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 #ifdef 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         x = prefix_roota(top, "/fs/kdbus");
 460         (void) mkdir_p(x, 0755);
 461
 462         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 463          * remount /sys read-only.
 464          */
 465         if (cg_ns_supported()) {
 466                 x = prefix_roota(top, "/fs/cgroup");
 467                 (void) mkdir_p(x, 0755);
 468         }
 469
 470         return mount_verbose(LOG_ERR, NULL, top, NULL,
 471                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 472 }
 473
 474 static int mkdir_userns(const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 475         int r;
 476
 477         assert(path);
 478
 479         r = mkdir(path, mode);
 480         if (r < 0 && errno != EEXIST)
 481                 return -errno;
 482
 483         if ((mask & MOUNT_USE_USERNS) == 0)
 484                 return 0;
 485
 486         if (mask & MOUNT_IN_USERNS)
 487                 return 0;
 488
 489         r = lchown(path, uid_shift, uid_shift);
 490         if (r < 0)
 491                 return -errno;
 492
 493         return 0;
 494 }
 495
 496 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 497         const char *p, *e;
 498         int r;
 499
 500         assert(path);
 501
 502         if (prefix && !path_startswith(path, prefix))
 503                 return -ENOTDIR;
 504
 505         /* create every parent directory in the path, except the last component */
 506         p = path + strspn(path, "/");
 507         for (;;) {
 508                 char t[strlen(path) + 1];
 509
 510                 e = p + strcspn(p, "/");
 511                 p = e + strspn(e, "/");
 512
 513                 /* Is this the last component? If so, then we're done */
 514                 if (*p == 0)
 515                         break;
 516
 517                 memcpy(t, path, e - path);
 518                 t[e-path] = 0;
 519
 520                 if (prefix && path_startswith(prefix, t))
 521                         continue;
 522
 523                 r = mkdir_userns(t, mode, mask, uid_shift);
 524                 if (r < 0)
 525                         return r;
 526         }
 527
 528         return mkdir_userns(path, mode, mask, uid_shift);
 529 }
 530
 531 int mount_all(const char *dest,
 532               MountSettingsMask mount_settings,
 533               uid_t uid_shift, uid_t uid_range,
 534               const char *selinux_apifs_context) {
 535
 536         typedef struct MountPoint {
 537                 const char *what;
 538                 const char *where;
 539                 const char *type;
 540                 const char *options;
 541                 unsigned long flags;
 542                 MountSettingsMask mount_settings;
 543         } MountPoint;
 544
 545         static const MountPoint mount_table[] = {
 546                 /* inner child mounts */
 547                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_IN_USERNS },
 548                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 549                 { "/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) */
 550                 { 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 */
 551                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                               MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ... */
 552                 { 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 */
 553
 554                 /* outer child mounts */
 555                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                                            MOUNT_FATAL },
 556                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS },
 557                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO },    /* skipped if above was mounted */
 558                 { "sysfs",               "/sys",                "sysfs", NULL,                  MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL },                          /* skipped if above was mounted */
 559
 560                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  MOUNT_FATAL },
 561                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 562                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 563 #ifdef HAVE_SELINUX
 564                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   0 },  /* Bind mount first */
 565                 { NULL,                  "/sys/fs/selinux",     NULL,     NULL,       MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, 0 },  /* Then, make it r/o */
 566 #endif
 567         };
 568
 569         unsigned k;
 570         int r;
 571         bool use_userns = (mount_settings & MOUNT_USE_USERNS);
 572         bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS);
 573         bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO);
 574         bool in_userns = (mount_settings & MOUNT_IN_USERNS);
 575
 576         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 577                 _cleanup_free_ char *where = NULL, *options = NULL;
 578                 const char *o;
 579                 bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL);
 580
 581                 if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS))
 582                         continue;
 583
 584                 if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS))
 585                         continue;
 586
 587                 if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO))
 588                         continue;
 589
 590                 r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where);
 591                 if (r < 0)
 592                         return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where);
 593
 594                 r = path_is_mount_point(where, NULL, 0);
 595                 if (r < 0 && r != -ENOENT)
 596                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 597
 598                 /* Skip this entry if it is not a remount. */
 599                 if (mount_table[k].what && r > 0)
 600                         continue;
 601
 602                 r = mkdir_userns_p(dest, where, 0755, mount_settings, uid_shift);
 603                 if (r < 0 && r != -EEXIST) {
 604                         if (fatal)
 605                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 606
 607                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 608                         continue;
 609                 }
 610
 611                 o = mount_table[k].options;
 612                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 613                         if (in_userns)
 614                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 615                         else
 616                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 617                         if (r < 0)
 618                                 return log_oom();
 619                         if (r > 0)
 620                                 o = options;
 621                 }
 622
 623                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 624                                   mount_table[k].what,
 625                                   where,
 626                                   mount_table[k].type,
 627                                   mount_table[k].flags,
 628                                   o);
 629                 if (r < 0 && fatal)
 630                         return r;
 631         }
 632
 633         return 0;
 634 }
 635
 636 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 637         const char *p = options;
 638         unsigned long flags = *mount_flags;
 639         char *opts = NULL;
 640         int r;
 641
 642         assert(options);
 643
 644         for (;;) {
 645                 _cleanup_free_ char *word = NULL;
 646
 647                 r = extract_first_word(&p, &word, ",", 0);
 648                 if (r < 0)
 649                         return log_error_errno(r, "Failed to extract mount option: %m");
 650                 if (r == 0)
 651                         break;
 652
 653                 if (streq(word, "rbind"))
 654                         flags |= MS_REC;
 655                 else if (streq(word, "norbind"))
 656                         flags &= ~MS_REC;
 657                 else {
 658                         log_error("Invalid bind mount option: %s", word);
 659                         return -EINVAL;
 660                 }
 661         }
 662
 663         *mount_flags = flags;
 664         /* in the future mount_opts will hold string options for mount(2) */
 665         *mount_opts = opts;
 666
 667         return 0;
 668 }
 669
 670 static int mount_bind(const char *dest, CustomMount *m) {
 671
 672         _cleanup_free_ char *mount_opts = NULL, *where = NULL;
 673         unsigned long mount_flags = MS_BIND | MS_REC;
 674         struct stat source_st, dest_st;
 675         int r;
 676
 677         assert(dest);
 678         assert(m);
 679
 680         if (m->options) {
 681                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 682                 if (r < 0)
 683                         return r;
 684         }
 685
 686         if (stat(m->source, &source_st) < 0)
 687                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 688
 689         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where);
 690         if (r < 0)
 691                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 692         if (r > 0) { /* Path exists already? */
 693
 694                 if (stat(where, &dest_st) < 0)
 695                         return log_error_errno(errno, "Failed to stat %s: %m", where);
 696
 697                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 698                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 699                         return -EINVAL;
 700                 }
 701
 702                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 703                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 704                         return -EINVAL;
 705                 }
 706
 707         } else { /* Path doesn't exist yet? */
 708                 r = mkdir_parents_label(where, 0755);
 709                 if (r < 0)
 710                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 711
 712                 /* Create the mount point. Any non-directory file can be
 713                 * mounted on any non-directory file (regular, fifo, socket,
 714                 * char, block).
 715                 */
 716                 if (S_ISDIR(source_st.st_mode))
 717                         r = mkdir_label(where, 0755);
 718                 else
 719                         r = touch(where);
 720                 if (r < 0)
 721                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 722
 723         }
 724
 725         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 726         if (r < 0)
 727                 return r;
 728
 729         if (m->read_only) {
 730                 r = bind_remount_recursive(where, true, NULL);
 731                 if (r < 0)
 732                         return log_error_errno(r, "Read-only bind mount failed: %m");
 733         }
 734
 735         return 0;
 736 }
 737
 738 static int mount_tmpfs(
 739                 const char *dest,
 740                 CustomMount *m,
 741                 bool userns, uid_t uid_shift, uid_t uid_range,
 742                 const char *selinux_apifs_context) {
 743
 744         const char *options;
 745         _cleanup_free_ char *buf = NULL, *where = NULL;
 746         int r;
 747
 748         assert(dest);
 749         assert(m);
 750
 751         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where);
 752         if (r < 0)
 753                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 754         if (r == 0) { /* Doesn't exist yet? */
 755                 r = mkdir_p_label(where, 0755);
 756                 if (r < 0)
 757                         return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 758         }
 759
 760         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 761         if (r < 0)
 762                 return log_oom();
 763         options = r > 0 ? buf : m->options;
 764
 765         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 766 }
 767
 768 static char *joined_and_escaped_lower_dirs(char **lower) {
 769         _cleanup_strv_free_ char **sv = NULL;
 770
 771         sv = strv_copy(lower);
 772         if (!sv)
 773                 return NULL;
 774
 775         strv_reverse(sv);
 776
 777         if (!strv_shell_escape(sv, ",:"))
 778                 return NULL;
 779
 780         return strv_join(sv, ":");
 781 }
 782
 783 static int mount_overlay(const char *dest, CustomMount *m) {
 784
 785         _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
 786         const char *options;
 787         int r;
 788
 789         assert(dest);
 790         assert(m);
 791
 792         r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where);
 793         if (r < 0)
 794                 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
 795         if (r == 0) { /* Doesn't exist yet? */
 796                 r = mkdir_label(where, 0755);
 797                 if (r < 0)
 798                         return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 799         }
 800
 801         (void) mkdir_p_label(m->source, 0755);
 802
 803         lower = joined_and_escaped_lower_dirs(m->lower);
 804         if (!lower)
 805                 return log_oom();
 806
 807         escaped_source = shell_escape(m->source, ",:");
 808         if (!escaped_source)
 809                 return log_oom();
 810
 811         if (m->read_only)
 812                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 813         else {
 814                 _cleanup_free_ char *escaped_work_dir = NULL;
 815
 816                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 817                 if (!escaped_work_dir)
 818                         return log_oom();
 819
 820                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 821         }
 822
 823         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 824 }
 825
 826 int mount_custom(
 827                 const char *dest,
 828                 CustomMount *mounts, unsigned n,
 829                 bool userns, uid_t uid_shift, uid_t uid_range,
 830                 const char *selinux_apifs_context) {
 831
 832         unsigned i;
 833         int r;
 834
 835         assert(dest);
 836
 837         for (i = 0; i < n; i++) {
 838                 CustomMount *m = mounts + i;
 839
 840                 switch (m->type) {
 841
 842                 case CUSTOM_MOUNT_BIND:
 843                         r = mount_bind(dest, m);
 844                         break;
 845
 846                 case CUSTOM_MOUNT_TMPFS:
 847                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 848                         break;
 849
 850                 case CUSTOM_MOUNT_OVERLAY:
 851                         r = mount_overlay(dest, m);
 852                         break;
 853
 854                 default:
 855                         assert_not_reached("Unknown custom mount type");
 856                 }
 857
 858                 if (r < 0)
 859                         return r;
 860         }
 861
 862         return 0;
 863 }
 864
 865 /* Retrieve existing subsystems. This function is called in a new cgroup
 866  * namespace.
 867  */
 868 static int get_controllers(Set *subsystems) {
 869         _cleanup_fclose_ FILE *f = NULL;
 870         char line[LINE_MAX];
 871
 872         assert(subsystems);
 873
 874         f = fopen("/proc/self/cgroup", "re");
 875         if (!f)
 876                 return errno == ENOENT ? -ESRCH : -errno;
 877
 878         FOREACH_LINE(line, f, return -errno) {
 879                 int r;
 880                 char *e, *l, *p;
 881
 882                 l = strchr(line, ':');
 883                 if (!l)
 884                         continue;
 885
 886                 l++;
 887                 e = strchr(l, ':');
 888                 if (!e)
 889                         continue;
 890
 891                 *e = 0;
 892
 893                 if (STR_IN_SET(l, "", "name=systemd", "name=unified"))
 894                         continue;
 895
 896                 p = strdup(l);
 897                 if (!p)
 898                         return -ENOMEM;
 899
 900                 r = set_consume(subsystems, p);
 901                 if (r < 0)
 902                         return r;
 903         }
 904
 905         return 0;
 906 }
 907
 908 static int mount_legacy_cgroup_hierarchy(
 909                 const char *dest,
 910                 const char *controller,
 911                 const char *hierarchy,
 912                 bool read_only) {
 913
 914         const char *to, *fstype, *opts;
 915         int r;
 916
 917         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 918
 919         r = path_is_mount_point(to, dest, 0);
 920         if (r < 0 && r != -ENOENT)
 921                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 922         if (r > 0)
 923                 return 0;
 924
 925         mkdir_p(to, 0755);
 926
 927         /* The superblock mount options of the mount point need to be
 928          * identical to the hosts', and hence writable... */
 929         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER_HYBRID)) {
 930                 fstype = "cgroup2";
 931                 opts = NULL;
 932         } else if (streq(controller, SYSTEMD_CGROUP_CONTROLLER_LEGACY)) {
 933                 fstype = "cgroup";
 934                 opts = "none,name=systemd,xattr";
 935         } else {
 936                 fstype = "cgroup";
 937                 opts = controller;
 938         }
 939
 940         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 941         if (r < 0)
 942                 return r;
 943
 944         /* ... hence let's only make the bind mount read-only, not the superblock. */
 945         if (read_only) {
 946                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 947                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 948                 if (r < 0)
 949                         return r;
 950         }
 951
 952         return 1;
 953 }
 954
 955 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 956 static int mount_legacy_cgns_supported(
 957                 const char *dest,
 958                 CGroupUnified unified_requested,
 959                 bool userns,
 960                 uid_t uid_shift,
 961                 uid_t uid_range,
 962                 const char *selinux_apifs_context) {
 963
 964         _cleanup_set_free_free_ Set *controllers = NULL;
 965         const char *cgroup_root = "/sys/fs/cgroup", *c;
 966         int r;
 967
 968         (void) mkdir_p(cgroup_root, 0755);
 969
 970         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 971         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 972         if (r < 0)
 973                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 974         if (r == 0) {
 975                 _cleanup_free_ char *options = NULL;
 976
 977                 /* When cgroup namespaces are enabled and user namespaces are
 978                  * used then the mount of the cgroupfs is done *inside* the new
 979                  * user namespace. We're root in the new user namespace and the
 980                  * kernel will happily translate our uid/gid to the correct
 981                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 982                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 983                  */
 984                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 985                 if (r < 0)
 986                         return log_oom();
 987
 988                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 989                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 990                 if (r < 0)
 991                         return r;
 992         }
 993
 994         r = cg_all_unified();
 995         if (r < 0)
 996                 return r;
 997         if (r > 0)
 998                 goto skip_controllers;
 999
1000         controllers = set_new(&string_hash_ops);
1001         if (!controllers)
1002                 return log_oom();
1003
1004         r = get_controllers(controllers);
1005         if (r < 0)
1006                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1007
1008         for (;;) {
1009                 _cleanup_free_ const char *controller = NULL;
1010
1011                 controller = set_steal_first(controllers);
1012                 if (!controller)
1013                         break;
1014
1015                 r = mount_legacy_cgroup_hierarchy("", controller, controller, !userns);
1016                 if (r < 0)
1017                         return r;
1018
1019                 /* When multiple hierarchies are co-mounted, make their
1020                  * constituting individual hierarchies a symlink to the
1021                  * co-mount.
1022                  */
1023                 c = controller;
1024                 for (;;) {
1025                         _cleanup_free_ char *target = NULL, *tok = NULL;
1026
1027                         r = extract_first_word(&c, &tok, ",", 0);
1028                         if (r < 0)
1029                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
1030                         if (r == 0)
1031                                 break;
1032
1033                         target = prefix_root("/sys/fs/cgroup", tok);
1034                         if (!target)
1035                                 return log_oom();
1036
1037                         if (streq(controller, tok))
1038                                 break;
1039
1040                         r = symlink_idempotent(controller, target);
1041                         if (r == -EINVAL)
1042                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1043                         if (r < 0)
1044                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1045                 }
1046         }
1047
1048 skip_controllers:
1049         if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
1050                 r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
1051                 if (r < 0)
1052                         return r;
1053         }
1054
1055         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
1056         if (r < 0)
1057                 return r;
1058
1059         if (!userns)
1060                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1061                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1062
1063         return 0;
1064 }
1065
1066 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
1067 static int mount_legacy_cgns_unsupported(
1068                 const char *dest,
1069                 CGroupUnified unified_requested,
1070                 bool userns,
1071                 uid_t uid_shift,
1072                 uid_t uid_range,
1073                 const char *selinux_apifs_context) {
1074
1075         _cleanup_set_free_free_ Set *controllers = NULL;
1076         const char *cgroup_root;
1077         int r;
1078
1079         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
1080
1081         (void) mkdir_p(cgroup_root, 0755);
1082
1083         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
1084         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
1085         if (r < 0)
1086                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
1087         if (r == 0) {
1088                 _cleanup_free_ char *options = NULL;
1089
1090                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
1091                 if (r < 0)
1092                         return log_oom();
1093
1094                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
1095                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
1096                 if (r < 0)
1097                         return r;
1098         }
1099
1100         r = cg_all_unified();
1101         if (r < 0)
1102                 return r;
1103         if (r > 0)
1104                 goto skip_controllers;
1105
1106         controllers = set_new(&string_hash_ops);
1107         if (!controllers)
1108                 return log_oom();
1109
1110         r = cg_kernel_controllers(controllers);
1111         if (r < 0)
1112                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1113
1114         for (;;) {
1115                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
1116
1117                 controller = set_steal_first(controllers);
1118                 if (!controller)
1119                         break;
1120
1121                 origin = prefix_root("/sys/fs/cgroup/", controller);
1122                 if (!origin)
1123                         return log_oom();
1124
1125                 r = readlink_malloc(origin, &combined);
1126                 if (r == -EINVAL) {
1127                         /* Not a symbolic link, but directly a single cgroup hierarchy */
1128
1129                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
1130                         if (r < 0)
1131                                 return r;
1132
1133                 } else if (r < 0)
1134                         return log_error_errno(r, "Failed to read link %s: %m", origin);
1135                 else {
1136                         _cleanup_free_ char *target = NULL;
1137
1138                         target = prefix_root(dest, origin);
1139                         if (!target)
1140                                 return log_oom();
1141
1142                         /* A symbolic link, a combination of controllers in one hierarchy */
1143
1144                         if (!filename_is_valid(combined)) {
1145                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
1146                                 continue;
1147                         }
1148
1149                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
1150                         if (r < 0)
1151                                 return r;
1152
1153                         r = symlink_idempotent(combined, target);
1154                         if (r == -EINVAL)
1155                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
1156                         if (r < 0)
1157                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1158                 }
1159         }
1160
1161 skip_controllers:
1162         if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
1163                 r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
1164                 if (r < 0)
1165                         return r;
1166         }
1167
1168         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
1169         if (r < 0)
1170                 return r;
1171
1172         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
1173                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
1174 }
1175
1176 static int mount_unified_cgroups(const char *dest) {
1177         const char *p;
1178         int r;
1179
1180         assert(dest);
1181
1182         p = prefix_roota(dest, "/sys/fs/cgroup");
1183
1184         (void) mkdir_p(p, 0755);
1185
1186         r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW);
1187         if (r < 0)
1188                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
1189         if (r > 0) {
1190                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
1191                 if (access(p, F_OK) >= 0)
1192                         return 0;
1193                 if (errno != ENOENT)
1194                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
1195
1196                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
1197                 return -EINVAL;
1198         }
1199
1200         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1201 }
1202
1203 int mount_cgroups(
1204                 const char *dest,
1205                 CGroupUnified unified_requested,
1206                 bool userns,
1207                 uid_t uid_shift,
1208                 uid_t uid_range,
1209                 const char *selinux_apifs_context,
1210                 bool use_cgns) {
1211
1212         if (unified_requested >= CGROUP_UNIFIED_ALL)
1213                 return mount_unified_cgroups(dest);
1214         else if (use_cgns)
1215                 return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1216
1217         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1218 }
1219
1220 static int mount_systemd_cgroup_writable_one(const char *systemd_own, const char *systemd_root)
1221 {
1222         int r;
1223
1224         /* Make our own cgroup a (writable) bind mount */
1225         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1226         if (r < 0)
1227                 return r;
1228
1229         /* And then remount the systemd cgroup root read-only */
1230         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1231                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1232 }
1233
1234 int mount_systemd_cgroup_writable(
1235                 const char *dest,
1236                 CGroupUnified unified_requested) {
1237
1238         _cleanup_free_ char *own_cgroup_path = NULL;
1239         int r;
1240
1241         assert(dest);
1242
1243         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1244         if (r < 0)
1245                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1246
1247         /* If we are living in the top-level, then there's nothing to do... */
1248         if (path_equal(own_cgroup_path, "/"))
1249                 return 0;
1250
1251         if (unified_requested >= CGROUP_UNIFIED_ALL)
1252                 return mount_systemd_cgroup_writable_one(strjoina(dest, "/sys/fs/cgroup", own_cgroup_path),
1253                                                          prefix_roota(dest, "/sys/fs/cgroup"));
1254
1255         if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
1256                 r = mount_systemd_cgroup_writable_one(strjoina(dest, "/sys/fs/cgroup/unified", own_cgroup_path),
1257                                                       prefix_roota(dest, "/sys/fs/cgroup/unified"));
1258                 if (r < 0)
1259                         return r;
1260         }
1261
1262         return mount_systemd_cgroup_writable_one(strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path),
1263                                                  prefix_roota(dest, "/sys/fs/cgroup/systemd"));
1264 }
1265
1266 int setup_volatile_state(
1267                 const char *directory,
1268                 VolatileMode mode,
1269                 bool userns, uid_t uid_shift, uid_t uid_range,
1270                 const char *selinux_apifs_context) {
1271
1272         _cleanup_free_ char *buf = NULL;
1273         const char *p, *options;
1274         int r;
1275
1276         assert(directory);
1277
1278         if (mode != VOLATILE_STATE)
1279                 return 0;
1280
1281         /* --volatile=state means we simply overmount /var
1282            with a tmpfs, and the rest read-only. */
1283
1284         r = bind_remount_recursive(directory, true, NULL);
1285         if (r < 0)
1286                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1287
1288         p = prefix_roota(directory, "/var");
1289         r = mkdir(p, 0755);
1290         if (r < 0 && errno != EEXIST)
1291                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1292
1293         options = "mode=755";
1294         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1295         if (r < 0)
1296                 return log_oom();
1297         if (r > 0)
1298                 options = buf;
1299
1300         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1301 }
1302
1303 int setup_volatile(
1304                 const char *directory,
1305                 VolatileMode mode,
1306                 bool userns, uid_t uid_shift, uid_t uid_range,
1307                 const char *selinux_apifs_context) {
1308
1309         bool tmpfs_mounted = false, bind_mounted = false;
1310         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1311         _cleanup_free_ char *buf = NULL;
1312         const char *f, *t, *options;
1313         int r;
1314
1315         assert(directory);
1316
1317         if (mode != VOLATILE_YES)
1318                 return 0;
1319
1320         /* --volatile=yes means we mount a tmpfs to the root dir, and
1321            the original /usr to use inside it, and that read-only. */
1322
1323         if (!mkdtemp(template))
1324                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1325
1326         options = "mode=755";
1327         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1328         if (r < 0)
1329                 return log_oom();
1330         if (r > 0)
1331                 options = buf;
1332
1333         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1334         if (r < 0)
1335                 goto fail;
1336
1337         tmpfs_mounted = true;
1338
1339         f = prefix_roota(directory, "/usr");
1340         t = prefix_roota(template, "/usr");
1341
1342         r = mkdir(t, 0755);
1343         if (r < 0 && errno != EEXIST) {
1344                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1345                 goto fail;
1346         }
1347
1348         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1349         if (r < 0)
1350                 goto fail;
1351
1352         bind_mounted = true;
1353
1354         r = bind_remount_recursive(t, true, NULL);
1355         if (r < 0) {
1356                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1357                 goto fail;
1358         }
1359
1360         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1361         if (r < 0)
1362                 goto fail;
1363
1364         (void) rmdir(template);
1365
1366         return 0;
1367
1368 fail:
1369         if (bind_mounted)
1370                 (void) umount_verbose(t);
1371
1372         if (tmpfs_mounted)
1373                 (void) umount_verbose(template);
1374         (void) rmdir(template);
1375         return r;
1376 }
1377
1378 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1379 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1380         _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1381         const char *p = s;
1382         int r;
1383
1384         assert(pivot_root_new);
1385         assert(pivot_root_old);
1386
1387         r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1388         if (r < 0)
1389                 return r;
1390         if (r == 0)
1391                 return -EINVAL;
1392
1393         if (isempty(p))
1394                 root_old = NULL;
1395         else {
1396                 root_old = strdup(p);
1397                 if (!root_old)
1398                         return -ENOMEM;
1399         }
1400
1401         if (!path_is_absolute(root_new))
1402                 return -EINVAL;
1403         if (root_old && !path_is_absolute(root_old))
1404                 return -EINVAL;
1405
1406         free_and_replace(*pivot_root_new, root_new);
1407         free_and_replace(*pivot_root_old, root_old);
1408
1409         return 0;
1410 }
1411
1412 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1413         _cleanup_free_ char *directory_pivot_root_new = NULL;
1414         _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1415         char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
1416         bool remove_pivot_tmp = false;
1417         int r;
1418
1419         assert(directory);
1420
1421         if (!pivot_root_new)
1422                 return 0;
1423
1424         /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1425          * If pivot_root_old is NULL, the existing / disappears.
1426          * This requires a temporary directory, pivot_tmp, which is
1427          * not a child of either.
1428          *
1429          * This is typically used for OSTree-style containers, where
1430          * the root partition contains several sysroots which could be
1431          * run. Normally, one would be chosen by the bootloader and
1432          * pivoted to / by initramfs.
1433          *
1434          * For example, for an OSTree deployment, pivot_root_new
1435          * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1436          * code doesn’t do the /var mount which OSTree expects: use
1437          * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1438          *
1439          * So in the OSTree case, we’ll end up with something like:
1440          *  - directory = /tmp/nspawn-root-123456
1441          *  - pivot_root_new = /ostree/deploy/os/deploy/123abc
1442          *  - pivot_root_old = /sysroot
1443          *  - directory_pivot_root_new =
1444          *       /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1445          *  - pivot_tmp = /tmp/nspawn-pivot-123456
1446          *  - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1447          *
1448          * Requires all file systems at directory and below to be mounted
1449          * MS_PRIVATE or MS_SLAVE so they can be moved.
1450          */
1451         directory_pivot_root_new = prefix_root(directory, pivot_root_new);
1452
1453         /* Remount directory_pivot_root_new to make it movable. */
1454         r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1455         if (r < 0)
1456                 goto done;
1457
1458         if (pivot_root_old) {
1459                 if (!mkdtemp(pivot_tmp)) {
1460                         r = log_error_errno(errno, "Failed to create temporary directory: %m");
1461                         goto done;
1462                 }
1463
1464                 remove_pivot_tmp = true;
1465                 pivot_tmp_pivot_root_old = prefix_root(pivot_tmp, pivot_root_old);
1466
1467                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1468                 if (r < 0)
1469                         goto done;
1470
1471                 r = mount_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1472                 if (r < 0)
1473                         goto done;
1474
1475                 r = mount_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1476                 if (r < 0)
1477                         goto done;
1478         } else {
1479                 r = mount_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1480                 if (r < 0)
1481                         goto done;
1482         }
1483
1484 done:
1485         if (remove_pivot_tmp)
1486                 (void) rmdir(pivot_tmp);
1487
1488         return r;
1489 }