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 static int allocate_temporary_source(CustomMount *m) {
 106         assert(m);
 107         assert(!m->source);
 108         assert(!m->rm_rf_tmpdir);
 109
 110         m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX");
 111         if (!m->rm_rf_tmpdir)
 112                 return log_oom();
 113
 114         if (!mkdtemp(m->rm_rf_tmpdir)) {
 115                 m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir);
 116                 return log_error_errno(errno, "Failed to acquire temporary directory: %m");
 117         }
 118
 119         m->source = path_join(m->rm_rf_tmpdir, "src");
 120         if (!m->source)
 121                 return log_oom();
 122
 123         if (mkdir(m->source, 0755) < 0)
 124                 return log_error_errno(errno, "Failed to create %s: %m", m->source);
 125
 126         return 0;
 127 }
 128
 129 int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) {
 130         size_t i;
 131         int r;
 132
 133         /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
 134          * parent process, so that we know the temporary directories to remove on exit before we fork off the
 135          * children. */
 136
 137         assert(l || n == 0);
 138
 139         /* Order the custom mounts, and make sure we have a working directory */
 140         typesafe_qsort(l, n, custom_mount_compare);
 141
 142         for (i = 0; i < n; i++) {
 143                 CustomMount *m = l + i;
 144
 145                 /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
 146                  * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
 147                  * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
 148                  * the inner child, not the outer one. Determine this here. */
 149                 m->in_userns = path_startswith(m->destination, "/proc");
 150
 151                 if (m->type == CUSTOM_MOUNT_BIND) {
 152                         if (m->source) {
 153                                 char *s;
 154
 155                                 s = resolve_source_path(dest, m->source);
 156                                 if (!s)
 157                                         return log_oom();
 158
 159                                 free_and_replace(m->source, s);
 160                         } else {
 161                                 /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
 162
 163                                 r = allocate_temporary_source(m);
 164                                 if (r < 0)
 165                                         return r;
 166                         }
 167                 }
 168
 169                 if (m->type == CUSTOM_MOUNT_OVERLAY) {
 170                         char **j;
 171
 172                         STRV_FOREACH(j, m->lower) {
 173                                 char *s;
 174
 175                                 s = resolve_source_path(dest, *j);
 176                                 if (!s)
 177                                         return log_oom();
 178
 179                                 free_and_replace(*j, s);
 180                         }
 181
 182                         if (m->source) {
 183                                 char *s;
 184
 185                                 s = resolve_source_path(dest, m->source);
 186                                 if (!s)
 187                                         return log_oom();
 188
 189                                 free_and_replace(m->source, s);
 190                         } else {
 191                                 r = allocate_temporary_source(m);
 192                                 if (r < 0)
 193                                         return r;
 194                         }
 195
 196                         if (m->work_dir) {
 197                                 char *s;
 198
 199                                 s = resolve_source_path(dest, m->work_dir);
 200                                 if (!s)
 201                                         return log_oom();
 202
 203                                 free_and_replace(m->work_dir, s);
 204                         } else {
 205                                 r = tempfn_random(m->source, NULL, &m->work_dir);
 206                                 if (r < 0)
 207                                         return log_error_errno(r, "Failed to acquire working directory: %m");
 208                         }
 209
 210                         (void) mkdir_label(m->work_dir, 0700);
 211                 }
 212         }
 213
 214         return 0;
 215 }
 216
 217 int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 218         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 219         const char *p = s;
 220         CustomMount *m;
 221         int r;
 222
 223         assert(l);
 224         assert(n);
 225
 226         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 227         if (r < 0)
 228                 return r;
 229         if (r == 0)
 230                 return -EINVAL;
 231         if (r == 1) {
 232                 destination = strdup(source[0] == '+' ? source+1 : source);
 233                 if (!destination)
 234                         return -ENOMEM;
 235         }
 236         if (r == 2 && !isempty(p)) {
 237                 opts = strdup(p);
 238                 if (!opts)
 239                         return -ENOMEM;
 240         }
 241
 242         if (isempty(source))
 243                 source = mfree(source);
 244         else if (!source_path_is_valid(source))
 245                 return -EINVAL;
 246
 247         if (!path_is_absolute(destination))
 248                 return -EINVAL;
 249
 250         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 251         if (!m)
 252                 return -ENOMEM;
 253
 254         m->source = TAKE_PTR(source);
 255         m->destination = TAKE_PTR(destination);
 256         m->read_only = read_only;
 257         m->options = TAKE_PTR(opts);
 258
 259         return 0;
 260 }
 261
 262 int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) {
 263         _cleanup_free_ char *path = NULL, *opts = NULL;
 264         const char *p = s;
 265         CustomMount *m;
 266         int r;
 267
 268         assert(l);
 269         assert(n);
 270         assert(s);
 271
 272         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 273         if (r < 0)
 274                 return r;
 275         if (r == 0)
 276                 return -EINVAL;
 277
 278         if (isempty(p))
 279                 opts = strdup("mode=0755");
 280         else
 281                 opts = strdup(p);
 282         if (!opts)
 283                 return -ENOMEM;
 284
 285         if (!path_is_absolute(path))
 286                 return -EINVAL;
 287
 288         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 289         if (!m)
 290                 return -ENOMEM;
 291
 292         m->destination = TAKE_PTR(path);
 293         m->options = TAKE_PTR(opts);
 294
 295         return 0;
 296 }
 297
 298 int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 299         _cleanup_free_ char *upper = NULL, *destination = NULL;
 300         _cleanup_strv_free_ char **lower = NULL;
 301         CustomMount *m;
 302         int k;
 303
 304         k = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 305         if (k < 0)
 306                 return k;
 307         if (k < 2)
 308                 return -EADDRNOTAVAIL;
 309         if (k == 2) {
 310                 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
 311                  * we'll also define the destination mount point the same as the upper. */
 312
 313                 if (!source_path_is_valid(lower[0]) ||
 314                     !source_path_is_valid(lower[1]))
 315                         return -EINVAL;
 316
 317                 upper = TAKE_PTR(lower[1]);
 318
 319                 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
 320                 if (!destination)
 321                         return -ENOMEM;
 322         } else {
 323                 char **i;
 324
 325                 /* If more than two parameters are specified, the last one is the destination, the second to last one
 326                  * the "upper", and all before that the "lower" directories. */
 327
 328                 destination = lower[k - 1];
 329                 upper = TAKE_PTR(lower[k - 2]);
 330
 331                 STRV_FOREACH(i, lower)
 332                         if (!source_path_is_valid(*i))
 333                                 return -EINVAL;
 334
 335                 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
 336                  * in /var/tmp */
 337                 if (isempty(upper))
 338                         upper = mfree(upper);
 339                 else if (!source_path_is_valid(upper))
 340                         return -EINVAL;
 341
 342                 if (!path_is_absolute(destination))
 343                         return -EINVAL;
 344         }
 345
 346         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 347         if (!m)
 348                 return -ENOMEM;
 349
 350         m->destination = TAKE_PTR(destination);
 351         m->source = TAKE_PTR(upper);
 352         m->lower = TAKE_PTR(lower);
 353         m->read_only = read_only;
 354
 355         return 0;
 356 }
 357
 358 int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) {
 359         _cleanup_free_ char *path = NULL;
 360         CustomMount *m;
 361
 362         assert(l);
 363         assert(n);
 364         assert(s);
 365
 366         if (!path_is_absolute(s))
 367                 return -EINVAL;
 368
 369         path = strdup(s);
 370         if (!path)
 371                 return -ENOMEM;
 372
 373         m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE);
 374         if (!m)
 375                 return -ENOMEM;
 376
 377         m->destination = TAKE_PTR(path);
 378         return 0;
 379 }
 380
 381 int tmpfs_patch_options(
 382                 const char *options,
 383                 uid_t uid_shift,
 384                 const char *selinux_apifs_context,
 385                 char **ret) {
 386
 387         char *buf = NULL;
 388
 389         if (uid_shift != UID_INVALID) {
 390                 if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 391                              strempty(options), options ? "," : "",
 392                              uid_shift, uid_shift) < 0)
 393                         return -ENOMEM;
 394
 395                 options = buf;
 396         }
 397
 398 #if HAVE_SELINUX
 399         if (selinux_apifs_context) {
 400                 char *t;
 401
 402                 t = strjoin(strempty(options), options ? "," : "",
 403                             "context=\"", selinux_apifs_context, "\"");
 404                 free(buf);
 405                 if (!t)
 406                         return -ENOMEM;
 407
 408                 buf = t;
 409         }
 410 #endif
 411
 412         if (!buf && options) {
 413                 buf = strdup(options);
 414                 if (!buf)
 415                         return -ENOMEM;
 416         }
 417         *ret = buf;
 418
 419         return !!buf;
 420 }
 421
 422 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 423         const char *full, *top, *x;
 424         int r;
 425         unsigned long extra_flags = 0;
 426
 427         top = prefix_roota(dest, "/sys");
 428         r = path_is_fs_type(top, SYSFS_MAGIC);
 429         if (r < 0)
 430                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 431         /* /sys might already be mounted as sysfs by the outer child in the
 432          * !netns case. In this case, it's all good. Don't touch it because we
 433          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 434          */
 435         if (r > 0)
 436                 return 0;
 437
 438         full = prefix_roota(top, "/full");
 439
 440         (void) mkdir(full, 0755);
 441
 442         if (FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO))
 443                 extra_flags |= MS_RDONLY;
 444
 445         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 446                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 447         if (r < 0)
 448                 return r;
 449
 450         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 451                 _cleanup_free_ char *from = NULL, *to = NULL;
 452
 453                 from = path_join(full, x);
 454                 if (!from)
 455                         return log_oom();
 456
 457                 to = path_join(top, x);
 458                 if (!to)
 459                         return log_oom();
 460
 461                 (void) mkdir(to, 0755);
 462
 463                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 464                 if (r < 0)
 465                         return r;
 466
 467                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 468                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 469                 if (r < 0)
 470                         return r;
 471         }
 472
 473         r = umount_verbose(full);
 474         if (r < 0)
 475                 return r;
 476
 477         if (rmdir(full) < 0)
 478                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 479
 480         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 481          * remount /sys read-only.
 482          */
 483         x = prefix_roota(top, "/fs/cgroup");
 484         (void) mkdir_p(x, 0755);
 485
 486         return mount_verbose(LOG_ERR, NULL, top, NULL,
 487                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 488 }
 489
 490 static int mkdir_userns(const char *path, mode_t mode, uid_t uid_shift) {
 491         int r;
 492
 493         assert(path);
 494
 495         r = mkdir_errno_wrapper(path, mode);
 496         if (r < 0 && r != -EEXIST)
 497                 return r;
 498
 499         if (uid_shift == UID_INVALID)
 500                 return 0;
 501
 502         if (lchown(path, uid_shift, uid_shift) < 0)
 503                 return -errno;
 504
 505         return 0;
 506 }
 507
 508 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, uid_t uid_shift) {
 509         const char *p, *e;
 510         int r;
 511
 512         assert(path);
 513
 514         if (prefix && !path_startswith(path, prefix))
 515                 return -ENOTDIR;
 516
 517         /* create every parent directory in the path, except the last component */
 518         p = path + strspn(path, "/");
 519         for (;;) {
 520                 char t[strlen(path) + 1];
 521
 522                 e = p + strcspn(p, "/");
 523                 p = e + strspn(e, "/");
 524
 525                 /* Is this the last component? If so, then we're done */
 526                 if (*p == 0)
 527                         break;
 528
 529                 memcpy(t, path, e - path);
 530                 t[e-path] = 0;
 531
 532                 if (prefix && path_startswith(prefix, t))
 533                         continue;
 534
 535                 r = mkdir_userns(t, mode, uid_shift);
 536                 if (r < 0)
 537                         return r;
 538         }
 539
 540         return mkdir_userns(path, mode, uid_shift);
 541 }
 542
 543 int mount_all(const char *dest,
 544               MountSettingsMask mount_settings,
 545               uid_t uid_shift,
 546               const char *selinux_apifs_context) {
 547
 548 #define PROC_INACCESSIBLE_REG(path)                                     \
 549         { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
 550           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 551         { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 552           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 553
 554 #define PROC_READ_ONLY(path)                                            \
 555         { (path), (path), NULL, NULL, MS_BIND,                          \
 556           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 557         { NULL,   (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 558           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 559
 560         typedef struct MountPoint {
 561                 const char *what;
 562                 const char *where;
 563                 const char *type;
 564                 const char *options;
 565                 unsigned long flags;
 566                 MountSettingsMask mount_settings;
 567         } MountPoint;
 568
 569         static const MountPoint mount_table[] = {
 570                 /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */
 571                 { "proc",            "/proc",           "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,
 572                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_MKDIR },
 573
 574                 { "/proc/sys",       "/proc/sys",       NULL,    NULL,        MS_BIND,
 575                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 576
 577                 { "/proc/sys/net",   "/proc/sys/net",   NULL,    NULL,        MS_BIND,
 578                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 579
 580                 { NULL,              "/proc/sys",       NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 581                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 582
 583                 /* Make these files inaccessible to container payloads: they potentially leak information about kernel
 584                  * internals or the host's execution environment to the container */
 585                 PROC_INACCESSIBLE_REG("/proc/kallsyms"),
 586                 PROC_INACCESSIBLE_REG("/proc/kcore"),
 587                 PROC_INACCESSIBLE_REG("/proc/keys"),
 588                 PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
 589                 PROC_INACCESSIBLE_REG("/proc/timer_list"),
 590
 591                 /* Make these directories read-only to container payloads: they show hardware information, and in some
 592                  * cases contain tunables the container really shouldn't have access to. */
 593                 PROC_READ_ONLY("/proc/acpi"),
 594                 PROC_READ_ONLY("/proc/apm"),
 595                 PROC_READ_ONLY("/proc/asound"),
 596                 PROC_READ_ONLY("/proc/bus"),
 597                 PROC_READ_ONLY("/proc/fs"),
 598                 PROC_READ_ONLY("/proc/irq"),
 599                 PROC_READ_ONLY("/proc/scsi"),
 600
 601                 { "mqueue",          "/dev/mqueue",     "mqueue", NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV,
 602                   MOUNT_IN_USERNS|MOUNT_MKDIR },
 603
 604                 /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
 605                 { "tmpfs",           "/tmp",            "tmpfs", "mode=1777" TMPFS_LIMITS_TMP,     MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 606                   MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR },
 607                 { "tmpfs",           "/sys",            "tmpfs", "mode=555" TMPFS_LIMITS_SYS,      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 608                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR },
 609                 { "sysfs",           "/sys",            "sysfs", NULL,                             MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
 610                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR },    /* skipped if above was mounted */
 611                 { "sysfs",           "/sys",            "sysfs", NULL,                             MS_NOSUID|MS_NOEXEC|MS_NODEV,
 612                   MOUNT_FATAL|MOUNT_MKDIR },                          /* skipped if above was mounted */
 613                 { "tmpfs",           "/dev",            "tmpfs", "mode=755" TMPFS_LIMITS_DEV,      MS_NOSUID|MS_STRICTATIME,
 614                   MOUNT_FATAL|MOUNT_MKDIR },
 615                 { "tmpfs",           "/dev/shm",        "tmpfs", "mode=1777" TMPFS_LIMITS_DEV_SHM, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 616                   MOUNT_FATAL|MOUNT_MKDIR },
 617                 { "tmpfs",           "/run",            "tmpfs", "mode=755" TMPFS_LIMITS_RUN,      MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 618                   MOUNT_FATAL|MOUNT_MKDIR },
 619
 620 #if HAVE_SELINUX
 621                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL,    NULL,                             MS_BIND,
 622                   0 },  /* Bind mount first */
 623                 { NULL,              "/sys/fs/selinux", NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 624                   0 },  /* Then, make it r/o */
 625 #endif
 626         };
 627
 628         bool use_userns = FLAGS_SET(mount_settings, MOUNT_USE_USERNS);
 629         bool netns = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_NETNS);
 630         bool ro = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO);
 631         bool in_userns = FLAGS_SET(mount_settings, MOUNT_IN_USERNS);
 632         bool tmpfs_tmp = FLAGS_SET(mount_settings, MOUNT_APPLY_TMPFS_TMP);
 633         size_t k;
 634         int r;
 635
 636         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 637                 _cleanup_free_ char *where = NULL, *options = NULL;
 638                 const char *o;
 639                 bool fatal = FLAGS_SET(mount_table[k].mount_settings, MOUNT_FATAL);
 640
 641                 if (in_userns != FLAGS_SET(mount_table[k].mount_settings, MOUNT_IN_USERNS))
 642                         continue;
 643
 644                 if (!netns && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_NETNS))
 645                         continue;
 646
 647                 if (!ro && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_RO))
 648                         continue;
 649
 650                 if (!tmpfs_tmp && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_TMPFS_TMP))
 651                         continue;
 652
 653                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL);
 654                 if (r < 0)
 655                         return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where);
 656
 657                 /* Skip this entry if it is not a remount. */
 658                 if (mount_table[k].what) {
 659                         r = path_is_mount_point(where, NULL, 0);
 660                         if (r < 0 && r != -ENOENT)
 661                                 return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 662                         if (r > 0)
 663                                 continue;
 664                 }
 665
 666                 if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_MKDIR)) {
 667                         r = mkdir_userns_p(dest, where, 0755, (use_userns && !in_userns) ? uid_shift : UID_INVALID);
 668                         if (r < 0 && r != -EEXIST) {
 669                                 if (fatal && r != -EROFS)
 670                                         return log_error_errno(r, "Failed to create directory %s: %m", where);
 671
 672                                 log_debug_errno(r, "Failed to create directory %s: %m", where);
 673
 674                                 /* If we failed mkdir() or chown() due to the root directory being read only,
 675                                  * attempt to mount this fs anyway and let mount_verbose log any errors */
 676                                 if (r != -EROFS)
 677                                         continue;
 678                         }
 679                 }
 680
 681                 o = mount_table[k].options;
 682                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 683                         r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
 684                         if (r < 0)
 685                                 return log_oom();
 686                         if (r > 0)
 687                                 o = options;
 688                 }
 689
 690                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 691                                   mount_table[k].what,
 692                                   where,
 693                                   mount_table[k].type,
 694                                   mount_table[k].flags,
 695                                   o);
 696                 if (r < 0 && fatal)
 697                         return r;
 698         }
 699
 700         return 0;
 701 }
 702
 703 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 704         const char *p = options;
 705         unsigned long flags = *mount_flags;
 706         char *opts = NULL;
 707         int r;
 708
 709         assert(options);
 710
 711         for (;;) {
 712                 _cleanup_free_ char *word = NULL;
 713
 714                 r = extract_first_word(&p, &word, ",", 0);
 715                 if (r < 0)
 716                         return log_error_errno(r, "Failed to extract mount option: %m");
 717                 if (r == 0)
 718                         break;
 719
 720                 if (streq(word, "rbind"))
 721                         flags |= MS_REC;
 722                 else if (streq(word, "norbind"))
 723                         flags &= ~MS_REC;
 724                 else {
 725                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 726                                                "Invalid bind mount option: %s",
 727                                                word);
 728                 }
 729         }
 730
 731         *mount_flags = flags;
 732         /* in the future mount_opts will hold string options for mount(2) */
 733         *mount_opts = opts;
 734
 735         return 0;
 736 }
 737
 738 static int mount_bind(const char *dest, CustomMount *m) {
 739         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 740         unsigned long mount_flags = MS_BIND | MS_REC;
 741         struct stat source_st, dest_st;
 742         int r;
 743
 744         assert(dest);
 745         assert(m);
 746
 747         if (m->options) {
 748                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 749                 if (r < 0)
 750                         return r;
 751         }
 752
 753         if (stat(m->source, &source_st) < 0)
 754                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 755
 756         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 757         if (r < 0)
 758                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 759         if (r > 0) { /* Path exists already? */
 760
 761                 if (stat(where, &dest_st) < 0)
 762                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 763
 764                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode))
 765                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 766                                                "Cannot bind mount directory %s on file %s.",
 767                                                m->source, where);
 768
 769                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode))
 770                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 771                                                "Cannot bind mount file %s on directory %s.",
 772                                                m->source, where);
 773
 774         } else { /* Path doesn't exist yet? */
 775                 r = mkdir_parents_label(where, 0755);
 776                 if (r < 0)
 777                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 778
 779                 /* Create the mount point. Any non-directory file can be
 780                 * mounted on any non-directory file (regular, fifo, socket,
 781                 * char, block).
 782                 */
 783                 if (S_ISDIR(source_st.st_mode))
 784                         r = mkdir_label(where, 0755);
 785                 else
 786                         r = touch(where);
 787                 if (r < 0)
 788                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 789         }
 790
 791         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 792         if (r < 0)
 793                 return r;
 794
 795         if (m->read_only) {
 796                 r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL);
 797                 if (r < 0)
 798                         return log_error_errno(r, "Read-only bind mount failed: %m");
 799         }
 800
 801         return 0;
 802 }
 803
 804 static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) {
 805
 806         const char *options;
 807         _cleanup_free_ char *buf = NULL, *where = NULL;
 808         int r;
 809
 810         assert(dest);
 811         assert(m);
 812
 813         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 814         if (r < 0)
 815                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 816         if (r == 0) { /* Doesn't exist yet? */
 817                 r = mkdir_p_label(where, 0755);
 818                 if (r < 0)
 819                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 820         }
 821
 822         r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
 823         if (r < 0)
 824                 return log_oom();
 825         options = r > 0 ? buf : m->options;
 826
 827         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 828 }
 829
 830 static char *joined_and_escaped_lower_dirs(char **lower) {
 831         _cleanup_strv_free_ char **sv = NULL;
 832
 833         sv = strv_copy(lower);
 834         if (!sv)
 835                 return NULL;
 836
 837         strv_reverse(sv);
 838
 839         if (!strv_shell_escape(sv, ",:"))
 840                 return NULL;
 841
 842         return strv_join(sv, ":");
 843 }
 844
 845 static int mount_overlay(const char *dest, CustomMount *m) {
 846         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 847         const char *options;
 848         int r;
 849
 850         assert(dest);
 851         assert(m);
 852
 853         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 854         if (r < 0)
 855                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 856         if (r == 0) { /* Doesn't exist yet? */
 857                 r = mkdir_label(where, 0755);
 858                 if (r < 0)
 859                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 860         }
 861
 862         (void) mkdir_p_label(m->source, 0755);
 863
 864         lower = joined_and_escaped_lower_dirs(m->lower);
 865         if (!lower)
 866                 return log_oom();
 867
 868         escaped_source = shell_escape(m->source, ",:");
 869         if (!escaped_source)
 870                 return log_oom();
 871
 872         if (m->read_only)
 873                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 874         else {
 875                 _cleanup_free_ char *escaped_work_dir = NULL;
 876
 877                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 878                 if (!escaped_work_dir)
 879                         return log_oom();
 880
 881                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 882         }
 883
 884         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 885 }
 886
 887 static int mount_inaccessible(const char *dest, CustomMount *m) {
 888         _cleanup_free_ char *where = NULL, *source = NULL;
 889         struct stat st;
 890         int r;
 891
 892         assert(dest);
 893         assert(m);
 894
 895         r = chase_symlinks_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st, NULL);
 896         if (r < 0) {
 897                 log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination);
 898                 return m->graceful ? 0 : r;
 899         }
 900
 901         r = mode_to_inaccessible_node("/run/systemd", st.st_mode, &source);
 902         if (r < 0)
 903                 return m->graceful ? 0 : r;
 904
 905         r = mount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL);
 906         if (r < 0)
 907                 return m->graceful ? 0 : r;
 908
 909         r = mount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL);
 910         if (r < 0) {
 911                 (void) umount_verbose(where);
 912                 return m->graceful ? 0 : r;
 913         }
 914
 915         return 0;
 916 }
 917
 918 static int mount_arbitrary(const char *dest, CustomMount *m) {
 919         _cleanup_free_ char *where = NULL;
 920         int r;
 921
 922         assert(dest);
 923         assert(m);
 924
 925         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 926         if (r < 0)
 927                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 928         if (r == 0) { /* Doesn't exist yet? */
 929                 r = mkdir_p_label(where, 0755);
 930                 if (r < 0)
 931                         return log_error_errno(r, "Creating mount point for mount %s failed: %m", where);
 932         }
 933
 934         return mount_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options);
 935 }
 936
 937 int mount_custom(
 938                 const char *dest,
 939                 CustomMount *mounts, size_t n,
 940                 uid_t uid_shift,
 941                 const char *selinux_apifs_context,
 942                 MountSettingsMask mount_settings) {
 943
 944         size_t i;
 945         int r;
 946
 947         assert(dest);
 948
 949         for (i = 0; i < n; i++) {
 950                 CustomMount *m = mounts + i;
 951
 952                 if (FLAGS_SET(mount_settings, MOUNT_IN_USERNS) != m->in_userns)
 953                         continue;
 954
 955                 if (FLAGS_SET(mount_settings, MOUNT_ROOT_ONLY) && !path_equal(m->destination, "/"))
 956                         continue;
 957
 958                 if (FLAGS_SET(mount_settings, MOUNT_NON_ROOT_ONLY) && path_equal(m->destination, "/"))
 959                         continue;
 960
 961                 switch (m->type) {
 962
 963                 case CUSTOM_MOUNT_BIND:
 964                         r = mount_bind(dest, m);
 965                         break;
 966
 967                 case CUSTOM_MOUNT_TMPFS:
 968                         r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context);
 969                         break;
 970
 971                 case CUSTOM_MOUNT_OVERLAY:
 972                         r = mount_overlay(dest, m);
 973                         break;
 974
 975                 case CUSTOM_MOUNT_INACCESSIBLE:
 976                         r = mount_inaccessible(dest, m);
 977                         break;
 978
 979                 case CUSTOM_MOUNT_ARBITRARY:
 980                         r = mount_arbitrary(dest, m);
 981                         break;
 982
 983                 default:
 984                         assert_not_reached("Unknown custom mount type");
 985                 }
 986
 987                 if (r < 0)
 988                         return r;
 989         }
 990
 991         return 0;
 992 }
 993
 994 bool has_custom_root_mount(const CustomMount *mounts, size_t n) {
 995         size_t i;
 996
 997         for (i = 0; i < n; i++) {
 998                 const CustomMount *m = mounts + i;
 999
1000                 if (path_equal(m->destination, "/"))
1001                         return true;
1002         }
1003
1004         return false;
1005 }
1006
1007 static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1008
1009         _cleanup_free_ char *buf = NULL;
1010         const char *p, *options;
1011         int r;
1012
1013         assert(directory);
1014
1015         /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
1016
1017         r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
1018         if (r < 0)
1019                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1020
1021         p = prefix_roota(directory, "/var");
1022         r = mkdir(p, 0755);
1023         if (r < 0 && errno != EEXIST)
1024                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1025
1026         options = "mode=755" TMPFS_LIMITS_VOLATILE_STATE;
1027         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1028         if (r < 0)
1029                 return log_oom();
1030         if (r > 0)
1031                 options = buf;
1032
1033         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1034 }
1035
1036 static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1037
1038         bool tmpfs_mounted = false, bind_mounted = false;
1039         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1040         _cleanup_free_ char *buf = NULL, *bindir = NULL;
1041         const char *f, *t, *options;
1042         struct stat st;
1043         int r;
1044
1045         assert(directory);
1046
1047         /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
1048          * that read-only. Before we start setting this up let's validate if the image has the /usr merge
1049          * implemented, and let's output a friendly log message if it hasn't. */
1050
1051         bindir = path_join(directory, "/bin");
1052         if (!bindir)
1053                 return log_oom();
1054         if (lstat(bindir, &st) < 0) {
1055                 if (errno != ENOENT)
1056                         return log_error_errno(errno, "Failed to stat /bin directory below image: %m");
1057
1058                 /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
1059                  * rest. */
1060         } else if (S_ISDIR(st.st_mode))
1061                 return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
1062                                        "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
1063                                        "Please work with your distribution and help them adopt the merged /usr scheme.");
1064         else if (!S_ISLNK(st.st_mode))
1065                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1066                                        "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).");
1067
1068         if (!mkdtemp(template))
1069                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1070
1071         options = "mode=755" TMPFS_LIMITS_ROOTFS;
1072         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1073         if (r < 0)
1074                 goto fail;
1075         if (r > 0)
1076                 options = buf;
1077
1078         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1079         if (r < 0)
1080                 goto fail;
1081
1082         tmpfs_mounted = true;
1083
1084         f = prefix_roota(directory, "/usr");
1085         t = prefix_roota(template, "/usr");
1086
1087         r = mkdir(t, 0755);
1088         if (r < 0 && errno != EEXIST) {
1089                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1090                 goto fail;
1091         }
1092
1093         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1094         if (r < 0)
1095                 goto fail;
1096
1097         bind_mounted = true;
1098
1099         r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL);
1100         if (r < 0) {
1101                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1102                 goto fail;
1103         }
1104
1105         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1106         if (r < 0)
1107                 goto fail;
1108
1109         (void) rmdir(template);
1110
1111         return 0;
1112
1113 fail:
1114         if (bind_mounted)
1115                 (void) umount_verbose(t);
1116
1117         if (tmpfs_mounted)
1118                 (void) umount_verbose(template);
1119         (void) rmdir(template);
1120         return r;
1121 }
1122
1123 static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1124
1125         _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL;
1126         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1127         const char *upper, *work, *options;
1128         bool tmpfs_mounted = false;
1129         int r;
1130
1131         assert(directory);
1132
1133         /* --volatile=overlay means we mount an overlayfs to the root dir. */
1134
1135         if (!mkdtemp(template))
1136                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1137
1138         options = "mode=755" TMPFS_LIMITS_ROOTFS;
1139         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1140         if (r < 0)
1141                 goto finish;
1142         if (r > 0)
1143                 options = buf;
1144
1145         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1146         if (r < 0)
1147                 goto finish;
1148
1149         tmpfs_mounted = true;
1150
1151         upper = strjoina(template, "/upper");
1152         work = strjoina(template, "/work");
1153
1154         if (mkdir(upper, 0755) < 0) {
1155                 r = log_error_errno(errno, "Failed to create %s: %m", upper);
1156                 goto finish;
1157         }
1158         if (mkdir(work, 0755) < 0) {
1159                 r = log_error_errno(errno, "Failed to create %s: %m", work);
1160                 goto finish;
1161         }
1162
1163         /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
1164          * that the kernel allows us to do that without going through some mount point rearrangements. */
1165
1166         escaped_directory = shell_escape(directory, ",:");
1167         escaped_upper = shell_escape(upper, ",:");
1168         escaped_work = shell_escape(work, ",:");
1169         if (!escaped_directory || !escaped_upper || !escaped_work) {
1170                 r = -ENOMEM;
1171                 goto finish;
1172         }
1173
1174         options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work);
1175         r = mount_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options);
1176
1177 finish:
1178         if (tmpfs_mounted)
1179                 (void) umount_verbose(template);
1180
1181         (void) rmdir(template);
1182         return r;
1183 }
1184
1185 int setup_volatile_mode(
1186                 const char *directory,
1187                 VolatileMode mode,
1188                 uid_t uid_shift,
1189                 const char *selinux_apifs_context) {
1190
1191         switch (mode) {
1192
1193         case VOLATILE_YES:
1194                 return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
1195
1196         case VOLATILE_STATE:
1197                 return setup_volatile_state(directory, uid_shift, selinux_apifs_context);
1198
1199         case VOLATILE_OVERLAY:
1200                 return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
1201
1202         default:
1203                 return 0;
1204         }
1205 }
1206
1207 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1208 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1209         _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1210         const char *p = s;
1211         int r;
1212
1213         assert(pivot_root_new);
1214         assert(pivot_root_old);
1215
1216         r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1217         if (r < 0)
1218                 return r;
1219         if (r == 0)
1220                 return -EINVAL;
1221
1222         if (isempty(p))
1223                 root_old = NULL;
1224         else {
1225                 root_old = strdup(p);
1226                 if (!root_old)
1227                         return -ENOMEM;
1228         }
1229
1230         if (!path_is_absolute(root_new))
1231                 return -EINVAL;
1232         if (root_old && !path_is_absolute(root_old))
1233                 return -EINVAL;
1234
1235         free_and_replace(*pivot_root_new, root_new);
1236         free_and_replace(*pivot_root_old, root_old);
1237
1238         return 0;
1239 }
1240
1241 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1242         _cleanup_free_ char *directory_pivot_root_new = NULL;
1243         _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1244         char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
1245         bool remove_pivot_tmp = false;
1246         int r;
1247
1248         assert(directory);
1249
1250         if (!pivot_root_new)
1251                 return 0;
1252
1253         /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1254          * If pivot_root_old is NULL, the existing / disappears.
1255          * This requires a temporary directory, pivot_tmp, which is
1256          * not a child of either.
1257          *
1258          * This is typically used for OSTree-style containers, where
1259          * the root partition contains several sysroots which could be
1260          * run. Normally, one would be chosen by the bootloader and
1261          * pivoted to / by initramfs.
1262          *
1263          * For example, for an OSTree deployment, pivot_root_new
1264          * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1265          * code doesn’t do the /var mount which OSTree expects: use
1266          * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1267          *
1268          * So in the OSTree case, we’ll end up with something like:
1269          *  - directory = /tmp/nspawn-root-123456
1270          *  - pivot_root_new = /ostree/deploy/os/deploy/123abc
1271          *  - pivot_root_old = /sysroot
1272          *  - directory_pivot_root_new =
1273          *       /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1274          *  - pivot_tmp = /tmp/nspawn-pivot-123456
1275          *  - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1276          *
1277          * Requires all file systems at directory and below to be mounted
1278          * MS_PRIVATE or MS_SLAVE so they can be moved.
1279          */
1280         directory_pivot_root_new = path_join(directory, pivot_root_new);
1281         if (!directory_pivot_root_new)
1282                 return log_oom();
1283
1284         /* Remount directory_pivot_root_new to make it movable. */
1285         r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1286         if (r < 0)
1287                 goto done;
1288
1289         if (pivot_root_old) {
1290                 if (!mkdtemp(pivot_tmp)) {
1291                         r = log_error_errno(errno, "Failed to create temporary directory: %m");
1292                         goto done;
1293                 }
1294
1295                 remove_pivot_tmp = true;
1296                 pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old);
1297                 if (!pivot_tmp_pivot_root_old) {
1298                         r = log_oom();
1299                         goto done;
1300                 }
1301
1302                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1303                 if (r < 0)
1304                         goto done;
1305
1306                 r = mount_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1307                 if (r < 0)
1308                         goto done;
1309
1310                 r = mount_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1311                 if (r < 0)
1312                         goto done;
1313         } else {
1314                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1315                 if (r < 0)
1316                         goto done;
1317         }
1318
1319 done:
1320         if (remove_pivot_tmp)
1321                 (void) rmdir(pivot_tmp);
1322
1323         return r;
1324 }