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 "cgroup-util.h"
  25 #include "escape.h"
  26 #include "fs-util.h"
  27 #include "label.h"
  28 #include "mkdir.h"
  29 #include "mount-util.h"
  30 #include "nspawn-mount.h"
  31 #include "parse-util.h"
  32 #include "path-util.h"
  33 #include "rm-rf.h"
  34 #include "set.h"
  35 #include "stat-util.h"
  36 #include "string-util.h"
  37 #include "strv.h"
  38 #include "user-util.h"
  39 #include "util.h"
  40
  41 CustomMount* custom_mount_add(CustomMount **l, unsigned *n, CustomMountType t) {
  42         CustomMount *c, *ret;
  43
  44         assert(l);
  45         assert(n);
  46         assert(t >= 0);
  47         assert(t < _CUSTOM_MOUNT_TYPE_MAX);
  48
  49         c = realloc(*l, (*n + 1) * sizeof(CustomMount));
  50         if (!c)
  51                 return NULL;
  52
  53         *l = c;
  54         ret = *l + *n;
  55         (*n)++;
  56
  57         *ret = (CustomMount) { .type = t };
  58
  59         return ret;
  60 }
  61
  62 void custom_mount_free_all(CustomMount *l, unsigned n) {
  63         unsigned i;
  64
  65         for (i = 0; i < n; i++) {
  66                 CustomMount *m = l + i;
  67
  68                 free(m->source);
  69                 free(m->destination);
  70                 free(m->options);
  71
  72                 if (m->work_dir) {
  73                         (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
  74                         free(m->work_dir);
  75                 }
  76
  77                 strv_free(m->lower);
  78         }
  79
  80         free(l);
  81 }
  82
  83 int custom_mount_compare(const void *a, const void *b) {
  84         const CustomMount *x = a, *y = b;
  85         int r;
  86
  87         r = path_compare(x->destination, y->destination);
  88         if (r != 0)
  89                 return r;
  90
  91         if (x->type < y->type)
  92                 return -1;
  93         if (x->type > y->type)
  94                 return 1;
  95
  96         return 0;
  97 }
  98
  99 int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 100         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 101         const char *p = s;
 102         CustomMount *m;
 103         int r;
 104
 105         assert(l);
 106         assert(n);
 107
 108         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 109         if (r < 0)
 110                 return r;
 111         if (r == 0)
 112                 return -EINVAL;
 113
 114         if (r == 1) {
 115                 destination = strdup(source);
 116                 if (!destination)
 117                         return -ENOMEM;
 118         }
 119
 120         if (r == 2 && !isempty(p)) {
 121                 opts = strdup(p);
 122                 if (!opts)
 123                         return -ENOMEM;
 124         }
 125
 126         if (!path_is_absolute(source))
 127                 return -EINVAL;
 128
 129         if (!path_is_absolute(destination))
 130                 return -EINVAL;
 131
 132         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 133         if (!m)
 134                 return log_oom();
 135
 136         m->source = source;
 137         m->destination = destination;
 138         m->read_only = read_only;
 139         m->options = opts;
 140
 141         source = destination = opts = NULL;
 142         return 0;
 143 }
 144
 145 int tmpfs_mount_parse(CustomMount **l, unsigned *n, const char *s) {
 146         _cleanup_free_ char *path = NULL, *opts = NULL;
 147         const char *p = s;
 148         CustomMount *m;
 149         int r;
 150
 151         assert(l);
 152         assert(n);
 153         assert(s);
 154
 155         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 156         if (r < 0)
 157                 return r;
 158         if (r == 0)
 159                 return -EINVAL;
 160
 161         if (isempty(p))
 162                 opts = strdup("mode=0755");
 163         else
 164                 opts = strdup(p);
 165         if (!opts)
 166                 return -ENOMEM;
 167
 168         if (!path_is_absolute(path))
 169                 return -EINVAL;
 170
 171         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 172         if (!m)
 173                 return -ENOMEM;
 174
 175         m->destination = path;
 176         m->options = opts;
 177
 178         path = opts = NULL;
 179         return 0;
 180 }
 181
 182 static int tmpfs_patch_options(
 183                 const char *options,
 184                 bool userns, uid_t uid_shift, uid_t uid_range,
 185                 const char *selinux_apifs_context,
 186                 char **ret) {
 187
 188         char *buf = NULL;
 189
 190         if (userns && uid_shift != 0) {
 191                 assert(uid_shift != UID_INVALID);
 192
 193                 if (options)
 194                         (void) asprintf(&buf, "%s,uid=" UID_FMT ",gid=" UID_FMT, options, uid_shift, uid_shift);
 195                 else
 196                         (void) asprintf(&buf, "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift);
 197                 if (!buf)
 198                         return -ENOMEM;
 199
 200                 options = buf;
 201         }
 202
 203 #ifdef HAVE_SELINUX
 204         if (selinux_apifs_context) {
 205                 char *t;
 206
 207                 if (options)
 208                         t = strjoin(options, ",context=\"", selinux_apifs_context, "\"", NULL);
 209                 else
 210                         t = strjoin("context=\"", selinux_apifs_context, "\"", NULL);
 211                 if (!t) {
 212                         free(buf);
 213                         return -ENOMEM;
 214                 }
 215
 216                 free(buf);
 217                 buf = t;
 218         }
 219 #endif
 220
 221         *ret = buf;
 222         return !!buf;
 223 }
 224
 225 int mount_sysfs(const char *dest) {
 226         const char *full, *top, *x;
 227         int r;
 228
 229         top = prefix_roota(dest, "/sys");
 230         r = path_check_fstype(top, SYSFS_MAGIC);
 231         if (r < 0)
 232                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 233         /* /sys might already be mounted as sysfs by the outer child in the
 234          * !netns case. In this case, it's all good. Don't touch it because we
 235          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 236          */
 237         if (r > 0)
 238                 return 0;
 239
 240         full = prefix_roota(top, "/full");
 241
 242         (void) mkdir(full, 0755);
 243
 244         if (mount("sysfs", full, "sysfs", MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
 245                 return log_error_errno(errno, "Failed to mount sysfs to %s: %m", full);
 246
 247         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 248                 _cleanup_free_ char *from = NULL, *to = NULL;
 249
 250                 from = prefix_root(full, x);
 251                 if (!from)
 252                         return log_oom();
 253
 254                 to = prefix_root(top, x);
 255                 if (!to)
 256                         return log_oom();
 257
 258                 (void) mkdir(to, 0755);
 259
 260                 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
 261                         return log_error_errno(errno, "Failed to mount /sys/%s into place: %m", x);
 262
 263                 if (mount(NULL, to, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL) < 0)
 264                         return log_error_errno(errno, "Failed to mount /sys/%s read-only: %m", x);
 265         }
 266
 267         if (umount(full) < 0)
 268                 return log_error_errno(errno, "Failed to unmount %s: %m", full);
 269
 270         if (rmdir(full) < 0)
 271                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 272
 273         x = prefix_roota(top, "/fs/kdbus");
 274         (void) mkdir(x, 0755);
 275
 276         if (mount(NULL, top, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL) < 0)
 277                 return log_error_errno(errno, "Failed to make %s read-only: %m", top);
 278
 279         return 0;
 280 }
 281
 282 int mount_all(const char *dest,
 283               bool use_userns, bool in_userns,
 284               bool use_netns,
 285               uid_t uid_shift, uid_t uid_range,
 286               const char *selinux_apifs_context) {
 287
 288         typedef struct MountPoint {
 289                 const char *what;
 290                 const char *where;
 291                 const char *type;
 292                 const char *options;
 293                 unsigned long flags;
 294                 bool fatal;
 295                 bool in_userns;
 296                 bool use_netns;
 297         } MountPoint;
 298
 299         static const MountPoint mount_table[] = {
 300                 { "proc",      "/proc",          "proc",   NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  true, false  },
 301                 { "/proc/sys", "/proc/sys",      NULL,     NULL,        MS_BIND,                                                   true,  true, false  },   /* Bind mount first */
 302                 { NULL,        "/proc/sys",      NULL,     NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, true,  true, false  },   /* Then, make it r/o */
 303                 { "tmpfs",     "/sys",           "tmpfs",  "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true },
 304                 { "sysfs",     "/sys",           "sysfs",  NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 305                 { "tmpfs",     "/dev",           "tmpfs",  "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 306                 { "tmpfs",     "/dev/shm",       "tmpfs",  "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 307                 { "tmpfs",     "/run",           "tmpfs",  "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 308                 { "tmpfs",     "/tmp",           "tmpfs",  "mode=1777", MS_STRICTATIME,                                            true,  false, false },
 309 #ifdef HAVE_SELINUX
 310                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL,     MS_BIND,                                                   false, false, false },  /* Bind mount first */
 311                 { NULL,              "/sys/fs/selinux", NULL, NULL,     MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, false, false, false },  /* Then, make it r/o */
 312 #endif
 313         };
 314
 315         unsigned k;
 316         int r;
 317
 318         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 319                 _cleanup_free_ char *where = NULL, *options = NULL;
 320                 const char *o;
 321
 322                 if (in_userns != mount_table[k].in_userns)
 323                         continue;
 324
 325                 if (!use_netns && mount_table[k].use_netns)
 326                         continue;
 327
 328                 where = prefix_root(dest, mount_table[k].where);
 329                 if (!where)
 330                         return log_oom();
 331
 332                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 333                 if (r < 0 && r != -ENOENT)
 334                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 335
 336                 /* Skip this entry if it is not a remount. */
 337                 if (mount_table[k].what && r > 0)
 338                         continue;
 339
 340                 r = mkdir_p(where, 0755);
 341                 if (r < 0) {
 342                         if (mount_table[k].fatal)
 343                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 344
 345                         log_warning_errno(r, "Failed to create directory %s: %m", where);
 346                         continue;
 347                 }
 348
 349                 o = mount_table[k].options;
 350                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 351                         r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, selinux_apifs_context, &options);
 352                         if (r < 0)
 353                                 return log_oom();
 354                         if (r > 0)
 355                                 o = options;
 356                 }
 357
 358                 if (mount(mount_table[k].what,
 359                           where,
 360                           mount_table[k].type,
 361                           mount_table[k].flags,
 362                           o) < 0) {
 363
 364                         if (mount_table[k].fatal)
 365                                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 366
 367                         log_warning_errno(errno, "mount(%s) failed, ignoring: %m", where);
 368                 }
 369         }
 370
 371         return 0;
 372 }
 373
 374 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 375         const char *p = options;
 376         unsigned long flags = *mount_flags;
 377         char *opts = NULL;
 378
 379         assert(options);
 380
 381         for (;;) {
 382                 _cleanup_free_ char *word = NULL;
 383                 int r = extract_first_word(&p, &word, ",", 0);
 384                 if (r < 0)
 385                         return log_error_errno(r, "Failed to extract mount option: %m");
 386                 if (r == 0)
 387                         break;
 388
 389                 if (streq(word, "rbind"))
 390                         flags |= MS_REC;
 391                 else if (streq(word, "norbind"))
 392                         flags &= ~MS_REC;
 393                 else {
 394                         log_error("Invalid bind mount option: %s", word);
 395                         return -EINVAL;
 396                 }
 397         }
 398
 399         *mount_flags = flags;
 400         /* in the future mount_opts will hold string options for mount(2) */
 401         *mount_opts = opts;
 402
 403         return 0;
 404 }
 405
 406 static int mount_bind(const char *dest, CustomMount *m) {
 407         struct stat source_st, dest_st;
 408         const char *where;
 409         unsigned long mount_flags = MS_BIND | MS_REC;
 410         _cleanup_free_ char *mount_opts = NULL;
 411         int r;
 412
 413         assert(m);
 414
 415         if (m->options) {
 416                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 417                 if (r < 0)
 418                         return r;
 419         }
 420
 421         if (stat(m->source, &source_st) < 0)
 422                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 423
 424         where = prefix_roota(dest, m->destination);
 425
 426         if (stat(where, &dest_st) >= 0) {
 427                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 428                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 429                         return -EINVAL;
 430                 }
 431
 432                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 433                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 434                         return -EINVAL;
 435                 }
 436
 437         } else if (errno == ENOENT) {
 438                 r = mkdir_parents_label(where, 0755);
 439                 if (r < 0)
 440                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 441
 442                 /* Create the mount point. Any non-directory file can be
 443                 * mounted on any non-directory file (regular, fifo, socket,
 444                 * char, block).
 445                 */
 446                 if (S_ISDIR(source_st.st_mode))
 447                         r = mkdir_label(where, 0755);
 448                 else
 449                         r = touch(where);
 450                 if (r < 0)
 451                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 452
 453         } else {
 454                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 455         }
 456
 457         if (mount(m->source, where, NULL, mount_flags, mount_opts) < 0)
 458                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 459
 460         if (m->read_only) {
 461                 r = bind_remount_recursive(where, true);
 462                 if (r < 0)
 463                         return log_error_errno(r, "Read-only bind mount failed: %m");
 464         }
 465
 466         return 0;
 467 }
 468
 469 static int mount_tmpfs(
 470                 const char *dest,
 471                 CustomMount *m,
 472                 bool userns, uid_t uid_shift, uid_t uid_range,
 473                 const char *selinux_apifs_context) {
 474
 475         const char *where, *options;
 476         _cleanup_free_ char *buf = NULL;
 477         int r;
 478
 479         assert(dest);
 480         assert(m);
 481
 482         where = prefix_roota(dest, m->destination);
 483
 484         r = mkdir_p_label(where, 0755);
 485         if (r < 0 && r != -EEXIST)
 486                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 487
 488         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, selinux_apifs_context, &buf);
 489         if (r < 0)
 490                 return log_oom();
 491         options = r > 0 ? buf : m->options;
 492
 493         if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options) < 0)
 494                 return log_error_errno(errno, "tmpfs mount to %s failed: %m", where);
 495
 496         return 0;
 497 }
 498
 499 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 500         _cleanup_strv_free_ char **sv = NULL;
 501
 502         sv = strv_copy(lower);
 503         if (!sv)
 504                 return NULL;
 505
 506         strv_reverse(sv);
 507
 508         if (!strv_shell_escape(sv, ",:"))
 509                 return NULL;
 510
 511         return strv_join(sv, ":");
 512 }
 513
 514 static int mount_overlay(const char *dest, CustomMount *m) {
 515         _cleanup_free_ char *lower = NULL;
 516         const char *where, *options;
 517         int r;
 518
 519         assert(dest);
 520         assert(m);
 521
 522         where = prefix_roota(dest, m->destination);
 523
 524         r = mkdir_label(where, 0755);
 525         if (r < 0 && r != -EEXIST)
 526                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 527
 528         (void) mkdir_p_label(m->source, 0755);
 529
 530         lower = joined_and_escaped_lower_dirs(m->lower);
 531         if (!lower)
 532                 return log_oom();
 533
 534         if (m->read_only) {
 535                 _cleanup_free_ char *escaped_source = NULL;
 536
 537                 escaped_source = shell_escape(m->source, ",:");
 538                 if (!escaped_source)
 539                         return log_oom();
 540
 541                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 542         } else {
 543                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 544
 545                 assert(m->work_dir);
 546                 (void) mkdir_label(m->work_dir, 0700);
 547
 548                 escaped_source = shell_escape(m->source, ",:");
 549                 if (!escaped_source)
 550                         return log_oom();
 551                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 552                 if (!escaped_work_dir)
 553                         return log_oom();
 554
 555                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 556         }
 557
 558         if (mount("overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options) < 0)
 559                 return log_error_errno(errno, "overlay mount to %s failed: %m", where);
 560
 561         return 0;
 562 }
 563
 564 int mount_custom(
 565                 const char *dest,
 566                 CustomMount *mounts, unsigned n,
 567                 bool userns, uid_t uid_shift, uid_t uid_range,
 568                 const char *selinux_apifs_context) {
 569
 570         unsigned i;
 571         int r;
 572
 573         assert(dest);
 574
 575         for (i = 0; i < n; i++) {
 576                 CustomMount *m = mounts + i;
 577
 578                 switch (m->type) {
 579
 580                 case CUSTOM_MOUNT_BIND:
 581                         r = mount_bind(dest, m);
 582                         break;
 583
 584                 case CUSTOM_MOUNT_TMPFS:
 585                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 586                         break;
 587
 588                 case CUSTOM_MOUNT_OVERLAY:
 589                         r = mount_overlay(dest, m);
 590                         break;
 591
 592                 default:
 593                         assert_not_reached("Unknown custom mount type");
 594                 }
 595
 596                 if (r < 0)
 597                         return r;
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, bool read_only) {
 604         char *to;
 605         int r;
 606
 607         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 608
 609         r = path_is_mount_point(to, 0);
 610         if (r < 0 && r != -ENOENT)
 611                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 612         if (r > 0)
 613                 return 0;
 614
 615         mkdir_p(to, 0755);
 616
 617         /* The superblock mount options of the mount point need to be
 618          * identical to the hosts', and hence writable... */
 619         if (mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller) < 0)
 620                 return log_error_errno(errno, "Failed to mount to %s: %m", to);
 621
 622         /* ... hence let's only make the bind mount read-only, not the
 623          * superblock. */
 624         if (read_only) {
 625                 if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 626                         return log_error_errno(errno, "Failed to remount %s read-only: %m", to);
 627         }
 628         return 1;
 629 }
 630
 631 static int mount_legacy_cgroups(
 632                 const char *dest,
 633                 bool userns, uid_t uid_shift, uid_t uid_range,
 634                 const char *selinux_apifs_context) {
 635
 636         _cleanup_set_free_free_ Set *controllers = NULL;
 637         const char *cgroup_root;
 638         int r;
 639
 640         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 641
 642         (void) mkdir_p(cgroup_root, 0755);
 643
 644         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 645         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 646         if (r < 0)
 647                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 648         if (r == 0) {
 649                 _cleanup_free_ char *options = NULL;
 650
 651                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, selinux_apifs_context, &options);
 652                 if (r < 0)
 653                         return log_oom();
 654
 655                 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
 656                         return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
 657         }
 658
 659         if (cg_unified() > 0)
 660                 goto skip_controllers;
 661
 662         controllers = set_new(&string_hash_ops);
 663         if (!controllers)
 664                 return log_oom();
 665
 666         r = cg_kernel_controllers(controllers);
 667         if (r < 0)
 668                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 669
 670         for (;;) {
 671                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 672
 673                 controller = set_steal_first(controllers);
 674                 if (!controller)
 675                         break;
 676
 677                 origin = prefix_root("/sys/fs/cgroup/", controller);
 678                 if (!origin)
 679                         return log_oom();
 680
 681                 r = readlink_malloc(origin, &combined);
 682                 if (r == -EINVAL) {
 683                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 684
 685                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
 686                         if (r < 0)
 687                                 return r;
 688
 689                 } else if (r < 0)
 690                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 691                 else {
 692                         _cleanup_free_ char *target = NULL;
 693
 694                         target = prefix_root(dest, origin);
 695                         if (!target)
 696                                 return log_oom();
 697
 698                         /* A symbolic link, a combination of controllers in one hierarchy */
 699
 700                         if (!filename_is_valid(combined)) {
 701                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 702                                 continue;
 703                         }
 704
 705                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
 706                         if (r < 0)
 707                                 return r;
 708
 709                         r = symlink_idempotent(combined, target);
 710                         if (r == -EINVAL) {
 711                                 log_error("Invalid existing symlink for combined hierarchy");
 712                                 return r;
 713                         }
 714                         if (r < 0)
 715                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 716                 }
 717         }
 718
 719 skip_controllers:
 720         r = mount_legacy_cgroup_hierarchy(dest, "none,name=systemd,xattr", "systemd", false);
 721         if (r < 0)
 722                 return r;
 723
 724         if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
 725                 return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
 726
 727         return 0;
 728 }
 729
 730 static int mount_unified_cgroups(const char *dest) {
 731         const char *p;
 732         int r;
 733
 734         assert(dest);
 735
 736         p = prefix_roota(dest, "/sys/fs/cgroup");
 737
 738         (void) mkdir_p(p, 0755);
 739
 740         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 741         if (r < 0)
 742                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 743         if (r > 0) {
 744                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 745                 if (access(p, F_OK) >= 0)
 746                         return 0;
 747                 if (errno != ENOENT)
 748                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 749
 750                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 751                 return -EINVAL;
 752         }
 753
 754         if (mount("cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
 755                 return log_error_errno(errno, "Failed to mount unified cgroup hierarchy to %s: %m", p);
 756
 757         return 0;
 758 }
 759
 760 int mount_cgroups(
 761                 const char *dest,
 762                 bool unified_requested,
 763                 bool userns, uid_t uid_shift, uid_t uid_range,
 764                 const char *selinux_apifs_context) {
 765
 766         if (unified_requested)
 767                 return mount_unified_cgroups(dest);
 768         else
 769                 return mount_legacy_cgroups(dest, userns, uid_shift, uid_range, selinux_apifs_context);
 770 }
 771
 772 int mount_systemd_cgroup_writable(
 773                 const char *dest,
 774                 bool unified_requested) {
 775
 776         _cleanup_free_ char *own_cgroup_path = NULL;
 777         const char *systemd_root, *systemd_own;
 778         int r;
 779
 780         assert(dest);
 781
 782         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
 783         if (r < 0)
 784                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
 785
 786         /* If we are living in the top-level, then there's nothing to do... */
 787         if (path_equal(own_cgroup_path, "/"))
 788                 return 0;
 789
 790         if (unified_requested) {
 791                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
 792                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
 793         } else {
 794                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
 795                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
 796         }
 797
 798         /* Make our own cgroup a (writable) bind mount */
 799         if (mount(systemd_own, systemd_own,  NULL, MS_BIND, NULL) < 0)
 800                 return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path);
 801
 802         /* And then remount the systemd cgroup root read-only */
 803         if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 804                 return log_error_errno(errno, "Failed to mount cgroup root read-only: %m");
 805
 806         return 0;
 807 }
 808
 809 int setup_volatile_state(
 810                 const char *directory,
 811                 VolatileMode mode,
 812                 bool userns, uid_t uid_shift, uid_t uid_range,
 813                 const char *selinux_apifs_context) {
 814
 815         _cleanup_free_ char *buf = NULL;
 816         const char *p, *options;
 817         int r;
 818
 819         assert(directory);
 820
 821         if (mode != VOLATILE_STATE)
 822                 return 0;
 823
 824         /* --volatile=state means we simply overmount /var
 825            with a tmpfs, and the rest read-only. */
 826
 827         r = bind_remount_recursive(directory, true);
 828         if (r < 0)
 829                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
 830
 831         p = prefix_roota(directory, "/var");
 832         r = mkdir(p, 0755);
 833         if (r < 0 && errno != EEXIST)
 834                 return log_error_errno(errno, "Failed to create %s: %m", directory);
 835
 836         options = "mode=755";
 837         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, selinux_apifs_context, &buf);
 838         if (r < 0)
 839                 return log_oom();
 840         if (r > 0)
 841                 options = buf;
 842
 843         if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, options) < 0)
 844                 return log_error_errno(errno, "Failed to mount tmpfs to /var: %m");
 845
 846         return 0;
 847 }
 848
 849 int setup_volatile(
 850                 const char *directory,
 851                 VolatileMode mode,
 852                 bool userns, uid_t uid_shift, uid_t uid_range,
 853                 const char *selinux_apifs_context) {
 854
 855         bool tmpfs_mounted = false, bind_mounted = false;
 856         char template[] = "/tmp/nspawn-volatile-XXXXXX";
 857         _cleanup_free_ char *buf = NULL;
 858         const char *f, *t, *options;
 859         int r;
 860
 861         assert(directory);
 862
 863         if (mode != VOLATILE_YES)
 864                 return 0;
 865
 866         /* --volatile=yes means we mount a tmpfs to the root dir, and
 867            the original /usr to use inside it, and that read-only. */
 868
 869         if (!mkdtemp(template))
 870                 return log_error_errno(errno, "Failed to create temporary directory: %m");
 871
 872         options = "mode=755";
 873         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, selinux_apifs_context, &buf);
 874         if (r < 0)
 875                 return log_oom();
 876         if (r > 0)
 877                 options = buf;
 878
 879         if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, options) < 0) {
 880                 r = log_error_errno(errno, "Failed to mount tmpfs for root directory: %m");
 881                 goto fail;
 882         }
 883
 884         tmpfs_mounted = true;
 885
 886         f = prefix_roota(directory, "/usr");
 887         t = prefix_roota(template, "/usr");
 888
 889         r = mkdir(t, 0755);
 890         if (r < 0 && errno != EEXIST) {
 891                 r = log_error_errno(errno, "Failed to create %s: %m", t);
 892                 goto fail;
 893         }
 894
 895         if (mount(f, t, NULL, MS_BIND|MS_REC, NULL) < 0) {
 896                 r = log_error_errno(errno, "Failed to create /usr bind mount: %m");
 897                 goto fail;
 898         }
 899
 900         bind_mounted = true;
 901
 902         r = bind_remount_recursive(t, true);
 903         if (r < 0) {
 904                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
 905                 goto fail;
 906         }
 907
 908         if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) {
 909                 r = log_error_errno(errno, "Failed to move root mount: %m");
 910                 goto fail;
 911         }
 912
 913         (void) rmdir(template);
 914
 915         return 0;
 916
 917 fail:
 918         if (bind_mounted)
 919                 (void) umount(t);
 920
 921         if (tmpfs_mounted)
 922                 (void) umount(template);
 923         (void) rmdir(template);
 924         return r;
 925 }
 926
 927 VolatileMode volatile_mode_from_string(const char *s) {
 928         int b;
 929
 930         if (isempty(s))
 931                 return _VOLATILE_MODE_INVALID;
 932
 933         b = parse_boolean(s);
 934         if (b > 0)
 935                 return VOLATILE_YES;
 936         if (b == 0)
 937                 return VOLATILE_NO;
 938
 939         if (streq(s, "state"))
 940                 return VOLATILE_STATE;
 941
 942         return _VOLATILE_MODE_INVALID;
 943 }