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                 { "/proc/sys/net",   "/proc/sys/net",   NULL,    NULL,        MS_BIND,                                                   true,  true,  true  },   /* (except for this) */
 303                 { 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 */
 304                 { "tmpfs",           "/sys",            "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true  },
 305                 { "sysfs",           "/sys",            "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 306                 { "tmpfs",           "/dev",            "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 307                 { "tmpfs",           "/dev/shm",        "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 308                 { "tmpfs",           "/run",            "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 309                 { "tmpfs",           "/tmp",            "tmpfs", "mode=1777", MS_STRICTATIME,                                            true,  false, false },
 310 #ifdef HAVE_SELINUX
 311                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL,     NULL,       MS_BIND,                                                   false, false, false },  /* Bind mount first */
 312                 { 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 */
 313 #endif
 314         };
 315
 316         unsigned k;
 317         int r;
 318
 319         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 320                 _cleanup_free_ char *where = NULL, *options = NULL;
 321                 const char *o;
 322
 323                 if (in_userns != mount_table[k].in_userns)
 324                         continue;
 325
 326                 if (!use_netns && mount_table[k].use_netns)
 327                         continue;
 328
 329                 where = prefix_root(dest, mount_table[k].where);
 330                 if (!where)
 331                         return log_oom();
 332
 333                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 334                 if (r < 0 && r != -ENOENT)
 335                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 336
 337                 /* Skip this entry if it is not a remount. */
 338                 if (mount_table[k].what && r > 0)
 339                         continue;
 340
 341                 r = mkdir_p(where, 0755);
 342                 if (r < 0) {
 343                         if (mount_table[k].fatal)
 344                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 345
 346                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 347                         continue;
 348                 }
 349
 350                 o = mount_table[k].options;
 351                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 352                         r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, selinux_apifs_context, &options);
 353                         if (r < 0)
 354                                 return log_oom();
 355                         if (r > 0)
 356                                 o = options;
 357                 }
 358
 359                 if (mount(mount_table[k].what,
 360                           where,
 361                           mount_table[k].type,
 362                           mount_table[k].flags,
 363                           o) < 0) {
 364
 365                         if (mount_table[k].fatal)
 366                                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 367
 368                         log_warning_errno(errno, "mount(%s) failed, ignoring: %m", where);
 369                 }
 370         }
 371
 372         return 0;
 373 }
 374
 375 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 376         const char *p = options;
 377         unsigned long flags = *mount_flags;
 378         char *opts = NULL;
 379
 380         assert(options);
 381
 382         for (;;) {
 383                 _cleanup_free_ char *word = NULL;
 384                 int r = extract_first_word(&p, &word, ",", 0);
 385                 if (r < 0)
 386                         return log_error_errno(r, "Failed to extract mount option: %m");
 387                 if (r == 0)
 388                         break;
 389
 390                 if (streq(word, "rbind"))
 391                         flags |= MS_REC;
 392                 else if (streq(word, "norbind"))
 393                         flags &= ~MS_REC;
 394                 else {
 395                         log_error("Invalid bind mount option: %s", word);
 396                         return -EINVAL;
 397                 }
 398         }
 399
 400         *mount_flags = flags;
 401         /* in the future mount_opts will hold string options for mount(2) */
 402         *mount_opts = opts;
 403
 404         return 0;
 405 }
 406
 407 static int mount_bind(const char *dest, CustomMount *m) {
 408         struct stat source_st, dest_st;
 409         const char *where;
 410         unsigned long mount_flags = MS_BIND | MS_REC;
 411         _cleanup_free_ char *mount_opts = NULL;
 412         int r;
 413
 414         assert(m);
 415
 416         if (m->options) {
 417                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 418                 if (r < 0)
 419                         return r;
 420         }
 421
 422         if (stat(m->source, &source_st) < 0)
 423                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 424
 425         where = prefix_roota(dest, m->destination);
 426
 427         if (stat(where, &dest_st) >= 0) {
 428                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 429                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 430                         return -EINVAL;
 431                 }
 432
 433                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 434                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 435                         return -EINVAL;
 436                 }
 437
 438         } else if (errno == ENOENT) {
 439                 r = mkdir_parents_label(where, 0755);
 440                 if (r < 0)
 441                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 442
 443                 /* Create the mount point. Any non-directory file can be
 444                 * mounted on any non-directory file (regular, fifo, socket,
 445                 * char, block).
 446                 */
 447                 if (S_ISDIR(source_st.st_mode))
 448                         r = mkdir_label(where, 0755);
 449                 else
 450                         r = touch(where);
 451                 if (r < 0)
 452                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 453
 454         } else {
 455                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 456         }
 457
 458         if (mount(m->source, where, NULL, mount_flags, mount_opts) < 0)
 459                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 460
 461         if (m->read_only) {
 462                 r = bind_remount_recursive(where, true);
 463                 if (r < 0)
 464                         return log_error_errno(r, "Read-only bind mount failed: %m");
 465         }
 466
 467         return 0;
 468 }
 469
 470 static int mount_tmpfs(
 471                 const char *dest,
 472                 CustomMount *m,
 473                 bool userns, uid_t uid_shift, uid_t uid_range,
 474                 const char *selinux_apifs_context) {
 475
 476         const char *where, *options;
 477         _cleanup_free_ char *buf = NULL;
 478         int r;
 479
 480         assert(dest);
 481         assert(m);
 482
 483         where = prefix_roota(dest, m->destination);
 484
 485         r = mkdir_p_label(where, 0755);
 486         if (r < 0 && r != -EEXIST)
 487                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 488
 489         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, selinux_apifs_context, &buf);
 490         if (r < 0)
 491                 return log_oom();
 492         options = r > 0 ? buf : m->options;
 493
 494         if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options) < 0)
 495                 return log_error_errno(errno, "tmpfs mount to %s failed: %m", where);
 496
 497         return 0;
 498 }
 499
 500 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 501         _cleanup_strv_free_ char **sv = NULL;
 502
 503         sv = strv_copy(lower);
 504         if (!sv)
 505                 return NULL;
 506
 507         strv_reverse(sv);
 508
 509         if (!strv_shell_escape(sv, ",:"))
 510                 return NULL;
 511
 512         return strv_join(sv, ":");
 513 }
 514
 515 static int mount_overlay(const char *dest, CustomMount *m) {
 516         _cleanup_free_ char *lower = NULL;
 517         const char *where, *options;
 518         int r;
 519
 520         assert(dest);
 521         assert(m);
 522
 523         where = prefix_roota(dest, m->destination);
 524
 525         r = mkdir_label(where, 0755);
 526         if (r < 0 && r != -EEXIST)
 527                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 528
 529         (void) mkdir_p_label(m->source, 0755);
 530
 531         lower = joined_and_escaped_lower_dirs(m->lower);
 532         if (!lower)
 533                 return log_oom();
 534
 535         if (m->read_only) {
 536                 _cleanup_free_ char *escaped_source = NULL;
 537
 538                 escaped_source = shell_escape(m->source, ",:");
 539                 if (!escaped_source)
 540                         return log_oom();
 541
 542                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 543         } else {
 544                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 545
 546                 assert(m->work_dir);
 547                 (void) mkdir_label(m->work_dir, 0700);
 548
 549                 escaped_source = shell_escape(m->source, ",:");
 550                 if (!escaped_source)
 551                         return log_oom();
 552                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 553                 if (!escaped_work_dir)
 554                         return log_oom();
 555
 556                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 557         }
 558
 559         if (mount("overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options) < 0)
 560                 return log_error_errno(errno, "overlay mount to %s failed: %m", where);
 561
 562         return 0;
 563 }
 564
 565 int mount_custom(
 566                 const char *dest,
 567                 CustomMount *mounts, unsigned n,
 568                 bool userns, uid_t uid_shift, uid_t uid_range,
 569                 const char *selinux_apifs_context) {
 570
 571         unsigned i;
 572         int r;
 573
 574         assert(dest);
 575
 576         for (i = 0; i < n; i++) {
 577                 CustomMount *m = mounts + i;
 578
 579                 switch (m->type) {
 580
 581                 case CUSTOM_MOUNT_BIND:
 582                         r = mount_bind(dest, m);
 583                         break;
 584
 585                 case CUSTOM_MOUNT_TMPFS:
 586                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 587                         break;
 588
 589                 case CUSTOM_MOUNT_OVERLAY:
 590                         r = mount_overlay(dest, m);
 591                         break;
 592
 593                 default:
 594                         assert_not_reached("Unknown custom mount type");
 595                 }
 596
 597                 if (r < 0)
 598                         return r;
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, bool read_only) {
 605         char *to;
 606         int r;
 607
 608         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 609
 610         r = path_is_mount_point(to, 0);
 611         if (r < 0 && r != -ENOENT)
 612                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 613         if (r > 0)
 614                 return 0;
 615
 616         mkdir_p(to, 0755);
 617
 618         /* The superblock mount options of the mount point need to be
 619          * identical to the hosts', and hence writable... */
 620         if (mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller) < 0)
 621                 return log_error_errno(errno, "Failed to mount to %s: %m", to);
 622
 623         /* ... hence let's only make the bind mount read-only, not the
 624          * superblock. */
 625         if (read_only) {
 626                 if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 627                         return log_error_errno(errno, "Failed to remount %s read-only: %m", to);
 628         }
 629         return 1;
 630 }
 631
 632 static int mount_legacy_cgroups(
 633                 const char *dest,
 634                 bool userns, uid_t uid_shift, uid_t uid_range,
 635                 const char *selinux_apifs_context) {
 636
 637         _cleanup_set_free_free_ Set *controllers = NULL;
 638         const char *cgroup_root;
 639         int r;
 640
 641         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 642
 643         (void) mkdir_p(cgroup_root, 0755);
 644
 645         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 646         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 647         if (r < 0)
 648                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 649         if (r == 0) {
 650                 _cleanup_free_ char *options = NULL;
 651
 652                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, selinux_apifs_context, &options);
 653                 if (r < 0)
 654                         return log_oom();
 655
 656                 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
 657                         return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
 658         }
 659
 660         if (cg_unified() > 0)
 661                 goto skip_controllers;
 662
 663         controllers = set_new(&string_hash_ops);
 664         if (!controllers)
 665                 return log_oom();
 666
 667         r = cg_kernel_controllers(controllers);
 668         if (r < 0)
 669                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 670
 671         for (;;) {
 672                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 673
 674                 controller = set_steal_first(controllers);
 675                 if (!controller)
 676                         break;
 677
 678                 origin = prefix_root("/sys/fs/cgroup/", controller);
 679                 if (!origin)
 680                         return log_oom();
 681
 682                 r = readlink_malloc(origin, &combined);
 683                 if (r == -EINVAL) {
 684                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 685
 686                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
 687                         if (r < 0)
 688                                 return r;
 689
 690                 } else if (r < 0)
 691                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 692                 else {
 693                         _cleanup_free_ char *target = NULL;
 694
 695                         target = prefix_root(dest, origin);
 696                         if (!target)
 697                                 return log_oom();
 698
 699                         /* A symbolic link, a combination of controllers in one hierarchy */
 700
 701                         if (!filename_is_valid(combined)) {
 702                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 703                                 continue;
 704                         }
 705
 706                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
 707                         if (r < 0)
 708                                 return r;
 709
 710                         r = symlink_idempotent(combined, target);
 711                         if (r == -EINVAL) {
 712                                 log_error("Invalid existing symlink for combined hierarchy");
 713                                 return r;
 714                         }
 715                         if (r < 0)
 716                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 717                 }
 718         }
 719
 720 skip_controllers:
 721         r = mount_legacy_cgroup_hierarchy(dest, "none,name=systemd,xattr", "systemd", false);
 722         if (r < 0)
 723                 return r;
 724
 725         if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
 726                 return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
 727
 728         return 0;
 729 }
 730
 731 static int mount_unified_cgroups(const char *dest) {
 732         const char *p;
 733         int r;
 734
 735         assert(dest);
 736
 737         p = prefix_roota(dest, "/sys/fs/cgroup");
 738
 739         (void) mkdir_p(p, 0755);
 740
 741         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 742         if (r < 0)
 743                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 744         if (r > 0) {
 745                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 746                 if (access(p, F_OK) >= 0)
 747                         return 0;
 748                 if (errno != ENOENT)
 749                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 750
 751                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 752                 return -EINVAL;
 753         }
 754
 755         if (mount("cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
 756                 return log_error_errno(errno, "Failed to mount unified cgroup hierarchy to %s: %m", p);
 757
 758         return 0;
 759 }
 760
 761 int mount_cgroups(
 762                 const char *dest,
 763                 bool unified_requested,
 764                 bool userns, uid_t uid_shift, uid_t uid_range,
 765                 const char *selinux_apifs_context) {
 766
 767         if (unified_requested)
 768                 return mount_unified_cgroups(dest);
 769         else
 770                 return mount_legacy_cgroups(dest, userns, uid_shift, uid_range, selinux_apifs_context);
 771 }
 772
 773 int mount_systemd_cgroup_writable(
 774                 const char *dest,
 775                 bool unified_requested) {
 776
 777         _cleanup_free_ char *own_cgroup_path = NULL;
 778         const char *systemd_root, *systemd_own;
 779         int r;
 780
 781         assert(dest);
 782
 783         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
 784         if (r < 0)
 785                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
 786
 787         /* If we are living in the top-level, then there's nothing to do... */
 788         if (path_equal(own_cgroup_path, "/"))
 789                 return 0;
 790
 791         if (unified_requested) {
 792                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
 793                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
 794         } else {
 795                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
 796                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
 797         }
 798
 799         /* Make our own cgroup a (writable) bind mount */
 800         if (mount(systemd_own, systemd_own,  NULL, MS_BIND, NULL) < 0)
 801                 return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path);
 802
 803         /* And then remount the systemd cgroup root read-only */
 804         if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 805                 return log_error_errno(errno, "Failed to mount cgroup root read-only: %m");
 806
 807         return 0;
 808 }
 809
 810 int setup_volatile_state(
 811                 const char *directory,
 812                 VolatileMode mode,
 813                 bool userns, uid_t uid_shift, uid_t uid_range,
 814                 const char *selinux_apifs_context) {
 815
 816         _cleanup_free_ char *buf = NULL;
 817         const char *p, *options;
 818         int r;
 819
 820         assert(directory);
 821
 822         if (mode != VOLATILE_STATE)
 823                 return 0;
 824
 825         /* --volatile=state means we simply overmount /var
 826            with a tmpfs, and the rest read-only. */
 827
 828         r = bind_remount_recursive(directory, true);
 829         if (r < 0)
 830                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
 831
 832         p = prefix_roota(directory, "/var");
 833         r = mkdir(p, 0755);
 834         if (r < 0 && errno != EEXIST)
 835                 return log_error_errno(errno, "Failed to create %s: %m", directory);
 836
 837         options = "mode=755";
 838         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, selinux_apifs_context, &buf);
 839         if (r < 0)
 840                 return log_oom();
 841         if (r > 0)
 842                 options = buf;
 843
 844         if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, options) < 0)
 845                 return log_error_errno(errno, "Failed to mount tmpfs to /var: %m");
 846
 847         return 0;
 848 }
 849
 850 int setup_volatile(
 851                 const char *directory,
 852                 VolatileMode mode,
 853                 bool userns, uid_t uid_shift, uid_t uid_range,
 854                 const char *selinux_apifs_context) {
 855
 856         bool tmpfs_mounted = false, bind_mounted = false;
 857         char template[] = "/tmp/nspawn-volatile-XXXXXX";
 858         _cleanup_free_ char *buf = NULL;
 859         const char *f, *t, *options;
 860         int r;
 861
 862         assert(directory);
 863
 864         if (mode != VOLATILE_YES)
 865                 return 0;
 866
 867         /* --volatile=yes means we mount a tmpfs to the root dir, and
 868            the original /usr to use inside it, and that read-only. */
 869
 870         if (!mkdtemp(template))
 871                 return log_error_errno(errno, "Failed to create temporary directory: %m");
 872
 873         options = "mode=755";
 874         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, selinux_apifs_context, &buf);
 875         if (r < 0)
 876                 return log_oom();
 877         if (r > 0)
 878                 options = buf;
 879
 880         if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, options) < 0) {
 881                 r = log_error_errno(errno, "Failed to mount tmpfs for root directory: %m");
 882                 goto fail;
 883         }
 884
 885         tmpfs_mounted = true;
 886
 887         f = prefix_roota(directory, "/usr");
 888         t = prefix_roota(template, "/usr");
 889
 890         r = mkdir(t, 0755);
 891         if (r < 0 && errno != EEXIST) {
 892                 r = log_error_errno(errno, "Failed to create %s: %m", t);
 893                 goto fail;
 894         }
 895
 896         if (mount(f, t, NULL, MS_BIND|MS_REC, NULL) < 0) {
 897                 r = log_error_errno(errno, "Failed to create /usr bind mount: %m");
 898                 goto fail;
 899         }
 900
 901         bind_mounted = true;
 902
 903         r = bind_remount_recursive(t, true);
 904         if (r < 0) {
 905                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
 906                 goto fail;
 907         }
 908
 909         if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) {
 910                 r = log_error_errno(errno, "Failed to move root mount: %m");
 911                 goto fail;
 912         }
 913
 914         (void) rmdir(template);
 915
 916         return 0;
 917
 918 fail:
 919         if (bind_mounted)
 920                 (void) umount(t);
 921
 922         if (tmpfs_mounted)
 923                 (void) umount(template);
 924         (void) rmdir(template);
 925         return r;
 926 }
 927
 928 VolatileMode volatile_mode_from_string(const char *s) {
 929         int b;
 930
 931         if (isempty(s))
 932                 return _VOLATILE_MODE_INVALID;
 933
 934         b = parse_boolean(s);
 935         if (b > 0)
 936                 return VOLATILE_YES;
 937         if (b == 0)
 938                 return VOLATILE_NO;
 939
 940         if (streq(s, "state"))
 941                 return VOLATILE_STATE;
 942
 943         return _VOLATILE_MODE_INVALID;
 944 }