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(*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                 strv_free(m->lower);
  79         }
  80
  81         free(l);
  82 }
  83
  84 int custom_mount_compare(const void *a, const void *b) {
  85         const CustomMount *x = a, *y = b;
  86         int r;
  87
  88         r = path_compare(x->destination, y->destination);
  89         if (r != 0)
  90                 return r;
  91
  92         if (x->type < y->type)
  93                 return -1;
  94         if (x->type > y->type)
  95                 return 1;
  96
  97         return 0;
  98 }
  99
 100 int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) {
 101         _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
 102         const char *p = s;
 103         CustomMount *m;
 104         int r;
 105
 106         assert(l);
 107         assert(n);
 108
 109         r = extract_many_words(&p, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
 110         if (r < 0)
 111                 return r;
 112         if (r == 0)
 113                 return -EINVAL;
 114
 115         if (r == 1) {
 116                 destination = strdup(source);
 117                 if (!destination)
 118                         return -ENOMEM;
 119         }
 120
 121         if (r == 2 && !isempty(p)) {
 122                 opts = strdup(p);
 123                 if (!opts)
 124                         return -ENOMEM;
 125         }
 126
 127         if (!path_is_absolute(source))
 128                 return -EINVAL;
 129
 130         if (!path_is_absolute(destination))
 131                 return -EINVAL;
 132
 133         m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
 134         if (!m)
 135                 return log_oom();
 136
 137         m->source = source;
 138         m->destination = destination;
 139         m->read_only = read_only;
 140         m->options = opts;
 141
 142         source = destination = opts = NULL;
 143         return 0;
 144 }
 145
 146 int tmpfs_mount_parse(CustomMount **l, unsigned *n, const char *s) {
 147         _cleanup_free_ char *path = NULL, *opts = NULL;
 148         const char *p = s;
 149         CustomMount *m;
 150         int r;
 151
 152         assert(l);
 153         assert(n);
 154         assert(s);
 155
 156         r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
 157         if (r < 0)
 158                 return r;
 159         if (r == 0)
 160                 return -EINVAL;
 161
 162         if (isempty(p))
 163                 opts = strdup("mode=0755");
 164         else
 165                 opts = strdup(p);
 166         if (!opts)
 167                 return -ENOMEM;
 168
 169         if (!path_is_absolute(path))
 170                 return -EINVAL;
 171
 172         m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
 173         if (!m)
 174                 return -ENOMEM;
 175
 176         m->destination = path;
 177         m->options = opts;
 178
 179         path = opts = NULL;
 180         return 0;
 181 }
 182
 183 static int tmpfs_patch_options(
 184                 const char *options,
 185                 bool userns,
 186                 uid_t uid_shift, uid_t uid_range,
 187                 bool patch_ids,
 188                 const char *selinux_apifs_context,
 189                 char **ret) {
 190
 191         char *buf = NULL;
 192
 193         if ((userns && uid_shift != 0) || patch_ids) {
 194                 assert(uid_shift != UID_INVALID);
 195
 196                 if (options)
 197                         (void) asprintf(&buf, "%s,uid=" UID_FMT ",gid=" UID_FMT, options, uid_shift, uid_shift);
 198                 else
 199                         (void) asprintf(&buf, "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift);
 200                 if (!buf)
 201                         return -ENOMEM;
 202
 203                 options = buf;
 204         }
 205
 206 #ifdef HAVE_SELINUX
 207         if (selinux_apifs_context) {
 208                 char *t;
 209
 210                 if (options)
 211                         t = strjoin(options, ",context=\"", selinux_apifs_context, "\"", NULL);
 212                 else
 213                         t = strjoin("context=\"", selinux_apifs_context, "\"", NULL);
 214                 if (!t) {
 215                         free(buf);
 216                         return -ENOMEM;
 217                 }
 218
 219                 free(buf);
 220                 buf = t;
 221         }
 222 #endif
 223
 224         if (!buf && options) {
 225                 buf = strdup(options);
 226                 if (!buf)
 227                         return -ENOMEM;
 228         }
 229         *ret = buf;
 230
 231         return !!buf;
 232 }
 233
 234 int mount_sysfs(const char *dest) {
 235         const char *full, *top, *x;
 236         int r;
 237
 238         top = prefix_roota(dest, "/sys");
 239         r = path_check_fstype(top, SYSFS_MAGIC);
 240         if (r < 0)
 241                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 242         /* /sys might already be mounted as sysfs by the outer child in the
 243          * !netns case. In this case, it's all good. Don't touch it because we
 244          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 245          */
 246         if (r > 0)
 247                 return 0;
 248
 249         full = prefix_roota(top, "/full");
 250
 251         (void) mkdir(full, 0755);
 252
 253         if (mount("sysfs", full, "sysfs", MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
 254                 return log_error_errno(errno, "Failed to mount sysfs to %s: %m", full);
 255
 256         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 257                 _cleanup_free_ char *from = NULL, *to = NULL;
 258
 259                 from = prefix_root(full, x);
 260                 if (!from)
 261                         return log_oom();
 262
 263                 to = prefix_root(top, x);
 264                 if (!to)
 265                         return log_oom();
 266
 267                 (void) mkdir(to, 0755);
 268
 269                 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
 270                         return log_error_errno(errno, "Failed to mount /sys/%s into place: %m", x);
 271
 272                 if (mount(NULL, to, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL) < 0)
 273                         return log_error_errno(errno, "Failed to mount /sys/%s read-only: %m", x);
 274         }
 275
 276         if (umount(full) < 0)
 277                 return log_error_errno(errno, "Failed to unmount %s: %m", full);
 278
 279         if (rmdir(full) < 0)
 280                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 281
 282         x = prefix_roota(top, "/fs/kdbus");
 283         (void) mkdir_p(x, 0755);
 284
 285         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 286          * remount /sys read-only.
 287          */
 288         if (cg_ns_supported()) {
 289                 x = prefix_roota(top, "/fs/cgroup");
 290                 (void) mkdir_p(x, 0755);
 291         }
 292
 293         if (mount(NULL, top, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL) < 0)
 294                 return log_error_errno(errno, "Failed to make %s read-only: %m", top);
 295
 296         return 0;
 297 }
 298
 299 int mount_all(const char *dest,
 300               bool use_userns, bool in_userns,
 301               bool use_netns,
 302               uid_t uid_shift, uid_t uid_range,
 303               const char *selinux_apifs_context) {
 304
 305         typedef struct MountPoint {
 306                 const char *what;
 307                 const char *where;
 308                 const char *type;
 309                 const char *options;
 310                 unsigned long flags;
 311                 bool fatal;
 312                 bool in_userns;
 313                 bool use_netns;
 314         } MountPoint;
 315
 316         static const MountPoint mount_table[] = {
 317                 { "proc",            "/proc",           "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  true,  false },
 318                 { "/proc/sys",       "/proc/sys",       NULL,    NULL,        MS_BIND,                                                   true,  true,  false },   /* Bind mount first ...*/
 319                 { "/proc/sys/net",   "/proc/sys/net",   NULL,    NULL,        MS_BIND,                                                   true,  true,  true  },   /* (except for this) */
 320                 { 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 */
 321                 { "tmpfs",           "/sys",            "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true  },
 322                 { "sysfs",           "/sys",            "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 323                 { "tmpfs",           "/dev",            "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 324                 { "tmpfs",           "/dev/shm",        "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 325                 { "tmpfs",           "/run",            "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 326                 { "tmpfs",           "/tmp",            "tmpfs", "mode=1777", MS_STRICTATIME,                                            true,  false, false },
 327 #ifdef HAVE_SELINUX
 328                 { "/sys/fs/selinux", "/sys/fs/selinux", NULL,     NULL,       MS_BIND,                                                   false, false, false },  /* Bind mount first */
 329                 { 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 */
 330 #endif
 331         };
 332
 333         unsigned k;
 334         int r;
 335
 336         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 337                 _cleanup_free_ char *where = NULL, *options = NULL;
 338                 const char *o;
 339
 340                 if (in_userns != mount_table[k].in_userns)
 341                         continue;
 342
 343                 if (!use_netns && mount_table[k].use_netns)
 344                         continue;
 345
 346                 where = prefix_root(dest, mount_table[k].where);
 347                 if (!where)
 348                         return log_oom();
 349
 350                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 351                 if (r < 0 && r != -ENOENT)
 352                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 353
 354                 /* Skip this entry if it is not a remount. */
 355                 if (mount_table[k].what && r > 0)
 356                         continue;
 357
 358                 r = mkdir_p(where, 0755);
 359                 if (r < 0) {
 360                         if (mount_table[k].fatal)
 361                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 362
 363                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 364                         continue;
 365                 }
 366
 367                 o = mount_table[k].options;
 368                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 369                         r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 370                         if (r < 0)
 371                                 return log_oom();
 372                         if (r > 0)
 373                                 o = options;
 374                 }
 375
 376                 if (mount(mount_table[k].what,
 377                           where,
 378                           mount_table[k].type,
 379                           mount_table[k].flags,
 380                           o) < 0) {
 381
 382                         if (mount_table[k].fatal)
 383                                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 384
 385                         log_warning_errno(errno, "mount(%s) failed, ignoring: %m", where);
 386                 }
 387         }
 388
 389         return 0;
 390 }
 391
 392 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 393         const char *p = options;
 394         unsigned long flags = *mount_flags;
 395         char *opts = NULL;
 396
 397         assert(options);
 398
 399         for (;;) {
 400                 _cleanup_free_ char *word = NULL;
 401                 int r = extract_first_word(&p, &word, ",", 0);
 402                 if (r < 0)
 403                         return log_error_errno(r, "Failed to extract mount option: %m");
 404                 if (r == 0)
 405                         break;
 406
 407                 if (streq(word, "rbind"))
 408                         flags |= MS_REC;
 409                 else if (streq(word, "norbind"))
 410                         flags &= ~MS_REC;
 411                 else {
 412                         log_error("Invalid bind mount option: %s", word);
 413                         return -EINVAL;
 414                 }
 415         }
 416
 417         *mount_flags = flags;
 418         /* in the future mount_opts will hold string options for mount(2) */
 419         *mount_opts = opts;
 420
 421         return 0;
 422 }
 423
 424 static int mount_bind(const char *dest, CustomMount *m) {
 425         struct stat source_st, dest_st;
 426         const char *where;
 427         unsigned long mount_flags = MS_BIND | MS_REC;
 428         _cleanup_free_ char *mount_opts = NULL;
 429         int r;
 430
 431         assert(m);
 432
 433         if (m->options) {
 434                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 435                 if (r < 0)
 436                         return r;
 437         }
 438
 439         if (stat(m->source, &source_st) < 0)
 440                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 441
 442         where = prefix_roota(dest, m->destination);
 443
 444         if (stat(where, &dest_st) >= 0) {
 445                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 446                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 447                         return -EINVAL;
 448                 }
 449
 450                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 451                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 452                         return -EINVAL;
 453                 }
 454
 455         } else if (errno == ENOENT) {
 456                 r = mkdir_parents_label(where, 0755);
 457                 if (r < 0)
 458                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 459
 460                 /* Create the mount point. Any non-directory file can be
 461                 * mounted on any non-directory file (regular, fifo, socket,
 462                 * char, block).
 463                 */
 464                 if (S_ISDIR(source_st.st_mode))
 465                         r = mkdir_label(where, 0755);
 466                 else
 467                         r = touch(where);
 468                 if (r < 0)
 469                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 470
 471         } else {
 472                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 473         }
 474
 475         if (mount(m->source, where, NULL, mount_flags, mount_opts) < 0)
 476                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 477
 478         if (m->read_only) {
 479                 r = bind_remount_recursive(where, true, NULL);
 480                 if (r < 0)
 481                         return log_error_errno(r, "Read-only bind mount failed: %m");
 482         }
 483
 484         return 0;
 485 }
 486
 487 static int mount_tmpfs(
 488                 const char *dest,
 489                 CustomMount *m,
 490                 bool userns, uid_t uid_shift, uid_t uid_range,
 491                 const char *selinux_apifs_context) {
 492
 493         const char *where, *options;
 494         _cleanup_free_ char *buf = NULL;
 495         int r;
 496
 497         assert(dest);
 498         assert(m);
 499
 500         where = prefix_roota(dest, m->destination);
 501
 502         r = mkdir_p_label(where, 0755);
 503         if (r < 0 && r != -EEXIST)
 504                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 505
 506         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 507         if (r < 0)
 508                 return log_oom();
 509         options = r > 0 ? buf : m->options;
 510
 511         if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options) < 0)
 512                 return log_error_errno(errno, "tmpfs mount to %s failed: %m", where);
 513
 514         return 0;
 515 }
 516
 517 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 518         _cleanup_strv_free_ char **sv = NULL;
 519
 520         sv = strv_copy(lower);
 521         if (!sv)
 522                 return NULL;
 523
 524         strv_reverse(sv);
 525
 526         if (!strv_shell_escape(sv, ",:"))
 527                 return NULL;
 528
 529         return strv_join(sv, ":");
 530 }
 531
 532 static int mount_overlay(const char *dest, CustomMount *m) {
 533         _cleanup_free_ char *lower = NULL;
 534         const char *where, *options;
 535         int r;
 536
 537         assert(dest);
 538         assert(m);
 539
 540         where = prefix_roota(dest, m->destination);
 541
 542         r = mkdir_label(where, 0755);
 543         if (r < 0 && r != -EEXIST)
 544                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 545
 546         (void) mkdir_p_label(m->source, 0755);
 547
 548         lower = joined_and_escaped_lower_dirs(m->lower);
 549         if (!lower)
 550                 return log_oom();
 551
 552         if (m->read_only) {
 553                 _cleanup_free_ char *escaped_source = NULL;
 554
 555                 escaped_source = shell_escape(m->source, ",:");
 556                 if (!escaped_source)
 557                         return log_oom();
 558
 559                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 560         } else {
 561                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 562
 563                 assert(m->work_dir);
 564                 (void) mkdir_label(m->work_dir, 0700);
 565
 566                 escaped_source = shell_escape(m->source, ",:");
 567                 if (!escaped_source)
 568                         return log_oom();
 569                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 570                 if (!escaped_work_dir)
 571                         return log_oom();
 572
 573                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 574         }
 575
 576         if (mount("overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options) < 0)
 577                 return log_error_errno(errno, "overlay mount to %s failed: %m", where);
 578
 579         return 0;
 580 }
 581
 582 int mount_custom(
 583                 const char *dest,
 584                 CustomMount *mounts, unsigned n,
 585                 bool userns, uid_t uid_shift, uid_t uid_range,
 586                 const char *selinux_apifs_context) {
 587
 588         unsigned i;
 589         int r;
 590
 591         assert(dest);
 592
 593         for (i = 0; i < n; i++) {
 594                 CustomMount *m = mounts + i;
 595
 596                 switch (m->type) {
 597
 598                 case CUSTOM_MOUNT_BIND:
 599                         r = mount_bind(dest, m);
 600                         break;
 601
 602                 case CUSTOM_MOUNT_TMPFS:
 603                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 604                         break;
 605
 606                 case CUSTOM_MOUNT_OVERLAY:
 607                         r = mount_overlay(dest, m);
 608                         break;
 609
 610                 default:
 611                         assert_not_reached("Unknown custom mount type");
 612                 }
 613
 614                 if (r < 0)
 615                         return r;
 616         }
 617
 618         return 0;
 619 }
 620
 621 /* Retrieve existing subsystems. This function is called in a new cgroup
 622  * namespace.
 623  */
 624 static int get_controllers(Set *subsystems) {
 625         _cleanup_fclose_ FILE *f = NULL;
 626         char line[LINE_MAX];
 627
 628         assert(subsystems);
 629
 630         f = fopen("/proc/self/cgroup", "re");
 631         if (!f)
 632                 return errno == ENOENT ? -ESRCH : -errno;
 633
 634         FOREACH_LINE(line, f, return -errno) {
 635                 int r;
 636                 char *e, *l, *p;
 637
 638                 truncate_nl(line);
 639
 640                 l = strchr(line, ':');
 641                 if (!l)
 642                         continue;
 643
 644                 l++;
 645                 e = strchr(l, ':');
 646                 if (!e)
 647                         continue;
 648
 649                 *e = 0;
 650
 651                 if (streq(l, "") || streq(l, "name=systemd"))
 652                         continue;
 653
 654                 p = strdup(l);
 655                 r = set_consume(subsystems, p);
 656                 if (r < 0)
 657                         return r;
 658         }
 659
 660         return 0;
 661 }
 662
 663 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy,
 664                                          CGroupUnified unified_requested, bool read_only) {
 665         char *to;
 666         int r;
 667
 668         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 669
 670         r = path_is_mount_point(to, 0);
 671         if (r < 0 && r != -ENOENT)
 672                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 673         if (r > 0)
 674                 return 0;
 675
 676         mkdir_p(to, 0755);
 677
 678         /* The superblock mount options of the mount point need to be
 679          * identical to the hosts', and hence writable... */
 680         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 681                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD)
 682                         r = mount("cgroup", to, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 683                 else
 684                         r = mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
 685         } else
 686                 r = mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller);
 687
 688         if (r < 0)
 689                 return log_error_errno(errno, "Failed to mount to %s: %m", to);
 690
 691         /* ... hence let's only make the bind mount read-only, not the
 692          * superblock. */
 693         if (read_only) {
 694                 if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 695                         return log_error_errno(errno, "Failed to remount %s read-only: %m", to);
 696         }
 697         return 1;
 698 }
 699
 700 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 701 static int mount_legacy_cgns_supported(
 702                 CGroupUnified unified_requested, bool userns, uid_t uid_shift,
 703                 uid_t uid_range, const char *selinux_apifs_context) {
 704         _cleanup_set_free_free_ Set *controllers = NULL;
 705         const char *cgroup_root = "/sys/fs/cgroup", *c;
 706         int r;
 707
 708         (void) mkdir_p(cgroup_root, 0755);
 709
 710         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 711         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 712         if (r < 0)
 713                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 714         if (r == 0) {
 715                 _cleanup_free_ char *options = NULL;
 716
 717                 /* When cgroup namespaces are enabled and user namespaces are
 718                  * used then the mount of the cgroupfs is done *inside* the new
 719                  * user namespace. We're root in the new user namespace and the
 720                  * kernel will happily translate our uid/gid to the correct
 721                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 722                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 723                  */
 724                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 725                 if (r < 0)
 726                         return log_oom();
 727
 728                 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
 729                         return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
 730         }
 731
 732         if (cg_all_unified() > 0)
 733                 goto skip_controllers;
 734
 735         controllers = set_new(&string_hash_ops);
 736         if (!controllers)
 737                 return log_oom();
 738
 739         r = get_controllers(controllers);
 740         if (r < 0)
 741                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 742
 743         for (;;) {
 744                 _cleanup_free_ const char *controller = NULL;
 745
 746                 controller = set_steal_first(controllers);
 747                 if (!controller)
 748                         break;
 749
 750                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 751                 if (r < 0)
 752                         return r;
 753
 754                 /* When multiple hierarchies are co-mounted, make their
 755                  * constituting individual hierarchies a symlink to the
 756                  * co-mount.
 757                  */
 758                 c = controller;
 759                 for (;;) {
 760                         _cleanup_free_ char *target = NULL, *tok = NULL;
 761
 762                         r = extract_first_word(&c, &tok, ",", 0);
 763                         if (r < 0)
 764                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 765                         if (r == 0)
 766                                 break;
 767
 768                         target = prefix_root("/sys/fs/cgroup", tok);
 769                         if (!target)
 770                                 return log_oom();
 771
 772                         if (streq(controller, tok))
 773                                 break;
 774
 775                         r = symlink_idempotent(controller, target);
 776                         if (r == -EINVAL)
 777                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 778                         if (r < 0)
 779                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 780                 }
 781         }
 782
 783 skip_controllers:
 784         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 785         if (r < 0)
 786                 return r;
 787
 788         if (!userns) {
 789                 if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
 790                         return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
 791         }
 792
 793         return 0;
 794 }
 795
 796 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 797 static int mount_legacy_cgns_unsupported(
 798                 const char *dest,
 799                 CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range,
 800                 const char *selinux_apifs_context) {
 801         _cleanup_set_free_free_ Set *controllers = NULL;
 802         const char *cgroup_root;
 803         int r;
 804
 805         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 806
 807         (void) mkdir_p(cgroup_root, 0755);
 808
 809         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 810         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 811         if (r < 0)
 812                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 813         if (r == 0) {
 814                 _cleanup_free_ char *options = NULL;
 815
 816                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 817                 if (r < 0)
 818                         return log_oom();
 819
 820                 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
 821                         return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
 822         }
 823
 824         if (cg_all_unified() > 0)
 825                 goto skip_controllers;
 826
 827         controllers = set_new(&string_hash_ops);
 828         if (!controllers)
 829                 return log_oom();
 830
 831         r = cg_kernel_controllers(controllers);
 832         if (r < 0)
 833                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 834
 835         for (;;) {
 836                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 837
 838                 controller = set_steal_first(controllers);
 839                 if (!controller)
 840                         break;
 841
 842                 origin = prefix_root("/sys/fs/cgroup/", controller);
 843                 if (!origin)
 844                         return log_oom();
 845
 846                 r = readlink_malloc(origin, &combined);
 847                 if (r == -EINVAL) {
 848                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 849
 850                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
 851                         if (r < 0)
 852                                 return r;
 853
 854                 } else if (r < 0)
 855                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 856                 else {
 857                         _cleanup_free_ char *target = NULL;
 858
 859                         target = prefix_root(dest, origin);
 860                         if (!target)
 861                                 return log_oom();
 862
 863                         /* A symbolic link, a combination of controllers in one hierarchy */
 864
 865                         if (!filename_is_valid(combined)) {
 866                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 867                                 continue;
 868                         }
 869
 870                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
 871                         if (r < 0)
 872                                 return r;
 873
 874                         r = symlink_idempotent(combined, target);
 875                         if (r == -EINVAL)
 876                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 877                         if (r < 0)
 878                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 879                 }
 880         }
 881
 882 skip_controllers:
 883         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 884         if (r < 0)
 885                 return r;
 886
 887         if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
 888                 return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
 889
 890         return 0;
 891 }
 892
 893 static int mount_unified_cgroups(const char *dest) {
 894         const char *p;
 895         int r;
 896
 897         assert(dest);
 898
 899         p = prefix_roota(dest, "/sys/fs/cgroup");
 900
 901         (void) mkdir_p(p, 0755);
 902
 903         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 904         if (r < 0)
 905                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 906         if (r > 0) {
 907                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 908                 if (access(p, F_OK) >= 0)
 909                         return 0;
 910                 if (errno != ENOENT)
 911                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 912
 913                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 914                 return -EINVAL;
 915         }
 916
 917         if (mount("cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
 918                 return log_error_errno(errno, "Failed to mount unified cgroup hierarchy to %s: %m", p);
 919
 920         return 0;
 921 }
 922
 923 int mount_cgroups(
 924                 const char *dest,
 925                 CGroupUnified unified_requested,
 926                 bool userns, uid_t uid_shift, uid_t uid_range,
 927                 const char *selinux_apifs_context,
 928                 bool use_cgns) {
 929
 930         if (unified_requested >= CGROUP_UNIFIED_ALL)
 931                 return mount_unified_cgroups(dest);
 932         else if (use_cgns && cg_ns_supported())
 933                 return mount_legacy_cgns_supported(unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 934
 935         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 936 }
 937
 938 int mount_systemd_cgroup_writable(
 939                 const char *dest,
 940                 CGroupUnified unified_requested) {
 941
 942         _cleanup_free_ char *own_cgroup_path = NULL;
 943         const char *systemd_root, *systemd_own;
 944         int r;
 945
 946         assert(dest);
 947
 948         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
 949         if (r < 0)
 950                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
 951
 952         /* If we are living in the top-level, then there's nothing to do... */
 953         if (path_equal(own_cgroup_path, "/"))
 954                 return 0;
 955
 956         if (unified_requested >= CGROUP_UNIFIED_ALL) {
 957                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
 958                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
 959         } else {
 960                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
 961                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
 962         }
 963
 964         /* Make our own cgroup a (writable) bind mount */
 965         if (mount(systemd_own, systemd_own,  NULL, MS_BIND, NULL) < 0)
 966                 return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path);
 967
 968         /* And then remount the systemd cgroup root read-only */
 969         if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 970                 return log_error_errno(errno, "Failed to mount cgroup root read-only: %m");
 971
 972         return 0;
 973 }
 974
 975 int setup_volatile_state(
 976                 const char *directory,
 977                 VolatileMode mode,
 978                 bool userns, uid_t uid_shift, uid_t uid_range,
 979                 const char *selinux_apifs_context) {
 980
 981         _cleanup_free_ char *buf = NULL;
 982         const char *p, *options;
 983         int r;
 984
 985         assert(directory);
 986
 987         if (mode != VOLATILE_STATE)
 988                 return 0;
 989
 990         /* --volatile=state means we simply overmount /var
 991            with a tmpfs, and the rest read-only. */
 992
 993         r = bind_remount_recursive(directory, true, NULL);
 994         if (r < 0)
 995                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
 996
 997         p = prefix_roota(directory, "/var");
 998         r = mkdir(p, 0755);
 999         if (r < 0 && errno != EEXIST)
1000                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1001
1002         options = "mode=755";
1003         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1004         if (r < 0)
1005                 return log_oom();
1006         if (r > 0)
1007                 options = buf;
1008
1009         if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, options) < 0)
1010                 return log_error_errno(errno, "Failed to mount tmpfs to /var: %m");
1011
1012         return 0;
1013 }
1014
1015 int setup_volatile(
1016                 const char *directory,
1017                 VolatileMode mode,
1018                 bool userns, uid_t uid_shift, uid_t uid_range,
1019                 const char *selinux_apifs_context) {
1020
1021         bool tmpfs_mounted = false, bind_mounted = false;
1022         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1023         _cleanup_free_ char *buf = NULL;
1024         const char *f, *t, *options;
1025         int r;
1026
1027         assert(directory);
1028
1029         if (mode != VOLATILE_YES)
1030                 return 0;
1031
1032         /* --volatile=yes means we mount a tmpfs to the root dir, and
1033            the original /usr to use inside it, and that read-only. */
1034
1035         if (!mkdtemp(template))
1036                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1037
1038         options = "mode=755";
1039         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1040         if (r < 0)
1041                 return log_oom();
1042         if (r > 0)
1043                 options = buf;
1044
1045         if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, options) < 0) {
1046                 r = log_error_errno(errno, "Failed to mount tmpfs for root directory: %m");
1047                 goto fail;
1048         }
1049
1050         tmpfs_mounted = true;
1051
1052         f = prefix_roota(directory, "/usr");
1053         t = prefix_roota(template, "/usr");
1054
1055         r = mkdir(t, 0755);
1056         if (r < 0 && errno != EEXIST) {
1057                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1058                 goto fail;
1059         }
1060
1061         if (mount(f, t, NULL, MS_BIND|MS_REC, NULL) < 0) {
1062                 r = log_error_errno(errno, "Failed to create /usr bind mount: %m");
1063                 goto fail;
1064         }
1065
1066         bind_mounted = true;
1067
1068         r = bind_remount_recursive(t, true, NULL);
1069         if (r < 0) {
1070                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1071                 goto fail;
1072         }
1073
1074         if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) {
1075                 r = log_error_errno(errno, "Failed to move root mount: %m");
1076                 goto fail;
1077         }
1078
1079         (void) rmdir(template);
1080
1081         return 0;
1082
1083 fail:
1084         if (bind_mounted)
1085                 (void) umount(t);
1086
1087         if (tmpfs_mounted)
1088                 (void) umount(template);
1089         (void) rmdir(template);
1090         return r;
1091 }
1092
1093 VolatileMode volatile_mode_from_string(const char *s) {
1094         int b;
1095
1096         if (isempty(s))
1097                 return _VOLATILE_MODE_INVALID;
1098
1099         b = parse_boolean(s);
1100         if (b > 0)
1101                 return VOLATILE_YES;
1102         if (b == 0)
1103                 return VOLATILE_NO;
1104
1105         if (streq(s, "state"))
1106                 return VOLATILE_STATE;
1107
1108         return _VOLATILE_MODE_INVALID;
1109 }