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                 if (m->rm_rf_tmpdir) {
  79                         (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL);
  80                         free(m->rm_rf_tmpdir);
  81                 }
  82
  83                 strv_free(m->lower);
  84         }
  85
  86         free(l);
  87 }
  88
  89 static int custom_mount_compare(const void *a, const void *b) {
  90         const CustomMount *x = a, *y = b;
  91         int r;
  92
  93         r = path_compare(x->destination, y->destination);
  94         if (r != 0)
  95                 return r;
  96
  97         if (x->type < y->type)
  98                 return -1;
  99         if (x->type > y->type)
 100                 return 1;
 101
 102         return 0;
 103 }
 104
 105 static bool source_path_is_valid(const char *p) {
 106         assert(p);
 107
 108         if (*p == '+')
 109                 p++;
 110
 111         return path_is_absolute(p);
 112 }
 113
 114 static char *resolve_source_path(const char *dest, const char *source) {
 115
 116         if (!source)
 117                 return NULL;
 118
 119         if (source[0] == '+')
 120                 return prefix_root(dest, source + 1);
 121
 122         return strdup(source);
 123 }
 124
 125 int custom_mount_prepare_all(const char *dest, CustomMount *l, unsigned n) {
 126         unsigned i;
 127         int r;
 128
 129         /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
 130          * parent process, so that we know the temporary directories to remove on exit before we fork off the
 131          * children. */
 132
 133         assert(l || n == 0);
 134
 135         /* Order the custom mounts, and make sure we have a working directory */
 136         qsort_safe(l, n, sizeof(CustomMount), custom_mount_compare);
 137
 138         for (i = 0; i < n; i++) {
 139                 CustomMount *m = l + i;
 140
 141                 if (m->source) {
 142                         char *s;
 143
 144                         s = resolve_source_path(dest, m->source);
 145                         if (!s)
 146                                 return log_oom();
 147
 148                         free(m->source);
 149                         m->source = s;
 150                 } else {
 151                         /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
 152
 153                         m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX");
 154                         if (!m->rm_rf_tmpdir)
 155                                 return log_oom();
 156
 157                         if (!mkdtemp(m->rm_rf_tmpdir)) {
 158                                 m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir);
 159                                 return log_error_errno(errno, "Failed to acquire temporary directory: %m");
 160                         }
 161
 162                         m->source = strjoin(m->rm_rf_tmpdir, "/src");
 163                         if (!m->source)
 164                                 return log_oom();
 165
 166                         if (mkdir(m->source, 0755) < 0)
 167                                 return log_error_errno(errno, "Failed to create %s: %m", m->source);
 168                 }
 169
 170                 if (m->type == CUSTOM_MOUNT_OVERLAY) {
 171                         char **j;
 172
 173                         STRV_FOREACH(j, m->lower) {
 174                                 char *s;
 175
 176                                 s = resolve_source_path(dest, *j);
 177                                 if (!s)
 178                                         return log_oom();
 179
 180                                 free(*j);
 181                                 *j = s;
 182                         }
 183
 184                         if (m->work_dir) {
 185                                 char *s;
 186
 187                                 s = resolve_source_path(dest, m->work_dir);
 188                                 if (!s)
 189                                         return log_oom();
 190
 191                                 free(m->work_dir);
 192                                 m->work_dir = s;
 193                         } else {
 194                                 assert(m->source);
 195
 196                                 r = tempfn_random(m->source, NULL, &m->work_dir);
 197                                 if (r < 0)
 198                                         return log_error_errno(r, "Failed to acquire working directory: %m");
 199                         }
 200
 201                         (void) mkdir_label(m->work_dir, 0700);
 202                 }
 203         }
 204
 205         return 0;
 206 }
 207
 208 int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 209         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 210         const char *p = s;
 211         CustomMount *m;
 212         int r;
 213
 214         assert(l);
 215         assert(n);
 216
 217         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 218         if (r < 0)
 219                 return r;
 220         if (r == 0)
 221                 return -EINVAL;
 222         if (r == 1) {
 223                 destination = strdup(source[0] == '+' ? source+1 : source);
 224                 if (!destination)
 225                         return -ENOMEM;
 226         }
 227         if (r == 2 && !isempty(p)) {
 228                 opts = strdup(p);
 229                 if (!opts)
 230                         return -ENOMEM;
 231         }
 232
 233         if (isempty(source))
 234                 source = NULL;
 235         else if (!source_path_is_valid(source))
 236                 return -EINVAL;
 237
 238         if (!path_is_absolute(destination))
 239                 return -EINVAL;
 240
 241         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 242         if (!m)
 243                 return -ENOMEM;
 244
 245         m->source = source;
 246         m->destination = destination;
 247         m->read_only = read_only;
 248         m->options = opts;
 249
 250         source = destination = opts = NULL;
 251         return 0;
 252 }
 253
 254 int tmpfs_mount_parse(CustomMount **l, unsigned *n, const char *s) {
 255         _cleanup_free_ char *path = NULL, *opts = NULL;
 256         const char *p = s;
 257         CustomMount *m;
 258         int r;
 259
 260         assert(l);
 261         assert(n);
 262         assert(s);
 263
 264         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 265         if (r < 0)
 266                 return r;
 267         if (r == 0)
 268                 return -EINVAL;
 269
 270         if (isempty(p))
 271                 opts = strdup("mode=0755");
 272         else
 273                 opts = strdup(p);
 274         if (!opts)
 275                 return -ENOMEM;
 276
 277         if (!path_is_absolute(path))
 278                 return -EINVAL;
 279
 280         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 281         if (!m)
 282                 return -ENOMEM;
 283
 284         m->destination = path;
 285         m->options = opts;
 286
 287         path = opts = NULL;
 288         return 0;
 289 }
 290
 291 int overlay_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 292         _cleanup_free_ char *upper = NULL, *destination = NULL;
 293         _cleanup_strv_free_ char **lower = NULL;
 294         CustomMount *m;
 295         int k;
 296
 297         k = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 298         if (k < 0)
 299                 return k;
 300         if (k < 2)
 301                 return -EADDRNOTAVAIL;
 302         if (k == 2) {
 303                 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
 304                  * we'll also define the destination mount point the same as the upper. */
 305
 306                 if (!source_path_is_valid(lower[0]) ||
 307                     !source_path_is_valid(lower[1]))
 308                         return -EINVAL;
 309
 310                 upper = lower[1];
 311                 lower[1] = NULL;
 312
 313                 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
 314                 if (!destination)
 315                         return -ENOMEM;
 316         } else {
 317                 char **i;
 318
 319                 /* If more than two parameters are specified, the last one is the destination, the second to last one
 320                  * the "upper", and all before that the "lower" directories. */
 321
 322                 destination = lower[k - 1];
 323                 upper = lower[k - 2];
 324                 lower[k - 2] = NULL;
 325
 326                 STRV_FOREACH(i, lower)
 327                         if (!source_path_is_valid(*i))
 328                                 return -EINVAL;
 329
 330                 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
 331                  * in /var/tmp */
 332                 if (isempty(upper))
 333                         upper = NULL;
 334                 else if (!source_path_is_valid(upper))
 335                         return -EINVAL;
 336
 337                 if (!path_is_absolute(destination))
 338                         return -EINVAL;
 339         }
 340
 341         m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
 342         if (!m)
 343                 return -ENOMEM;
 344
 345         m->destination = destination;
 346         m->source = upper;
 347         m->lower = lower;
 348         m->read_only = read_only;
 349
 350         upper = destination = NULL;
 351         lower = NULL;
 352
 353         return 0;
 354 }
 355
 356 static int tmpfs_patch_options(
 357                 const char *options,
 358                 bool userns,
 359                 uid_t uid_shift, uid_t uid_range,
 360                 bool patch_ids,
 361                 const char *selinux_apifs_context,
 362                 char **ret) {
 363
 364         char *buf = NULL;
 365
 366         if ((userns && uid_shift != 0) || patch_ids) {
 367                 assert(uid_shift != UID_INVALID);
 368
 369                 if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 370                              options ?: "", options ? "," : "",
 371                              uid_shift, uid_shift) < 0)
 372                         return -ENOMEM;
 373
 374                 options = buf;
 375         }
 376
 377 #if HAVE_SELINUX
 378         if (selinux_apifs_context) {
 379                 char *t;
 380
 381                 t = strjoin(options ?: "", options ? "," : "",
 382                             "context=\"", selinux_apifs_context, "\"");
 383                 free(buf);
 384                 if (!t)
 385                         return -ENOMEM;
 386
 387                 buf = t;
 388         }
 389 #endif
 390
 391         if (!buf && options) {
 392                 buf = strdup(options);
 393                 if (!buf)
 394                         return -ENOMEM;
 395         }
 396         *ret = buf;
 397
 398         return !!buf;
 399 }
 400
 401 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
 402         const char *full, *top, *x;
 403         int r;
 404         unsigned long extra_flags = 0;
 405
 406         top = prefix_roota(dest, "/sys");
 407         r = path_check_fstype(top, SYSFS_MAGIC);
 408         if (r < 0)
 409                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 410         /* /sys might already be mounted as sysfs by the outer child in the
 411          * !netns case. In this case, it's all good. Don't touch it because we
 412          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 413          */
 414         if (r > 0)
 415                 return 0;
 416
 417         full = prefix_roota(top, "/full");
 418
 419         (void) mkdir(full, 0755);
 420
 421         if (mount_settings & MOUNT_APPLY_APIVFS_RO)
 422                 extra_flags |= MS_RDONLY;
 423
 424         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 425                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 426         if (r < 0)
 427                 return r;
 428
 429         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 430                 _cleanup_free_ char *from = NULL, *to = NULL;
 431
 432                 from = prefix_root(full, x);
 433                 if (!from)
 434                         return log_oom();
 435
 436                 to = prefix_root(top, x);
 437                 if (!to)
 438                         return log_oom();
 439
 440                 (void) mkdir(to, 0755);
 441
 442                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 443                 if (r < 0)
 444                         return r;
 445
 446                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 447                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 448                 if (r < 0)
 449                         return r;
 450         }
 451
 452         r = umount_verbose(full);
 453         if (r < 0)
 454                 return r;
 455
 456         if (rmdir(full) < 0)
 457                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 458
 459         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 460          * remount /sys read-only.
 461          */
 462         if (cg_ns_supported()) {
 463                 x = prefix_roota(top, "/fs/cgroup");
 464                 (void) mkdir_p(x, 0755);
 465         }
 466
 467         return mount_verbose(LOG_ERR, NULL, top, NULL,
 468                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 469 }
 470
 471 static int mkdir_userns(const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 472         int r;
 473
 474         assert(path);
 475
 476         r = mkdir(path, mode);
 477         if (r < 0 && errno != EEXIST)
 478                 return -errno;
 479
 480         if ((mask & MOUNT_USE_USERNS) == 0)
 481                 return 0;
 482
 483         if (mask & MOUNT_IN_USERNS)
 484                 return 0;
 485
 486         r = lchown(path, uid_shift, uid_shift);
 487         if (r < 0)
 488                 return -errno;
 489
 490         return 0;
 491 }
 492
 493 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 494         const char *p, *e;
 495         int r;
 496
 497         assert(path);
 498
 499         if (prefix && !path_startswith(path, prefix))
 500                 return -ENOTDIR;
 501
 502         /* create every parent directory in the path, except the last component */
 503         p = path + strspn(path, "/");
 504         for (;;) {
 505                 char t[strlen(path) + 1];
 506
 507                 e = p + strcspn(p, "/");
 508                 p = e + strspn(e, "/");
 509
 510                 /* Is this the last component? If so, then we're done */
 511                 if (*p == 0)
 512                         break;
 513
 514                 memcpy(t, path, e - path);
 515                 t[e-path] = 0;
 516
 517                 if (prefix && path_startswith(prefix, t))
 518                         continue;
 519
 520                 r = mkdir_userns(t, mode, mask, uid_shift);
 521                 if (r < 0)
 522                         return r;
 523         }
 524
 525         return mkdir_userns(path, mode, mask, uid_shift);
 526 }
 527
 528 int mount_all(const char *dest,
 529               MountSettingsMask mount_settings,
 530               uid_t uid_shift, uid_t uid_range,
 531               const char *selinux_apifs_context) {
 532
 533         typedef struct MountPoint {
 534                 const char *what;
 535                 const char *where;
 536                 const char *type;
 537                 const char *options;
 538                 unsigned long flags;
 539                 MountSettingsMask mount_settings;
 540         } MountPoint;
 541
 542         static const MountPoint mount_table[] = {
 543                 /* inner child mounts */
 544                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_IN_USERNS },
 545                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 546                 { "/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) */
 547                 { 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 */
 548                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                               MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 549                 { 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 */
 550
 551                 /* outer child mounts */
 552                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                                            MOUNT_FATAL },
 553                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS },
 554                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO },    /* skipped if above was mounted */
 555                 { "sysfs",               "/sys",                "sysfs", NULL,                  MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL },                          /* skipped if above was mounted */
 556
 557                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  MOUNT_FATAL },
 558                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 559                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 560 #if HAVE_SELINUX
 561                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   0 },  /* Bind mount first */
 562                 { NULL,                  "/sys/fs/selinux",     NULL,     NULL,       MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, 0 },  /* Then, make it r/o */
 563 #endif
 564         };
 565
 566         unsigned k;
 567         int r;
 568         bool use_userns = (mount_settings & MOUNT_USE_USERNS);
 569         bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS);
 570         bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO);
 571         bool in_userns = (mount_settings & MOUNT_IN_USERNS);
 572
 573         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 574                 _cleanup_free_ char *where = NULL, *options = NULL;
 575                 const char *o;
 576                 bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL);
 577
 578                 if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS))
 579                         continue;
 580
 581                 if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS))
 582                         continue;
 583
 584                 if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO))
 585                         continue;
 586
 587                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where);
 588                 if (r < 0)
 589                         return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where);
 590
 591                 r = path_is_mount_point(where, NULL, 0);
 592                 if (r < 0 && r != -ENOENT)
 593                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 594
 595                 /* Skip this entry if it is not a remount. */
 596                 if (mount_table[k].what && r > 0)
 597                         continue;
 598
 599                 r = mkdir_userns_p(dest, where, 0755, mount_settings, uid_shift);
 600                 if (r < 0 && r != -EEXIST) {
 601                         if (fatal)
 602                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 603
 604                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 605                         continue;
 606                 }
 607
 608                 o = mount_table[k].options;
 609                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 610                         if (in_userns)
 611                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 612                         else
 613                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 614                         if (r < 0)
 615                                 return log_oom();
 616                         if (r > 0)
 617                                 o = options;
 618                 }
 619
 620                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 621                                   mount_table[k].what,
 622                                   where,
 623                                   mount_table[k].type,
 624                                   mount_table[k].flags,
 625                                   o);
 626                 if (r < 0 && fatal)
 627                         return r;
 628         }
 629
 630         return 0;
 631 }
 632
 633 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 634         const char *p = options;
 635         unsigned long flags = *mount_flags;
 636         char *opts = NULL;
 637         int r;
 638
 639         assert(options);
 640
 641         for (;;) {
 642                 _cleanup_free_ char *word = NULL;
 643
 644                 r = extract_first_word(&p, &word, ",", 0);
 645                 if (r < 0)
 646                         return log_error_errno(r, "Failed to extract mount option: %m");
 647                 if (r == 0)
 648                         break;
 649
 650                 if (streq(word, "rbind"))
 651                         flags |= MS_REC;
 652                 else if (streq(word, "norbind"))
 653                         flags &= ~MS_REC;
 654                 else {
 655                         log_error("Invalid bind mount option: %s", word);
 656                         return -EINVAL;
 657                 }
 658         }
 659
 660         *mount_flags = flags;
 661         /* in the future mount_opts will hold string options for mount(2) */
 662         *mount_opts = opts;
 663
 664         return 0;
 665 }
 666
 667 static int mount_bind(const char *dest, CustomMount *m) {
 668
 669         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 670         unsigned long mount_flags = MS_BIND | MS_REC;
 671         struct stat source_st, dest_st;
 672         int r;
 673
 674         assert(dest);
 675         assert(m);
 676
 677         if (m->options) {
 678                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 679                 if (r < 0)
 680                         return r;
 681         }
 682
 683         if (stat(m->source, &source_st) < 0)
 684                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 685
 686         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where);
 687         if (r < 0)
 688                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 689         if (r > 0) { /* Path exists already? */
 690
 691                 if (stat(where, &dest_st) < 0)
 692                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 693
 694                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 695                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 696                         return -EINVAL;
 697                 }
 698
 699                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 700                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 701                         return -EINVAL;
 702                 }
 703
 704         } else { /* Path doesn't exist yet? */
 705                 r = mkdir_parents_label(where, 0755);
 706                 if (r < 0)
 707                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 708
 709                 /* Create the mount point. Any non-directory file can be
 710                 * mounted on any non-directory file (regular, fifo, socket,
 711                 * char, block).
 712                 */
 713                 if (S_ISDIR(source_st.st_mode))
 714                         r = mkdir_label(where, 0755);
 715                 else
 716                         r = touch(where);
 717                 if (r < 0)
 718                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 719
 720         }
 721
 722         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 723         if (r < 0)
 724                 return r;
 725
 726         if (m->read_only) {
 727                 r = bind_remount_recursive(where, true, NULL);
 728                 if (r < 0)
 729                         return log_error_errno(r, "Read-only bind mount failed: %m");
 730         }
 731
 732         return 0;
 733 }
 734
 735 static int mount_tmpfs(
 736                 const char *dest,
 737                 CustomMount *m,
 738                 bool userns, uid_t uid_shift, uid_t uid_range,
 739                 const char *selinux_apifs_context) {
 740
 741         const char *options;
 742         _cleanup_free_ char *buf = NULL, *where = NULL;
 743         int r;
 744
 745         assert(dest);
 746         assert(m);
 747
 748         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where);
 749         if (r < 0)
 750                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 751         if (r == 0) { /* Doesn't exist yet? */
 752                 r = mkdir_p_label(where, 0755);
 753                 if (r < 0)
 754                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 755         }
 756
 757         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 758         if (r < 0)
 759                 return log_oom();
 760         options = r > 0 ? buf : m->options;
 761
 762         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 763 }
 764
 765 static char *joined_and_escaped_lower_dirs(char **lower) {
 766         _cleanup_strv_free_ char **sv = NULL;
 767
 768         sv = strv_copy(lower);
 769         if (!sv)
 770                 return NULL;
 771
 772         strv_reverse(sv);
 773
 774         if (!strv_shell_escape(sv, ",:"))
 775                 return NULL;
 776
 777         return strv_join(sv, ":");
 778 }
 779
 780 static int mount_overlay(const char *dest, CustomMount *m) {
 781
 782         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 783         const char *options;
 784         int r;
 785
 786         assert(dest);
 787         assert(m);
 788
 789         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where);
 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_label(where, 0755);
 794                 if (r < 0)
 795                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 796         }
 797
 798         (void) mkdir_p_label(m->source, 0755);
 799
 800         lower = joined_and_escaped_lower_dirs(m->lower);
 801         if (!lower)
 802                 return log_oom();
 803
 804         escaped_source = shell_escape(m->source, ",:");
 805         if (!escaped_source)
 806                 return log_oom();
 807
 808         if (m->read_only)
 809                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 810         else {
 811                 _cleanup_free_ char *escaped_work_dir = NULL;
 812
 813                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 814                 if (!escaped_work_dir)
 815                         return log_oom();
 816
 817                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 818         }
 819
 820         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 821 }
 822
 823 int mount_custom(
 824                 const char *dest,
 825                 CustomMount *mounts, unsigned n,
 826                 bool userns, uid_t uid_shift, uid_t uid_range,
 827                 const char *selinux_apifs_context) {
 828
 829         unsigned i;
 830         int r;
 831
 832         assert(dest);
 833
 834         for (i = 0; i < n; i++) {
 835                 CustomMount *m = mounts + i;
 836
 837                 switch (m->type) {
 838
 839                 case CUSTOM_MOUNT_BIND:
 840                         r = mount_bind(dest, m);
 841                         break;
 842
 843                 case CUSTOM_MOUNT_TMPFS:
 844                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 845                         break;
 846
 847                 case CUSTOM_MOUNT_OVERLAY:
 848                         r = mount_overlay(dest, m);
 849                         break;
 850
 851                 default:
 852                         assert_not_reached("Unknown custom mount type");
 853                 }
 854
 855                 if (r < 0)
 856                         return r;
 857         }
 858
 859         return 0;
 860 }
 861
 862 /* Retrieve existing subsystems. This function is called in a new cgroup
 863  * namespace.
 864  */
 865 static int get_controllers(Set *subsystems) {
 866         _cleanup_fclose_ FILE *f = NULL;
 867         char line[LINE_MAX];
 868
 869         assert(subsystems);
 870
 871         f = fopen("/proc/self/cgroup", "re");
 872         if (!f)
 873                 return errno == ENOENT ? -ESRCH : -errno;
 874
 875         FOREACH_LINE(line, f, return -errno) {
 876                 int r;
 877                 char *e, *l, *p;
 878
 879                 l = strchr(line, ':');
 880                 if (!l)
 881                         continue;
 882
 883                 l++;
 884                 e = strchr(l, ':');
 885                 if (!e)
 886                         continue;
 887
 888                 *e = 0;
 889
 890                 if (STR_IN_SET(l, "", "name=systemd", "name=unified"))
 891                         continue;
 892
 893                 p = strdup(l);
 894                 if (!p)
 895                         return -ENOMEM;
 896
 897                 r = set_consume(subsystems, p);
 898                 if (r < 0)
 899                         return r;
 900         }
 901
 902         return 0;
 903 }
 904
 905 static int mount_legacy_cgroup_hierarchy(
 906                 const char *dest,
 907                 const char *controller,
 908                 const char *hierarchy,
 909                 bool read_only) {
 910
 911         const char *to, *fstype, *opts;
 912         int r;
 913
 914         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 915
 916         r = path_is_mount_point(to, dest, 0);
 917         if (r < 0 && r != -ENOENT)
 918                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 919         if (r > 0)
 920                 return 0;
 921
 922         mkdir_p(to, 0755);
 923
 924         /* The superblock mount options of the mount point need to be
 925          * identical to the hosts', and hence writable... */
 926         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER_HYBRID)) {
 927                 fstype = "cgroup2";
 928                 opts = NULL;
 929         } else if (streq(controller, SYSTEMD_CGROUP_CONTROLLER_LEGACY)) {
 930                 fstype = "cgroup";
 931                 opts = "none,name=systemd,xattr";
 932         } else {
 933                 fstype = "cgroup";
 934                 opts = controller;
 935         }
 936
 937         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 938         if (r < 0)
 939                 return r;
 940
 941         /* ... hence let's only make the bind mount read-only, not the superblock. */
 942         if (read_only) {
 943                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 944                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 945                 if (r < 0)
 946                         return r;
 947         }
 948
 949         return 1;
 950 }
 951
 952 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 953 static int mount_legacy_cgns_supported(
 954                 const char *dest,
 955                 CGroupUnified unified_requested,
 956                 bool userns,
 957                 uid_t uid_shift,
 958                 uid_t uid_range,
 959                 const char *selinux_apifs_context) {
 960
 961         _cleanup_set_free_free_ Set *controllers = NULL;
 962         const char *cgroup_root = "/sys/fs/cgroup", *c;
 963         int r;
 964
 965         (void) mkdir_p(cgroup_root, 0755);
 966
 967         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 968         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 969         if (r < 0)
 970                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 971         if (r == 0) {
 972                 _cleanup_free_ char *options = NULL;
 973
 974                 /* When cgroup namespaces are enabled and user namespaces are
 975                  * used then the mount of the cgroupfs is done *inside* the new
 976                  * user namespace. We're root in the new user namespace and the
 977                  * kernel will happily translate our uid/gid to the correct
 978                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 979                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 980                  */
 981                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 982                 if (r < 0)
 983                         return log_oom();
 984
 985                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 986                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 987                 if (r < 0)
 988                         return r;
 989         }
 990
 991         r = cg_all_unified();
 992         if (r < 0)
 993                 return r;
 994         if (r > 0)
 995                 goto skip_controllers;
 996
 997         controllers = set_new(&string_hash_ops);
 998         if (!controllers)
 999                 return log_oom();
1000
1001         r = get_controllers(controllers);
1002         if (r < 0)
1003                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1004
1005         for (;;) {
1006                 _cleanup_free_ const char *controller = NULL;
1007
1008                 controller = set_steal_first(controllers);
1009                 if (!controller)
1010                         break;
1011
1012                 r = mount_legacy_cgroup_hierarchy("", controller, controller, !userns);
1013                 if (r < 0)
1014                         return r;
1015
1016                 /* When multiple hierarchies are co-mounted, make their
1017                  * constituting individual hierarchies a symlink to the
1018                  * co-mount.
1019                  */
1020                 c = controller;
1021                 for (;;) {
1022                         _cleanup_free_ char *target = NULL, *tok = NULL;
1023
1024                         r = extract_first_word(&c, &tok, ",", 0);
1025                         if (r < 0)
1026                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
1027                         if (r == 0)
1028                                 break;
1029
1030                         target = prefix_root("/sys/fs/cgroup", tok);
1031                         if (!target)
1032                                 return log_oom();
1033
1034                         if (streq(controller, tok))
1035                                 break;
1036
1037                         r = symlink_idempotent(controller, target);
1038                         if (r == -EINVAL)
1039                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1040                         if (r < 0)
1041                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1042                 }
1043         }
1044
1045 skip_controllers:
1046         if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
1047                 r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
1048                 if (r < 0)
1049                         return r;
1050         }
1051
1052         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
1053         if (r < 0)
1054                 return r;
1055
1056         if (!userns)
1057                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1058                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1059
1060         return 0;
1061 }
1062
1063 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
1064 static int mount_legacy_cgns_unsupported(
1065                 const char *dest,
1066                 CGroupUnified unified_requested,
1067                 bool userns,
1068                 uid_t uid_shift,
1069                 uid_t uid_range,
1070                 const char *selinux_apifs_context) {
1071
1072         _cleanup_set_free_free_ Set *controllers = NULL;
1073         const char *cgroup_root;
1074         int r;
1075
1076         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
1077
1078         (void) mkdir_p(cgroup_root, 0755);
1079
1080         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
1081         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
1082         if (r < 0)
1083                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
1084         if (r == 0) {
1085                 _cleanup_free_ char *options = NULL;
1086
1087                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
1088                 if (r < 0)
1089                         return log_oom();
1090
1091                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
1092                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
1093                 if (r < 0)
1094                         return r;
1095         }
1096
1097         r = cg_all_unified();
1098         if (r < 0)
1099                 return r;
1100         if (r > 0)
1101                 goto skip_controllers;
1102
1103         controllers = set_new(&string_hash_ops);
1104         if (!controllers)
1105                 return log_oom();
1106
1107         r = cg_kernel_controllers(controllers);
1108         if (r < 0)
1109                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1110
1111         for (;;) {
1112                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
1113
1114                 controller = set_steal_first(controllers);
1115                 if (!controller)
1116                         break;
1117
1118                 origin = prefix_root("/sys/fs/cgroup/", controller);
1119                 if (!origin)
1120                         return log_oom();
1121
1122                 r = readlink_malloc(origin, &combined);
1123                 if (r == -EINVAL) {
1124                         /* Not a symbolic link, but directly a single cgroup hierarchy */
1125
1126                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
1127                         if (r < 0)
1128                                 return r;
1129
1130                 } else if (r < 0)
1131                         return log_error_errno(r, "Failed to read link %s: %m", origin);
1132                 else {
1133                         _cleanup_free_ char *target = NULL;
1134
1135                         target = prefix_root(dest, origin);
1136                         if (!target)
1137                                 return log_oom();
1138
1139                         /* A symbolic link, a combination of controllers in one hierarchy */
1140
1141                         if (!filename_is_valid(combined)) {
1142                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
1143                                 continue;
1144                         }
1145
1146                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
1147                         if (r < 0)
1148                                 return r;
1149
1150                         r = symlink_idempotent(combined, target);
1151                         if (r == -EINVAL)
1152                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1153                         if (r < 0)
1154                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1155                 }
1156         }
1157
1158 skip_controllers:
1159         if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
1160                 r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
1161                 if (r < 0)
1162                         return r;
1163         }
1164
1165         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
1166         if (r < 0)
1167                 return r;
1168
1169         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1170                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1171 }
1172
1173 static int mount_unified_cgroups(const char *dest) {
1174         const char *p;
1175         int r;
1176
1177         assert(dest);
1178
1179         p = prefix_roota(dest, "/sys/fs/cgroup");
1180
1181         (void) mkdir_p(p, 0755);
1182
1183         r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW);
1184         if (r < 0)
1185                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
1186         if (r > 0) {
1187                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
1188                 if (access(p, F_OK) >= 0)
1189                         return 0;
1190                 if (errno != ENOENT)
1191                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
1192
1193                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
1194                 return -EINVAL;
1195         }
1196
1197         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1198 }
1199
1200 int mount_cgroups(
1201                 const char *dest,
1202                 CGroupUnified unified_requested,
1203                 bool userns,
1204                 uid_t uid_shift,
1205                 uid_t uid_range,
1206                 const char *selinux_apifs_context,
1207                 bool use_cgns) {
1208
1209         if (unified_requested >= CGROUP_UNIFIED_ALL)
1210                 return mount_unified_cgroups(dest);
1211         else if (use_cgns)
1212                 return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1213
1214         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1215 }
1216
1217 static int mount_systemd_cgroup_writable_one(const char *systemd_own, const char *systemd_root)
1218 {
1219         int r;
1220
1221         /* Make our own cgroup a (writable) bind mount */
1222         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1223         if (r < 0)
1224                 return r;
1225
1226         /* And then remount the systemd cgroup root read-only */
1227         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1228                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1229 }
1230
1231 int mount_systemd_cgroup_writable(
1232                 const char *dest,
1233                 CGroupUnified unified_requested) {
1234
1235         _cleanup_free_ char *own_cgroup_path = NULL;
1236         int r;
1237
1238         assert(dest);
1239
1240         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1241         if (r < 0)
1242                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1243
1244         /* If we are living in the top-level, then there's nothing to do... */
1245         if (path_equal(own_cgroup_path, "/"))
1246                 return 0;
1247
1248         if (unified_requested >= CGROUP_UNIFIED_ALL)
1249                 return mount_systemd_cgroup_writable_one(strjoina(dest, "/sys/fs/cgroup", own_cgroup_path),
1250                                                          prefix_roota(dest, "/sys/fs/cgroup"));
1251
1252         if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
1253                 r = mount_systemd_cgroup_writable_one(strjoina(dest, "/sys/fs/cgroup/unified", own_cgroup_path),
1254                                                       prefix_roota(dest, "/sys/fs/cgroup/unified"));
1255                 if (r < 0)
1256                         return r;
1257         }
1258
1259         return mount_systemd_cgroup_writable_one(strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path),
1260                                                  prefix_roota(dest, "/sys/fs/cgroup/systemd"));
1261 }
1262
1263 int setup_volatile_state(
1264                 const char *directory,
1265                 VolatileMode mode,
1266                 bool userns, uid_t uid_shift, uid_t uid_range,
1267                 const char *selinux_apifs_context) {
1268
1269         _cleanup_free_ char *buf = NULL;
1270         const char *p, *options;
1271         int r;
1272
1273         assert(directory);
1274
1275         if (mode != VOLATILE_STATE)
1276                 return 0;
1277
1278         /* --volatile=state means we simply overmount /var
1279            with a tmpfs, and the rest read-only. */
1280
1281         r = bind_remount_recursive(directory, true, NULL);
1282         if (r < 0)
1283                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1284
1285         p = prefix_roota(directory, "/var");
1286         r = mkdir(p, 0755);
1287         if (r < 0 && errno != EEXIST)
1288                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1289
1290         options = "mode=755";
1291         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1292         if (r < 0)
1293                 return log_oom();
1294         if (r > 0)
1295                 options = buf;
1296
1297         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1298 }
1299
1300 int setup_volatile(
1301                 const char *directory,
1302                 VolatileMode mode,
1303                 bool userns, uid_t uid_shift, uid_t uid_range,
1304                 const char *selinux_apifs_context) {
1305
1306         bool tmpfs_mounted = false, bind_mounted = false;
1307         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1308         _cleanup_free_ char *buf = NULL;
1309         const char *f, *t, *options;
1310         int r;
1311
1312         assert(directory);
1313
1314         if (mode != VOLATILE_YES)
1315                 return 0;
1316
1317         /* --volatile=yes means we mount a tmpfs to the root dir, and
1318            the original /usr to use inside it, and that read-only. */
1319
1320         if (!mkdtemp(template))
1321                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1322
1323         options = "mode=755";
1324         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1325         if (r < 0)
1326                 return log_oom();
1327         if (r > 0)
1328                 options = buf;
1329
1330         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1331         if (r < 0)
1332                 goto fail;
1333
1334         tmpfs_mounted = true;
1335
1336         f = prefix_roota(directory, "/usr");
1337         t = prefix_roota(template, "/usr");
1338
1339         r = mkdir(t, 0755);
1340         if (r < 0 && errno != EEXIST) {
1341                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1342                 goto fail;
1343         }
1344
1345         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1346         if (r < 0)
1347                 goto fail;
1348
1349         bind_mounted = true;
1350
1351         r = bind_remount_recursive(t, true, NULL);
1352         if (r < 0) {
1353                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1354                 goto fail;
1355         }
1356
1357         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1358         if (r < 0)
1359                 goto fail;
1360
1361         (void) rmdir(template);
1362
1363         return 0;
1364
1365 fail:
1366         if (bind_mounted)
1367                 (void) umount_verbose(t);
1368
1369         if (tmpfs_mounted)
1370                 (void) umount_verbose(template);
1371         (void) rmdir(template);
1372         return r;
1373 }
1374
1375 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1376 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1377         _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1378         const char *p = s;
1379         int r;
1380
1381         assert(pivot_root_new);
1382         assert(pivot_root_old);
1383
1384         r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1385         if (r < 0)
1386                 return r;
1387         if (r == 0)
1388                 return -EINVAL;
1389
1390         if (isempty(p))
1391                 root_old = NULL;
1392         else {
1393                 root_old = strdup(p);
1394                 if (!root_old)
1395                         return -ENOMEM;
1396         }
1397
1398         if (!path_is_absolute(root_new))
1399                 return -EINVAL;
1400         if (root_old && !path_is_absolute(root_old))
1401                 return -EINVAL;
1402
1403         free_and_replace(*pivot_root_new, root_new);
1404         free_and_replace(*pivot_root_old, root_old);
1405
1406         return 0;
1407 }
1408
1409 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1410         _cleanup_free_ char *directory_pivot_root_new = NULL;
1411         _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1412         char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
1413         bool remove_pivot_tmp = false;
1414         int r;
1415
1416         assert(directory);
1417
1418         if (!pivot_root_new)
1419                 return 0;
1420
1421         /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1422          * If pivot_root_old is NULL, the existing / disappears.
1423          * This requires a temporary directory, pivot_tmp, which is
1424          * not a child of either.
1425          *
1426          * This is typically used for OSTree-style containers, where
1427          * the root partition contains several sysroots which could be
1428          * run. Normally, one would be chosen by the bootloader and
1429          * pivoted to / by initramfs.
1430          *
1431          * For example, for an OSTree deployment, pivot_root_new
1432          * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1433          * code doesn’t do the /var mount which OSTree expects: use
1434          * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1435          *
1436          * So in the OSTree case, we’ll end up with something like:
1437          *  - directory = /tmp/nspawn-root-123456
1438          *  - pivot_root_new = /ostree/deploy/os/deploy/123abc
1439          *  - pivot_root_old = /sysroot
1440          *  - directory_pivot_root_new =
1441          *       /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1442          *  - pivot_tmp = /tmp/nspawn-pivot-123456
1443          *  - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1444          *
1445          * Requires all file systems at directory and below to be mounted
1446          * MS_PRIVATE or MS_SLAVE so they can be moved.
1447          */
1448         directory_pivot_root_new = prefix_root(directory, pivot_root_new);
1449
1450         /* Remount directory_pivot_root_new to make it movable. */
1451         r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1452         if (r < 0)
1453                 goto done;
1454
1455         if (pivot_root_old) {
1456                 if (!mkdtemp(pivot_tmp)) {
1457                         r = log_error_errno(errno, "Failed to create temporary directory: %m");
1458                         goto done;
1459                 }
1460
1461                 remove_pivot_tmp = true;
1462                 pivot_tmp_pivot_root_old = prefix_root(pivot_tmp, pivot_root_old);
1463
1464                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1465                 if (r < 0)
1466                         goto done;
1467
1468                 r = mount_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1469                 if (r < 0)
1470                         goto done;
1471
1472                 r = mount_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1473                 if (r < 0)
1474                         goto done;
1475         } else {
1476                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1477                 if (r < 0)
1478                         goto done;
1479         }
1480
1481 done:
1482         if (remove_pivot_tmp)
1483                 (void) rmdir(pivot_tmp);
1484
1485         return r;
1486 }