src/nspawn/nspawn-mount.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2015 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <sys/mount.h>
  21 #include <linux/magic.h>
  22
  23 #include "alloc-util.h"
  24 #include "escape.h"
  25 #include "fd-util.h"
  26 #include "fileio.h"
  27 #include "fs-util.h"
  28 #include "label.h"
  29 #include "mkdir.h"
  30 #include "mount-util.h"
  31 #include "nspawn-mount.h"
  32 #include "parse-util.h"
  33 #include "path-util.h"
  34 #include "rm-rf.h"
  35 #include "set.h"
  36 #include "stat-util.h"
  37 #include "string-util.h"
  38 #include "strv.h"
  39 #include "user-util.h"
  40 #include "util.h"
  41
  42 CustomMount* custom_mount_add(CustomMount **l, unsigned *n, CustomMountType t) {
  43         CustomMount *c, *ret;
  44
  45         assert(l);
  46         assert(n);
  47         assert(t >= 0);
  48         assert(t < _CUSTOM_MOUNT_TYPE_MAX);
  49
  50         c = realloc_multiply(*l, (*n + 1), sizeof(CustomMount));
  51         if (!c)
  52                 return NULL;
  53
  54         *l = c;
  55         ret = *l + *n;
  56         (*n)++;
  57
  58         *ret = (CustomMount) { .type = t };
  59
  60         return ret;
  61 }
  62
  63 void custom_mount_free_all(CustomMount *l, unsigned n) {
  64         unsigned i;
  65
  66         for (i = 0; i < n; i++) {
  67                 CustomMount *m = l + i;
  68
  69                 free(m->source);
  70                 free(m->destination);
  71                 free(m->options);
  72
  73                 if (m->work_dir) {
  74                         (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
  75                         free(m->work_dir);
  76                 }
  77
  78                 strv_free(m->lower);
  79         }
  80
  81         free(l);
  82 }
  83
  84 static int custom_mount_compare(const void *a, const void *b) {
  85         const CustomMount *x = a, *y = b;
  86         int r;
  87
  88         r = path_compare(x->destination, y->destination);
  89         if (r != 0)
  90                 return r;
  91
  92         if (x->type < y->type)
  93                 return -1;
  94         if (x->type > y->type)
  95                 return 1;
  96
  97         return 0;
  98 }
  99
 100 static bool source_path_is_valid(const char *p) {
 101         assert(p);
 102
 103         if (*p == '+')
 104                 p++;
 105
 106         return path_is_absolute(p);
 107 }
 108
 109 static char *resolve_source_path(const char *dest, const char *source) {
 110
 111         if (!source)
 112                 return NULL;
 113
 114         if (source[0] == '+')
 115                 return prefix_root(dest, source + 1);
 116
 117         return strdup(source);
 118 }
 119
 120 int custom_mount_prepare_all(const char *dest, CustomMount *l, unsigned n) {
 121         unsigned i;
 122         int r;
 123
 124         /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
 125          * parent process, so that we know the temporary directories to remove on exit before we fork off the
 126          * children. */
 127
 128         assert(l || n == 0);
 129
 130         /* Order the custom mounts, and make sure we have a working directory */
 131         qsort_safe(l, n, sizeof(CustomMount), custom_mount_compare);
 132
 133         for (i = 0; i < n; i++) {
 134                 CustomMount *m = l + i;
 135
 136                 if (m->source) {
 137                         char *s;
 138
 139                         s = resolve_source_path(dest, m->source);
 140                         if (!s)
 141                                 return log_oom();
 142
 143                         free(m->source);
 144                         m->source = s;
 145                 }
 146
 147                 if (m->type == CUSTOM_MOUNT_OVERLAY) {
 148                         char **j;
 149
 150                         STRV_FOREACH(j, m->lower) {
 151                                 char *s;
 152
 153                                 s = resolve_source_path(dest, *j);
 154                                 if (!s)
 155                                         return log_oom();
 156
 157                                 free(*j);
 158                                 *j = s;
 159                         }
 160
 161                         if (m->work_dir) {
 162                                 char *s;
 163
 164                                 s = resolve_source_path(dest, m->work_dir);
 165                                 if (!s)
 166                                         return log_oom();
 167
 168                                 free(m->work_dir);
 169                                 m->work_dir = s;
 170                         } else {
 171                                 assert(m->source);
 172
 173                                 r = tempfn_random(m->source, NULL, &m->work_dir);
 174                                 if (r < 0)
 175                                         return log_error_errno(r, "Failed to acquire working directory: %m");
 176                         }
 177
 178                         (void) mkdir_label(m->work_dir, 0700);
 179                 }
 180         }
 181
 182         return 0;
 183 }
 184
 185 int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 186         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 187         const char *p = s;
 188         CustomMount *m;
 189         int r;
 190
 191         assert(l);
 192         assert(n);
 193
 194         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 195         if (r < 0)
 196                 return r;
 197         if (r == 0)
 198                 return -EINVAL;
 199         if (r == 1) {
 200                 destination = strdup(source[0] == '+' ? source+1 : source);
 201                 if (!destination)
 202                         return -ENOMEM;
 203         }
 204         if (r == 2 && !isempty(p)) {
 205                 opts = strdup(p);
 206                 if (!opts)
 207                         return -ENOMEM;
 208         }
 209
 210         if (!source_path_is_valid(source))
 211                 return -EINVAL;
 212         if (!path_is_absolute(destination))
 213                 return -EINVAL;
 214
 215         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 216         if (!m)
 217                 return -ENOMEM;
 218
 219         m->source = source;
 220         m->destination = destination;
 221         m->read_only = read_only;
 222         m->options = opts;
 223
 224         source = destination = opts = NULL;
 225         return 0;
 226 }
 227
 228 int tmpfs_mount_parse(CustomMount **l, unsigned *n, const char *s) {
 229         _cleanup_free_ char *path = NULL, *opts = NULL;
 230         const char *p = s;
 231         CustomMount *m;
 232         int r;
 233
 234         assert(l);
 235         assert(n);
 236         assert(s);
 237
 238         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 239         if (r < 0)
 240                 return r;
 241         if (r == 0)
 242                 return -EINVAL;
 243
 244         if (isempty(p))
 245                 opts = strdup("mode=0755");
 246         else
 247                 opts = strdup(p);
 248         if (!opts)
 249                 return -ENOMEM;
 250
 251         if (!path_is_absolute(path))
 252                 return -EINVAL;
 253
 254         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 255         if (!m)
 256                 return -ENOMEM;
 257
 258         m->destination = path;
 259         m->options = opts;
 260
 261         path = opts = NULL;
 262         return 0;
 263 }
 264
 265 int overlay_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 266         _cleanup_free_ char *upper = NULL, *destination = NULL;
 267         _cleanup_strv_free_ char **lower = NULL;
 268         CustomMount *m;
 269         int k;
 270
 271         k = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 272         if (k < 0)
 273                 return k;
 274         if (k < 2)
 275                 return -EADDRNOTAVAIL;
 276         if (k == 2) {
 277                 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
 278                  * we'll also define the destination mount point the same as the upper. */
 279
 280                 if (!source_path_is_valid(lower[0]) ||
 281                     !source_path_is_valid(lower[1]))
 282                         return -EINVAL;
 283
 284                 upper = lower[1];
 285                 lower[1] = NULL;
 286
 287                 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
 288                 if (!destination)
 289                         return -ENOMEM;
 290         } else {
 291                 int i;
 292
 293                 /* If more than two parameters are specified, the last one is the destination, the second to last one
 294                  * the "upper", and all before that the "lower" directories. */
 295
 296                 for (i = 0; i < k - 1; i++)
 297                         if (!source_path_is_valid(lower[i]))
 298                                 return -EINVAL;
 299
 300                 destination = lower[k - 1];
 301                 upper = lower[k - 2];
 302                 lower[k - 2] = NULL;
 303
 304                 if (!path_is_absolute(destination))
 305                         return -EINVAL;
 306         }
 307
 308         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 309         if (!m)
 310                 return -ENOMEM;
 311
 312         m->destination = destination;
 313         m->source = upper;
 314         m->lower = lower;
 315         m->read_only = read_only;
 316
 317         upper = destination = NULL;
 318         lower = NULL;
 319
 320         return 0;
 321 }
 322
 323 static int tmpfs_patch_options(
 324                 const char *options,
 325                 bool userns,
 326                 uid_t uid_shift, uid_t uid_range,
 327                 bool patch_ids,
 328                 const char *selinux_apifs_context,
 329                 char **ret) {
 330
 331         char *buf = NULL;
 332
 333         if ((userns && uid_shift != 0) || patch_ids) {
 334                 assert(uid_shift != UID_INVALID);
 335
 336                 if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 337                              options ?: "", options ? "," : "",
 338                              uid_shift, uid_shift) < 0)
 339                         return -ENOMEM;
 340
 341                 options = buf;
 342         }
 343
 344 #ifdef HAVE_SELINUX
 345         if (selinux_apifs_context) {
 346                 char *t;
 347
 348                 t = strjoin(options ?: "", options ? "," : "",
 349                             "context=\"", selinux_apifs_context, "\"");
 350                 free(buf);
 351                 if (!t)
 352                         return -ENOMEM;
 353
 354                 buf = t;
 355         }
 356 #endif
 357
 358         if (!buf && options) {
 359                 buf = strdup(options);
 360                 if (!buf)
 361                         return -ENOMEM;
 362         }
 363         *ret = buf;
 364
 365         return !!buf;
 366 }
 367
 368 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 369         const char *full, *top, *x;
 370         int r;
 371         unsigned long extra_flags = 0;
 372
 373         top = prefix_roota(dest, "/sys");
 374         r = path_check_fstype(top, SYSFS_MAGIC);
 375         if (r < 0)
 376                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 377         /* /sys might already be mounted as sysfs by the outer child in the
 378          * !netns case. In this case, it's all good. Don't touch it because we
 379          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 380          */
 381         if (r > 0)
 382                 return 0;
 383
 384         full = prefix_roota(top, "/full");
 385
 386         (void) mkdir(full, 0755);
 387
 388         if (mount_settings & MOUNT_APPLY_APIVFS_RO)
 389                 extra_flags |= MS_RDONLY;
 390
 391         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 392                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 393         if (r < 0)
 394                 return r;
 395
 396         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 397                 _cleanup_free_ char *from = NULL, *to = NULL;
 398
 399                 from = prefix_root(full, x);
 400                 if (!from)
 401                         return log_oom();
 402
 403                 to = prefix_root(top, x);
 404                 if (!to)
 405                         return log_oom();
 406
 407                 (void) mkdir(to, 0755);
 408
 409                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 410                 if (r < 0)
 411                         return r;
 412
 413                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 414                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 415                 if (r < 0)
 416                         return r;
 417         }
 418
 419         r = umount_verbose(full);
 420         if (r < 0)
 421                 return r;
 422
 423         if (rmdir(full) < 0)
 424                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 425
 426         x = prefix_roota(top, "/fs/kdbus");
 427         (void) mkdir_p(x, 0755);
 428
 429         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 430          * remount /sys read-only.
 431          */
 432         if (cg_ns_supported()) {
 433                 x = prefix_roota(top, "/fs/cgroup");
 434                 (void) mkdir_p(x, 0755);
 435         }
 436
 437         return mount_verbose(LOG_ERR, NULL, top, NULL,
 438                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 439 }
 440
 441 static int mkdir_userns(const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 442         int r;
 443
 444         assert(path);
 445
 446         r = mkdir(path, mode);
 447         if (r < 0 && errno != EEXIST)
 448                 return -errno;
 449
 450         if ((mask & MOUNT_USE_USERNS) == 0)
 451                 return 0;
 452
 453         if (mask & MOUNT_IN_USERNS)
 454                 return 0;
 455
 456         r = lchown(path, uid_shift, uid_shift);
 457         if (r < 0)
 458                 return -errno;
 459
 460         return 0;
 461 }
 462
 463 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 464         const char *p, *e;
 465         int r;
 466
 467         assert(path);
 468
 469         if (prefix && !path_startswith(path, prefix))
 470                 return -ENOTDIR;
 471
 472         /* create every parent directory in the path, except the last component */
 473         p = path + strspn(path, "/");
 474         for (;;) {
 475                 char t[strlen(path) + 1];
 476
 477                 e = p + strcspn(p, "/");
 478                 p = e + strspn(e, "/");
 479
 480                 /* Is this the last component? If so, then we're done */
 481                 if (*p == 0)
 482                         break;
 483
 484                 memcpy(t, path, e - path);
 485                 t[e-path] = 0;
 486
 487                 if (prefix && path_startswith(prefix, t))
 488                         continue;
 489
 490                 r = mkdir_userns(t, mode, mask, uid_shift);
 491                 if (r < 0)
 492                         return r;
 493         }
 494
 495         return mkdir_userns(path, mode, mask, uid_shift);
 496 }
 497
 498 int mount_all(const char *dest,
 499               MountSettingsMask mount_settings,
 500               uid_t uid_shift, uid_t uid_range,
 501               const char *selinux_apifs_context) {
 502
 503         typedef struct MountPoint {
 504                 const char *what;
 505                 const char *where;
 506                 const char *type;
 507                 const char *options;
 508                 unsigned long flags;
 509                 MountSettingsMask mount_settings;
 510         } MountPoint;
 511
 512         static const MountPoint mount_table[] = {
 513                 /* inner child mounts */
 514                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_IN_USERNS },
 515                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ...*/
 516                 { "/proc/sys/net",       "/proc/sys/net",       NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 517                 { NULL,                  "/proc/sys",           NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 518                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                               MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ...*/
 519                 { NULL,                  "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,             MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 520                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            MOUNT_FATAL|MOUNT_IN_USERNS },
 521
 522                 /* outer child mounts */
 523                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS },
 524                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO },    /* skipped if above was mounted */
 525                 { "sysfs",               "/sys",                "sysfs", NULL,                  MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL },                          /* skipped if above was mounted */
 526
 527                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  MOUNT_FATAL },
 528                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 529                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 530 #ifdef HAVE_SELINUX
 531                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   0 },  /* Bind mount first */
 532                 { NULL,                  "/sys/fs/selinux",     NULL,     NULL,       MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, 0 },  /* Then, make it r/o */
 533 #endif
 534         };
 535
 536         unsigned k;
 537         int r;
 538         bool use_userns = (mount_settings & MOUNT_USE_USERNS);
 539         bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS);
 540         bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO);
 541         bool in_userns = (mount_settings & MOUNT_IN_USERNS);
 542
 543         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 544                 _cleanup_free_ char *where = NULL, *options = NULL;
 545                 const char *o;
 546                 bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL);
 547
 548                 if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS))
 549                         continue;
 550
 551                 if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS))
 552                         continue;
 553
 554                 if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO))
 555                         continue;
 556
 557                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NON_EXISTING|CHASE_PREFIX_ROOT, &where);
 558                 if (r < 0)
 559                         return log_error_errno(r, "Failed to resolve %s: %m", mount_table[k].where);
 560
 561                 r = path_is_mount_point(where, NULL, 0);
 562                 if (r < 0 && r != -ENOENT)
 563                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 564
 565                 /* Skip this entry if it is not a remount. */
 566                 if (mount_table[k].what && r > 0)
 567                         continue;
 568
 569                 r = mkdir_userns_p(dest, where, 0755, mount_settings, uid_shift);
 570                 if (r < 0 && r != -EEXIST) {
 571                         if (fatal)
 572                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 573
 574                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 575                         continue;
 576                 }
 577
 578                 o = mount_table[k].options;
 579                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 580                         if (in_userns)
 581                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 582                         else
 583                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 584                         if (r < 0)
 585                                 return log_oom();
 586                         if (r > 0)
 587                                 o = options;
 588                 }
 589
 590                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 591                                   mount_table[k].what,
 592                                   where,
 593                                   mount_table[k].type,
 594                                   mount_table[k].flags,
 595                                   o);
 596                 if (r < 0 && fatal)
 597                         return r;
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 604         const char *p = options;
 605         unsigned long flags = *mount_flags;
 606         char *opts = NULL;
 607         int r;
 608
 609         assert(options);
 610
 611         for (;;) {
 612                 _cleanup_free_ char *word = NULL;
 613
 614                 r = extract_first_word(&p, &word, ",", 0);
 615                 if (r < 0)
 616                         return log_error_errno(r, "Failed to extract mount option: %m");
 617                 if (r == 0)
 618                         break;
 619
 620                 if (streq(word, "rbind"))
 621                         flags |= MS_REC;
 622                 else if (streq(word, "norbind"))
 623                         flags &= ~MS_REC;
 624                 else {
 625                         log_error("Invalid bind mount option: %s", word);
 626                         return -EINVAL;
 627                 }
 628         }
 629
 630         *mount_flags = flags;
 631         /* in the future mount_opts will hold string options for mount(2) */
 632         *mount_opts = opts;
 633
 634         return 0;
 635 }
 636
 637 static int mount_bind(const char *dest, CustomMount *m) {
 638
 639         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 640         unsigned long mount_flags = MS_BIND | MS_REC;
 641         struct stat source_st, dest_st;
 642         int r;
 643
 644         assert(dest);
 645         assert(m);
 646
 647         if (m->options) {
 648                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 649                 if (r < 0)
 650                         return r;
 651         }
 652
 653         if (stat(m->source, &source_st) < 0)
 654                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 655
 656         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NON_EXISTING, &where);
 657         if (r < 0)
 658                 return log_error_errno(r, "Failed to resolve %s: %m", m->destination);
 659         if (r > 0) { /* Path exists already? */
 660
 661                 if (stat(where, &dest_st) < 0)
 662                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 663
 664                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 665                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 666                         return -EINVAL;
 667                 }
 668
 669                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 670                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 671                         return -EINVAL;
 672                 }
 673
 674         } else { /* Path doesn't exist yet? */
 675                 r = mkdir_parents_label(where, 0755);
 676                 if (r < 0)
 677                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 678
 679                 /* Create the mount point. Any non-directory file can be
 680                 * mounted on any non-directory file (regular, fifo, socket,
 681                 * char, block).
 682                 */
 683                 if (S_ISDIR(source_st.st_mode))
 684                         r = mkdir_label(where, 0755);
 685                 else
 686                         r = touch(where);
 687                 if (r < 0)
 688                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 689
 690         }
 691
 692         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 693         if (r < 0)
 694                 return r;
 695
 696         if (m->read_only) {
 697                 r = bind_remount_recursive(where, true, NULL);
 698                 if (r < 0)
 699                         return log_error_errno(r, "Read-only bind mount failed: %m");
 700         }
 701
 702         return 0;
 703 }
 704
 705 static int mount_tmpfs(
 706                 const char *dest,
 707                 CustomMount *m,
 708                 bool userns, uid_t uid_shift, uid_t uid_range,
 709                 const char *selinux_apifs_context) {
 710
 711         const char *options;
 712         _cleanup_free_ char *buf = NULL, *where = NULL;
 713         int r;
 714
 715         assert(dest);
 716         assert(m);
 717
 718         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NON_EXISTING, &where);
 719         if (r < 0)
 720                 return log_error_errno(r, "Failed to resolve %s: %m", m->destination);
 721         if (r == 0) { /* Doesn't exist yet? */
 722                 r = mkdir_p_label(where, 0755);
 723                 if (r < 0)
 724                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 725         }
 726
 727         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 728         if (r < 0)
 729                 return log_oom();
 730         options = r > 0 ? buf : m->options;
 731
 732         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 733 }
 734
 735 static char *joined_and_escaped_lower_dirs(char **lower) {
 736         _cleanup_strv_free_ char **sv = NULL;
 737
 738         sv = strv_copy(lower);
 739         if (!sv)
 740                 return NULL;
 741
 742         strv_reverse(sv);
 743
 744         if (!strv_shell_escape(sv, ",:"))
 745                 return NULL;
 746
 747         return strv_join(sv, ":");
 748 }
 749
 750 static int mount_overlay(const char *dest, CustomMount *m) {
 751
 752         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 753         const char *options;
 754         int r;
 755
 756         assert(dest);
 757         assert(m);
 758
 759         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NON_EXISTING, &where);
 760         if (r < 0)
 761                 return log_error_errno(r, "Failed to resolve %s: %m", m->destination);
 762         if (r == 0) { /* Doesn't exist yet? */
 763                 r = mkdir_label(where, 0755);
 764                 if (r < 0)
 765                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 766         }
 767
 768         (void) mkdir_p_label(m->source, 0755);
 769
 770         lower = joined_and_escaped_lower_dirs(m->lower);
 771         if (!lower)
 772                 return log_oom();
 773
 774         escaped_source = shell_escape(m->source, ",:");
 775         if (!escaped_source)
 776                 return log_oom();
 777
 778         if (m->read_only)
 779                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 780         else {
 781                 _cleanup_free_ char *escaped_work_dir = NULL;
 782
 783                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 784                 if (!escaped_work_dir)
 785                         return log_oom();
 786
 787                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 788         }
 789
 790         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 791 }
 792
 793 int mount_custom(
 794                 const char *dest,
 795                 CustomMount *mounts, unsigned n,
 796                 bool userns, uid_t uid_shift, uid_t uid_range,
 797                 const char *selinux_apifs_context) {
 798
 799         unsigned i;
 800         int r;
 801
 802         assert(dest);
 803
 804         for (i = 0; i < n; i++) {
 805                 CustomMount *m = mounts + i;
 806
 807                 switch (m->type) {
 808
 809                 case CUSTOM_MOUNT_BIND:
 810                         r = mount_bind(dest, m);
 811                         break;
 812
 813                 case CUSTOM_MOUNT_TMPFS:
 814                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 815                         break;
 816
 817                 case CUSTOM_MOUNT_OVERLAY:
 818                         r = mount_overlay(dest, m);
 819                         break;
 820
 821                 default:
 822                         assert_not_reached("Unknown custom mount type");
 823                 }
 824
 825                 if (r < 0)
 826                         return r;
 827         }
 828
 829         return 0;
 830 }
 831
 832 /* Retrieve existing subsystems. This function is called in a new cgroup
 833  * namespace.
 834  */
 835 static int get_controllers(Set *subsystems) {
 836         _cleanup_fclose_ FILE *f = NULL;
 837         char line[LINE_MAX];
 838
 839         assert(subsystems);
 840
 841         f = fopen("/proc/self/cgroup", "re");
 842         if (!f)
 843                 return errno == ENOENT ? -ESRCH : -errno;
 844
 845         FOREACH_LINE(line, f, return -errno) {
 846                 int r;
 847                 char *e, *l, *p;
 848
 849                 l = strchr(line, ':');
 850                 if (!l)
 851                         continue;
 852
 853                 l++;
 854                 e = strchr(l, ':');
 855                 if (!e)
 856                         continue;
 857
 858                 *e = 0;
 859
 860                 if (STR_IN_SET(l, "", "name=systemd"))
 861                         continue;
 862
 863                 p = strdup(l);
 864                 if (!p)
 865                         return -ENOMEM;
 866
 867                 r = set_consume(subsystems, p);
 868                 if (r < 0)
 869                         return r;
 870         }
 871
 872         return 0;
 873 }
 874
 875 static int mount_legacy_cgroup_hierarchy(
 876                 const char *dest,
 877                 const char *controller,
 878                 const char *hierarchy,
 879                 CGroupUnified unified_requested,
 880                 bool read_only) {
 881
 882         const char *to, *fstype, *opts;
 883         int r;
 884
 885         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 886
 887         r = path_is_mount_point(to, dest, 0);
 888         if (r < 0 && r != -ENOENT)
 889                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 890         if (r > 0)
 891                 return 0;
 892
 893         mkdir_p(to, 0755);
 894
 895         /* The superblock mount options of the mount point need to be
 896          * identical to the hosts', and hence writable... */
 897         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 898                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
 899                         fstype = "cgroup2";
 900                         opts = NULL;
 901                 } else {
 902                         fstype = "cgroup";
 903                         opts = "none,name=systemd,xattr";
 904                 }
 905         } else {
 906                 fstype = "cgroup";
 907                 opts = controller;
 908         }
 909
 910         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 911         if (r < 0)
 912                 return r;
 913
 914         /* ... hence let's only make the bind mount read-only, not the superblock. */
 915         if (read_only) {
 916                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 917                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 918                 if (r < 0)
 919                         return r;
 920         }
 921
 922         return 1;
 923 }
 924
 925 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 926 static int mount_legacy_cgns_supported(
 927                 const char *dest,
 928                 CGroupUnified unified_requested,
 929                 bool userns,
 930                 uid_t uid_shift,
 931                 uid_t uid_range,
 932                 const char *selinux_apifs_context) {
 933
 934         _cleanup_set_free_free_ Set *controllers = NULL;
 935         const char *cgroup_root = "/sys/fs/cgroup", *c;
 936         int r;
 937
 938         (void) mkdir_p(cgroup_root, 0755);
 939
 940         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 941         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 942         if (r < 0)
 943                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 944         if (r == 0) {
 945                 _cleanup_free_ char *options = NULL;
 946
 947                 /* When cgroup namespaces are enabled and user namespaces are
 948                  * used then the mount of the cgroupfs is done *inside* the new
 949                  * user namespace. We're root in the new user namespace and the
 950                  * kernel will happily translate our uid/gid to the correct
 951                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 952                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 953                  */
 954                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 955                 if (r < 0)
 956                         return log_oom();
 957
 958                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 959                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 960                 if (r < 0)
 961                         return r;
 962         }
 963
 964         if (cg_all_unified() > 0)
 965                 goto skip_controllers;
 966
 967         controllers = set_new(&string_hash_ops);
 968         if (!controllers)
 969                 return log_oom();
 970
 971         r = get_controllers(controllers);
 972         if (r < 0)
 973                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 974
 975         for (;;) {
 976                 _cleanup_free_ const char *controller = NULL;
 977
 978                 controller = set_steal_first(controllers);
 979                 if (!controller)
 980                         break;
 981
 982                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 983                 if (r < 0)
 984                         return r;
 985
 986                 /* When multiple hierarchies are co-mounted, make their
 987                  * constituting individual hierarchies a symlink to the
 988                  * co-mount.
 989                  */
 990                 c = controller;
 991                 for (;;) {
 992                         _cleanup_free_ char *target = NULL, *tok = NULL;
 993
 994                         r = extract_first_word(&c, &tok, ",", 0);
 995                         if (r < 0)
 996                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 997                         if (r == 0)
 998                                 break;
 999
1000                         target = prefix_root("/sys/fs/cgroup", tok);
1001                         if (!target)
1002                                 return log_oom();
1003
1004                         if (streq(controller, tok))
1005                                 break;
1006
1007                         r = symlink_idempotent(controller, target);
1008                         if (r == -EINVAL)
1009                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1010                         if (r < 0)
1011                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1012                 }
1013         }
1014
1015 skip_controllers:
1016         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
1017         if (r < 0)
1018                 return r;
1019
1020         if (!userns)
1021                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1022                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1023
1024         return 0;
1025 }
1026
1027 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
1028 static int mount_legacy_cgns_unsupported(
1029                 const char *dest,
1030                 CGroupUnified unified_requested,
1031                 bool userns,
1032                 uid_t uid_shift,
1033                 uid_t uid_range,
1034                 const char *selinux_apifs_context) {
1035
1036         _cleanup_set_free_free_ Set *controllers = NULL;
1037         const char *cgroup_root;
1038         int r;
1039
1040         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
1041
1042         (void) mkdir_p(cgroup_root, 0755);
1043
1044         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
1045         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
1046         if (r < 0)
1047                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
1048         if (r == 0) {
1049                 _cleanup_free_ char *options = NULL;
1050
1051                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
1052                 if (r < 0)
1053                         return log_oom();
1054
1055                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
1056                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
1057                 if (r < 0)
1058                         return r;
1059         }
1060
1061         if (cg_all_unified() > 0)
1062                 goto skip_controllers;
1063
1064         controllers = set_new(&string_hash_ops);
1065         if (!controllers)
1066                 return log_oom();
1067
1068         r = cg_kernel_controllers(controllers);
1069         if (r < 0)
1070                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1071
1072         for (;;) {
1073                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
1074
1075                 controller = set_steal_first(controllers);
1076                 if (!controller)
1077                         break;
1078
1079                 origin = prefix_root("/sys/fs/cgroup/", controller);
1080                 if (!origin)
1081                         return log_oom();
1082
1083                 r = readlink_malloc(origin, &combined);
1084                 if (r == -EINVAL) {
1085                         /* Not a symbolic link, but directly a single cgroup hierarchy */
1086
1087                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
1088                         if (r < 0)
1089                                 return r;
1090
1091                 } else if (r < 0)
1092                         return log_error_errno(r, "Failed to read link %s: %m", origin);
1093                 else {
1094                         _cleanup_free_ char *target = NULL;
1095
1096                         target = prefix_root(dest, origin);
1097                         if (!target)
1098                                 return log_oom();
1099
1100                         /* A symbolic link, a combination of controllers in one hierarchy */
1101
1102                         if (!filename_is_valid(combined)) {
1103                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
1104                                 continue;
1105                         }
1106
1107                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
1108                         if (r < 0)
1109                                 return r;
1110
1111                         r = symlink_idempotent(combined, target);
1112                         if (r == -EINVAL)
1113                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1114                         if (r < 0)
1115                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1116                 }
1117         }
1118
1119 skip_controllers:
1120         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
1121         if (r < 0)
1122                 return r;
1123
1124         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1125                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1126 }
1127
1128 static int mount_unified_cgroups(const char *dest) {
1129         const char *p;
1130         int r;
1131
1132         assert(dest);
1133
1134         p = prefix_roota(dest, "/sys/fs/cgroup");
1135
1136         (void) mkdir_p(p, 0755);
1137
1138         r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW);
1139         if (r < 0)
1140                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
1141         if (r > 0) {
1142                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
1143                 if (access(p, F_OK) >= 0)
1144                         return 0;
1145                 if (errno != ENOENT)
1146                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
1147
1148                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
1149                 return -EINVAL;
1150         }
1151
1152         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1153 }
1154
1155 int mount_cgroups(
1156                 const char *dest,
1157                 CGroupUnified unified_requested,
1158                 bool userns,
1159                 uid_t uid_shift,
1160                 uid_t uid_range,
1161                 const char *selinux_apifs_context,
1162                 bool use_cgns) {
1163
1164         if (unified_requested >= CGROUP_UNIFIED_ALL)
1165                 return mount_unified_cgroups(dest);
1166         else if (use_cgns)
1167                 return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1168
1169         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1170 }
1171
1172 int mount_systemd_cgroup_writable(
1173                 const char *dest,
1174                 CGroupUnified unified_requested) {
1175
1176         _cleanup_free_ char *own_cgroup_path = NULL;
1177         const char *systemd_root, *systemd_own;
1178         int r;
1179
1180         assert(dest);
1181
1182         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1183         if (r < 0)
1184                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1185
1186         /* If we are living in the top-level, then there's nothing to do... */
1187         if (path_equal(own_cgroup_path, "/"))
1188                 return 0;
1189
1190         if (unified_requested >= CGROUP_UNIFIED_ALL) {
1191                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
1192                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
1193         } else {
1194                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1195                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
1196         }
1197
1198         /* Make our own cgroup a (writable) bind mount */
1199         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1200         if (r < 0)
1201                 return r;
1202
1203         /* And then remount the systemd cgroup root read-only */
1204         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1205                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1206 }
1207
1208 int setup_volatile_state(
1209                 const char *directory,
1210                 VolatileMode mode,
1211                 bool userns, uid_t uid_shift, uid_t uid_range,
1212                 const char *selinux_apifs_context) {
1213
1214         _cleanup_free_ char *buf = NULL;
1215         const char *p, *options;
1216         int r;
1217
1218         assert(directory);
1219
1220         if (mode != VOLATILE_STATE)
1221                 return 0;
1222
1223         /* --volatile=state means we simply overmount /var
1224            with a tmpfs, and the rest read-only. */
1225
1226         r = bind_remount_recursive(directory, true, NULL);
1227         if (r < 0)
1228                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1229
1230         p = prefix_roota(directory, "/var");
1231         r = mkdir(p, 0755);
1232         if (r < 0 && errno != EEXIST)
1233                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1234
1235         options = "mode=755";
1236         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1237         if (r < 0)
1238                 return log_oom();
1239         if (r > 0)
1240                 options = buf;
1241
1242         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1243 }
1244
1245 int setup_volatile(
1246                 const char *directory,
1247                 VolatileMode mode,
1248                 bool userns, uid_t uid_shift, uid_t uid_range,
1249                 const char *selinux_apifs_context) {
1250
1251         bool tmpfs_mounted = false, bind_mounted = false;
1252         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1253         _cleanup_free_ char *buf = NULL;
1254         const char *f, *t, *options;
1255         int r;
1256
1257         assert(directory);
1258
1259         if (mode != VOLATILE_YES)
1260                 return 0;
1261
1262         /* --volatile=yes means we mount a tmpfs to the root dir, and
1263            the original /usr to use inside it, and that read-only. */
1264
1265         if (!mkdtemp(template))
1266                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1267
1268         options = "mode=755";
1269         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1270         if (r < 0)
1271                 return log_oom();
1272         if (r > 0)
1273                 options = buf;
1274
1275         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1276         if (r < 0)
1277                 goto fail;
1278
1279         tmpfs_mounted = true;
1280
1281         f = prefix_roota(directory, "/usr");
1282         t = prefix_roota(template, "/usr");
1283
1284         r = mkdir(t, 0755);
1285         if (r < 0 && errno != EEXIST) {
1286                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1287                 goto fail;
1288         }
1289
1290         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1291         if (r < 0)
1292                 goto fail;
1293
1294         bind_mounted = true;
1295
1296         r = bind_remount_recursive(t, true, NULL);
1297         if (r < 0) {
1298                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1299                 goto fail;
1300         }
1301
1302         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1303         if (r < 0)
1304                 goto fail;
1305
1306         (void) rmdir(template);
1307
1308         return 0;
1309
1310 fail:
1311         if (bind_mounted)
1312                 (void) umount_verbose(t);
1313
1314         if (tmpfs_mounted)
1315                 (void) umount_verbose(template);
1316         (void) rmdir(template);
1317         return r;
1318 }
1319
1320 VolatileMode volatile_mode_from_string(const char *s) {
1321         int b;
1322
1323         if (isempty(s))
1324                 return _VOLATILE_MODE_INVALID;
1325
1326         b = parse_boolean(s);
1327         if (b > 0)
1328                 return VOLATILE_YES;
1329         if (b == 0)
1330                 return VOLATILE_NO;
1331
1332         if (streq(s, "state"))
1333                 return VOLATILE_STATE;
1334
1335         return _VOLATILE_MODE_INVALID;
1336 }