src/nspawn/nspawn-mount.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <sys/mount.h>
   4 #include <linux/magic.h>
   5
   6 #include "alloc-util.h"
   7 #include "chase-symlinks.h"
   8 #include "escape.h"
   9 #include "fd-util.h"
  10 #include "format-util.h"
  11 #include "fs-util.h"
  12 #include "label.h"
  13 #include "mkdir-label.h"
  14 #include "mount-util.h"
  15 #include "mountpoint-util.h"
  16 #include "nspawn-mount.h"
  17 #include "parse-util.h"
  18 #include "path-util.h"
  19 #include "rm-rf.h"
  20 #include "set.h"
  21 #include "sort-util.h"
  22 #include "stat-util.h"
  23 #include "string-util.h"
  24 #include "strv.h"
  25 #include "tmpfile-util.h"
  26 #include "user-util.h"
  27
  28 CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t) {
  29         CustomMount *c, *ret;
  30
  31         assert(l);
  32         assert(n);
  33         assert(t >= 0);
  34         assert(t < _CUSTOM_MOUNT_TYPE_MAX);
  35
  36         c = reallocarray(*l, *n + 1, sizeof(CustomMount));
  37         if (!c)
  38                 return NULL;
  39
  40         *l = c;
  41         ret = *l + *n;
  42         (*n)++;
  43
  44         *ret = (CustomMount) {
  45                 .type = t
  46         };
  47
  48         return ret;
  49 }
  50
  51 void custom_mount_free_all(CustomMount *l, size_t n) {
  52         for (size_t i = 0; i < n; i++) {
  53                 CustomMount *m = l + i;
  54
  55                 free(m->source);
  56                 free(m->destination);
  57                 free(m->options);
  58
  59                 if (m->work_dir) {
  60                         (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
  61                         free(m->work_dir);
  62                 }
  63
  64                 if (m->rm_rf_tmpdir) {
  65                         (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL);
  66                         free(m->rm_rf_tmpdir);
  67                 }
  68
  69                 strv_free(m->lower);
  70                 free(m->type_argument);
  71         }
  72
  73         free(l);
  74 }
  75
  76 static int custom_mount_compare(const CustomMount *a, const CustomMount *b) {
  77         int r;
  78
  79         r = path_compare(a->destination, b->destination);
  80         if (r != 0)
  81                 return r;
  82
  83         return CMP(a->type, b->type);
  84 }
  85
  86 static int source_path_parse(const char *p, char **ret) {
  87         assert(p);
  88         assert(ret);
  89
  90         if (isempty(p))
  91                 return -EINVAL;
  92
  93         if (*p == '+') {
  94                 if (!path_is_absolute(p + 1))
  95                         return -EINVAL;
  96
  97                 char *s = strdup(p);
  98                 if (!s)
  99                         return -ENOMEM;
 100
 101                 *ret = TAKE_PTR(s);
 102                 return 0;
 103         }
 104
 105         return path_make_absolute_cwd(p, ret);
 106 }
 107
 108 static int source_path_parse_nullable(const char *p, char **ret) {
 109         assert(p);
 110         assert(ret);
 111
 112         if (isempty(p)) {
 113                 *ret = NULL;
 114                 return 0;
 115         }
 116
 117         return source_path_parse(p, ret);
 118 }
 119
 120 static char *resolve_source_path(const char *dest, const char *source) {
 121         if (!source)
 122                 return NULL;
 123
 124         if (source[0] == '+')
 125                 return path_join(dest, source + 1);
 126
 127         return strdup(source);
 128 }
 129
 130 static int allocate_temporary_source(CustomMount *m) {
 131         assert(m);
 132         assert(!m->source);
 133         assert(!m->rm_rf_tmpdir);
 134
 135         m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX");
 136         if (!m->rm_rf_tmpdir)
 137                 return log_oom();
 138
 139         if (!mkdtemp(m->rm_rf_tmpdir)) {
 140                 m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir);
 141                 return log_error_errno(errno, "Failed to acquire temporary directory: %m");
 142         }
 143
 144         m->source = path_join(m->rm_rf_tmpdir, "src");
 145         if (!m->source)
 146                 return log_oom();
 147
 148         if (mkdir(m->source, 0755) < 0)
 149                 return log_error_errno(errno, "Failed to create %s: %m", m->source);
 150
 151         return 0;
 152 }
 153
 154 int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) {
 155         int r;
 156
 157         /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
 158          * parent process, so that we know the temporary directories to remove on exit before we fork off the
 159          * children. */
 160
 161         assert(l || n == 0);
 162
 163         /* Order the custom mounts, and make sure we have a working directory */
 164         typesafe_qsort(l, n, custom_mount_compare);
 165
 166         for (size_t i = 0; i < n; i++) {
 167                 CustomMount *m = l + i;
 168
 169                 /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
 170                  * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
 171                  * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
 172                  * the inner child, not the outer one. Determine this here. */
 173                 m->in_userns = path_startswith(m->destination, "/proc");
 174
 175                 if (m->type == CUSTOM_MOUNT_BIND) {
 176                         if (m->source) {
 177                                 char *s;
 178
 179                                 s = resolve_source_path(dest, m->source);
 180                                 if (!s)
 181                                         return log_oom();
 182
 183                                 free_and_replace(m->source, s);
 184                         } else {
 185                                 /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
 186
 187                                 r = allocate_temporary_source(m);
 188                                 if (r < 0)
 189                                         return r;
 190                         }
 191                 }
 192
 193                 if (m->type == CUSTOM_MOUNT_OVERLAY) {
 194                         STRV_FOREACH(j, m->lower) {
 195                                 char *s;
 196
 197                                 s = resolve_source_path(dest, *j);
 198                                 if (!s)
 199                                         return log_oom();
 200
 201                                 free_and_replace(*j, s);
 202                         }
 203
 204                         if (m->source) {
 205                                 char *s;
 206
 207                                 s = resolve_source_path(dest, m->source);
 208                                 if (!s)
 209                                         return log_oom();
 210
 211                                 free_and_replace(m->source, s);
 212                         } else {
 213                                 r = allocate_temporary_source(m);
 214                                 if (r < 0)
 215                                         return r;
 216                         }
 217
 218                         if (m->work_dir) {
 219                                 char *s;
 220
 221                                 s = resolve_source_path(dest, m->work_dir);
 222                                 if (!s)
 223                                         return log_oom();
 224
 225                                 free_and_replace(m->work_dir, s);
 226                         } else {
 227                                 r = tempfn_random(m->source, NULL, &m->work_dir);
 228                                 if (r < 0)
 229                                         return log_error_errno(r, "Failed to acquire working directory: %m");
 230                         }
 231
 232                         (void) mkdir_label(m->work_dir, 0700);
 233                 }
 234         }
 235
 236         return 0;
 237 }
 238
 239 int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 240         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL, *p = NULL;
 241         CustomMount *m;
 242         int r;
 243
 244         assert(l);
 245         assert(n);
 246
 247         r = extract_many_words(&s, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 248         if (r < 0)
 249                 return r;
 250         if (r == 0)
 251                 return -EINVAL;
 252         if (r == 1) {
 253                 destination = strdup(source[0] == '+' ? source+1 : source);
 254                 if (!destination)
 255                         return -ENOMEM;
 256         }
 257         if (r == 2 && !isempty(s)) {
 258                 opts = strdup(s);
 259                 if (!opts)
 260                         return -ENOMEM;
 261         }
 262
 263         r = source_path_parse_nullable(source, &p);
 264         if (r < 0)
 265                 return r;
 266
 267         if (!path_is_absolute(destination))
 268                 return -EINVAL;
 269
 270         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 271         if (!m)
 272                 return -ENOMEM;
 273
 274         m->source = TAKE_PTR(p);
 275         m->destination = TAKE_PTR(destination);
 276         m->read_only = read_only;
 277         m->options = TAKE_PTR(opts);
 278
 279         return 0;
 280 }
 281
 282 int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) {
 283         _cleanup_free_ char *path = NULL, *opts = NULL;
 284         const char *p = ASSERT_PTR(s);
 285         CustomMount *m;
 286         int r;
 287
 288         assert(l);
 289         assert(n);
 290
 291         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 292         if (r < 0)
 293                 return r;
 294         if (r == 0)
 295                 return -EINVAL;
 296
 297         if (isempty(p))
 298                 opts = strdup("mode=0755");
 299         else
 300                 opts = strdup(p);
 301         if (!opts)
 302                 return -ENOMEM;
 303
 304         if (!path_is_absolute(path))
 305                 return -EINVAL;
 306
 307         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 308         if (!m)
 309                 return -ENOMEM;
 310
 311         m->destination = TAKE_PTR(path);
 312         m->options = TAKE_PTR(opts);
 313
 314         return 0;
 315 }
 316
 317 int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 318         _cleanup_free_ char *upper = NULL, *destination = NULL;
 319         _cleanup_strv_free_ char **lower = NULL;
 320         CustomMount *m;
 321         int r, k;
 322
 323         k = strv_split_full(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 324         if (k < 0)
 325                 return k;
 326         if (k < 2)
 327                 return -EADDRNOTAVAIL;
 328         if (k == 2) {
 329                 _cleanup_free_ char *p = NULL;
 330
 331                 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
 332                  * we'll also define the destination mount point the same as the upper. */
 333
 334                 r = source_path_parse(lower[0], &p);
 335                 if (r < 0)
 336                         return r;
 337
 338                 free_and_replace(lower[0], p);
 339
 340                 r = source_path_parse(lower[1], &p);
 341                 if (r < 0)
 342                         return r;
 343
 344                 free_and_replace(lower[1], p);
 345
 346                 upper = TAKE_PTR(lower[1]);
 347
 348                 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
 349                 if (!destination)
 350                         return -ENOMEM;
 351         } else {
 352                 _cleanup_free_ char *p = NULL;
 353
 354                 /* If more than two parameters are specified, the last one is the destination, the second to last one
 355                  * the "upper", and all before that the "lower" directories. */
 356
 357                 destination = lower[k - 1];
 358                 upper = TAKE_PTR(lower[k - 2]);
 359
 360                 STRV_FOREACH(i, lower) {
 361                         r = source_path_parse(*i, &p);
 362                         if (r < 0)
 363                                 return r;
 364
 365                         free_and_replace(*i, p);
 366                 }
 367
 368                 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
 369                  * in /var/tmp */
 370                 r = source_path_parse_nullable(upper, &p);
 371                 if (r < 0)
 372                         return r;
 373
 374                 free_and_replace(upper, p);
 375
 376                 if (!path_is_absolute(destination))
 377                         return -EINVAL;
 378         }
 379
 380         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 381         if (!m)
 382                 return -ENOMEM;
 383
 384         m->destination = TAKE_PTR(destination);
 385         m->source = TAKE_PTR(upper);
 386         m->lower = TAKE_PTR(lower);
 387         m->read_only = read_only;
 388
 389         return 0;
 390 }
 391
 392 int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) {
 393         _cleanup_free_ char *path = NULL;
 394         CustomMount *m;
 395
 396         assert(l);
 397         assert(n);
 398         assert(s);
 399
 400         if (!path_is_absolute(s))
 401                 return -EINVAL;
 402
 403         path = strdup(s);
 404         if (!path)
 405                 return -ENOMEM;
 406
 407         m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE);
 408         if (!m)
 409                 return -ENOMEM;
 410
 411         m->destination = TAKE_PTR(path);
 412         return 0;
 413 }
 414
 415 int tmpfs_patch_options(
 416                 const char *options,
 417                 uid_t uid_shift,
 418                 const char *selinux_apifs_context,
 419                 char **ret) {
 420
 421         _cleanup_free_ char *buf = NULL;
 422
 423         assert(ret);
 424
 425         if (options) {
 426                 buf = strdup(options);
 427                 if (!buf)
 428                         return -ENOMEM;
 429         }
 430
 431         if (uid_shift != UID_INVALID)
 432                 if (strextendf_with_separator(&buf, ",", "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift) < 0)
 433                         return -ENOMEM;
 434
 435 #if HAVE_SELINUX
 436         if (selinux_apifs_context)
 437                 if (strextendf_with_separator(&buf, ",", "context=\"%s\"", selinux_apifs_context) < 0)
 438                         return -ENOMEM;
 439 #endif
 440
 441         *ret = TAKE_PTR(buf);
 442         return !!*ret;
 443 }
 444
 445 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 446         const char *full, *top;
 447         int r;
 448         unsigned long extra_flags = 0;
 449
 450         top = prefix_roota(dest, "/sys");
 451         r = path_is_fs_type(top, SYSFS_MAGIC);
 452         if (r < 0)
 453                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 454         /* /sys might already be mounted as sysfs by the outer child in the
 455          * !netns case. In this case, it's all good. Don't touch it because we
 456          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 457          */
 458         if (r > 0)
 459                 return 0;
 460
 461         full = prefix_roota(top, "/full");
 462
 463         (void) mkdir(full, 0755);
 464
 465         if (FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO))
 466                 extra_flags |= MS_RDONLY;
 467
 468         r = mount_nofollow_verbose(LOG_ERR, "sysfs", full, "sysfs",
 469                                    MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 470         if (r < 0)
 471                 return r;
 472
 473         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 474                 _cleanup_free_ char *from = NULL, *to = NULL;
 475
 476                 from = path_join(full, x);
 477                 if (!from)
 478                         return log_oom();
 479
 480                 to = path_join(top, x);
 481                 if (!to)
 482                         return log_oom();
 483
 484                 (void) mkdir(to, 0755);
 485
 486                 r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 487                 if (r < 0)
 488                         return r;
 489
 490                 r = mount_nofollow_verbose(LOG_ERR, NULL, to, NULL,
 491                                            MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 492                 if (r < 0)
 493                         return r;
 494         }
 495
 496         r = umount_verbose(LOG_ERR, full, UMOUNT_NOFOLLOW);
 497         if (r < 0)
 498                 return r;
 499
 500         if (rmdir(full) < 0)
 501                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 502
 503         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 504          * remount /sys read-only.
 505          */
 506         const char *x = prefix_roota(top, "/fs/cgroup");
 507         (void) mkdir_p(x, 0755);
 508
 509         return mount_nofollow_verbose(LOG_ERR, NULL, top, NULL,
 510                                       MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 511 }
 512
 513 int mount_all(const char *dest,
 514               MountSettingsMask mount_settings,
 515               uid_t uid_shift,
 516               const char *selinux_apifs_context) {
 517
 518 #define PROC_INACCESSIBLE_REG(path)                                     \
 519         { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
 520           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 521         { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 522           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 523
 524 #define PROC_READ_ONLY(path)                                            \
 525         { (path), (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         typedef struct MountPoint {
 531                 const char *what;
 532                 const char *where;
 533                 const char *type;
 534                 const char *options;
 535                 unsigned long flags;
 536                 MountSettingsMask mount_settings;
 537         } MountPoint;
 538
 539         static const MountPoint mount_table[] = {
 540                 /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */
 541                 { "proc",            "/proc",           "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,
 542                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_MKDIR|MOUNT_FOLLOW_SYMLINKS }, /* we follow symlinks here since not following them requires /proc/ already being mounted, which we don't have here. */
 543
 544                 { "/proc/sys",       "/proc/sys",       NULL,    NULL,        MS_BIND,
 545                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 546
 547                 { "/proc/sys/net",   "/proc/sys/net",   NULL,    NULL,        MS_BIND,
 548                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 549
 550                 { NULL,              "/proc/sys",       NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 551                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 552
 553                 /* Make these files inaccessible to container payloads: they potentially leak information about kernel
 554                  * internals or the host's execution environment to the container */
 555                 PROC_INACCESSIBLE_REG("/proc/kallsyms"),
 556                 PROC_INACCESSIBLE_REG("/proc/kcore"),
 557                 PROC_INACCESSIBLE_REG("/proc/keys"),
 558                 PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
 559                 PROC_INACCESSIBLE_REG("/proc/timer_list"),
 560
 561                 /* Make these directories read-only to container payloads: they show hardware information, and in some
 562                  * cases contain tunables the container really shouldn't have access to. */
 563                 PROC_READ_ONLY("/proc/acpi"),
 564                 PROC_READ_ONLY("/proc/apm"),
 565                 PROC_READ_ONLY("/proc/asound"),
 566                 PROC_READ_ONLY("/proc/bus"),
 567                 PROC_READ_ONLY("/proc/fs"),
 568                 PROC_READ_ONLY("/proc/irq"),
 569                 PROC_READ_ONLY("/proc/scsi"),
 570
 571                 { "mqueue",                 "/dev/mqueue",                  "mqueue", NULL,                            MS_NOSUID|MS_NOEXEC|MS_NODEV,
 572                   MOUNT_IN_USERNS|MOUNT_MKDIR },
 573
 574                 /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
 575                 { "tmpfs",                  "/tmp",                         "tmpfs", "mode=1777" NESTED_TMPFS_LIMITS,  MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 576                   MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR },
 577                 { "tmpfs",                  "/sys",                         "tmpfs", "mode=555" TMPFS_LIMITS_SYS,      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 578                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR },
 579                 { "sysfs",                  "/sys",                         "sysfs", NULL,                             MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
 580                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR },    /* skipped if above was mounted */
 581                 { "sysfs",                  "/sys",                         "sysfs", NULL,                             MS_NOSUID|MS_NOEXEC|MS_NODEV,
 582                   MOUNT_FATAL|MOUNT_MKDIR },                          /* skipped if above was mounted */
 583                 { "tmpfs",                  "/dev",                         "tmpfs", "mode=755" TMPFS_LIMITS_PRIVATE_DEV, MS_NOSUID|MS_STRICTATIME,
 584                   MOUNT_FATAL|MOUNT_MKDIR },
 585                 { "tmpfs",                  "/dev/shm",                     "tmpfs", "mode=1777" NESTED_TMPFS_LIMITS,  MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 586                   MOUNT_FATAL|MOUNT_MKDIR },
 587                 { "tmpfs",                  "/run",                         "tmpfs", "mode=755" TMPFS_LIMITS_RUN,      MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 588                   MOUNT_FATAL|MOUNT_MKDIR },
 589                 { "/run/host",              "/run/host",                    NULL,    NULL,                             MS_BIND,
 590                   MOUNT_FATAL|MOUNT_MKDIR|MOUNT_PREFIX_ROOT }, /* Prepare this so that we can make it read-only when we are done */
 591                 { "/etc/os-release",        "/run/host/os-release",         NULL,    NULL,                             MS_BIND,
 592                   MOUNT_TOUCH }, /* As per kernel interface requirements, bind mount first (creating mount points) and make read-only later */
 593                 { "/usr/lib/os-release",    "/run/host/os-release",         NULL,    NULL,                             MS_BIND,
 594                   MOUNT_FATAL }, /* If /etc/os-release doesn't exist use the version in /usr/lib as fallback */
 595                 { NULL,                     "/run/host/os-release",         NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 596                   MOUNT_FATAL },
 597                 { NULL,                     "/run/host",                    NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 598                   MOUNT_FATAL|MOUNT_IN_USERNS },
 599 #if HAVE_SELINUX
 600                 { "/sys/fs/selinux",        "/sys/fs/selinux",              NULL,    NULL,                             MS_BIND,
 601                   MOUNT_MKDIR },  /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */
 602                 { NULL,                     "/sys/fs/selinux",              NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 603                   0 },            /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */
 604 #endif
 605         };
 606
 607         bool use_userns = FLAGS_SET(mount_settings, MOUNT_USE_USERNS);
 608         bool netns = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_NETNS);
 609         bool ro = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO);
 610         bool in_userns = FLAGS_SET(mount_settings, MOUNT_IN_USERNS);
 611         bool tmpfs_tmp = FLAGS_SET(mount_settings, MOUNT_APPLY_TMPFS_TMP);
 612         int r;
 613
 614         for (size_t k = 0; k < ELEMENTSOF(mount_table); k++) {
 615                 _cleanup_free_ char *where = NULL, *options = NULL, *prefixed = NULL;
 616                 bool fatal = FLAGS_SET(mount_table[k].mount_settings, MOUNT_FATAL);
 617                 const char *o;
 618
 619                 if (in_userns != FLAGS_SET(mount_table[k].mount_settings, MOUNT_IN_USERNS))
 620                         continue;
 621
 622                 if (!netns && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_NETNS))
 623                         continue;
 624
 625                 if (!ro && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_RO))
 626                         continue;
 627
 628                 if (!tmpfs_tmp && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_TMPFS_TMP))
 629                         continue;
 630
 631                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL);
 632                 if (r < 0)
 633                         return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where);
 634
 635                 /* Skip this entry if it is not a remount. */
 636                 if (mount_table[k].what) {
 637                         r = path_is_mount_point(where, NULL, 0);
 638                         if (r < 0 && r != -ENOENT)
 639                                 return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 640                         if (r > 0)
 641                                 continue;
 642                 }
 643
 644                 if ((mount_table[k].mount_settings & (MOUNT_MKDIR|MOUNT_TOUCH)) != 0) {
 645                         uid_t u = (use_userns && !in_userns) ? uid_shift : UID_INVALID;
 646
 647                         if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH))
 648                                 r = mkdir_parents_safe(dest, where, 0755, u, u, 0);
 649                         else
 650                                 r = mkdir_p_safe(dest, where, 0755, u, u, 0);
 651                         if (r < 0 && r != -EEXIST) {
 652                                 if (fatal && r != -EROFS)
 653                                         return log_error_errno(r, "Failed to create directory %s: %m", where);
 654
 655                                 log_debug_errno(r, "Failed to create directory %s: %m", where);
 656
 657                                 /* If we failed mkdir() or chown() due to the root directory being read only,
 658                                  * attempt to mount this fs anyway and let mount_verbose log any errors */
 659                                 if (r != -EROFS)
 660                                         continue;
 661                         }
 662                 }
 663
 664                 if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH)) {
 665                         r = touch(where);
 666                         if (r < 0 && r != -EEXIST) {
 667                                 if (fatal && r != -EROFS)
 668                                         return log_error_errno(r, "Failed to create file %s: %m", where);
 669
 670                                 log_debug_errno(r, "Failed to create file %s: %m", where);
 671                                 if (r != -EROFS)
 672                                         continue;
 673                         }
 674                 }
 675
 676                 o = mount_table[k].options;
 677                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 678                         r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
 679                         if (r < 0)
 680                                 return log_oom();
 681                         if (r > 0)
 682                                 o = options;
 683                 }
 684
 685                 if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_PREFIX_ROOT)) {
 686                         /* Optionally prefix the mount source with the root dir. This is useful in bind
 687                          * mounts to be created within the container image before we transition into it. Note
 688                          * that MOUNT_IN_USERNS is run after we transitioned hence prefixing is not ncessary
 689                          * for those. */
 690                         r = chase_symlinks(mount_table[k].what, dest, CHASE_PREFIX_ROOT, &prefixed, NULL);
 691                         if (r < 0)
 692                                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].what);
 693                 }
 694
 695                 r = mount_verbose_full(
 696                                 fatal ? LOG_ERR : LOG_DEBUG,
 697                                 prefixed ?: mount_table[k].what,
 698                                 where,
 699                                 mount_table[k].type,
 700                                 mount_table[k].flags,
 701                                 o,
 702                                 FLAGS_SET(mount_table[k].mount_settings, MOUNT_FOLLOW_SYMLINKS));
 703                 if (r < 0 && fatal)
 704                         return r;
 705         }
 706
 707         return 0;
 708 }
 709
 710 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts, RemountIdmapping *idmapping) {
 711         unsigned long flags = *mount_flags;
 712         char *opts = NULL;
 713         RemountIdmapping new_idmapping = *idmapping;
 714         int r;
 715
 716         assert(options);
 717
 718         for (;;) {
 719                 _cleanup_free_ char *word = NULL;
 720
 721                 r = extract_first_word(&options, &word, ",", 0);
 722                 if (r < 0)
 723                         return log_error_errno(r, "Failed to extract mount option: %m");
 724                 if (r == 0)
 725                         break;
 726
 727                 if (streq(word, "rbind"))
 728                         flags |= MS_REC;
 729                 else if (streq(word, "norbind"))
 730                         flags &= ~MS_REC;
 731                 else if (streq(word, "idmap"))
 732                         new_idmapping = REMOUNT_IDMAPPING_HOST_ROOT;
 733                 else if (streq(word, "noidmap"))
 734                         new_idmapping = REMOUNT_IDMAPPING_NONE;
 735                 else if (streq(word, "rootidmap"))
 736                         new_idmapping = REMOUNT_IDMAPPING_HOST_OWNER;
 737                 else
 738                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 739                                                "Invalid bind mount option: %s", word);
 740         }
 741
 742         *mount_flags = flags;
 743         *idmapping = new_idmapping;
 744         /* in the future mount_opts will hold string options for mount(2) */
 745         *mount_opts = opts;
 746
 747         return 0;
 748 }
 749
 750 static int mount_bind(const char *dest, CustomMount *m, uid_t uid_shift, uid_t uid_range) {
 751         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 752         unsigned long mount_flags = MS_BIND | MS_REC;
 753         struct stat source_st, dest_st;
 754         int r;
 755         RemountIdmapping idmapping = REMOUNT_IDMAPPING_NONE;
 756
 757         assert(dest);
 758         assert(m);
 759
 760         if (m->options) {
 761                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts, &idmapping);
 762                 if (r < 0)
 763                         return r;
 764         }
 765
 766         /* If this is a bind mount from a temporary sources change ownership of the source to the container's
 767          * root UID. Otherwise it would always show up as "nobody" if user namespacing is used. */
 768         if (m->rm_rf_tmpdir && chown(m->source, uid_shift, uid_shift) < 0)
 769                 return log_error_errno(errno, "Failed to chown %s: %m", m->source);
 770
 771         if (stat(m->source, &source_st) < 0)
 772                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 773
 774         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 775         if (r < 0)
 776                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 777         if (r > 0) { /* Path exists already? */
 778
 779                 if (stat(where, &dest_st) < 0)
 780                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 781
 782                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode))
 783                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 784                                                "Cannot bind mount directory %s on file %s.",
 785                                                m->source, where);
 786
 787                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode))
 788                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 789                                                "Cannot bind mount file %s on directory %s.",
 790                                                m->source, where);
 791
 792         } else { /* Path doesn't exist yet? */
 793                 r = mkdir_parents_label(where, 0755);
 794                 if (r < 0)
 795                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 796
 797                 /* Create the mount point. Any non-directory file can be
 798                 * mounted on any non-directory file (regular, fifo, socket,
 799                 * char, block).
 800                 */
 801                 if (S_ISDIR(source_st.st_mode))
 802                         r = mkdir_label(where, 0755);
 803                 else
 804                         r = touch(where);
 805                 if (r < 0)
 806                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 807         }
 808
 809         r = mount_nofollow_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 810         if (r < 0)
 811                 return r;
 812
 813         if (m->read_only) {
 814                 r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL);
 815                 if (r < 0)
 816                         return log_error_errno(r, "Read-only bind mount failed: %m");
 817         }
 818
 819         if (idmapping != REMOUNT_IDMAPPING_NONE) {
 820                 r = remount_idmap(where, uid_shift, uid_range, source_st.st_uid, idmapping);
 821                 if (r < 0)
 822                         return log_error_errno(r, "Failed to map ids for bind mount %s: %m", where);
 823         }
 824
 825         return 0;
 826 }
 827
 828 static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) {
 829         const char *options;
 830         _cleanup_free_ char *buf = NULL, *where = NULL;
 831         int r;
 832
 833         assert(dest);
 834         assert(m);
 835
 836         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 837         if (r < 0)
 838                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 839         if (r == 0) { /* Doesn't exist yet? */
 840                 r = mkdir_p_label(where, 0755);
 841                 if (r < 0)
 842                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 843         }
 844
 845         r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
 846         if (r < 0)
 847                 return log_oom();
 848         options = r > 0 ? buf : m->options;
 849
 850         return mount_nofollow_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 851 }
 852
 853 static char *joined_and_escaped_lower_dirs(char **lower) {
 854         _cleanup_strv_free_ char **sv = NULL;
 855
 856         sv = strv_copy(lower);
 857         if (!sv)
 858                 return NULL;
 859
 860         strv_reverse(sv);
 861
 862         if (!strv_shell_escape(sv, ",:"))
 863                 return NULL;
 864
 865         return strv_join(sv, ":");
 866 }
 867
 868 static int mount_overlay(const char *dest, CustomMount *m) {
 869         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 870         const char *options;
 871         int r;
 872
 873         assert(dest);
 874         assert(m);
 875
 876         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 877         if (r < 0)
 878                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 879         if (r == 0) { /* Doesn't exist yet? */
 880                 r = mkdir_label(where, 0755);
 881                 if (r < 0)
 882                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 883         }
 884
 885         (void) mkdir_p_label(m->source, 0755);
 886
 887         lower = joined_and_escaped_lower_dirs(m->lower);
 888         if (!lower)
 889                 return log_oom();
 890
 891         escaped_source = shell_escape(m->source, ",:");
 892         if (!escaped_source)
 893                 return log_oom();
 894
 895         if (m->read_only)
 896                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 897         else {
 898                 _cleanup_free_ char *escaped_work_dir = NULL;
 899
 900                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 901                 if (!escaped_work_dir)
 902                         return log_oom();
 903
 904                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 905         }
 906
 907         return mount_nofollow_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 908 }
 909
 910 static int mount_inaccessible(const char *dest, CustomMount *m) {
 911         _cleanup_free_ char *where = NULL, *source = NULL;
 912         struct stat st;
 913         int r;
 914
 915         assert(dest);
 916         assert(m);
 917
 918         r = chase_symlinks_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st, NULL);
 919         if (r < 0) {
 920                 log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination);
 921                 return m->graceful ? 0 : r;
 922         }
 923
 924         r = mode_to_inaccessible_node(NULL, st.st_mode, &source);
 925         if (r < 0)
 926                 return m->graceful ? 0 : r;
 927
 928         r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL);
 929         if (r < 0)
 930                 return m->graceful ? 0 : r;
 931
 932         r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL);
 933         if (r < 0) {
 934                 (void) umount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, where, UMOUNT_NOFOLLOW);
 935                 return m->graceful ? 0 : r;
 936         }
 937
 938         return 0;
 939 }
 940
 941 static int mount_arbitrary(const char *dest, CustomMount *m) {
 942         _cleanup_free_ char *where = NULL;
 943         int r;
 944
 945         assert(dest);
 946         assert(m);
 947
 948         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 949         if (r < 0)
 950                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 951         if (r == 0) { /* Doesn't exist yet? */
 952                 r = mkdir_p_label(where, 0755);
 953                 if (r < 0)
 954                         return log_error_errno(r, "Creating mount point for mount %s failed: %m", where);
 955         }
 956
 957         return mount_nofollow_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options);
 958 }
 959
 960 int mount_custom(
 961                 const char *dest,
 962                 CustomMount *mounts, size_t n,
 963                 uid_t uid_shift,
 964                 uid_t uid_range,
 965                 const char *selinux_apifs_context,
 966                 MountSettingsMask mount_settings) {
 967         int r;
 968
 969         assert(dest);
 970
 971         for (size_t i = 0; i < n; i++) {
 972                 CustomMount *m = mounts + i;
 973
 974                 if (FLAGS_SET(mount_settings, MOUNT_IN_USERNS) != m->in_userns)
 975                         continue;
 976
 977                 if (FLAGS_SET(mount_settings, MOUNT_ROOT_ONLY) && !path_equal(m->destination, "/"))
 978                         continue;
 979
 980                 if (FLAGS_SET(mount_settings, MOUNT_NON_ROOT_ONLY) && path_equal(m->destination, "/"))
 981                         continue;
 982
 983                 switch (m->type) {
 984
 985                 case CUSTOM_MOUNT_BIND:
 986                         r = mount_bind(dest, m, uid_shift, uid_range);
 987                         break;
 988
 989                 case CUSTOM_MOUNT_TMPFS:
 990                         r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context);
 991                         break;
 992
 993                 case CUSTOM_MOUNT_OVERLAY:
 994                         r = mount_overlay(dest, m);
 995                         break;
 996
 997                 case CUSTOM_MOUNT_INACCESSIBLE:
 998                         r = mount_inaccessible(dest, m);
 999                         break;
1000
1001                 case CUSTOM_MOUNT_ARBITRARY:
1002                         r = mount_arbitrary(dest, m);
1003                         break;
1004
1005                 default:
1006                         assert_not_reached();
1007                 }
1008
1009                 if (r < 0)
1010                         return r;
1011         }
1012
1013         return 0;
1014 }
1015
1016 bool has_custom_root_mount(const CustomMount *mounts, size_t n) {
1017         for (size_t i = 0; i < n; i++)
1018                 if (path_equal(mounts[i].destination, "/"))
1019                         return true;
1020
1021         return false;
1022 }
1023
1024 static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1025         _cleanup_free_ char *buf = NULL;
1026         const char *p, *options;
1027         int r;
1028
1029         assert(directory);
1030
1031         /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
1032
1033         r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
1034         if (r < 0)
1035                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1036
1037         p = prefix_roota(directory, "/var");
1038         r = mkdir(p, 0755);
1039         if (r < 0 && errno != EEXIST)
1040                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1041
1042         options = "mode=755" TMPFS_LIMITS_VOLATILE_STATE;
1043         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1044         if (r < 0)
1045                 return log_oom();
1046         if (r > 0)
1047                 options = buf;
1048
1049         return mount_nofollow_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1050 }
1051
1052 static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1053         bool tmpfs_mounted = false, bind_mounted = false;
1054         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1055         _cleanup_free_ char *buf = NULL, *bindir = NULL;
1056         const char *f, *t, *options;
1057         struct stat st;
1058         int r;
1059
1060         assert(directory);
1061
1062         /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
1063          * that read-only. Before we start setting this up let's validate if the image has the /usr merge
1064          * implemented, and let's output a friendly log message if it hasn't. */
1065
1066         bindir = path_join(directory, "/bin");
1067         if (!bindir)
1068                 return log_oom();
1069         if (lstat(bindir, &st) < 0) {
1070                 if (errno != ENOENT)
1071                         return log_error_errno(errno, "Failed to stat /bin directory below image: %m");
1072
1073                 /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
1074                  * rest. */
1075         } else if (S_ISDIR(st.st_mode))
1076                 return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
1077                                        "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
1078                                        "Please work with your distribution and help them adopt the merged /usr scheme.");
1079         else if (!S_ISLNK(st.st_mode))
1080                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1081                                        "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).");
1082
1083         if (!mkdtemp(template))
1084                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1085
1086         options = "mode=755" TMPFS_LIMITS_ROOTFS;
1087         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1088         if (r < 0)
1089                 goto fail;
1090         if (r > 0)
1091                 options = buf;
1092
1093         r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1094         if (r < 0)
1095                 goto fail;
1096
1097         tmpfs_mounted = true;
1098
1099         f = prefix_roota(directory, "/usr");
1100         t = prefix_roota(template, "/usr");
1101
1102         r = mkdir(t, 0755);
1103         if (r < 0 && errno != EEXIST) {
1104                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1105                 goto fail;
1106         }
1107
1108         r = mount_nofollow_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1109         if (r < 0)
1110                 goto fail;
1111
1112         bind_mounted = true;
1113
1114         r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL);
1115         if (r < 0) {
1116                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1117                 goto fail;
1118         }
1119
1120         r = mount_nofollow_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1121         if (r < 0)
1122                 goto fail;
1123
1124         (void) rmdir(template);
1125
1126         return 0;
1127
1128 fail:
1129         if (bind_mounted)
1130                 (void) umount_verbose(LOG_ERR, t, UMOUNT_NOFOLLOW);
1131
1132         if (tmpfs_mounted)
1133                 (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW);
1134
1135         (void) rmdir(template);
1136         return r;
1137 }
1138
1139 static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1140         _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL;
1141         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1142         const char *upper, *work, *options;
1143         bool tmpfs_mounted = false;
1144         int r;
1145
1146         assert(directory);
1147
1148         /* --volatile=overlay means we mount an overlayfs to the root dir. */
1149
1150         if (!mkdtemp(template))
1151                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1152
1153         options = "mode=755" TMPFS_LIMITS_ROOTFS;
1154         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1155         if (r < 0)
1156                 goto finish;
1157         if (r > 0)
1158                 options = buf;
1159
1160         r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1161         if (r < 0)
1162                 goto finish;
1163
1164         tmpfs_mounted = true;
1165
1166         upper = strjoina(template, "/upper");
1167         work = strjoina(template, "/work");
1168
1169         if (mkdir(upper, 0755) < 0) {
1170                 r = log_error_errno(errno, "Failed to create %s: %m", upper);
1171                 goto finish;
1172         }
1173         if (mkdir(work, 0755) < 0) {
1174                 r = log_error_errno(errno, "Failed to create %s: %m", work);
1175                 goto finish;
1176         }
1177
1178         /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
1179          * that the kernel allows us to do that without going through some mount point rearrangements. */
1180
1181         escaped_directory = shell_escape(directory, ",:");
1182         escaped_upper = shell_escape(upper, ",:");
1183         escaped_work = shell_escape(work, ",:");
1184         if (!escaped_directory || !escaped_upper || !escaped_work) {
1185                 r = -ENOMEM;
1186                 goto finish;
1187         }
1188
1189         options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work);
1190         r = mount_nofollow_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options);
1191
1192 finish:
1193         if (tmpfs_mounted)
1194                 (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW);
1195
1196         (void) rmdir(template);
1197         return r;
1198 }
1199
1200 int setup_volatile_mode(
1201                 const char *directory,
1202                 VolatileMode mode,
1203                 uid_t uid_shift,
1204                 const char *selinux_apifs_context) {
1205
1206         switch (mode) {
1207
1208         case VOLATILE_YES:
1209                 return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
1210
1211         case VOLATILE_STATE:
1212                 return setup_volatile_state(directory, uid_shift, selinux_apifs_context);
1213
1214         case VOLATILE_OVERLAY:
1215                 return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
1216
1217         default:
1218                 return 0;
1219         }
1220 }
1221
1222 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1223 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1224         _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1225         const char *p = s;
1226         int r;
1227
1228         assert(pivot_root_new);
1229         assert(pivot_root_old);
1230
1231         r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1232         if (r < 0)
1233                 return r;
1234         if (r == 0)
1235                 return -EINVAL;
1236
1237         if (isempty(p))
1238                 root_old = NULL;
1239         else {
1240                 root_old = strdup(p);
1241                 if (!root_old)
1242                         return -ENOMEM;
1243         }
1244
1245         if (!path_is_absolute(root_new))
1246                 return -EINVAL;
1247         if (root_old && !path_is_absolute(root_old))
1248                 return -EINVAL;
1249
1250         free_and_replace(*pivot_root_new, root_new);
1251         free_and_replace(*pivot_root_old, root_old);
1252
1253         return 0;
1254 }
1255
1256 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1257         _cleanup_free_ char *directory_pivot_root_new = NULL;
1258         _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1259         char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
1260         bool remove_pivot_tmp = false;
1261         int r;
1262
1263         assert(directory);
1264
1265         if (!pivot_root_new)
1266                 return 0;
1267
1268         /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1269          * If pivot_root_old is NULL, the existing / disappears.
1270          * This requires a temporary directory, pivot_tmp, which is
1271          * not a child of either.
1272          *
1273          * This is typically used for OSTree-style containers, where
1274          * the root partition contains several sysroots which could be
1275          * run. Normally, one would be chosen by the bootloader and
1276          * pivoted to / by initramfs.
1277          *
1278          * For example, for an OSTree deployment, pivot_root_new
1279          * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1280          * code doesn’t do the /var mount which OSTree expects: use
1281          * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1282          *
1283          * So in the OSTree case, we’ll end up with something like:
1284          *  - directory = /tmp/nspawn-root-123456
1285          *  - pivot_root_new = /ostree/deploy/os/deploy/123abc
1286          *  - pivot_root_old = /sysroot
1287          *  - directory_pivot_root_new =
1288          *       /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1289          *  - pivot_tmp = /tmp/nspawn-pivot-123456
1290          *  - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1291          *
1292          * Requires all file systems at directory and below to be mounted
1293          * MS_PRIVATE or MS_SLAVE so they can be moved.
1294          */
1295         directory_pivot_root_new = path_join(directory, pivot_root_new);
1296         if (!directory_pivot_root_new)
1297                 return log_oom();
1298
1299         /* Remount directory_pivot_root_new to make it movable. */
1300         r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1301         if (r < 0)
1302                 goto done;
1303
1304         if (pivot_root_old) {
1305                 if (!mkdtemp(pivot_tmp)) {
1306                         r = log_error_errno(errno, "Failed to create temporary directory: %m");
1307                         goto done;
1308                 }
1309
1310                 remove_pivot_tmp = true;
1311                 pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old);
1312                 if (!pivot_tmp_pivot_root_old) {
1313                         r = log_oom();
1314                         goto done;
1315                 }
1316
1317                 r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1318                 if (r < 0)
1319                         goto done;
1320
1321                 r = mount_nofollow_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1322                 if (r < 0)
1323                         goto done;
1324
1325                 r = mount_nofollow_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1326                 if (r < 0)
1327                         goto done;
1328         } else {
1329                 r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1330                 if (r < 0)
1331                         goto done;
1332         }
1333
1334 done:
1335         if (remove_pivot_tmp)
1336                 (void) rmdir(pivot_tmp);
1337
1338         return r;
1339 }