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                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                   false, true,  false },   /* Bind mount first ...*/
 322                 { NULL,                  "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, false, true,  false },   /* ... then, make it r/o */
 323                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true  },
 324                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 325                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 326                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 327                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 328                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            true,  true,  false },
 329 #ifdef HAVE_SELINUX
 330                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   false, false, false },  /* Bind mount first */
 331                 { 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 */
 332 #endif
 333         };
 334
 335         unsigned k;
 336         int r;
 337
 338         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 339                 _cleanup_free_ char *where = NULL, *options = NULL;
 340                 const char *o;
 341
 342                 if (in_userns != mount_table[k].in_userns)
 343                         continue;
 344
 345                 if (!use_netns && mount_table[k].use_netns)
 346                         continue;
 347
 348                 where = prefix_root(dest, mount_table[k].where);
 349                 if (!where)
 350                         return log_oom();
 351
 352                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 353                 if (r < 0 && r != -ENOENT)
 354                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 355
 356                 /* Skip this entry if it is not a remount. */
 357                 if (mount_table[k].what && r > 0)
 358                         continue;
 359
 360                 r = mkdir_p(where, 0755);
 361                 if (r < 0 && r != -EEXIST) {
 362                         if (mount_table[k].fatal)
 363                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 364
 365                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 366                         continue;
 367                 }
 368
 369                 o = mount_table[k].options;
 370                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 371                         if (in_userns)
 372                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 373                         else
 374                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 375                         if (r < 0)
 376                                 return log_oom();
 377                         if (r > 0)
 378                                 o = options;
 379                 }
 380
 381                 if (mount(mount_table[k].what,
 382                           where,
 383                           mount_table[k].type,
 384                           mount_table[k].flags,
 385                           o) < 0) {
 386
 387                         if (mount_table[k].fatal)
 388                                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 389
 390                         log_warning_errno(errno, "mount(%s) failed, ignoring: %m", where);
 391                 }
 392         }
 393
 394         return 0;
 395 }
 396
 397 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 398         const char *p = options;
 399         unsigned long flags = *mount_flags;
 400         char *opts = NULL;
 401
 402         assert(options);
 403
 404         for (;;) {
 405                 _cleanup_free_ char *word = NULL;
 406                 int r = extract_first_word(&p, &word, ",", 0);
 407                 if (r < 0)
 408                         return log_error_errno(r, "Failed to extract mount option: %m");
 409                 if (r == 0)
 410                         break;
 411
 412                 if (streq(word, "rbind"))
 413                         flags |= MS_REC;
 414                 else if (streq(word, "norbind"))
 415                         flags &= ~MS_REC;
 416                 else {
 417                         log_error("Invalid bind mount option: %s", word);
 418                         return -EINVAL;
 419                 }
 420         }
 421
 422         *mount_flags = flags;
 423         /* in the future mount_opts will hold string options for mount(2) */
 424         *mount_opts = opts;
 425
 426         return 0;
 427 }
 428
 429 static int mount_bind(const char *dest, CustomMount *m) {
 430         struct stat source_st, dest_st;
 431         const char *where;
 432         unsigned long mount_flags = MS_BIND | MS_REC;
 433         _cleanup_free_ char *mount_opts = NULL;
 434         int r;
 435
 436         assert(m);
 437
 438         if (m->options) {
 439                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 440                 if (r < 0)
 441                         return r;
 442         }
 443
 444         if (stat(m->source, &source_st) < 0)
 445                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 446
 447         where = prefix_roota(dest, m->destination);
 448
 449         if (stat(where, &dest_st) >= 0) {
 450                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 451                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 452                         return -EINVAL;
 453                 }
 454
 455                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 456                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 457                         return -EINVAL;
 458                 }
 459
 460         } else if (errno == ENOENT) {
 461                 r = mkdir_parents_label(where, 0755);
 462                 if (r < 0)
 463                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 464
 465                 /* Create the mount point. Any non-directory file can be
 466                 * mounted on any non-directory file (regular, fifo, socket,
 467                 * char, block).
 468                 */
 469                 if (S_ISDIR(source_st.st_mode))
 470                         r = mkdir_label(where, 0755);
 471                 else
 472                         r = touch(where);
 473                 if (r < 0)
 474                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 475
 476         } else {
 477                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 478         }
 479
 480         if (mount(m->source, where, NULL, mount_flags, mount_opts) < 0)
 481                 return log_error_errno(errno, "mount(%s) failed: %m", where);
 482
 483         if (m->read_only) {
 484                 r = bind_remount_recursive(where, true, NULL);
 485                 if (r < 0)
 486                         return log_error_errno(r, "Read-only bind mount failed: %m");
 487         }
 488
 489         return 0;
 490 }
 491
 492 static int mount_tmpfs(
 493                 const char *dest,
 494                 CustomMount *m,
 495                 bool userns, uid_t uid_shift, uid_t uid_range,
 496                 const char *selinux_apifs_context) {
 497
 498         const char *where, *options;
 499         _cleanup_free_ char *buf = NULL;
 500         int r;
 501
 502         assert(dest);
 503         assert(m);
 504
 505         where = prefix_roota(dest, m->destination);
 506
 507         r = mkdir_p_label(where, 0755);
 508         if (r < 0 && r != -EEXIST)
 509                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 510
 511         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 512         if (r < 0)
 513                 return log_oom();
 514         options = r > 0 ? buf : m->options;
 515
 516         if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options) < 0)
 517                 return log_error_errno(errno, "tmpfs mount to %s failed: %m", where);
 518
 519         return 0;
 520 }
 521
 522 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 523         _cleanup_strv_free_ char **sv = NULL;
 524
 525         sv = strv_copy(lower);
 526         if (!sv)
 527                 return NULL;
 528
 529         strv_reverse(sv);
 530
 531         if (!strv_shell_escape(sv, ",:"))
 532                 return NULL;
 533
 534         return strv_join(sv, ":");
 535 }
 536
 537 static int mount_overlay(const char *dest, CustomMount *m) {
 538         _cleanup_free_ char *lower = NULL;
 539         const char *where, *options;
 540         int r;
 541
 542         assert(dest);
 543         assert(m);
 544
 545         where = prefix_roota(dest, m->destination);
 546
 547         r = mkdir_label(where, 0755);
 548         if (r < 0 && r != -EEXIST)
 549                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 550
 551         (void) mkdir_p_label(m->source, 0755);
 552
 553         lower = joined_and_escaped_lower_dirs(m->lower);
 554         if (!lower)
 555                 return log_oom();
 556
 557         if (m->read_only) {
 558                 _cleanup_free_ char *escaped_source = NULL;
 559
 560                 escaped_source = shell_escape(m->source, ",:");
 561                 if (!escaped_source)
 562                         return log_oom();
 563
 564                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 565         } else {
 566                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 567
 568                 assert(m->work_dir);
 569                 (void) mkdir_label(m->work_dir, 0700);
 570
 571                 escaped_source = shell_escape(m->source, ",:");
 572                 if (!escaped_source)
 573                         return log_oom();
 574                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 575                 if (!escaped_work_dir)
 576                         return log_oom();
 577
 578                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 579         }
 580
 581         if (mount("overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options) < 0)
 582                 return log_error_errno(errno, "overlay mount to %s failed: %m", where);
 583
 584         return 0;
 585 }
 586
 587 int mount_custom(
 588                 const char *dest,
 589                 CustomMount *mounts, unsigned n,
 590                 bool userns, uid_t uid_shift, uid_t uid_range,
 591                 const char *selinux_apifs_context) {
 592
 593         unsigned i;
 594         int r;
 595
 596         assert(dest);
 597
 598         for (i = 0; i < n; i++) {
 599                 CustomMount *m = mounts + i;
 600
 601                 switch (m->type) {
 602
 603                 case CUSTOM_MOUNT_BIND:
 604                         r = mount_bind(dest, m);
 605                         break;
 606
 607                 case CUSTOM_MOUNT_TMPFS:
 608                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 609                         break;
 610
 611                 case CUSTOM_MOUNT_OVERLAY:
 612                         r = mount_overlay(dest, m);
 613                         break;
 614
 615                 default:
 616                         assert_not_reached("Unknown custom mount type");
 617                 }
 618
 619                 if (r < 0)
 620                         return r;
 621         }
 622
 623         return 0;
 624 }
 625
 626 /* Retrieve existing subsystems. This function is called in a new cgroup
 627  * namespace.
 628  */
 629 static int get_controllers(Set *subsystems) {
 630         _cleanup_fclose_ FILE *f = NULL;
 631         char line[LINE_MAX];
 632
 633         assert(subsystems);
 634
 635         f = fopen("/proc/self/cgroup", "re");
 636         if (!f)
 637                 return errno == ENOENT ? -ESRCH : -errno;
 638
 639         FOREACH_LINE(line, f, return -errno) {
 640                 int r;
 641                 char *e, *l, *p;
 642
 643                 truncate_nl(line);
 644
 645                 l = strchr(line, ':');
 646                 if (!l)
 647                         continue;
 648
 649                 l++;
 650                 e = strchr(l, ':');
 651                 if (!e)
 652                         continue;
 653
 654                 *e = 0;
 655
 656                 if (streq(l, "") || streq(l, "name=systemd"))
 657                         continue;
 658
 659                 p = strdup(l);
 660                 r = set_consume(subsystems, p);
 661                 if (r < 0)
 662                         return r;
 663         }
 664
 665         return 0;
 666 }
 667
 668 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy,
 669                                          CGroupUnified unified_requested, bool read_only) {
 670         char *to;
 671         int r;
 672
 673         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 674
 675         r = path_is_mount_point(to, 0);
 676         if (r < 0 && r != -ENOENT)
 677                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 678         if (r > 0)
 679                 return 0;
 680
 681         mkdir_p(to, 0755);
 682
 683         /* The superblock mount options of the mount point need to be
 684          * identical to the hosts', and hence writable... */
 685         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 686                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD)
 687                         r = mount("cgroup", to, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 688                 else
 689                         r = mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
 690         } else
 691                 r = mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller);
 692
 693         if (r < 0)
 694                 return log_error_errno(errno, "Failed to mount to %s: %m", to);
 695
 696         /* ... hence let's only make the bind mount read-only, not the
 697          * superblock. */
 698         if (read_only) {
 699                 if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 700                         return log_error_errno(errno, "Failed to remount %s read-only: %m", to);
 701         }
 702         return 1;
 703 }
 704
 705 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 706 static int mount_legacy_cgns_supported(
 707                 CGroupUnified unified_requested, bool userns, uid_t uid_shift,
 708                 uid_t uid_range, const char *selinux_apifs_context) {
 709         _cleanup_set_free_free_ Set *controllers = NULL;
 710         const char *cgroup_root = "/sys/fs/cgroup", *c;
 711         int r;
 712
 713         (void) mkdir_p(cgroup_root, 0755);
 714
 715         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 716         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 717         if (r < 0)
 718                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 719         if (r == 0) {
 720                 _cleanup_free_ char *options = NULL;
 721
 722                 /* When cgroup namespaces are enabled and user namespaces are
 723                  * used then the mount of the cgroupfs is done *inside* the new
 724                  * user namespace. We're root in the new user namespace and the
 725                  * kernel will happily translate our uid/gid to the correct
 726                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 727                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 728                  */
 729                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 730                 if (r < 0)
 731                         return log_oom();
 732
 733                 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
 734                         return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
 735         }
 736
 737         if (cg_all_unified() > 0)
 738                 goto skip_controllers;
 739
 740         controllers = set_new(&string_hash_ops);
 741         if (!controllers)
 742                 return log_oom();
 743
 744         r = get_controllers(controllers);
 745         if (r < 0)
 746                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 747
 748         for (;;) {
 749                 _cleanup_free_ const char *controller = NULL;
 750
 751                 controller = set_steal_first(controllers);
 752                 if (!controller)
 753                         break;
 754
 755                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 756                 if (r < 0)
 757                         return r;
 758
 759                 /* When multiple hierarchies are co-mounted, make their
 760                  * constituting individual hierarchies a symlink to the
 761                  * co-mount.
 762                  */
 763                 c = controller;
 764                 for (;;) {
 765                         _cleanup_free_ char *target = NULL, *tok = NULL;
 766
 767                         r = extract_first_word(&c, &tok, ",", 0);
 768                         if (r < 0)
 769                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 770                         if (r == 0)
 771                                 break;
 772
 773                         target = prefix_root("/sys/fs/cgroup", tok);
 774                         if (!target)
 775                                 return log_oom();
 776
 777                         if (streq(controller, tok))
 778                                 break;
 779
 780                         r = symlink_idempotent(controller, target);
 781                         if (r == -EINVAL)
 782                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 783                         if (r < 0)
 784                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 785                 }
 786         }
 787
 788 skip_controllers:
 789         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 790         if (r < 0)
 791                 return r;
 792
 793         if (!userns) {
 794                 if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
 795                         return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
 796         }
 797
 798         return 0;
 799 }
 800
 801 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 802 static int mount_legacy_cgns_unsupported(
 803                 const char *dest,
 804                 CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range,
 805                 const char *selinux_apifs_context) {
 806         _cleanup_set_free_free_ Set *controllers = NULL;
 807         const char *cgroup_root;
 808         int r;
 809
 810         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 811
 812         (void) mkdir_p(cgroup_root, 0755);
 813
 814         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 815         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 816         if (r < 0)
 817                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 818         if (r == 0) {
 819                 _cleanup_free_ char *options = NULL;
 820
 821                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 822                 if (r < 0)
 823                         return log_oom();
 824
 825                 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
 826                         return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
 827         }
 828
 829         if (cg_all_unified() > 0)
 830                 goto skip_controllers;
 831
 832         controllers = set_new(&string_hash_ops);
 833         if (!controllers)
 834                 return log_oom();
 835
 836         r = cg_kernel_controllers(controllers);
 837         if (r < 0)
 838                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 839
 840         for (;;) {
 841                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 842
 843                 controller = set_steal_first(controllers);
 844                 if (!controller)
 845                         break;
 846
 847                 origin = prefix_root("/sys/fs/cgroup/", controller);
 848                 if (!origin)
 849                         return log_oom();
 850
 851                 r = readlink_malloc(origin, &combined);
 852                 if (r == -EINVAL) {
 853                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 854
 855                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
 856                         if (r < 0)
 857                                 return r;
 858
 859                 } else if (r < 0)
 860                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 861                 else {
 862                         _cleanup_free_ char *target = NULL;
 863
 864                         target = prefix_root(dest, origin);
 865                         if (!target)
 866                                 return log_oom();
 867
 868                         /* A symbolic link, a combination of controllers in one hierarchy */
 869
 870                         if (!filename_is_valid(combined)) {
 871                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 872                                 continue;
 873                         }
 874
 875                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
 876                         if (r < 0)
 877                                 return r;
 878
 879                         r = symlink_idempotent(combined, target);
 880                         if (r == -EINVAL)
 881                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 882                         if (r < 0)
 883                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 884                 }
 885         }
 886
 887 skip_controllers:
 888         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 889         if (r < 0)
 890                 return r;
 891
 892         if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
 893                 return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
 894
 895         return 0;
 896 }
 897
 898 static int mount_unified_cgroups(const char *dest) {
 899         const char *p;
 900         int r;
 901
 902         assert(dest);
 903
 904         p = prefix_roota(dest, "/sys/fs/cgroup");
 905
 906         (void) mkdir_p(p, 0755);
 907
 908         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 909         if (r < 0)
 910                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 911         if (r > 0) {
 912                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 913                 if (access(p, F_OK) >= 0)
 914                         return 0;
 915                 if (errno != ENOENT)
 916                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 917
 918                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 919                 return -EINVAL;
 920         }
 921
 922         if (mount("cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
 923                 return log_error_errno(errno, "Failed to mount unified cgroup hierarchy to %s: %m", p);
 924
 925         return 0;
 926 }
 927
 928 int mount_cgroups(
 929                 const char *dest,
 930                 CGroupUnified unified_requested,
 931                 bool userns, uid_t uid_shift, uid_t uid_range,
 932                 const char *selinux_apifs_context,
 933                 bool use_cgns) {
 934
 935         if (unified_requested >= CGROUP_UNIFIED_ALL)
 936                 return mount_unified_cgroups(dest);
 937         else if (use_cgns && cg_ns_supported())
 938                 return mount_legacy_cgns_supported(unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 939
 940         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 941 }
 942
 943 int mount_systemd_cgroup_writable(
 944                 const char *dest,
 945                 CGroupUnified unified_requested) {
 946
 947         _cleanup_free_ char *own_cgroup_path = NULL;
 948         const char *systemd_root, *systemd_own;
 949         int r;
 950
 951         assert(dest);
 952
 953         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
 954         if (r < 0)
 955                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
 956
 957         /* If we are living in the top-level, then there's nothing to do... */
 958         if (path_equal(own_cgroup_path, "/"))
 959                 return 0;
 960
 961         if (unified_requested >= CGROUP_UNIFIED_ALL) {
 962                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
 963                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
 964         } else {
 965                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
 966                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
 967         }
 968
 969         /* Make our own cgroup a (writable) bind mount */
 970         if (mount(systemd_own, systemd_own,  NULL, MS_BIND, NULL) < 0)
 971                 return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path);
 972
 973         /* And then remount the systemd cgroup root read-only */
 974         if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
 975                 return log_error_errno(errno, "Failed to mount cgroup root read-only: %m");
 976
 977         return 0;
 978 }
 979
 980 int setup_volatile_state(
 981                 const char *directory,
 982                 VolatileMode mode,
 983                 bool userns, uid_t uid_shift, uid_t uid_range,
 984                 const char *selinux_apifs_context) {
 985
 986         _cleanup_free_ char *buf = NULL;
 987         const char *p, *options;
 988         int r;
 989
 990         assert(directory);
 991
 992         if (mode != VOLATILE_STATE)
 993                 return 0;
 994
 995         /* --volatile=state means we simply overmount /var
 996            with a tmpfs, and the rest read-only. */
 997
 998         r = bind_remount_recursive(directory, true, NULL);
 999         if (r < 0)
1000                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1001
1002         p = prefix_roota(directory, "/var");
1003         r = mkdir(p, 0755);
1004         if (r < 0 && errno != EEXIST)
1005                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1006
1007         options = "mode=755";
1008         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1009         if (r < 0)
1010                 return log_oom();
1011         if (r > 0)
1012                 options = buf;
1013
1014         if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, options) < 0)
1015                 return log_error_errno(errno, "Failed to mount tmpfs to /var: %m");
1016
1017         return 0;
1018 }
1019
1020 int setup_volatile(
1021                 const char *directory,
1022                 VolatileMode mode,
1023                 bool userns, uid_t uid_shift, uid_t uid_range,
1024                 const char *selinux_apifs_context) {
1025
1026         bool tmpfs_mounted = false, bind_mounted = false;
1027         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1028         _cleanup_free_ char *buf = NULL;
1029         const char *f, *t, *options;
1030         int r;
1031
1032         assert(directory);
1033
1034         if (mode != VOLATILE_YES)
1035                 return 0;
1036
1037         /* --volatile=yes means we mount a tmpfs to the root dir, and
1038            the original /usr to use inside it, and that read-only. */
1039
1040         if (!mkdtemp(template))
1041                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1042
1043         options = "mode=755";
1044         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1045         if (r < 0)
1046                 return log_oom();
1047         if (r > 0)
1048                 options = buf;
1049
1050         if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, options) < 0) {
1051                 r = log_error_errno(errno, "Failed to mount tmpfs for root directory: %m");
1052                 goto fail;
1053         }
1054
1055         tmpfs_mounted = true;
1056
1057         f = prefix_roota(directory, "/usr");
1058         t = prefix_roota(template, "/usr");
1059
1060         r = mkdir(t, 0755);
1061         if (r < 0 && errno != EEXIST) {
1062                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1063                 goto fail;
1064         }
1065
1066         if (mount(f, t, NULL, MS_BIND|MS_REC, NULL) < 0) {
1067                 r = log_error_errno(errno, "Failed to create /usr bind mount: %m");
1068                 goto fail;
1069         }
1070
1071         bind_mounted = true;
1072
1073         r = bind_remount_recursive(t, true, NULL);
1074         if (r < 0) {
1075                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1076                 goto fail;
1077         }
1078
1079         if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) {
1080                 r = log_error_errno(errno, "Failed to move root mount: %m");
1081                 goto fail;
1082         }
1083
1084         (void) rmdir(template);
1085
1086         return 0;
1087
1088 fail:
1089         if (bind_mounted)
1090                 (void) umount(t);
1091
1092         if (tmpfs_mounted)
1093                 (void) umount(template);
1094         (void) rmdir(template);
1095         return r;
1096 }
1097
1098 VolatileMode volatile_mode_from_string(const char *s) {
1099         int b;
1100
1101         if (isempty(s))
1102                 return _VOLATILE_MODE_INVALID;
1103
1104         b = parse_boolean(s);
1105         if (b > 0)
1106                 return VOLATILE_YES;
1107         if (b == 0)
1108                 return VOLATILE_NO;
1109
1110         if (streq(s, "state"))
1111                 return VOLATILE_STATE;
1112
1113         return _VOLATILE_MODE_INVALID;
1114 }