268                 if (r < 0) {
2269                         log_error("Failed to mount root directory: %s", strerror(-r));
2270                         return r;
2271                 }
2272         }
2273
2274         if (home_device) {
2275                 r = mount_device(home_device, arg_directory, "/home", home_device_rw);
2276                 if (r < 0) {
2277                         log_error("Failed to mount home directory: %s", strerror(-r));
2278                         return r;
2279                 }
2280         }
2281
2282         if (srv_device) {
2283                 r = mount_device(srv_device, arg_directory, "/srv", srv_device_rw);
2284                 if (r < 0) {
2285                         log_error("Failed to mount server data directory: %s", strerror(-r));
2286                         return r;
2287                 }
2288         }
2289
2290         return 0;
2291 }
2292
2293 static void loop_remove(int nr, int *image_fd) {
2294         _cleanup_close_ int control = -1;
2295
2296         if (nr < 0)
2297                 return;
2298
2299         if (image_fd && *image_fd >= 0) {
2300                 ioctl(*image_fd, LOOP_CLR_FD);
2301                 close_nointr_nofail(*image_fd);
2302                 *image_fd = -1;
2303         }
2304
2305         control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
2306         if (control < 0)
2307                 return;
2308
2309         ioctl(control, LOOP_CTL_REMOVE, nr);
2310 }
2311
2312 static int spawn_getent(const char *database, const char *key, pid_t *rpid) {
2313         int pipe_fds[2];
2314         pid_t pid;
2315
2316         assert(database);
2317         assert(key);
2318         assert(rpid);
2319
2320         if (pipe2(pipe_fds, O_CLOEXEC) < 0) {
2321                 log_error("Failed to allocate pipe: %m");
2322                 return -errno;
2323         }
2324
2325         pid = fork();
2326         if (pid < 0) {
2327                 log_error("Failed to fork getent child: %m");
2328                 return -errno;
2329         } else if (pid == 0) {
2330                 int nullfd;
2331                 char *empty_env = NULL;
2332
2333                 if (dup3(pipe_fds[1], STDOUT_FILENO, 0) < 0)
2334                         _exit(EXIT_FAILURE);
2335
2336                 if (pipe_fds[0] > 2)
2337                         close_nointr_nofail(pipe_fds[0]);
2338                 if (pipe_fds[1] > 2)
2339                         close_nointr_nofail(pipe_fds[1]);
2340
2341                 nullfd = open("/dev/null", O_RDWR);
2342                 if (nullfd < 0)
2343                         _exit(EXIT_FAILURE);
2344
2345                 if (dup3(nullfd, STDIN_FILENO, 0) < 0)
2346                         _exit(EXIT_FAILURE);
2347
2348                 if (dup3(nullfd, STDERR_FILENO, 0) < 0)
2349                         _exit(EXIT_FAILURE);
2350
2351                 if (nullfd > 2)
2352                         close_nointr_nofail(nullfd);
2353
2354                 reset_all_signal_handlers();
2355                 close_all_fds(NULL, 0);
2356
2357                 execle("/usr/bin/getent", "getent", database, key, NULL, &empty_env);
2358                 execle("/bin/getent", "getent", database, key, NULL, &empty_env);
2359                 _exit(EXIT_FAILURE);
2360         }
2361
2362         close_nointr_nofail(pipe_fds[1]);
2363         pipe_fds[1] = -1;
2364
2365         *rpid = pid;
2366
2367         return pipe_fds[0];
2368 }
2369
2370 static int change_uid_gid(char **_home) {
2371
2372         _cleanup_strv_free_ char **passwd = NULL;
2373         char line[LINE_MAX], *w, *x, *state, *u, *g, *h;
2374         _cleanup_free_ uid_t *uids = NULL;
2375         _cleanup_free_ char *home = NULL;
2376         _cleanup_fclose_ FILE *f = NULL;
2377         _cleanup_close_ int fd = -1;
2378         unsigned n_uids = 0;
2379         size_t sz, l;
2380         uid_t uid;
2381         gid_t gid;
2382         pid_t pid;
2383         int r;
2384
2385         assert(_home);
2386
2387         if (!arg_user || streq(arg_user, "root") || streq(arg_user, "0")) {
2388                 /* Reset everything fully to 0, just in case */
2389
2390                 if (setgroups(0, NULL) < 0) {
2391                         log_error("setgroups() failed: %m");
2392                         return -errno;
2393                 }
2394
2395                 if (setresgid(0, 0, 0) < 0) {
2396                         log_error("setregid() failed: %m");
2397                         return -errno;
2398                 }
2399
2400                 if (setresuid(0, 0, 0) < 0) {
2401                         log_error("setreuid() failed: %m");
2402                         return -errno;
2403                 }
2404
2405                 *_home = NULL;
2406                 return 0;
2407         }
2408
2409         /* First, get user credentials */
2410         fd = spawn_getent("passwd", arg_user, &pid);
2411         if (fd < 0)
2412                 return fd;
2413
2414         f = fdopen(fd, "r");
2415         if (!f)
2416                 return log_oom();
2417         fd = -1;
2418
2419         if (!fgets(line, sizeof(line), f)) {
2420
2421                 if (!ferror(f)) {
2422                         log_error("Failed to resolve user %s.", arg_user);
2423                         return -ESRCH;
2424                 }
2425
2426                 log_error("Failed to read from getent: %m");
2427                 return -errno;
2428         }
2429
2430         truncate_nl(line);
2431
2432         wait_for_terminate_and_warn("getent passwd", pid);
2433
2434         x = strchr(line, ':');
2435         if (!x) {
2436                 log_error("/etc/passwd entry has invalid user field.");
2437                 return -EIO;
2438         }
2439
2440         u = strchr(x+1, ':');
2441         if (!u) {
2442                 log_error("/etc/passwd entry has invalid password field.");
2443                 return -EIO;
2444         }
2445
2446         u++;
2447         g = strchr(u, ':');
2448         if (!g) {
2449                 log_error("/etc/passwd entry has invalid UID field.");
2450                 return -EIO;
2451         }
2452
2453         *g = 0;
2454         g++;
2455         x = strchr(g, ':');
2456         if (!x) {
2457                 log_error("/etc/passwd entry has invalid GID field.");
2458                 return -EIO;
2459         }
2460
2461         *x = 0;
2462         h = strchr(x+1, ':');
2463         if (!h) {
2464                 log_error("/etc/passwd entry has invalid GECOS field.");
2465                 return -EIO;
2466         }
2467
2468         h++;
2469         x = strchr(h, ':');
2470         if (!x) {
2471                 log_error("/etc/passwd entry has invalid home directory field.");
2472                 return -EIO;
2473         }
2474
2475         *x = 0;
2476
2477         r = parse_uid(u, &uid);
2478         if (r < 0) {
2479                 log_error("Failed to parse UID of user.");
2480                 return -EIO;
2481         }
2482
2483         r = parse_gid(g, &gid);
2484         if (r < 0) {
2485                 log_error("Failed to parse GID of user.");
2486                 return -EIO;
2487         }
2488
2489         home = strdup(h);
2490         if (!home)
2491                 return log_oom();
2492
2493         /* Second, get group memberships */
2494         fd = spawn_getent("initgroups", arg_user, &pid);
2495         if (fd < 0)
2496                 return fd;
2497
2498         fclose(f);
2499         f = fdopen(fd, "r");
2500         if (!f)
2501                 return log_oom();
2502         fd = -1;
2503
2504         if (!fgets(line, sizeof(line), f)) {
2505                 if (!ferror(f)) {
2506                         log_error("Failed to resolve user %s.", arg_user);
2507                         return -ESRCH;
2508                 }
2509
2510                 log_error("Failed to read from getent: %m");
2511                 return -errno;
2512         }
2513
2514         truncate_nl(line);
2515
2516         wait_for_terminate_and_warn("getent initgroups", pid);
2517
2518         /* Skip over the username and subsequent separator whitespace */
2519         x = line;
2520         x += strcspn(x, WHITESPACE);
2521         x += strspn(x, WHITESPACE);
2522
2523         FOREACH_WORD(w, l, x, state) {
2524                 char c[l+1];
2525
2526                 memcpy(c, w, l);
2527                 c[l] = 0;
2528
2529                 if (!GREEDY_REALLOC(uids, sz, n_uids+1))
2530                         return log_oom();
2531
2532                 r = parse_uid(c, &uids[n_uids++]);
2533                 if (r < 0) {
2534                         log_error("Failed to parse group data from getent.");
2535                         return -EIO;
2536                 }
2537         }
2538
2539         r = mkdir_parents(home, 0775);
2540         if (r < 0) {
2541                 log_error("Failed to make home root directory: %s", strerror(-r));
2542                 return r;
2543         }
2544
2545         r = mkdir_safe(home, 0755, uid, gid);
2546         if (r < 0 && r != -EEXIST) {
2547                 log_error("Failed to make home directory: %s", strerror(-r));
2548                 return r;
2549         }
2550
2551         fchown(STDIN_FILENO, uid, gid);
2552         fchown(STDOUT_FILENO, uid, gid);
2553         fchown(STDERR_FILENO, uid, gid);
2554
2555         if (setgroups(n_uids, uids) < 0) {
2556                 log_error("Failed to set auxiliary groups: %m");
2557                 return -errno;
2558         }
2559
2560         if (setresgid(gid, gid, gid) < 0) {
2561                 log_error("setregid() failed: %m");
2562                 return -errno;
2563         }
2564
2565         if (setresuid(uid, uid, uid) < 0) {
2566                 log_error("setreuid() failed: %m");
2567                 return -errno;
2568         }
2569
2570         if (_home) {
2571                 *_home = home;
2572                 home = NULL;
2573         }
2574
2575         return 0;
2576 }
2577
2578 int main(int argc, char *argv[]) {
2579
2580         _cleanup_free_ char *kdbus_domain = NULL, *device_path = NULL, *root_device = NULL, *home_device = NULL, *srv_device = NULL;
2581         bool root_device_rw = true, home_device_rw = true, srv_device_rw = true;
2582         _cleanup_close_ int master = -1, kdbus_fd = -1, image_fd = -1;
2583         _cleanup_close_pipe_ int kmsg_socket_pair[2] = { -1, -1 };
2584         _cleanup_fdset_free_ FDSet *fds = NULL;
2585         int r = EXIT_FAILURE, k, n_fd_passed, loop_nr = -1;
2586         const char *console = NULL;
2587         char veth_name[IFNAMSIZ];
2588         bool secondary = false;
2589         pid_t pid = 0;
2590         sigset_t mask;
2591
2592         log_parse_environment();
2593         log_open();
2594
2595         k = parse_argv(argc, argv);
2596         if (k < 0)
2597                 goto finish;
2598         else if (k == 0) {
2599                 r = EXIT_SUCCESS;
2600                 goto finish;
2601         }
2602
2603         if (!arg_image) {
2604                 if (arg_directory) {
2605                         char *p;
2606
2607                         p = path_make_absolute_cwd(arg_directory);
2608                         free(arg_directory);
2609                         arg_directory = p;
2610                 } else
2611                         arg_directory = get_current_dir_name();
2612
2613                 if (!arg_directory) {
2614                         log_error("Failed to determine path, please use -D.");
2615                         goto finish;
2616                 }
2617                 path_kill_slashes(arg_directory);
2618         }
2619
2620         if (!arg_machine) {
2621                 arg_machine = strdup(basename(arg_image ? arg_image : arg_directory));
2622                 if (!arg_machine) {
2623                         log_oom();
2624                         goto finish;
2625                 }
2626
2627                 hostname_cleanup(arg_machine, false);
2628                 if (isempty(arg_machine)) {
2629                         log_error("Failed to determine machine name automatically, please use -M.");
2630                         goto finish;
2631                 }
2632         }
2633
2634         if (geteuid() != 0) {
2635                 log_error("Need to be root.");
2636                 goto finish;
2637         }
2638
2639         if (sd_booted() <= 0) {
2640                 log_error("Not running on a systemd system.");
2641                 goto finish;
2642         }
2643
2644         log_close();
2645         n_fd_passed = sd_listen_fds(false);
2646         if (n_fd_passed > 0) {
2647                 k = fdset_new_listen_fds(&fds, false);
2648                 if (k < 0) {
2649                         log_error("Failed to collect file descriptors: %s", strerror(-k));
2650                         goto finish;
2651                 }
2652         }
2653         fdset_close_others(fds);
2654         log_open();
2655
2656         if (arg_directory) {
2657                 if (path_equal(arg_directory, "/")) {
2658                         log_error("Spawning container on root directory not supported.");
2659                         goto finish;
2660                 }
2661
2662                 if (arg_boot) {
2663                         if (path_is_os_tree(arg_directory) <= 0) {
2664                                 log_error("Directory %s doesn't look like an OS root directory (/etc/os-release is missing). Refusing.", arg_directory);
2665                                 goto finish;
2666                         }
2667                 } else {
2668                         const char *p;
2669
2670                         p = strappenda(arg_directory,
2671                                        argc > optind && path_is_absolute(argv[optind]) ? argv[optind] : "/usr/bin/");
2672                         if (access(p, F_OK) < 0) {
2673                                 log_error("Directory %s lacks the binary to execute or doesn't look like a binary tree. Refusing.", arg_directory);
2674                                 goto finish;
2675
2676                         }
2677                 }
2678         } else {
2679                 char template[] = "/tmp/nspawn-root-XXXXXX";
2680
2681                 if (!mkdtemp(template)) {
2682                         log_error("Failed to create temporary directory: %m");
2683                         r = -errno;
2684                         goto finish;
2685                 }
2686
2687                 arg_directory = strdup(template);
2688                 if (!arg_directory) {
2689                         r = log_oom();
2690                         goto finish;
2691                 }
2692
2693                 image_fd = setup_image(&device_path, &loop_nr);
2694                 if (image_fd < 0) {
2695                         r = image_fd;
2696                         goto finish;
2697                 }
2698
2699                 r = dissect_image(image_fd, &root_device, &root_device_rw, &home_device, &home_device_rw, &srv_device, &srv_device_rw, &secondary);
2700                 if (r < 0)
2701                         goto finish;
2702         }
2703
2704         master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
2705         if (master < 0) {
2706                 log_error("Failed to acquire pseudo tty: %m");
2707                 goto finish;
2708         }
2709
2710         console = ptsname(master);
2711         if (!console) {
2712                 log_error("Failed to determine tty name: %m");
2713                 goto finish;
2714         }
2715
2716         if (!arg_quiet)
2717                 log_info("Spawning container %s on %s. Press ^] three times within 1s to abort execution.", arg_machine, arg_image ? arg_image : arg_directory);
2718
2719         if (unlockpt(master) < 0) {
2720                 log_error("Failed to unlock tty: %m");
2721                 goto finish;
2722         }
2723
2724         if (access("/dev/kdbus/control", F_OK) >= 0) {
2725
2726                 if (arg_share_system) {
2727                         kdbus_domain = strdup("/dev/kdbus");
2728                         if (!kdbus_domain) {
2729                                 log_oom();
2730                                 goto finish;
2731                         }
2732                 } else {
2733                         const char *ns;
2734
2735                         ns = strappenda("machine-", arg_machine);
2736                         kdbus_fd = bus_kernel_create_domain(ns, &kdbus_domain);
2737                         if (r < 0)
2738                                 log_debug("Failed to create kdbus domain: %s", strerror(-r));
2739                         else
2740                                 log_debug("Successfully created kdbus domain as %s", kdbus_domain);
2741                 }
2742         }
2743
2744         if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_NONBLOCK|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
2745                 log_error("Failed to create kmsg socket pair: %m");
2746                 goto finish;
2747         }
2748
2749         sd_notify(0, "READY=1");
2750
2751         assert_se(sigemptyset(&mask) == 0);
2752         sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
2753         assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
2754
2755         for (;;) {
2756                 int parent_ready_fd = -1, child_ready_fd = -1;
2757                 siginfo_t status;
2758                 eventfd_t x;
2759
2760                 parent_ready_fd = eventfd(0, EFD_CLOEXEC);
2761                 if (parent_ready_fd < 0) {
2762                         log_error("Failed to create event fd: %m");
2763                         goto finish;
2764                 }
2765
2766                 child_ready_fd = eventfd(0, EFD_CLOEXEC);
2767                 if (child_ready_fd < 0) {
2768                         log_error("Failed to create event fd: %m");
2769                         goto finish;
2770                 }
2771
2772                 pid = syscall(__NR_clone,
2773                               SIGCHLD|CLONE_NEWNS|
2774                               (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS)|
2775                               (arg_private_network ? CLONE_NEWNET : 0), NULL);
2776                 if (pid < 0) {
2777                         if (errno == EINVAL)
2778                                 log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
2779                         else
2780                                 log_error("clone() failed: %m");
2781
2782                         goto finish;
2783                 }
2784
2785                 if (pid == 0) {
2786                         /* child */
2787                         _cleanup_free_ char *home = NULL;
2788                         unsigned n_env = 2;
2789                         const char *envp[] = {
2790                                 "PATH=" DEFAULT_PATH_SPLIT_USR,
2791                                 "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
2792                                 NULL, /* TERM */
2793                                 NULL, /* HOME */
2794                                 NULL, /* USER */
2795                                 NULL, /* LOGNAME */
2796                                 NULL, /* container_uuid */
2797                                 NULL, /* LISTEN_FDS */
2798                                 NULL, /* LISTEN_PID */
2799                                 NULL
2800                         };
2801                         char **env_use;
2802
2803                         envp[n_env] = strv_find_prefix(environ, "TERM=");
2804                         if (envp[n_env])
2805                                 n_env ++;
2806
2807                         close_nointr_nofail(master);
2808                         master = -1;
2809
2810                         close_nointr(STDIN_FILENO);
2811                         close_nointr(STDOUT_FILENO);
2812                         close_nointr(STDERR_FILENO);
2813
2814                         close_nointr_nofail(kmsg_socket_pair[0]);
2815                         kmsg_socket_pair[0] = -1;
2816
2817                         reset_all_signal_handlers();
2818
2819                         assert_se(sigemptyset(&mask) == 0);
2820                         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
2821
2822                         k = open_terminal(console, O_RDWR);
2823                         if (k != STDIN_FILENO) {
2824                                 if (k >= 0) {
2825                                         close_nointr_nofail(k);
2826                                         k = -EINVAL;
2827                                 }
2828
2829                                 log_error("Failed to open console: %s", strerror(-k));
2830                                 goto child_fail;
2831                         }
2832
2833                         if (dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
2834                             dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO) {
2835                                 log_error("Failed to duplicate console: %m");
2836                                 goto child_fail;
2837                         }
2838
2839                         if (setsid() < 0) {
2840                                 log_error("setsid() failed: %m");
2841                                 goto child_fail;
2842                         }
2843
2844                         if (reset_audit_loginuid() < 0)
2845                                 goto child_fail;
2846
2847                         if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0) {
2848                                 log_error("PR_SET_PDEATHSIG failed: %m");
2849                                 goto child_fail;
2850                         }
2851
2852                         /* Mark everything as slave, so that we still
2853                          * receive mounts from the real root, but don't
2854                          * propagate mounts to the real root. */
2855                         if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
2856                                 log_error("MS_SLAVE|MS_REC failed: %m");
2857                                 goto child_fail;
2858                         }
2859
2860                         if (mount_devices(arg_directory,
2861                                           root_device, root_device_rw,
2862                                           home_device, home_device_rw,
2863                                           srv_device, srv_device_rw) < 0)
2864                                 goto child_fail;
2865
2866                         /* Turn directory into bind mount */
2867                         if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REC, NULL) < 0) {
2868                                 log_error("Failed to make bind mount.");
2869                                 goto child_fail;
2870                         }
2871
2872                         if (arg_read_only)
2873                                 if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY|MS_REC, NULL) < 0) {
2874                                         log_error("Failed to make read-only.");
2875                                         goto child_fail;
2876                                 }
2877
2878                         if (mount_all(arg_directory) < 0)
2879                                 goto child_fail;
2880
2881                         if (copy_devnodes(arg_directory) < 0)
2882                                 goto child_fail;
2883
2884                         if (setup_ptmx(arg_directory) < 0)
2885                                 goto child_fail;
2886
2887                         dev_setup(arg_directory);
2888
2889                         if (audit_still_doesnt_work_in_containers() < 0)
2890                                 goto child_fail;
2891
2892                         if (setup_dev_console(arg_directory, console) < 0)
2893                                 goto child_fail;
2894
2895                         if (setup_kmsg(arg_directory, kmsg_socket_pair[1]) < 0)
2896                                 goto child_fail;
2897
2898                         close_nointr_nofail(kmsg_socket_pair[1]);
2899                         kmsg_socket_pair[1] = -1;
2900
2901                         if (setup_boot_id(arg_directory) < 0)
2902                                 goto child_fail;
2903
2904                         if (setup_timezone(arg_directory) < 0)
2905                                 goto child_fail;
2906
2907                         if (setup_resolv_conf(arg_directory) < 0)
2908                                 goto child_fail;
2909
2910                         if (setup_journal(arg_directory) < 0)
2911                                 goto child_fail;
2912
2913                         if (mount_binds(arg_directory, arg_bind, 0) < 0)
2914                                 goto child_fail;
2915
2916                         if (mount_binds(arg_directory, arg_bind_ro, MS_RDONLY) < 0)
2917                                 goto child_fail;
2918
2919                         if (setup_kdbus(arg_directory, kdbus_domain) < 0)
2920                                 goto child_fail;
2921
2922                         /* Tell the parent that we are ready, and that
2923                          * it can cgroupify us to that we lack access
2924                          * to certain devices and resources. */
2925                         eventfd_write(child_ready_fd, 1);
2926                         close_nointr_nofail(child_ready_fd);
2927                         child_ready_fd = -1;
2928
2929                         if (chdir(arg_directory) < 0) {
2930                                 log_error("chdir(%s) failed: %m", arg_directory);
2931                                 goto child_fail;
2932                         }
2933
2934                         if (mount(arg_directory, "/", NULL, MS_MOVE, NULL) < 0) {
2935                                 log_error("mount(MS_MOVE) failed: %m");
2936                                 goto child_fail;
2937                         }
2938
2939                         if (chroot(".") < 0) {
2940                                 log_error("chroot() failed: %m");
2941                                 goto child_fail;
2942                         }
2943
2944                         if (chdir("/") < 0) {
2945                                 log_error("chdir() failed: %m");
2946                                 goto child_fail;
2947                         }
2948
2949                         umask(0022);
2950
2951                         if (arg_private_network)
2952                                 loopback_setup();
2953
2954                         if (drop_capabilities() < 0) {
2955                                 log_error("drop_capabilities() failed: %m");
2956                                 goto child_fail;
2957                         }
2958
2959                         r = change_uid_gid(&home);
2960                         if (r < 0)
2961                                 goto child_fail;
2962
2963                         if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
2964                             (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
2965                             (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)) {
2966                                 log_oom();
2967                                 goto child_fail;
2968                         }
2969
2970                         if (!sd_id128_equal(arg_uuid, SD_ID128_NULL)) {
2971                                 if (asprintf((char**)(envp + n_env++), "container_uuid=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid)) < 0) {
2972                                         log_oom();
2973                                         goto child_fail;
2974                                 }
2975                         }
2976
2977                         if (fdset_size(fds) > 0) {
2978                                 k = fdset_cloexec(fds, false);
2979                                 if (k < 0) {
2980                                         log_error("Failed to unset O_CLOEXEC for file descriptors.");
2981                                         goto child_fail;
2982                                 }
2983
2984                                 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", n_fd_passed) < 0) ||
2985                                     (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0)) {
2986                                         log_oom();
2987                                         goto child_fail;
2988                                 }
2989                         }
2990
2991                         setup_hostname();
2992
2993                         if (arg_personality != 0xffffffffLU) {
2994                                 if (personality(arg_personality) < 0) {
2995                                         log_error("personality() failed: %m");
2996                                         goto child_fail;
2997                                 }
2998                         } else if (secondary) {
2999                                 if (personality(PER_LINUX32) < 0) {
3000                                         log_error("personality() failed: %m");
3001                                         goto child_fail;
3002                                 }
3003                         }
3004
3005 #ifdef HAVE_SELINUX
3006                         if (arg_selinux_context)
3007                                 if (setexeccon((security_context_t) arg_selinux_context) < 0) {
3008                                         log_error("setexeccon(\"%s\") failed: %m", arg_selinux_context);
3009                                         goto child_fail;
3010                                 }
3011 #endif
3012
3013                         if (!strv_isempty(arg_setenv)) {
3014                                 char **n;
3015
3016                                 n = strv_env_merge(2, envp, arg_setenv);
3017                                 if (!n) {
3018                                         log_oom();
3019                                         goto child_fail;
3020                                 }
3021
3022                                 env_use = n;
3023                         } else
3024                                 env_use = (char**) envp;
3025
3026                         /* Wait until the parent is ready with the setup, too... */
3027                         eventfd_read(parent_ready_fd, &x);
3028                         close_nointr_nofail(parent_ready_fd);
3029                         parent_ready_fd = -1;
3030
3031                         if (arg_boot) {
3032                                 char **a;
3033                                 size_t l;
3034
3035                                 /* Automatically search for the init system */
3036
3037                                 l = 1 + argc - optind;
3038                                 a = newa(char*, l + 1);
3039                                 memcpy(a + 1, argv + optind, l * sizeof(char*));
3040
3041                                 a[0] = (char*) "/usr/lib/systemd/systemd";
3042                                 execve(a[0], a, env_use);
3043
3044                                 a[0] = (char*) "/lib/systemd/systemd";
3045                                 execve(a[0], a, env_use);
3046
3047                                 a[0] = (char*) "/sbin/init";
3048                                 execve(a[0], a, env_use);
3049                         } else if (argc > optind)
3050                                 execvpe(argv[optind], argv + optind, env_use);
3051                         else {
3052                                 chdir(home ? home : "/root");
3053                                 execle("/bin/bash", "-bash", NULL, env_use);
3054                                 execle("/bin/sh", "-sh", NULL, env_use);
3055                         }
3056
3057                         log_error("execv() failed: %m");
3058
3059                 child_fail:
3060                         _exit(EXIT_FAILURE);
3061                 }
3062
3063                 fdset_free(fds);
3064                 fds = NULL;
3065
3066                 /* Wait until the child reported that it is ready with
3067                  * all it needs to do with priviliges. After we got
3068                  * the notification we can make the process join its
3069                  * cgroup which might limit what it can do */
3070                 eventfd_read(child_ready_fd, &x);
3071
3072                 r = register_machine(pid);
3073                 if (r < 0)
3074                         goto finish;
3075
3076                 r = move_network_interfaces(pid);
3077                 if (r < 0)
3078                         goto finish;
3079
3080                 r = setup_veth(pid, veth_name);
3081                 if (r < 0)
3082                         goto finish;
3083
3084                 r = setup_bridge(veth_name);
3085                 if (r < 0)
3086                         goto finish;
3087
3088                 r = setup_macvlan(pid);
3089                 if (r < 0)
3090                         goto finish;
3091
3092                 /* Notify the child that the parent is ready with all
3093                  * its setup, and thtat the child can now hand over
3094                  * control to the code to run inside the container. */
3095                 eventfd_write(parent_ready_fd, 1);
3096
3097                 k = process_pty(master, &mask, arg_boot ? pid : 0, SIGRTMIN+3);
3098                 if (k < 0) {
3099                         r = EXIT_FAILURE;
3100                         break;
3101                 }
3102
3103                 if (!arg_quiet)
3104                         putc('\n', stdout);
3105
3106                 /* Kill if it is not dead yet anyway */
3107                 terminate_machine(pid);
3108
3109                 /* Redundant, but better safe than sorry */
3110                 kill(pid, SIGKILL);
3111
3112                 k = wait_for_terminate(pid, &status);
3113                 pid = 0;
3114
3115                 if (k < 0) {
3116                         r = EXIT_FAILURE;
3117                         break;
3118                 }
3119
3120                 if (status.si_code == CLD_EXITED) {
3121                         r = status.si_status;
3122                         if (status.si_status != 0) {
3123                                 log_error("Container %s failed with error code %i.", arg_machine, status.si_status);
3124                                 break;
3125                         }
3126
3127                         if (!arg_quiet)
3128                                 log_debug("Container %s exited successfully.", arg_machine);
3129                         break;
3130                 } else if (status.si_code == CLD_KILLED &&
3131                            status.si_status == SIGINT) {
3132
3133                         if (!arg_quiet)
3134                                 log_info("Container %s has been shut down.", arg_machine);
3135                         r = 0;
3136                         break;
3137                 } else if (status.si_code == CLD_KILLED &&
3138                            status.si_status == SIGHUP) {
3139
3140                         if (!arg_quiet)
3141                                 log_info("Container %s is being rebooted.", arg_machine);
3142                         continue;
3143                 } else if (status.si_code == CLD_KILLED ||
3144                            status.si_code == CLD_DUMPED) {
3145
3146                         log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
3147                         r = EXIT_FAILURE;
3148                         break;
3149                 } else {
3150                         log_error("Container %s failed due to unknown reason.", arg_machine);
3151                         r = EXIT_FAILURE;
3152                         break;
3153                 }
3154         }
3155
3156 finish:
3157         loop_remove(loop_nr, &image_fd);
3158
3159         if (pid > 0)
3160                 kill(pid, SIGKILL);
3161
3162         free(arg_directory);
3163         free(arg_machine);
3164         free(arg_user);
3165         strv_free(arg_setenv);
3166         strv_free(arg_network_interfaces);
3167         strv_free(arg_network_macvlan);
3168         strv_free(arg_bind);
3169         strv_free(arg_bind_ro);
3170
3171         return r;
3172 }