src/nspawn/nspawn-mount.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <sys/mount.h>
   4 #include <linux/magic.h>
   5
   6 #include "alloc-util.h"
   7 #include "escape.h"
   8 #include "fd-util.h"
   9 #include "format-util.h"
  10 #include "fs-util.h"
  11 #include "label.h"
  12 #include "mkdir.h"
  13 #include "mount-util.h"
  14 #include "mountpoint-util.h"
  15 #include "nspawn-mount.h"
  16 #include "parse-util.h"
  17 #include "path-util.h"
  18 #include "rm-rf.h"
  19 #include "set.h"
  20 #include "sort-util.h"
  21 #include "stat-util.h"
  22 #include "string-util.h"
  23 #include "strv.h"
  24 #include "tmpfile-util.h"
  25 #include "user-util.h"
  26
  27 CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t) {
  28         CustomMount *c, *ret;
  29
  30         assert(l);
  31         assert(n);
  32         assert(t >= 0);
  33         assert(t < _CUSTOM_MOUNT_TYPE_MAX);
  34
  35         c = reallocarray(*l, *n + 1, sizeof(CustomMount));
  36         if (!c)
  37                 return NULL;
  38
  39         *l = c;
  40         ret = *l + *n;
  41         (*n)++;
  42
  43         *ret = (CustomMount) { .type = t };
  44
  45         return ret;
  46 }
  47
  48 void custom_mount_free_all(CustomMount *l, size_t n) {
  49         size_t i;
  50
  51         for (i = 0; i < n; i++) {
  52                 CustomMount *m = l + i;
  53
  54                 free(m->source);
  55                 free(m->destination);
  56                 free(m->options);
  57
  58                 if (m->work_dir) {
  59                         (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
  60                         free(m->work_dir);
  61                 }
  62
  63                 if (m->rm_rf_tmpdir) {
  64                         (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL);
  65                         free(m->rm_rf_tmpdir);
  66                 }
  67
  68                 strv_free(m->lower);
  69                 free(m->type_argument);
  70         }
  71
  72         free(l);
  73 }
  74
  75 static int custom_mount_compare(const CustomMount *a, const CustomMount *b) {
  76         int r;
  77
  78         r = path_compare(a->destination, b->destination);
  79         if (r != 0)
  80                 return r;
  81
  82         return CMP(a->type, b->type);
  83 }
  84
  85 static bool source_path_is_valid(const char *p) {
  86         assert(p);
  87
  88         if (*p == '+')
  89                 p++;
  90
  91         return path_is_absolute(p);
  92 }
  93
  94 static char *resolve_source_path(const char *dest, const char *source) {
  95
  96         if (!source)
  97                 return NULL;
  98
  99         if (source[0] == '+')
 100                 return path_join(dest, source + 1);
 101
 102         return strdup(source);
 103 }
 104
 105 int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) {
 106         size_t i;
 107         int r;
 108
 109         /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
 110          * parent process, so that we know the temporary directories to remove on exit before we fork off the
 111          * children. */
 112
 113         assert(l || n == 0);
 114
 115         /* Order the custom mounts, and make sure we have a working directory */
 116         typesafe_qsort(l, n, custom_mount_compare);
 117
 118         for (i = 0; i < n; i++) {
 119                 CustomMount *m = l + i;
 120
 121                 /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
 122                  * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
 123                  * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
 124                  * the inner child, not the outer one. Determine this here. */
 125                 m->in_userns = path_startswith(m->destination, "/proc");
 126
 127                 if (m->type == CUSTOM_MOUNT_BIND) {
 128                         if (m->source) {
 129                                 char *s;
 130
 131                                 s = resolve_source_path(dest, m->source);
 132                                 if (!s)
 133                                         return log_oom();
 134
 135                                 free_and_replace(m->source, s);
 136                         } else {
 137                                 /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
 138
 139                                 m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX");
 140                                 if (!m->rm_rf_tmpdir)
 141                                         return log_oom();
 142
 143                                 if (!mkdtemp(m->rm_rf_tmpdir)) {
 144                                         m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir);
 145                                         return log_error_errno(errno, "Failed to acquire temporary directory: %m");
 146                                 }
 147
 148                                 m->source = path_join(m->rm_rf_tmpdir, "src");
 149                                 if (!m->source)
 150                                         return log_oom();
 151
 152                                 if (mkdir(m->source, 0755) < 0)
 153                                         return log_error_errno(errno, "Failed to create %s: %m", m->source);
 154                         }
 155                 }
 156
 157                 if (m->type == CUSTOM_MOUNT_OVERLAY) {
 158                         char **j;
 159
 160                         STRV_FOREACH(j, m->lower) {
 161                                 char *s;
 162
 163                                 s = resolve_source_path(dest, *j);
 164                                 if (!s)
 165                                         return log_oom();
 166
 167                                 free_and_replace(*j, s);
 168                         }
 169
 170                         if (m->work_dir) {
 171                                 char *s;
 172
 173                                 s = resolve_source_path(dest, m->work_dir);
 174                                 if (!s)
 175                                         return log_oom();
 176
 177                                 free_and_replace(m->work_dir, s);
 178                         } else {
 179                                 assert(m->source);
 180
 181                                 r = tempfn_random(m->source, NULL, &m->work_dir);
 182                                 if (r < 0)
 183                                         return log_error_errno(r, "Failed to acquire working directory: %m");
 184                         }
 185
 186                         (void) mkdir_label(m->work_dir, 0700);
 187                 }
 188         }
 189
 190         return 0;
 191 }
 192
 193 int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 194         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 195         const char *p = s;
 196         CustomMount *m;
 197         int r;
 198
 199         assert(l);
 200         assert(n);
 201
 202         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 203         if (r < 0)
 204                 return r;
 205         if (r == 0)
 206                 return -EINVAL;
 207         if (r == 1) {
 208                 destination = strdup(source[0] == '+' ? source+1 : source);
 209                 if (!destination)
 210                         return -ENOMEM;
 211         }
 212         if (r == 2 && !isempty(p)) {
 213                 opts = strdup(p);
 214                 if (!opts)
 215                         return -ENOMEM;
 216         }
 217
 218         if (isempty(source))
 219                 source = mfree(source);
 220         else if (!source_path_is_valid(source))
 221                 return -EINVAL;
 222
 223         if (!path_is_absolute(destination))
 224                 return -EINVAL;
 225
 226         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 227         if (!m)
 228                 return -ENOMEM;
 229
 230         m->source = TAKE_PTR(source);
 231         m->destination = TAKE_PTR(destination);
 232         m->read_only = read_only;
 233         m->options = TAKE_PTR(opts);
 234
 235         return 0;
 236 }
 237
 238 int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) {
 239         _cleanup_free_ char *path = NULL, *opts = NULL;
 240         const char *p = s;
 241         CustomMount *m;
 242         int r;
 243
 244         assert(l);
 245         assert(n);
 246         assert(s);
 247
 248         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 249         if (r < 0)
 250                 return r;
 251         if (r == 0)
 252                 return -EINVAL;
 253
 254         if (isempty(p))
 255                 opts = strdup("mode=0755");
 256         else
 257                 opts = strdup(p);
 258         if (!opts)
 259                 return -ENOMEM;
 260
 261         if (!path_is_absolute(path))
 262                 return -EINVAL;
 263
 264         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 265         if (!m)
 266                 return -ENOMEM;
 267
 268         m->destination = TAKE_PTR(path);
 269         m->options = TAKE_PTR(opts);
 270
 271         return 0;
 272 }
 273
 274 int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 275         _cleanup_free_ char *upper = NULL, *destination = NULL;
 276         _cleanup_strv_free_ char **lower = NULL;
 277         CustomMount *m;
 278         int k;
 279
 280         k = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 281         if (k < 0)
 282                 return k;
 283         if (k < 2)
 284                 return -EADDRNOTAVAIL;
 285         if (k == 2) {
 286                 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
 287                  * we'll also define the destination mount point the same as the upper. */
 288
 289                 if (!source_path_is_valid(lower[0]) ||
 290                     !source_path_is_valid(lower[1]))
 291                         return -EINVAL;
 292
 293                 upper = TAKE_PTR(lower[1]);
 294
 295                 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
 296                 if (!destination)
 297                         return -ENOMEM;
 298         } else {
 299                 char **i;
 300
 301                 /* If more than two parameters are specified, the last one is the destination, the second to last one
 302                  * the "upper", and all before that the "lower" directories. */
 303
 304                 destination = lower[k - 1];
 305                 upper = TAKE_PTR(lower[k - 2]);
 306
 307                 STRV_FOREACH(i, lower)
 308                         if (!source_path_is_valid(*i))
 309                                 return -EINVAL;
 310
 311                 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
 312                  * in /var/tmp */
 313                 if (isempty(upper))
 314                         upper = mfree(upper);
 315                 else if (!source_path_is_valid(upper))
 316                         return -EINVAL;
 317
 318                 if (!path_is_absolute(destination))
 319                         return -EINVAL;
 320         }
 321
 322         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 323         if (!m)
 324                 return -ENOMEM;
 325
 326         m->destination = TAKE_PTR(destination);
 327         m->source = TAKE_PTR(upper);
 328         m->lower = TAKE_PTR(lower);
 329         m->read_only = read_only;
 330
 331         return 0;
 332 }
 333
 334 int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) {
 335         _cleanup_free_ char *path = NULL;
 336         CustomMount *m;
 337
 338         assert(l);
 339         assert(n);
 340         assert(s);
 341
 342         if (!path_is_absolute(s))
 343                 return -EINVAL;
 344
 345         path = strdup(s);
 346         if (!path)
 347                 return -ENOMEM;
 348
 349         m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE);
 350         if (!m)
 351                 return -ENOMEM;
 352
 353         m->destination = TAKE_PTR(path);
 354         return 0;
 355 }
 356
 357 int tmpfs_patch_options(
 358                 const char *options,
 359                 uid_t uid_shift,
 360                 const char *selinux_apifs_context,
 361                 char **ret) {
 362
 363         char *buf = NULL;
 364
 365         if (uid_shift != UID_INVALID) {
 366                 if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 367                              strempty(options), options ? "," : "",
 368                              uid_shift, uid_shift) < 0)
 369                         return -ENOMEM;
 370
 371                 options = buf;
 372         }
 373
 374 #if HAVE_SELINUX
 375         if (selinux_apifs_context) {
 376                 char *t;
 377
 378                 t = strjoin(strempty(options), options ? "," : "",
 379                             "context=\"", selinux_apifs_context, "\"");
 380                 free(buf);
 381                 if (!t)
 382                         return -ENOMEM;
 383
 384                 buf = t;
 385         }
 386 #endif
 387
 388         if (!buf && options) {
 389                 buf = strdup(options);
 390                 if (!buf)
 391                         return -ENOMEM;
 392         }
 393         *ret = buf;
 394
 395         return !!buf;
 396 }
 397
 398 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 399         const char *full, *top, *x;
 400         int r;
 401         unsigned long extra_flags = 0;
 402
 403         top = prefix_roota(dest, "/sys");
 404         r = path_is_fs_type(top, SYSFS_MAGIC);
 405         if (r < 0)
 406                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 407         /* /sys might already be mounted as sysfs by the outer child in the
 408          * !netns case. In this case, it's all good. Don't touch it because we
 409          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 410          */
 411         if (r > 0)
 412                 return 0;
 413
 414         full = prefix_roota(top, "/full");
 415
 416         (void) mkdir(full, 0755);
 417
 418         if (mount_settings & MOUNT_APPLY_APIVFS_RO)
 419                 extra_flags |= MS_RDONLY;
 420
 421         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 422                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 423         if (r < 0)
 424                 return r;
 425
 426         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 427                 _cleanup_free_ char *from = NULL, *to = NULL;
 428
 429                 from = path_join(full, x);
 430                 if (!from)
 431                         return log_oom();
 432
 433                 to = path_join(top, x);
 434                 if (!to)
 435                         return log_oom();
 436
 437                 (void) mkdir(to, 0755);
 438
 439                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 440                 if (r < 0)
 441                         return r;
 442
 443                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 444                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 445                 if (r < 0)
 446                         return r;
 447         }
 448
 449         r = umount_verbose(full);
 450         if (r < 0)
 451                 return r;
 452
 453         if (rmdir(full) < 0)
 454                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 455
 456         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 457          * remount /sys read-only.
 458          */
 459         x = prefix_roota(top, "/fs/cgroup");
 460         (void) mkdir_p(x, 0755);
 461
 462         return mount_verbose(LOG_ERR, NULL, top, NULL,
 463                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 464 }
 465
 466 static int mkdir_userns(const char *path, mode_t mode, uid_t uid_shift) {
 467         int r;
 468
 469         assert(path);
 470
 471         r = mkdir_errno_wrapper(path, mode);
 472         if (r < 0 && r != -EEXIST)
 473                 return r;
 474
 475         if (uid_shift == UID_INVALID)
 476                 return 0;
 477
 478         if (lchown(path, uid_shift, uid_shift) < 0)
 479                 return -errno;
 480
 481         return 0;
 482 }
 483
 484 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, uid_t uid_shift) {
 485         const char *p, *e;
 486         int r;
 487
 488         assert(path);
 489
 490         if (prefix && !path_startswith(path, prefix))
 491                 return -ENOTDIR;
 492
 493         /* create every parent directory in the path, except the last component */
 494         p = path + strspn(path, "/");
 495         for (;;) {
 496                 char t[strlen(path) + 1];
 497
 498                 e = p + strcspn(p, "/");
 499                 p = e + strspn(e, "/");
 500
 501                 /* Is this the last component? If so, then we're done */
 502                 if (*p == 0)
 503                         break;
 504
 505                 memcpy(t, path, e - path);
 506                 t[e-path] = 0;
 507
 508                 if (prefix && path_startswith(prefix, t))
 509                         continue;
 510
 511                 r = mkdir_userns(t, mode, uid_shift);
 512                 if (r < 0)
 513                         return r;
 514         }
 515
 516         return mkdir_userns(path, mode, uid_shift);
 517 }
 518
 519 int mount_all(const char *dest,
 520               MountSettingsMask mount_settings,
 521               uid_t uid_shift,
 522               const char *selinux_apifs_context) {
 523
 524 #define PROC_INACCESSIBLE_REG(path)                                     \
 525         { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
 526           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 527         { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 528           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 529
 530 #define PROC_READ_ONLY(path)                                            \
 531         { (path), (path), NULL, NULL, MS_BIND,                          \
 532           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 533         { NULL,   (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 534           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 535
 536         typedef struct MountPoint {
 537                 const char *what;
 538                 const char *where;
 539                 const char *type;
 540                 const char *options;
 541                 unsigned long flags;
 542                 MountSettingsMask mount_settings;
 543         } MountPoint;
 544
 545         static const MountPoint mount_table[] = {
 546                 /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */
 547                 { "proc",            "/proc",           "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,
 548                   MOUNT_FATAL|MOUNT_IN_USERNS },
 549
 550                 { "/proc/sys",       "/proc/sys",       NULL,    NULL,        MS_BIND,
 551                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 552
 553                 { "/proc/sys/net",   "/proc/sys/net",   NULL,    NULL,        MS_BIND,
 554                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 555
 556                 { NULL,              "/proc/sys",       NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 557                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 558
 559                 /* Make these files inaccessible to container payloads: they potentially leak information about kernel
 560                  * internals or the host's execution environment to the container */
 561                 PROC_INACCESSIBLE_REG("/proc/kallsyms"),
 562                 PROC_INACCESSIBLE_REG("/proc/kcore"),
 563                 PROC_INACCESSIBLE_REG("/proc/keys"),
 564                 PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
 565                 PROC_INACCESSIBLE_REG("/proc/timer_list"),
 566
 567                 /* Make these directories read-only to container payloads: they show hardware information, and in some
 568                  * cases contain tunables the container really shouldn't have access to. */
 569                 PROC_READ_ONLY("/proc/acpi"),
 570                 PROC_READ_ONLY("/proc/apm"),
 571                 PROC_READ_ONLY("/proc/asound"),
 572                 PROC_READ_ONLY("/proc/bus"),
 573                 PROC_READ_ONLY("/proc/fs"),
 574                 PROC_READ_ONLY("/proc/irq"),
 575                 PROC_READ_ONLY("/proc/scsi"),
 576
 577                 { "mqueue",          "/dev/mqueue",     "mqueue", NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV,
 578                   MOUNT_IN_USERNS },
 579
 580                 /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
 581                 { "tmpfs",           "/tmp",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 582                   MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP },
 583                 { "tmpfs",           "/sys",            "tmpfs", "mode=555",  MS_NOSUID|MS_NOEXEC|MS_NODEV,
 584                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS },
 585                 { "sysfs",           "/sys",            "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
 586                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO },    /* skipped if above was mounted */
 587                 { "sysfs",           "/sys",            "sysfs", NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,
 588                   MOUNT_FATAL },                          /* skipped if above was mounted */
 589                 { "tmpfs",           "/dev",            "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,
 590                   MOUNT_FATAL },
 591                 { "tmpfs",           "/dev/shm",        "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 592                   MOUNT_FATAL },
 593                 { "tmpfs",           "/run",            "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 594                   MOUNT_FATAL },
 595
 596 #if HAVE_SELINUX
 597                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL,    NULL,        MS_BIND,
 598                   0 },  /* Bind mount first */
 599                 { NULL,              "/sys/fs/selinux", NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 600                   0 },  /* Then, make it r/o */
 601 #endif
 602         };
 603
 604         bool use_userns = (mount_settings & MOUNT_USE_USERNS);
 605         bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS);
 606         bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO);
 607         bool in_userns = (mount_settings & MOUNT_IN_USERNS);
 608         bool tmpfs_tmp = (mount_settings & MOUNT_APPLY_TMPFS_TMP);
 609         size_t k;
 610         int r;
 611
 612         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 613                 _cleanup_free_ char *where = NULL, *options = NULL;
 614                 const char *o;
 615                 bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL);
 616
 617                 if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS))
 618                         continue;
 619
 620                 if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS))
 621                         continue;
 622
 623                 if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO))
 624                         continue;
 625
 626                 if (!tmpfs_tmp && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_TMPFS_TMP))
 627                         continue;
 628
 629                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL);
 630                 if (r < 0)
 631                         return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where);
 632
 633                 /* Skip this entry if it is not a remount. */
 634                 if (mount_table[k].what) {
 635                         r = path_is_mount_point(where, NULL, 0);
 636                         if (r < 0 && r != -ENOENT)
 637                                 return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 638                         if (r > 0)
 639                                 continue;
 640                 }
 641
 642                 r = mkdir_userns_p(dest, where, 0755, (use_userns && !in_userns) ? uid_shift : UID_INVALID);
 643                 if (r < 0 && r != -EEXIST) {
 644                         if (fatal && r != -EROFS)
 645                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 646
 647                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 648                         /* If we failed mkdir() or chown() due to the root
 649                          * directory being read only, attempt to mount this fs
 650                          * anyway and let mount_verbose log any errors */
 651                         if (r != -EROFS)
 652                                 continue;
 653                 }
 654
 655                 o = mount_table[k].options;
 656                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 657                         r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
 658                         if (r < 0)
 659                                 return log_oom();
 660                         if (r > 0)
 661                                 o = options;
 662                 }
 663
 664                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 665                                   mount_table[k].what,
 666                                   where,
 667                                   mount_table[k].type,
 668                                   mount_table[k].flags,
 669                                   o);
 670                 if (r < 0 && fatal)
 671                         return r;
 672         }
 673
 674         return 0;
 675 }
 676
 677 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 678         const char *p = options;
 679         unsigned long flags = *mount_flags;
 680         char *opts = NULL;
 681         int r;
 682
 683         assert(options);
 684
 685         for (;;) {
 686                 _cleanup_free_ char *word = NULL;
 687
 688                 r = extract_first_word(&p, &word, ",", 0);
 689                 if (r < 0)
 690                         return log_error_errno(r, "Failed to extract mount option: %m");
 691                 if (r == 0)
 692                         break;
 693
 694                 if (streq(word, "rbind"))
 695                         flags |= MS_REC;
 696                 else if (streq(word, "norbind"))
 697                         flags &= ~MS_REC;
 698                 else {
 699                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 700                                                "Invalid bind mount option: %s",
 701                                                word);
 702                 }
 703         }
 704
 705         *mount_flags = flags;
 706         /* in the future mount_opts will hold string options for mount(2) */
 707         *mount_opts = opts;
 708
 709         return 0;
 710 }
 711
 712 static int mount_bind(const char *dest, CustomMount *m) {
 713         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 714         unsigned long mount_flags = MS_BIND | MS_REC;
 715         struct stat source_st, dest_st;
 716         int r;
 717
 718         assert(dest);
 719         assert(m);
 720
 721         if (m->options) {
 722                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 723                 if (r < 0)
 724                         return r;
 725         }
 726
 727         if (stat(m->source, &source_st) < 0)
 728                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 729
 730         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 731         if (r < 0)
 732                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 733         if (r > 0) { /* Path exists already? */
 734
 735                 if (stat(where, &dest_st) < 0)
 736                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 737
 738                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode))
 739                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 740                                                "Cannot bind mount directory %s on file %s.",
 741                                                m->source, where);
 742
 743                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode))
 744                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 745                                                "Cannot bind mount file %s on directory %s.",
 746                                                m->source, where);
 747
 748         } else { /* Path doesn't exist yet? */
 749                 r = mkdir_parents_label(where, 0755);
 750                 if (r < 0)
 751                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 752
 753                 /* Create the mount point. Any non-directory file can be
 754                 * mounted on any non-directory file (regular, fifo, socket,
 755                 * char, block).
 756                 */
 757                 if (S_ISDIR(source_st.st_mode))
 758                         r = mkdir_label(where, 0755);
 759                 else
 760                         r = touch(where);
 761                 if (r < 0)
 762                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 763         }
 764
 765         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 766         if (r < 0)
 767                 return r;
 768
 769         if (m->read_only) {
 770                 r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL);
 771                 if (r < 0)
 772                         return log_error_errno(r, "Read-only bind mount failed: %m");
 773         }
 774
 775         return 0;
 776 }
 777
 778 static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) {
 779
 780         const char *options;
 781         _cleanup_free_ char *buf = NULL, *where = NULL;
 782         int r;
 783
 784         assert(dest);
 785         assert(m);
 786
 787         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 788         if (r < 0)
 789                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 790         if (r == 0) { /* Doesn't exist yet? */
 791                 r = mkdir_p_label(where, 0755);
 792                 if (r < 0)
 793                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 794         }
 795
 796         r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
 797         if (r < 0)
 798                 return log_oom();
 799         options = r > 0 ? buf : m->options;
 800
 801         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 802 }
 803
 804 static char *joined_and_escaped_lower_dirs(char **lower) {
 805         _cleanup_strv_free_ char **sv = NULL;
 806
 807         sv = strv_copy(lower);
 808         if (!sv)
 809                 return NULL;
 810
 811         strv_reverse(sv);
 812
 813         if (!strv_shell_escape(sv, ",:"))
 814                 return NULL;
 815
 816         return strv_join(sv, ":");
 817 }
 818
 819 static int mount_overlay(const char *dest, CustomMount *m) {
 820         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 821         const char *options;
 822         int r;
 823
 824         assert(dest);
 825         assert(m);
 826
 827         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 828         if (r < 0)
 829                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 830         if (r == 0) { /* Doesn't exist yet? */
 831                 r = mkdir_label(where, 0755);
 832                 if (r < 0)
 833                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 834         }
 835
 836         (void) mkdir_p_label(m->source, 0755);
 837
 838         lower = joined_and_escaped_lower_dirs(m->lower);
 839         if (!lower)
 840                 return log_oom();
 841
 842         escaped_source = shell_escape(m->source, ",:");
 843         if (!escaped_source)
 844                 return log_oom();
 845
 846         if (m->read_only)
 847                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 848         else {
 849                 _cleanup_free_ char *escaped_work_dir = NULL;
 850
 851                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 852                 if (!escaped_work_dir)
 853                         return log_oom();
 854
 855                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 856         }
 857
 858         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 859 }
 860
 861 static int mount_inaccessible(const char *dest, CustomMount *m) {
 862         _cleanup_free_ char *where = NULL;
 863         const char *source;
 864         struct stat st;
 865         int r;
 866
 867         assert(dest);
 868         assert(m);
 869
 870         r = chase_symlinks_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st, NULL);
 871         if (r < 0) {
 872                 log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination);
 873                 return m->graceful ? 0 : r;
 874         }
 875
 876         assert_se(source = mode_to_inaccessible_node(st.st_mode));
 877
 878         r = mount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL);
 879         if (r < 0)
 880                 return m->graceful ? 0 : r;
 881
 882         r = mount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL);
 883         if (r < 0) {
 884                 umount_verbose(where);
 885                 return m->graceful ? 0 : r;
 886         }
 887
 888         return 0;
 889 }
 890
 891 static int mount_arbitrary(const char *dest, CustomMount *m) {
 892         _cleanup_free_ char *where = NULL;
 893         int r;
 894
 895         assert(dest);
 896         assert(m);
 897
 898         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 899         if (r < 0)
 900                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 901         if (r == 0) { /* Doesn't exist yet? */
 902                 r = mkdir_p_label(where, 0755);
 903                 if (r < 0)
 904                         return log_error_errno(r, "Creating mount point for mount %s failed: %m", where);
 905         }
 906
 907         return mount_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options);
 908 }
 909
 910 int mount_custom(
 911                 const char *dest,
 912                 CustomMount *mounts, size_t n,
 913                 uid_t uid_shift,
 914                 const char *selinux_apifs_context,
 915                 MountSettingsMask mount_settings) {
 916
 917         size_t i;
 918         int r;
 919
 920         assert(dest);
 921
 922         for (i = 0; i < n; i++) {
 923                 CustomMount *m = mounts + i;
 924
 925                 if ((mount_settings & MOUNT_IN_USERNS) != m->in_userns)
 926                         continue;
 927
 928                 if (mount_settings & MOUNT_ROOT_ONLY && !path_equal(m->destination, "/"))
 929                         continue;
 930
 931                 if (mount_settings & MOUNT_NON_ROOT_ONLY && path_equal(m->destination, "/"))
 932                         continue;
 933
 934                 switch (m->type) {
 935
 936                 case CUSTOM_MOUNT_BIND:
 937                         r = mount_bind(dest, m);
 938                         break;
 939
 940                 case CUSTOM_MOUNT_TMPFS:
 941                         r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context);
 942                         break;
 943
 944                 case CUSTOM_MOUNT_OVERLAY:
 945                         r = mount_overlay(dest, m);
 946                         break;
 947
 948                 case CUSTOM_MOUNT_INACCESSIBLE:
 949                         r = mount_inaccessible(dest, m);
 950                         break;
 951
 952                 case CUSTOM_MOUNT_ARBITRARY:
 953                         r = mount_arbitrary(dest, m);
 954                         break;
 955
 956                 default:
 957                         assert_not_reached("Unknown custom mount type");
 958                 }
 959
 960                 if (r < 0)
 961                         return r;
 962         }
 963
 964         return 0;
 965 }
 966
 967 static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
 968
 969         _cleanup_free_ char *buf = NULL;
 970         const char *p, *options;
 971         int r;
 972
 973         assert(directory);
 974
 975         /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
 976
 977         r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
 978         if (r < 0)
 979                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
 980
 981         p = prefix_roota(directory, "/var");
 982         r = mkdir(p, 0755);
 983         if (r < 0 && errno != EEXIST)
 984                 return log_error_errno(errno, "Failed to create %s: %m", directory);
 985
 986         options = "mode=755";
 987         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
 988         if (r < 0)
 989                 return log_oom();
 990         if (r > 0)
 991                 options = buf;
 992
 993         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
 994 }
 995
 996 static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
 997
 998         bool tmpfs_mounted = false, bind_mounted = false;
 999         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1000         _cleanup_free_ char *buf = NULL, *bindir = NULL;
1001         const char *f, *t, *options;
1002         struct stat st;
1003         int r;
1004
1005         assert(directory);
1006
1007         /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
1008          * that read-only. Before we start setting this up let's validate if the image has the /usr merge
1009          * implemented, and let's output a friendly log message if it hasn't. */
1010
1011         bindir = path_join(directory, "/bin");
1012         if (!bindir)
1013                 return log_oom();
1014         if (lstat(bindir, &st) < 0) {
1015                 if (errno != ENOENT)
1016                         return log_error_errno(errno, "Failed to stat /bin directory below image: %m");
1017
1018                 /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
1019                  * rest. */
1020         } else if (S_ISDIR(st.st_mode))
1021                 return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
1022                                        "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
1023                                        "Please work with your distribution and help them adopt the merged /usr scheme.");
1024         else if (!S_ISLNK(st.st_mode))
1025                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1026                                        "Error starting image: if --volatile=yes is used /bin must be a symlink (for merged /usr support) or non-existent (in which case a symlink is created automatically).");
1027
1028         if (!mkdtemp(template))
1029                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1030
1031         options = "mode=755";
1032         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1033         if (r < 0)
1034                 goto fail;
1035         if (r > 0)
1036                 options = buf;
1037
1038         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1039         if (r < 0)
1040                 goto fail;
1041
1042         tmpfs_mounted = true;
1043
1044         f = prefix_roota(directory, "/usr");
1045         t = prefix_roota(template, "/usr");
1046
1047         r = mkdir(t, 0755);
1048         if (r < 0 && errno != EEXIST) {
1049                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1050                 goto fail;
1051         }
1052
1053         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1054         if (r < 0)
1055                 goto fail;
1056
1057         bind_mounted = true;
1058
1059         r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL);
1060         if (r < 0) {
1061                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1062                 goto fail;
1063         }
1064
1065         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1066         if (r < 0)
1067                 goto fail;
1068
1069         (void) rmdir(template);
1070
1071         return 0;
1072
1073 fail:
1074         if (bind_mounted)
1075                 (void) umount_verbose(t);
1076
1077         if (tmpfs_mounted)
1078                 (void) umount_verbose(template);
1079         (void) rmdir(template);
1080         return r;
1081 }
1082
1083 static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1084
1085         _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL;
1086         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1087         const char *upper, *work, *options;
1088         bool tmpfs_mounted = false;
1089         int r;
1090
1091         assert(directory);
1092
1093         /* --volatile=overlay means we mount an overlayfs to the root dir. */
1094
1095         if (!mkdtemp(template))
1096                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1097
1098         options = "mode=755";
1099         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1100         if (r < 0)
1101                 goto finish;
1102         if (r > 0)
1103                 options = buf;
1104
1105         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1106         if (r < 0)
1107                 goto finish;
1108
1109         tmpfs_mounted = true;
1110
1111         upper = strjoina(template, "/upper");
1112         work = strjoina(template, "/work");
1113
1114         if (mkdir(upper, 0755) < 0) {
1115                 r = log_error_errno(errno, "Failed to create %s: %m", upper);
1116                 goto finish;
1117         }
1118         if (mkdir(work, 0755) < 0) {
1119                 r = log_error_errno(errno, "Failed to create %s: %m", work);
1120                 goto finish;
1121         }
1122
1123         /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
1124          * that the kernel allows us to do that without going through some mount point rearrangements. */
1125
1126         escaped_directory = shell_escape(directory, ",:");
1127         escaped_upper = shell_escape(upper, ",:");
1128         escaped_work = shell_escape(work, ",:");
1129         if (!escaped_directory || !escaped_upper || !escaped_work) {
1130                 r = -ENOMEM;
1131                 goto finish;
1132         }
1133
1134         options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work);
1135         r = mount_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options);
1136
1137 finish:
1138         if (tmpfs_mounted)
1139                 (void) umount_verbose(template);
1140
1141         (void) rmdir(template);
1142         return r;
1143 }
1144
1145 int setup_volatile_mode(
1146                 const char *directory,
1147                 VolatileMode mode,
1148                 uid_t uid_shift,
1149                 const char *selinux_apifs_context) {
1150
1151         switch (mode) {
1152
1153         case VOLATILE_YES:
1154                 return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
1155
1156         case VOLATILE_STATE:
1157                 return setup_volatile_state(directory, uid_shift, selinux_apifs_context);
1158
1159         case VOLATILE_OVERLAY:
1160                 return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
1161
1162         default:
1163                 return 0;
1164         }
1165 }
1166
1167 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1168 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1169         _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1170         const char *p = s;
1171         int r;
1172
1173         assert(pivot_root_new);
1174         assert(pivot_root_old);
1175
1176         r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1177         if (r < 0)
1178                 return r;
1179         if (r == 0)
1180                 return -EINVAL;
1181
1182         if (isempty(p))
1183                 root_old = NULL;
1184         else {
1185                 root_old = strdup(p);
1186                 if (!root_old)
1187                         return -ENOMEM;
1188         }
1189
1190         if (!path_is_absolute(root_new))
1191                 return -EINVAL;
1192         if (root_old && !path_is_absolute(root_old))
1193                 return -EINVAL;
1194
1195         free_and_replace(*pivot_root_new, root_new);
1196         free_and_replace(*pivot_root_old, root_old);
1197
1198         return 0;
1199 }
1200
1201 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1202         _cleanup_free_ char *directory_pivot_root_new = NULL;
1203         _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1204         char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
1205         bool remove_pivot_tmp = false;
1206         int r;
1207
1208         assert(directory);
1209
1210         if (!pivot_root_new)
1211                 return 0;
1212
1213         /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1214          * If pivot_root_old is NULL, the existing / disappears.
1215          * This requires a temporary directory, pivot_tmp, which is
1216          * not a child of either.
1217          *
1218          * This is typically used for OSTree-style containers, where
1219          * the root partition contains several sysroots which could be
1220          * run. Normally, one would be chosen by the bootloader and
1221          * pivoted to / by initramfs.
1222          *
1223          * For example, for an OSTree deployment, pivot_root_new
1224          * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1225          * code doesn’t do the /var mount which OSTree expects: use
1226          * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1227          *
1228          * So in the OSTree case, we’ll end up with something like:
1229          *  - directory = /tmp/nspawn-root-123456
1230          *  - pivot_root_new = /ostree/deploy/os/deploy/123abc
1231          *  - pivot_root_old = /sysroot
1232          *  - directory_pivot_root_new =
1233          *       /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1234          *  - pivot_tmp = /tmp/nspawn-pivot-123456
1235          *  - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1236          *
1237          * Requires all file systems at directory and below to be mounted
1238          * MS_PRIVATE or MS_SLAVE so they can be moved.
1239          */
1240         directory_pivot_root_new = path_join(directory, pivot_root_new);
1241         if (!directory_pivot_root_new)
1242                 return log_oom();
1243
1244         /* Remount directory_pivot_root_new to make it movable. */
1245         r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1246         if (r < 0)
1247                 goto done;
1248
1249         if (pivot_root_old) {
1250                 if (!mkdtemp(pivot_tmp)) {
1251                         r = log_error_errno(errno, "Failed to create temporary directory: %m");
1252                         goto done;
1253                 }
1254
1255                 remove_pivot_tmp = true;
1256                 pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old);
1257                 if (!pivot_tmp_pivot_root_old) {
1258                         r = log_oom();
1259                         goto done;
1260                 }
1261
1262                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1263                 if (r < 0)
1264                         goto done;
1265
1266                 r = mount_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1267                 if (r < 0)
1268                         goto done;
1269
1270                 r = mount_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1271                 if (r < 0)
1272                         goto done;
1273         } else {
1274                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1275                 if (r < 0)
1276                         goto done;
1277         }
1278
1279 done:
1280         if (remove_pivot_tmp)
1281                 (void) rmdir(pivot_tmp);
1282
1283         return r;
1284 }