src/nspawn/nspawn-mount.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <sys/mount.h>
   4 #include <linux/magic.h>
   5
   6 #include "alloc-util.h"
   7 #include "escape.h"
   8 #include "fd-util.h"
   9 #include "format-util.h"
  10 #include "fs-util.h"
  11 #include "label.h"
  12 #include "mkdir.h"
  13 #include "mount-util.h"
  14 #include "mountpoint-util.h"
  15 #include "nspawn-mount.h"
  16 #include "parse-util.h"
  17 #include "path-util.h"
  18 #include "rm-rf.h"
  19 #include "set.h"
  20 #include "sort-util.h"
  21 #include "stat-util.h"
  22 #include "string-util.h"
  23 #include "strv.h"
  24 #include "tmpfile-util.h"
  25 #include "user-util.h"
  26
  27 CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t) {
  28         CustomMount *c, *ret;
  29
  30         assert(l);
  31         assert(n);
  32         assert(t >= 0);
  33         assert(t < _CUSTOM_MOUNT_TYPE_MAX);
  34
  35         c = reallocarray(*l, *n + 1, sizeof(CustomMount));
  36         if (!c)
  37                 return NULL;
  38
  39         *l = c;
  40         ret = *l + *n;
  41         (*n)++;
  42
  43         *ret = (CustomMount) { .type = t };
  44
  45         return ret;
  46 }
  47
  48 void custom_mount_free_all(CustomMount *l, size_t n) {
  49         size_t i;
  50
  51         for (i = 0; i < n; i++) {
  52                 CustomMount *m = l + i;
  53
  54                 free(m->source);
  55                 free(m->destination);
  56                 free(m->options);
  57
  58                 if (m->work_dir) {
  59                         (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
  60                         free(m->work_dir);
  61                 }
  62
  63                 if (m->rm_rf_tmpdir) {
  64                         (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL);
  65                         free(m->rm_rf_tmpdir);
  66                 }
  67
  68                 strv_free(m->lower);
  69                 free(m->type_argument);
  70         }
  71
  72         free(l);
  73 }
  74
  75 static int custom_mount_compare(const CustomMount *a, const CustomMount *b) {
  76         int r;
  77
  78         r = path_compare(a->destination, b->destination);
  79         if (r != 0)
  80                 return r;
  81
  82         return CMP(a->type, b->type);
  83 }
  84
  85 static bool source_path_is_valid(const char *p) {
  86         assert(p);
  87
  88         if (*p == '+')
  89                 p++;
  90
  91         return path_is_absolute(p);
  92 }
  93
  94 static char *resolve_source_path(const char *dest, const char *source) {
  95
  96         if (!source)
  97                 return NULL;
  98
  99         if (source[0] == '+')
 100                 return path_join(dest, source + 1);
 101
 102         return strdup(source);
 103 }
 104
 105 static int allocate_temporary_source(CustomMount *m) {
 106         assert(m);
 107         assert(!m->source);
 108         assert(!m->rm_rf_tmpdir);
 109
 110         m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX");
 111         if (!m->rm_rf_tmpdir)
 112                 return log_oom();
 113
 114         if (!mkdtemp(m->rm_rf_tmpdir)) {
 115                 m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir);
 116                 return log_error_errno(errno, "Failed to acquire temporary directory: %m");
 117         }
 118
 119         m->source = path_join(m->rm_rf_tmpdir, "src");
 120         if (!m->source)
 121                 return log_oom();
 122
 123         if (mkdir(m->source, 0755) < 0)
 124                 return log_error_errno(errno, "Failed to create %s: %m", m->source);
 125
 126         return 0;
 127 }
 128
 129 int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) {
 130         size_t i;
 131         int r;
 132
 133         /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
 134          * parent process, so that we know the temporary directories to remove on exit before we fork off the
 135          * children. */
 136
 137         assert(l || n == 0);
 138
 139         /* Order the custom mounts, and make sure we have a working directory */
 140         typesafe_qsort(l, n, custom_mount_compare);
 141
 142         for (i = 0; i < n; i++) {
 143                 CustomMount *m = l + i;
 144
 145                 /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
 146                  * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
 147                  * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
 148                  * the inner child, not the outer one. Determine this here. */
 149                 m->in_userns = path_startswith(m->destination, "/proc");
 150
 151                 if (m->type == CUSTOM_MOUNT_BIND) {
 152                         if (m->source) {
 153                                 char *s;
 154
 155                                 s = resolve_source_path(dest, m->source);
 156                                 if (!s)
 157                                         return log_oom();
 158
 159                                 free_and_replace(m->source, s);
 160                         } else {
 161                                 /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
 162
 163                                 r = allocate_temporary_source(m);
 164                                 if (r < 0)
 165                                         return r;
 166                         }
 167                 }
 168
 169                 if (m->type == CUSTOM_MOUNT_OVERLAY) {
 170                         char **j;
 171
 172                         STRV_FOREACH(j, m->lower) {
 173                                 char *s;
 174
 175                                 s = resolve_source_path(dest, *j);
 176                                 if (!s)
 177                                         return log_oom();
 178
 179                                 free_and_replace(*j, s);
 180                         }
 181
 182                         if (m->source) {
 183                                 char *s;
 184
 185                                 s = resolve_source_path(dest, m->source);
 186                                 if (!s)
 187                                         return log_oom();
 188
 189                                 free_and_replace(m->source, s);
 190                         } else {
 191                                 r = allocate_temporary_source(m);
 192                                 if (r < 0)
 193                                         return r;
 194                         }
 195
 196                         if (m->work_dir) {
 197                                 char *s;
 198
 199                                 s = resolve_source_path(dest, m->work_dir);
 200                                 if (!s)
 201                                         return log_oom();
 202
 203                                 free_and_replace(m->work_dir, s);
 204                         } else {
 205                                 r = tempfn_random(m->source, NULL, &m->work_dir);
 206                                 if (r < 0)
 207                                         return log_error_errno(r, "Failed to acquire working directory: %m");
 208                         }
 209
 210                         (void) mkdir_label(m->work_dir, 0700);
 211                 }
 212         }
 213
 214         return 0;
 215 }
 216
 217 int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 218         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 219         const char *p = s;
 220         CustomMount *m;
 221         int r;
 222
 223         assert(l);
 224         assert(n);
 225
 226         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 227         if (r < 0)
 228                 return r;
 229         if (r == 0)
 230                 return -EINVAL;
 231         if (r == 1) {
 232                 destination = strdup(source[0] == '+' ? source+1 : source);
 233                 if (!destination)
 234                         return -ENOMEM;
 235         }
 236         if (r == 2 && !isempty(p)) {
 237                 opts = strdup(p);
 238                 if (!opts)
 239                         return -ENOMEM;
 240         }
 241
 242         if (isempty(source))
 243                 source = mfree(source);
 244         else if (!source_path_is_valid(source))
 245                 return -EINVAL;
 246
 247         if (!path_is_absolute(destination))
 248                 return -EINVAL;
 249
 250         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 251         if (!m)
 252                 return -ENOMEM;
 253
 254         m->source = TAKE_PTR(source);
 255         m->destination = TAKE_PTR(destination);
 256         m->read_only = read_only;
 257         m->options = TAKE_PTR(opts);
 258
 259         return 0;
 260 }
 261
 262 int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) {
 263         _cleanup_free_ char *path = NULL, *opts = NULL;
 264         const char *p = s;
 265         CustomMount *m;
 266         int r;
 267
 268         assert(l);
 269         assert(n);
 270         assert(s);
 271
 272         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 273         if (r < 0)
 274                 return r;
 275         if (r == 0)
 276                 return -EINVAL;
 277
 278         if (isempty(p))
 279                 opts = strdup("mode=0755");
 280         else
 281                 opts = strdup(p);
 282         if (!opts)
 283                 return -ENOMEM;
 284
 285         if (!path_is_absolute(path))
 286                 return -EINVAL;
 287
 288         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 289         if (!m)
 290                 return -ENOMEM;
 291
 292         m->destination = TAKE_PTR(path);
 293         m->options = TAKE_PTR(opts);
 294
 295         return 0;
 296 }
 297
 298 int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
 299         _cleanup_free_ char *upper = NULL, *destination = NULL;
 300         _cleanup_strv_free_ char **lower = NULL;
 301         CustomMount *m;
 302         int k;
 303
 304         k = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 305         if (k < 0)
 306                 return k;
 307         if (k < 2)
 308                 return -EADDRNOTAVAIL;
 309         if (k == 2) {
 310                 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
 311                  * we'll also define the destination mount point the same as the upper. */
 312
 313                 if (!source_path_is_valid(lower[0]) ||
 314                     !source_path_is_valid(lower[1]))
 315                         return -EINVAL;
 316
 317                 upper = TAKE_PTR(lower[1]);
 318
 319                 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
 320                 if (!destination)
 321                         return -ENOMEM;
 322         } else {
 323                 char **i;
 324
 325                 /* If more than two parameters are specified, the last one is the destination, the second to last one
 326                  * the "upper", and all before that the "lower" directories. */
 327
 328                 destination = lower[k - 1];
 329                 upper = TAKE_PTR(lower[k - 2]);
 330
 331                 STRV_FOREACH(i, lower)
 332                         if (!source_path_is_valid(*i))
 333                                 return -EINVAL;
 334
 335                 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
 336                  * in /var/tmp */
 337                 if (isempty(upper))
 338                         upper = mfree(upper);
 339                 else if (!source_path_is_valid(upper))
 340                         return -EINVAL;
 341
 342                 if (!path_is_absolute(destination))
 343                         return -EINVAL;
 344         }
 345
 346         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 347         if (!m)
 348                 return -ENOMEM;
 349
 350         m->destination = TAKE_PTR(destination);
 351         m->source = TAKE_PTR(upper);
 352         m->lower = TAKE_PTR(lower);
 353         m->read_only = read_only;
 354
 355         return 0;
 356 }
 357
 358 int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) {
 359         _cleanup_free_ char *path = NULL;
 360         CustomMount *m;
 361
 362         assert(l);
 363         assert(n);
 364         assert(s);
 365
 366         if (!path_is_absolute(s))
 367                 return -EINVAL;
 368
 369         path = strdup(s);
 370         if (!path)
 371                 return -ENOMEM;
 372
 373         m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE);
 374         if (!m)
 375                 return -ENOMEM;
 376
 377         m->destination = TAKE_PTR(path);
 378         return 0;
 379 }
 380
 381 int tmpfs_patch_options(
 382                 const char *options,
 383                 uid_t uid_shift,
 384                 const char *selinux_apifs_context,
 385                 char **ret) {
 386
 387         char *buf = NULL;
 388
 389         if (uid_shift != UID_INVALID) {
 390                 if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 391                              strempty(options), options ? "," : "",
 392                              uid_shift, uid_shift) < 0)
 393                         return -ENOMEM;
 394
 395                 options = buf;
 396         }
 397
 398 #if HAVE_SELINUX
 399         if (selinux_apifs_context) {
 400                 char *t;
 401
 402                 t = strjoin(strempty(options), options ? "," : "",
 403                             "context=\"", selinux_apifs_context, "\"");
 404                 free(buf);
 405                 if (!t)
 406                         return -ENOMEM;
 407
 408                 buf = t;
 409         }
 410 #endif
 411
 412         if (!buf && options) {
 413                 buf = strdup(options);
 414                 if (!buf)
 415                         return -ENOMEM;
 416         }
 417         *ret = buf;
 418
 419         return !!buf;
 420 }
 421
 422 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 423         const char *full, *top, *x;
 424         int r;
 425         unsigned long extra_flags = 0;
 426
 427         top = prefix_roota(dest, "/sys");
 428         r = path_is_fs_type(top, SYSFS_MAGIC);
 429         if (r < 0)
 430                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 431         /* /sys might already be mounted as sysfs by the outer child in the
 432          * !netns case. In this case, it's all good. Don't touch it because we
 433          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 434          */
 435         if (r > 0)
 436                 return 0;
 437
 438         full = prefix_roota(top, "/full");
 439
 440         (void) mkdir(full, 0755);
 441
 442         if (FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO))
 443                 extra_flags |= MS_RDONLY;
 444
 445         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 446                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 447         if (r < 0)
 448                 return r;
 449
 450         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 451                 _cleanup_free_ char *from = NULL, *to = NULL;
 452
 453                 from = path_join(full, x);
 454                 if (!from)
 455                         return log_oom();
 456
 457                 to = path_join(top, x);
 458                 if (!to)
 459                         return log_oom();
 460
 461                 (void) mkdir(to, 0755);
 462
 463                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 464                 if (r < 0)
 465                         return r;
 466
 467                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 468                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 469                 if (r < 0)
 470                         return r;
 471         }
 472
 473         r = umount_verbose(full);
 474         if (r < 0)
 475                 return r;
 476
 477         if (rmdir(full) < 0)
 478                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 479
 480         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 481          * remount /sys read-only.
 482          */
 483         x = prefix_roota(top, "/fs/cgroup");
 484         (void) mkdir_p(x, 0755);
 485
 486         return mount_verbose(LOG_ERR, NULL, top, NULL,
 487                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 488 }
 489
 490 int mount_all(const char *dest,
 491               MountSettingsMask mount_settings,
 492               uid_t uid_shift,
 493               const char *selinux_apifs_context) {
 494
 495 #define PROC_INACCESSIBLE_REG(path)                                     \
 496         { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
 497           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 498         { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 499           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 500
 501 #define PROC_READ_ONLY(path)                                            \
 502         { (path), (path), NULL, NULL, MS_BIND,                          \
 503           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
 504         { NULL,   (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
 505           MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
 506
 507         typedef struct MountPoint {
 508                 const char *what;
 509                 const char *where;
 510                 const char *type;
 511                 const char *options;
 512                 unsigned long flags;
 513                 MountSettingsMask mount_settings;
 514         } MountPoint;
 515
 516         static const MountPoint mount_table[] = {
 517                 /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */
 518                 { "proc",            "/proc",           "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,
 519                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_MKDIR },
 520
 521                 { "/proc/sys",       "/proc/sys",       NULL,    NULL,        MS_BIND,
 522                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 523
 524                 { "/proc/sys/net",   "/proc/sys/net",   NULL,    NULL,        MS_BIND,
 525                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 526
 527                 { NULL,              "/proc/sys",       NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 528                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 529
 530                 /* Make these files inaccessible to container payloads: they potentially leak information about kernel
 531                  * internals or the host's execution environment to the container */
 532                 PROC_INACCESSIBLE_REG("/proc/kallsyms"),
 533                 PROC_INACCESSIBLE_REG("/proc/kcore"),
 534                 PROC_INACCESSIBLE_REG("/proc/keys"),
 535                 PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
 536                 PROC_INACCESSIBLE_REG("/proc/timer_list"),
 537
 538                 /* Make these directories read-only to container payloads: they show hardware information, and in some
 539                  * cases contain tunables the container really shouldn't have access to. */
 540                 PROC_READ_ONLY("/proc/acpi"),
 541                 PROC_READ_ONLY("/proc/apm"),
 542                 PROC_READ_ONLY("/proc/asound"),
 543                 PROC_READ_ONLY("/proc/bus"),
 544                 PROC_READ_ONLY("/proc/fs"),
 545                 PROC_READ_ONLY("/proc/irq"),
 546                 PROC_READ_ONLY("/proc/scsi"),
 547
 548                 { "mqueue",                 "/dev/mqueue",                  "mqueue", NULL,                            MS_NOSUID|MS_NOEXEC|MS_NODEV,
 549                   MOUNT_IN_USERNS|MOUNT_MKDIR },
 550
 551                 /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
 552                 { "tmpfs",                  "/tmp",                         "tmpfs", "mode=1777" TMPFS_LIMITS_TMP,     MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 553                   MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR },
 554                 { "tmpfs",                  "/sys",                         "tmpfs", "mode=555" TMPFS_LIMITS_SYS,      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 555                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR },
 556                 { "sysfs",                  "/sys",                         "sysfs", NULL,                             MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
 557                   MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR },    /* skipped if above was mounted */
 558                 { "sysfs",                  "/sys",                         "sysfs", NULL,                             MS_NOSUID|MS_NOEXEC|MS_NODEV,
 559                   MOUNT_FATAL|MOUNT_MKDIR },                          /* skipped if above was mounted */
 560                 { "tmpfs",                  "/dev",                         "tmpfs", "mode=755" TMPFS_LIMITS_DEV,      MS_NOSUID|MS_STRICTATIME,
 561                   MOUNT_FATAL|MOUNT_MKDIR },
 562                 { "tmpfs",                  "/dev/shm",                     "tmpfs", "mode=1777" TMPFS_LIMITS_DEV_SHM, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 563                   MOUNT_FATAL|MOUNT_MKDIR },
 564                 { "tmpfs",                  "/run",                         "tmpfs", "mode=755" TMPFS_LIMITS_RUN,      MS_NOSUID|MS_NODEV|MS_STRICTATIME,
 565                   MOUNT_FATAL|MOUNT_MKDIR },
 566                 { "/usr/lib/os-release",    "/run/host/usr/lib/os-release", NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
 567                   MOUNT_FATAL|MOUNT_MKDIR|MOUNT_TOUCH },
 568                 { "/etc/os-release",        "/run/host/etc/os-release",     NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
 569                   MOUNT_MKDIR|MOUNT_TOUCH },
 570
 571 #if HAVE_SELINUX
 572                 { "/sys/fs/selinux",        "/sys/fs/selinux",              NULL,    NULL,                             MS_BIND,
 573                   MOUNT_MKDIR },  /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */
 574                 { NULL,                     "/sys/fs/selinux",              NULL,    NULL,                             MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
 575                   0 },            /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */
 576 #endif
 577         };
 578
 579         bool use_userns = FLAGS_SET(mount_settings, MOUNT_USE_USERNS);
 580         bool netns = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_NETNS);
 581         bool ro = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO);
 582         bool in_userns = FLAGS_SET(mount_settings, MOUNT_IN_USERNS);
 583         bool tmpfs_tmp = FLAGS_SET(mount_settings, MOUNT_APPLY_TMPFS_TMP);
 584         size_t k;
 585         int r;
 586
 587         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 588                 _cleanup_free_ char *where = NULL, *options = NULL;
 589                 const char *o;
 590                 struct stat source_st;
 591                 bool fatal = FLAGS_SET(mount_table[k].mount_settings, MOUNT_FATAL);
 592
 593                 if (in_userns != FLAGS_SET(mount_table[k].mount_settings, MOUNT_IN_USERNS))
 594                         continue;
 595
 596                 if (!netns && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_NETNS))
 597                         continue;
 598
 599                 if (!ro && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_RO))
 600                         continue;
 601
 602                 if (!tmpfs_tmp && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_TMPFS_TMP))
 603                         continue;
 604
 605                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL);
 606                 if (r < 0)
 607                         return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where);
 608
 609                 /* Skip this entry if it is not a remount. */
 610                 if (mount_table[k].what) {
 611                         r = path_is_mount_point(where, NULL, 0);
 612                         if (r < 0 && r != -ENOENT)
 613                                 return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 614                         if (r > 0)
 615                                 continue;
 616
 617                         /* Shortcut for optional bind mounts: if the source can't be found skip ahead to avoid creating
 618                          * empty and unused directories. */
 619                         if (!fatal && FLAGS_SET(mount_table[k].mount_settings, MOUNT_MKDIR) && FLAGS_SET(mount_table[k].flags, MS_BIND)) {
 620                                 r = stat(mount_table[k].what, &source_st);
 621                                 if (r < 0) {
 622                                         if (errno == ENOENT)
 623                                                 continue;
 624                                         return log_error_errno(errno, "Failed to stat %s: %m", mount_table[k].what);
 625                                 }
 626                         }
 627                 }
 628
 629                 if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_MKDIR)) {
 630                         uid_t u = (use_userns && !in_userns) ? uid_shift : UID_INVALID;
 631
 632                         if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH))
 633                                 r = mkdir_parents_safe(dest, where, 0755, u, u, 0);
 634                         else
 635                                 r = mkdir_p_safe(dest, where, 0755, u, u, 0);
 636                         if (r < 0 && r != -EEXIST) {
 637                                 if (fatal && r != -EROFS)
 638                                         return log_error_errno(r, "Failed to create directory %s: %m", where);
 639
 640                                 log_debug_errno(r, "Failed to create directory %s: %m", where);
 641
 642                                 /* If we failed mkdir() or chown() due to the root directory being read only,
 643                                  * attempt to mount this fs anyway and let mount_verbose log any errors */
 644                                 if (r != -EROFS)
 645                                         continue;
 646                         }
 647                         if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH)) {
 648                                 r = touch(where);
 649                                 if (r < 0 && r != -EEXIST) {
 650                                         if (fatal)
 651                                                 return log_error_errno(r, "Failed to create mount point %s: %m", where);
 652                                         log_debug_errno(r, "Failed to create mount point %s: %m", where);
 653                                 }
 654                         }
 655                 }
 656
 657                 o = mount_table[k].options;
 658                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 659                         r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
 660                         if (r < 0)
 661                                 return log_oom();
 662                         if (r > 0)
 663                                 o = options;
 664                 }
 665
 666                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 667                                   mount_table[k].what,
 668                                   where,
 669                                   mount_table[k].type,
 670                                   mount_table[k].flags,
 671                                   o);
 672                 if (r < 0 && fatal)
 673                         return r;
 674         }
 675
 676         return 0;
 677 }
 678
 679 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 680         const char *p = options;
 681         unsigned long flags = *mount_flags;
 682         char *opts = NULL;
 683         int r;
 684
 685         assert(options);
 686
 687         for (;;) {
 688                 _cleanup_free_ char *word = NULL;
 689
 690                 r = extract_first_word(&p, &word, ",", 0);
 691                 if (r < 0)
 692                         return log_error_errno(r, "Failed to extract mount option: %m");
 693                 if (r == 0)
 694                         break;
 695
 696                 if (streq(word, "rbind"))
 697                         flags |= MS_REC;
 698                 else if (streq(word, "norbind"))
 699                         flags &= ~MS_REC;
 700                 else {
 701                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 702                                                "Invalid bind mount option: %s",
 703                                                word);
 704                 }
 705         }
 706
 707         *mount_flags = flags;
 708         /* in the future mount_opts will hold string options for mount(2) */
 709         *mount_opts = opts;
 710
 711         return 0;
 712 }
 713
 714 static int mount_bind(const char *dest, CustomMount *m) {
 715         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 716         unsigned long mount_flags = MS_BIND | MS_REC;
 717         struct stat source_st, dest_st;
 718         int r;
 719
 720         assert(dest);
 721         assert(m);
 722
 723         if (m->options) {
 724                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 725                 if (r < 0)
 726                         return r;
 727         }
 728
 729         if (stat(m->source, &source_st) < 0)
 730                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 731
 732         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 733         if (r < 0)
 734                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 735         if (r > 0) { /* Path exists already? */
 736
 737                 if (stat(where, &dest_st) < 0)
 738                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 739
 740                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode))
 741                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 742                                                "Cannot bind mount directory %s on file %s.",
 743                                                m->source, where);
 744
 745                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode))
 746                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 747                                                "Cannot bind mount file %s on directory %s.",
 748                                                m->source, where);
 749
 750         } else { /* Path doesn't exist yet? */
 751                 r = mkdir_parents_label(where, 0755);
 752                 if (r < 0)
 753                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 754
 755                 /* Create the mount point. Any non-directory file can be
 756                 * mounted on any non-directory file (regular, fifo, socket,
 757                 * char, block).
 758                 */
 759                 if (S_ISDIR(source_st.st_mode))
 760                         r = mkdir_label(where, 0755);
 761                 else
 762                         r = touch(where);
 763                 if (r < 0)
 764                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 765         }
 766
 767         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 768         if (r < 0)
 769                 return r;
 770
 771         if (m->read_only) {
 772                 r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL);
 773                 if (r < 0)
 774                         return log_error_errno(r, "Read-only bind mount failed: %m");
 775         }
 776
 777         return 0;
 778 }
 779
 780 static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) {
 781
 782         const char *options;
 783         _cleanup_free_ char *buf = NULL, *where = NULL;
 784         int r;
 785
 786         assert(dest);
 787         assert(m);
 788
 789         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 790         if (r < 0)
 791                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 792         if (r == 0) { /* Doesn't exist yet? */
 793                 r = mkdir_p_label(where, 0755);
 794                 if (r < 0)
 795                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 796         }
 797
 798         r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
 799         if (r < 0)
 800                 return log_oom();
 801         options = r > 0 ? buf : m->options;
 802
 803         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 804 }
 805
 806 static char *joined_and_escaped_lower_dirs(char **lower) {
 807         _cleanup_strv_free_ char **sv = NULL;
 808
 809         sv = strv_copy(lower);
 810         if (!sv)
 811                 return NULL;
 812
 813         strv_reverse(sv);
 814
 815         if (!strv_shell_escape(sv, ",:"))
 816                 return NULL;
 817
 818         return strv_join(sv, ":");
 819 }
 820
 821 static int mount_overlay(const char *dest, CustomMount *m) {
 822         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 823         const char *options;
 824         int r;
 825
 826         assert(dest);
 827         assert(m);
 828
 829         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 830         if (r < 0)
 831                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 832         if (r == 0) { /* Doesn't exist yet? */
 833                 r = mkdir_label(where, 0755);
 834                 if (r < 0)
 835                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 836         }
 837
 838         (void) mkdir_p_label(m->source, 0755);
 839
 840         lower = joined_and_escaped_lower_dirs(m->lower);
 841         if (!lower)
 842                 return log_oom();
 843
 844         escaped_source = shell_escape(m->source, ",:");
 845         if (!escaped_source)
 846                 return log_oom();
 847
 848         if (m->read_only)
 849                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 850         else {
 851                 _cleanup_free_ char *escaped_work_dir = NULL;
 852
 853                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 854                 if (!escaped_work_dir)
 855                         return log_oom();
 856
 857                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 858         }
 859
 860         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 861 }
 862
 863 static int mount_inaccessible(const char *dest, CustomMount *m) {
 864         _cleanup_free_ char *where = NULL, *source = NULL;
 865         struct stat st;
 866         int r;
 867
 868         assert(dest);
 869         assert(m);
 870
 871         r = chase_symlinks_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st, NULL);
 872         if (r < 0) {
 873                 log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination);
 874                 return m->graceful ? 0 : r;
 875         }
 876
 877         r = mode_to_inaccessible_node(NULL, st.st_mode, &source);
 878         if (r < 0)
 879                 return m->graceful ? 0 : r;
 880
 881         r = mount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL);
 882         if (r < 0)
 883                 return m->graceful ? 0 : r;
 884
 885         r = mount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL);
 886         if (r < 0) {
 887                 (void) umount_verbose(where);
 888                 return m->graceful ? 0 : r;
 889         }
 890
 891         return 0;
 892 }
 893
 894 static int mount_arbitrary(const char *dest, CustomMount *m) {
 895         _cleanup_free_ char *where = NULL;
 896         int r;
 897
 898         assert(dest);
 899         assert(m);
 900
 901         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
 902         if (r < 0)
 903                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 904         if (r == 0) { /* Doesn't exist yet? */
 905                 r = mkdir_p_label(where, 0755);
 906                 if (r < 0)
 907                         return log_error_errno(r, "Creating mount point for mount %s failed: %m", where);
 908         }
 909
 910         return mount_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options);
 911 }
 912
 913 int mount_custom(
 914                 const char *dest,
 915                 CustomMount *mounts, size_t n,
 916                 uid_t uid_shift,
 917                 const char *selinux_apifs_context,
 918                 MountSettingsMask mount_settings) {
 919
 920         size_t i;
 921         int r;
 922
 923         assert(dest);
 924
 925         for (i = 0; i < n; i++) {
 926                 CustomMount *m = mounts + i;
 927
 928                 if (FLAGS_SET(mount_settings, MOUNT_IN_USERNS) != m->in_userns)
 929                         continue;
 930
 931                 if (FLAGS_SET(mount_settings, MOUNT_ROOT_ONLY) && !path_equal(m->destination, "/"))
 932                         continue;
 933
 934                 if (FLAGS_SET(mount_settings, MOUNT_NON_ROOT_ONLY) && path_equal(m->destination, "/"))
 935                         continue;
 936
 937                 switch (m->type) {
 938
 939                 case CUSTOM_MOUNT_BIND:
 940                         r = mount_bind(dest, m);
 941                         break;
 942
 943                 case CUSTOM_MOUNT_TMPFS:
 944                         r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context);
 945                         break;
 946
 947                 case CUSTOM_MOUNT_OVERLAY:
 948                         r = mount_overlay(dest, m);
 949                         break;
 950
 951                 case CUSTOM_MOUNT_INACCESSIBLE:
 952                         r = mount_inaccessible(dest, m);
 953                         break;
 954
 955                 case CUSTOM_MOUNT_ARBITRARY:
 956                         r = mount_arbitrary(dest, m);
 957                         break;
 958
 959                 default:
 960                         assert_not_reached("Unknown custom mount type");
 961                 }
 962
 963                 if (r < 0)
 964                         return r;
 965         }
 966
 967         return 0;
 968 }
 969
 970 bool has_custom_root_mount(const CustomMount *mounts, size_t n) {
 971         size_t i;
 972
 973         for (i = 0; i < n; i++) {
 974                 const CustomMount *m = mounts + i;
 975
 976                 if (path_equal(m->destination, "/"))
 977                         return true;
 978         }
 979
 980         return false;
 981 }
 982
 983 static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
 984
 985         _cleanup_free_ char *buf = NULL;
 986         const char *p, *options;
 987         int r;
 988
 989         assert(directory);
 990
 991         /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
 992
 993         r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
 994         if (r < 0)
 995                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
 996
 997         p = prefix_roota(directory, "/var");
 998         r = mkdir(p, 0755);
 999         if (r < 0 && errno != EEXIST)
1000                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1001
1002         options = "mode=755" TMPFS_LIMITS_VOLATILE_STATE;
1003         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1004         if (r < 0)
1005                 return log_oom();
1006         if (r > 0)
1007                 options = buf;
1008
1009         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1010 }
1011
1012 static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1013
1014         bool tmpfs_mounted = false, bind_mounted = false;
1015         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1016         _cleanup_free_ char *buf = NULL, *bindir = NULL;
1017         const char *f, *t, *options;
1018         struct stat st;
1019         int r;
1020
1021         assert(directory);
1022
1023         /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
1024          * that read-only. Before we start setting this up let's validate if the image has the /usr merge
1025          * implemented, and let's output a friendly log message if it hasn't. */
1026
1027         bindir = path_join(directory, "/bin");
1028         if (!bindir)
1029                 return log_oom();
1030         if (lstat(bindir, &st) < 0) {
1031                 if (errno != ENOENT)
1032                         return log_error_errno(errno, "Failed to stat /bin directory below image: %m");
1033
1034                 /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
1035                  * rest. */
1036         } else if (S_ISDIR(st.st_mode))
1037                 return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
1038                                        "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
1039                                        "Please work with your distribution and help them adopt the merged /usr scheme.");
1040         else if (!S_ISLNK(st.st_mode))
1041                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1042                                        "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).");
1043
1044         if (!mkdtemp(template))
1045                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1046
1047         options = "mode=755" TMPFS_LIMITS_ROOTFS;
1048         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1049         if (r < 0)
1050                 goto fail;
1051         if (r > 0)
1052                 options = buf;
1053
1054         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1055         if (r < 0)
1056                 goto fail;
1057
1058         tmpfs_mounted = true;
1059
1060         f = prefix_roota(directory, "/usr");
1061         t = prefix_roota(template, "/usr");
1062
1063         r = mkdir(t, 0755);
1064         if (r < 0 && errno != EEXIST) {
1065                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1066                 goto fail;
1067         }
1068
1069         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1070         if (r < 0)
1071                 goto fail;
1072
1073         bind_mounted = true;
1074
1075         r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL);
1076         if (r < 0) {
1077                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1078                 goto fail;
1079         }
1080
1081         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1082         if (r < 0)
1083                 goto fail;
1084
1085         (void) rmdir(template);
1086
1087         return 0;
1088
1089 fail:
1090         if (bind_mounted)
1091                 (void) umount_verbose(t);
1092
1093         if (tmpfs_mounted)
1094                 (void) umount_verbose(template);
1095         (void) rmdir(template);
1096         return r;
1097 }
1098
1099 static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1100
1101         _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL;
1102         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1103         const char *upper, *work, *options;
1104         bool tmpfs_mounted = false;
1105         int r;
1106
1107         assert(directory);
1108
1109         /* --volatile=overlay means we mount an overlayfs to the root dir. */
1110
1111         if (!mkdtemp(template))
1112                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1113
1114         options = "mode=755" TMPFS_LIMITS_ROOTFS;
1115         r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1116         if (r < 0)
1117                 goto finish;
1118         if (r > 0)
1119                 options = buf;
1120
1121         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1122         if (r < 0)
1123                 goto finish;
1124
1125         tmpfs_mounted = true;
1126
1127         upper = strjoina(template, "/upper");
1128         work = strjoina(template, "/work");
1129
1130         if (mkdir(upper, 0755) < 0) {
1131                 r = log_error_errno(errno, "Failed to create %s: %m", upper);
1132                 goto finish;
1133         }
1134         if (mkdir(work, 0755) < 0) {
1135                 r = log_error_errno(errno, "Failed to create %s: %m", work);
1136                 goto finish;
1137         }
1138
1139         /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
1140          * that the kernel allows us to do that without going through some mount point rearrangements. */
1141
1142         escaped_directory = shell_escape(directory, ",:");
1143         escaped_upper = shell_escape(upper, ",:");
1144         escaped_work = shell_escape(work, ",:");
1145         if (!escaped_directory || !escaped_upper || !escaped_work) {
1146                 r = -ENOMEM;
1147                 goto finish;
1148         }
1149
1150         options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work);
1151         r = mount_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options);
1152
1153 finish:
1154         if (tmpfs_mounted)
1155                 (void) umount_verbose(template);
1156
1157         (void) rmdir(template);
1158         return r;
1159 }
1160
1161 int setup_volatile_mode(
1162                 const char *directory,
1163                 VolatileMode mode,
1164                 uid_t uid_shift,
1165                 const char *selinux_apifs_context) {
1166
1167         switch (mode) {
1168
1169         case VOLATILE_YES:
1170                 return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
1171
1172         case VOLATILE_STATE:
1173                 return setup_volatile_state(directory, uid_shift, selinux_apifs_context);
1174
1175         case VOLATILE_OVERLAY:
1176                 return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
1177
1178         default:
1179                 return 0;
1180         }
1181 }
1182
1183 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1184 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1185         _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1186         const char *p = s;
1187         int r;
1188
1189         assert(pivot_root_new);
1190         assert(pivot_root_old);
1191
1192         r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1193         if (r < 0)
1194                 return r;
1195         if (r == 0)
1196                 return -EINVAL;
1197
1198         if (isempty(p))
1199                 root_old = NULL;
1200         else {
1201                 root_old = strdup(p);
1202                 if (!root_old)
1203                         return -ENOMEM;
1204         }
1205
1206         if (!path_is_absolute(root_new))
1207                 return -EINVAL;
1208         if (root_old && !path_is_absolute(root_old))
1209                 return -EINVAL;
1210
1211         free_and_replace(*pivot_root_new, root_new);
1212         free_and_replace(*pivot_root_old, root_old);
1213
1214         return 0;
1215 }
1216
1217 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1218         _cleanup_free_ char *directory_pivot_root_new = NULL;
1219         _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1220         char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
1221         bool remove_pivot_tmp = false;
1222         int r;
1223
1224         assert(directory);
1225
1226         if (!pivot_root_new)
1227                 return 0;
1228
1229         /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1230          * If pivot_root_old is NULL, the existing / disappears.
1231          * This requires a temporary directory, pivot_tmp, which is
1232          * not a child of either.
1233          *
1234          * This is typically used for OSTree-style containers, where
1235          * the root partition contains several sysroots which could be
1236          * run. Normally, one would be chosen by the bootloader and
1237          * pivoted to / by initramfs.
1238          *
1239          * For example, for an OSTree deployment, pivot_root_new
1240          * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1241          * code doesn’t do the /var mount which OSTree expects: use
1242          * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1243          *
1244          * So in the OSTree case, we’ll end up with something like:
1245          *  - directory = /tmp/nspawn-root-123456
1246          *  - pivot_root_new = /ostree/deploy/os/deploy/123abc
1247          *  - pivot_root_old = /sysroot
1248          *  - directory_pivot_root_new =
1249          *       /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1250          *  - pivot_tmp = /tmp/nspawn-pivot-123456
1251          *  - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1252          *
1253          * Requires all file systems at directory and below to be mounted
1254          * MS_PRIVATE or MS_SLAVE so they can be moved.
1255          */
1256         directory_pivot_root_new = path_join(directory, pivot_root_new);
1257         if (!directory_pivot_root_new)
1258                 return log_oom();
1259
1260         /* Remount directory_pivot_root_new to make it movable. */
1261         r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1262         if (r < 0)
1263                 goto done;
1264
1265         if (pivot_root_old) {
1266                 if (!mkdtemp(pivot_tmp)) {
1267                         r = log_error_errno(errno, "Failed to create temporary directory: %m");
1268                         goto done;
1269                 }
1270
1271                 remove_pivot_tmp = true;
1272                 pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old);
1273                 if (!pivot_tmp_pivot_root_old) {
1274                         r = log_oom();
1275                         goto done;
1276                 }
1277
1278                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1279                 if (r < 0)
1280                         goto done;
1281
1282                 r = mount_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1283                 if (r < 0)
1284                         goto done;
1285
1286                 r = mount_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1287                 if (r < 0)
1288                         goto done;
1289         } else {
1290                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1291                 if (r < 0)
1292                         goto done;
1293         }
1294
1295 done:
1296         if (remove_pivot_tmp)
1297                 (void) rmdir(pivot_tmp);
1298
1299         return r;
1300 }