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         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 254                           MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 255         if (r < 0)
 256                 return r;
 257
 258         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 259                 _cleanup_free_ char *from = NULL, *to = NULL;
 260
 261                 from = prefix_root(full, x);
 262                 if (!from)
 263                         return log_oom();
 264
 265                 to = prefix_root(top, x);
 266                 if (!to)
 267                         return log_oom();
 268
 269                 (void) mkdir(to, 0755);
 270
 271                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 272                 if (r < 0)
 273                         return r;
 274
 275                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 276                                   MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL);
 277                 if (r < 0)
 278                         return r;
 279         }
 280
 281         r = umount_verbose(full);
 282         if (r < 0)
 283                 return r;
 284
 285         if (rmdir(full) < 0)
 286                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 287
 288         x = prefix_roota(top, "/fs/kdbus");
 289         (void) mkdir_p(x, 0755);
 290
 291         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 292          * remount /sys read-only.
 293          */
 294         if (cg_ns_supported()) {
 295                 x = prefix_roota(top, "/fs/cgroup");
 296                 (void) mkdir_p(x, 0755);
 297         }
 298
 299         return mount_verbose(LOG_ERR, NULL, top, NULL,
 300                              MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL);
 301 }
 302
 303 int mount_all(const char *dest,
 304               bool use_userns, bool in_userns,
 305               bool use_netns,
 306               uid_t uid_shift, uid_t uid_range,
 307               const char *selinux_apifs_context) {
 308
 309         typedef struct MountPoint {
 310                 const char *what;
 311                 const char *where;
 312                 const char *type;
 313                 const char *options;
 314                 unsigned long flags;
 315                 bool fatal;
 316                 bool in_userns;
 317                 bool use_netns;
 318         } MountPoint;
 319
 320         static const MountPoint mount_table[] = {
 321                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  true,  false },
 322                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   true,  true,  false },   /* Bind mount first ...*/
 323                 { "/proc/sys/net",       "/proc/sys/net",       NULL,    NULL,        MS_BIND,                                                   true,  true,  true  },   /* (except for this) */
 324                 { 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 */
 325                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                   false, true,  false },   /* Bind mount first ...*/
 326                 { 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 */
 327                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true  },
 328                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 329                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 330                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 331                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 332                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            true,  true,  false },
 333 #ifdef HAVE_SELINUX
 334                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   false, false, false },  /* Bind mount first */
 335                 { 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 */
 336 #endif
 337         };
 338
 339         unsigned k;
 340         int r;
 341
 342         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 343                 _cleanup_free_ char *where = NULL, *options = NULL;
 344                 const char *o;
 345
 346                 if (in_userns != mount_table[k].in_userns)
 347                         continue;
 348
 349                 if (!use_netns && mount_table[k].use_netns)
 350                         continue;
 351
 352                 where = prefix_root(dest, mount_table[k].where);
 353                 if (!where)
 354                         return log_oom();
 355
 356                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 357                 if (r < 0 && r != -ENOENT)
 358                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 359
 360                 /* Skip this entry if it is not a remount. */
 361                 if (mount_table[k].what && r > 0)
 362                         continue;
 363
 364                 r = mkdir_p(where, 0755);
 365                 if (r < 0 && r != -EEXIST) {
 366                         if (mount_table[k].fatal)
 367                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 368
 369                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 370                         continue;
 371                 }
 372
 373                 o = mount_table[k].options;
 374                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 375                         if (in_userns)
 376                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 377                         else
 378                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 379                         if (r < 0)
 380                                 return log_oom();
 381                         if (r > 0)
 382                                 o = options;
 383                 }
 384
 385                 r = mount_verbose(mount_table[k].fatal ? LOG_ERR : LOG_WARNING,
 386                                   mount_table[k].what,
 387                                   where,
 388                                   mount_table[k].type,
 389                                   mount_table[k].flags,
 390                                   o);
 391                 if (r < 0 && mount_table[k].fatal)
 392                         return r;
 393         }
 394
 395         return 0;
 396 }
 397
 398 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 399         const char *p = options;
 400         unsigned long flags = *mount_flags;
 401         char *opts = NULL;
 402
 403         assert(options);
 404
 405         for (;;) {
 406                 _cleanup_free_ char *word = NULL;
 407                 int r = extract_first_word(&p, &word, ",", 0);
 408                 if (r < 0)
 409                         return log_error_errno(r, "Failed to extract mount option: %m");
 410                 if (r == 0)
 411                         break;
 412
 413                 if (streq(word, "rbind"))
 414                         flags |= MS_REC;
 415                 else if (streq(word, "norbind"))
 416                         flags &= ~MS_REC;
 417                 else {
 418                         log_error("Invalid bind mount option: %s", word);
 419                         return -EINVAL;
 420                 }
 421         }
 422
 423         *mount_flags = flags;
 424         /* in the future mount_opts will hold string options for mount(2) */
 425         *mount_opts = opts;
 426
 427         return 0;
 428 }
 429
 430 static int mount_bind(const char *dest, CustomMount *m) {
 431         struct stat source_st, dest_st;
 432         const char *where;
 433         unsigned long mount_flags = MS_BIND | MS_REC;
 434         _cleanup_free_ char *mount_opts = NULL;
 435         int r;
 436
 437         assert(m);
 438
 439         if (m->options) {
 440                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 441                 if (r < 0)
 442                         return r;
 443         }
 444
 445         if (stat(m->source, &source_st) < 0)
 446                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 447
 448         where = prefix_roota(dest, m->destination);
 449
 450         if (stat(where, &dest_st) >= 0) {
 451                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 452                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 453                         return -EINVAL;
 454                 }
 455
 456                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 457                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 458                         return -EINVAL;
 459                 }
 460
 461         } else if (errno == ENOENT) {
 462                 r = mkdir_parents_label(where, 0755);
 463                 if (r < 0)
 464                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 465
 466                 /* Create the mount point. Any non-directory file can be
 467                 * mounted on any non-directory file (regular, fifo, socket,
 468                 * char, block).
 469                 */
 470                 if (S_ISDIR(source_st.st_mode))
 471                         r = mkdir_label(where, 0755);
 472                 else
 473                         r = touch(where);
 474                 if (r < 0)
 475                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 476
 477         } else
 478                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 479
 480         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 481         if (r < 0)
 482                 return r;
 483
 484         if (m->read_only) {
 485                 r = bind_remount_recursive(where, true, NULL);
 486                 if (r < 0)
 487                         return log_error_errno(r, "Read-only bind mount failed: %m");
 488         }
 489
 490         return 0;
 491 }
 492
 493 static int mount_tmpfs(
 494                 const char *dest,
 495                 CustomMount *m,
 496                 bool userns, uid_t uid_shift, uid_t uid_range,
 497                 const char *selinux_apifs_context) {
 498
 499         const char *where, *options;
 500         _cleanup_free_ char *buf = NULL;
 501         int r;
 502
 503         assert(dest);
 504         assert(m);
 505
 506         where = prefix_roota(dest, m->destination);
 507
 508         r = mkdir_p_label(where, 0755);
 509         if (r < 0 && r != -EEXIST)
 510                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 511
 512         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 513         if (r < 0)
 514                 return log_oom();
 515         options = r > 0 ? buf : m->options;
 516
 517         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 518 }
 519
 520 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 521         _cleanup_strv_free_ char **sv = NULL;
 522
 523         sv = strv_copy(lower);
 524         if (!sv)
 525                 return NULL;
 526
 527         strv_reverse(sv);
 528
 529         if (!strv_shell_escape(sv, ",:"))
 530                 return NULL;
 531
 532         return strv_join(sv, ":");
 533 }
 534
 535 static int mount_overlay(const char *dest, CustomMount *m) {
 536         _cleanup_free_ char *lower = NULL;
 537         const char *where, *options;
 538         int r;
 539
 540         assert(dest);
 541         assert(m);
 542
 543         where = prefix_roota(dest, m->destination);
 544
 545         r = mkdir_label(where, 0755);
 546         if (r < 0 && r != -EEXIST)
 547                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 548
 549         (void) mkdir_p_label(m->source, 0755);
 550
 551         lower = joined_and_escaped_lower_dirs(m->lower);
 552         if (!lower)
 553                 return log_oom();
 554
 555         if (m->read_only) {
 556                 _cleanup_free_ char *escaped_source = NULL;
 557
 558                 escaped_source = shell_escape(m->source, ",:");
 559                 if (!escaped_source)
 560                         return log_oom();
 561
 562                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 563         } else {
 564                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 565
 566                 assert(m->work_dir);
 567                 (void) mkdir_label(m->work_dir, 0700);
 568
 569                 escaped_source = shell_escape(m->source, ",:");
 570                 if (!escaped_source)
 571                         return log_oom();
 572                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 573                 if (!escaped_work_dir)
 574                         return log_oom();
 575
 576                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 577         }
 578
 579         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 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                 l = strchr(line, ':');
 639                 if (!l)
 640                         continue;
 641
 642                 l++;
 643                 e = strchr(l, ':');
 644                 if (!e)
 645                         continue;
 646
 647                 *e = 0;
 648
 649                 if (STR_IN_SET(l, "", "name=systemd"))
 650                         continue;
 651
 652                 p = strdup(l);
 653                 if (!p)
 654                         return -ENOMEM;
 655
 656                 r = set_consume(subsystems, p);
 657                 if (r < 0)
 658                         return r;
 659         }
 660
 661         return 0;
 662 }
 663
 664 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy,
 665                                          CGroupUnified unified_requested, bool read_only) {
 666         const char *to, *fstype, *opts;
 667         int r;
 668
 669         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 670
 671         r = path_is_mount_point(to, 0);
 672         if (r < 0 && r != -ENOENT)
 673                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 674         if (r > 0)
 675                 return 0;
 676
 677         mkdir_p(to, 0755);
 678
 679         /* The superblock mount options of the mount point need to be
 680          * identical to the hosts', and hence writable... */
 681         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 682                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
 683                         fstype = "cgroup2";
 684                         opts = NULL;
 685                 } else {
 686                         fstype = "cgroup";
 687                         opts = "none,name=systemd,xattr";
 688                 }
 689         } else {
 690                 fstype = "cgroup";
 691                 opts = controller;
 692         }
 693
 694         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 695         if (r < 0)
 696                 return r;
 697
 698         /* ... hence let's only make the bind mount read-only, not the superblock. */
 699         if (read_only) {
 700                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 701                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 702                 if (r < 0)
 703                         return r;
 704         }
 705
 706         return 1;
 707 }
 708
 709 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 710 static int mount_legacy_cgns_supported(
 711                 CGroupUnified unified_requested, bool userns, uid_t uid_shift,
 712                 uid_t uid_range, const char *selinux_apifs_context) {
 713         _cleanup_set_free_free_ Set *controllers = NULL;
 714         const char *cgroup_root = "/sys/fs/cgroup", *c;
 715         int r;
 716
 717         (void) mkdir_p(cgroup_root, 0755);
 718
 719         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 720         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 721         if (r < 0)
 722                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 723         if (r == 0) {
 724                 _cleanup_free_ char *options = NULL;
 725
 726                 /* When cgroup namespaces are enabled and user namespaces are
 727                  * used then the mount of the cgroupfs is done *inside* the new
 728                  * user namespace. We're root in the new user namespace and the
 729                  * kernel will happily translate our uid/gid to the correct
 730                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 731                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 732                  */
 733                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 734                 if (r < 0)
 735                         return log_oom();
 736
 737                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 738                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 739                 if (r < 0)
 740                         return r;
 741         }
 742
 743         if (cg_all_unified() > 0)
 744                 goto skip_controllers;
 745
 746         controllers = set_new(&string_hash_ops);
 747         if (!controllers)
 748                 return log_oom();
 749
 750         r = get_controllers(controllers);
 751         if (r < 0)
 752                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 753
 754         for (;;) {
 755                 _cleanup_free_ const char *controller = NULL;
 756
 757                 controller = set_steal_first(controllers);
 758                 if (!controller)
 759                         break;
 760
 761                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 762                 if (r < 0)
 763                         return r;
 764
 765                 /* When multiple hierarchies are co-mounted, make their
 766                  * constituting individual hierarchies a symlink to the
 767                  * co-mount.
 768                  */
 769                 c = controller;
 770                 for (;;) {
 771                         _cleanup_free_ char *target = NULL, *tok = NULL;
 772
 773                         r = extract_first_word(&c, &tok, ",", 0);
 774                         if (r < 0)
 775                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 776                         if (r == 0)
 777                                 break;
 778
 779                         target = prefix_root("/sys/fs/cgroup", tok);
 780                         if (!target)
 781                                 return log_oom();
 782
 783                         if (streq(controller, tok))
 784                                 break;
 785
 786                         r = symlink_idempotent(controller, target);
 787                         if (r == -EINVAL)
 788                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 789                         if (r < 0)
 790                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 791                 }
 792         }
 793
 794 skip_controllers:
 795         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 796         if (r < 0)
 797                 return r;
 798
 799         if (!userns)
 800                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 801                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 802
 803         return 0;
 804 }
 805
 806 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 807 static int mount_legacy_cgns_unsupported(
 808                 const char *dest,
 809                 CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range,
 810                 const char *selinux_apifs_context) {
 811         _cleanup_set_free_free_ Set *controllers = NULL;
 812         const char *cgroup_root;
 813         int r;
 814
 815         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 816
 817         (void) mkdir_p(cgroup_root, 0755);
 818
 819         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 820         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 821         if (r < 0)
 822                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 823         if (r == 0) {
 824                 _cleanup_free_ char *options = NULL;
 825
 826                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 827                 if (r < 0)
 828                         return log_oom();
 829
 830                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 831                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 832                 if (r < 0)
 833                         return r;
 834         }
 835
 836         if (cg_all_unified() > 0)
 837                 goto skip_controllers;
 838
 839         controllers = set_new(&string_hash_ops);
 840         if (!controllers)
 841                 return log_oom();
 842
 843         r = cg_kernel_controllers(controllers);
 844         if (r < 0)
 845                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 846
 847         for (;;) {
 848                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 849
 850                 controller = set_steal_first(controllers);
 851                 if (!controller)
 852                         break;
 853
 854                 origin = prefix_root("/sys/fs/cgroup/", controller);
 855                 if (!origin)
 856                         return log_oom();
 857
 858                 r = readlink_malloc(origin, &combined);
 859                 if (r == -EINVAL) {
 860                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 861
 862                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
 863                         if (r < 0)
 864                                 return r;
 865
 866                 } else if (r < 0)
 867                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 868                 else {
 869                         _cleanup_free_ char *target = NULL;
 870
 871                         target = prefix_root(dest, origin);
 872                         if (!target)
 873                                 return log_oom();
 874
 875                         /* A symbolic link, a combination of controllers in one hierarchy */
 876
 877                         if (!filename_is_valid(combined)) {
 878                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 879                                 continue;
 880                         }
 881
 882                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
 883                         if (r < 0)
 884                                 return r;
 885
 886                         r = symlink_idempotent(combined, target);
 887                         if (r == -EINVAL)
 888                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 889                         if (r < 0)
 890                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 891                 }
 892         }
 893
 894 skip_controllers:
 895         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 896         if (r < 0)
 897                 return r;
 898
 899         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 900                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 901 }
 902
 903 static int mount_unified_cgroups(const char *dest) {
 904         const char *p;
 905         int r;
 906
 907         assert(dest);
 908
 909         p = prefix_roota(dest, "/sys/fs/cgroup");
 910
 911         (void) mkdir_p(p, 0755);
 912
 913         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 914         if (r < 0)
 915                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 916         if (r > 0) {
 917                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 918                 if (access(p, F_OK) >= 0)
 919                         return 0;
 920                 if (errno != ENOENT)
 921                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 922
 923                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 924                 return -EINVAL;
 925         }
 926
 927         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 928 }
 929
 930 int mount_cgroups(
 931                 const char *dest,
 932                 CGroupUnified unified_requested,
 933                 bool userns, uid_t uid_shift, uid_t uid_range,
 934                 const char *selinux_apifs_context,
 935                 bool use_cgns) {
 936
 937         if (unified_requested >= CGROUP_UNIFIED_ALL)
 938                 return mount_unified_cgroups(dest);
 939         else if (use_cgns)
 940                 return mount_legacy_cgns_supported(unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 941
 942         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 943 }
 944
 945 int mount_systemd_cgroup_writable(
 946                 const char *dest,
 947                 CGroupUnified unified_requested) {
 948
 949         _cleanup_free_ char *own_cgroup_path = NULL;
 950         const char *systemd_root, *systemd_own;
 951         int r;
 952
 953         assert(dest);
 954
 955         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
 956         if (r < 0)
 957                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
 958
 959         /* If we are living in the top-level, then there's nothing to do... */
 960         if (path_equal(own_cgroup_path, "/"))
 961                 return 0;
 962
 963         if (unified_requested >= CGROUP_UNIFIED_ALL) {
 964                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
 965                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
 966         } else {
 967                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
 968                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
 969         }
 970
 971         /* Make our own cgroup a (writable) bind mount */
 972         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
 973         if (r < 0)
 974                 return r;
 975
 976         /* And then remount the systemd cgroup root read-only */
 977         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
 978                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 979 }
 980
 981 int setup_volatile_state(
 982                 const char *directory,
 983                 VolatileMode mode,
 984                 bool userns, uid_t uid_shift, uid_t uid_range,
 985                 const char *selinux_apifs_context) {
 986
 987         _cleanup_free_ char *buf = NULL;
 988         const char *p, *options;
 989         int r;
 990
 991         assert(directory);
 992
 993         if (mode != VOLATILE_STATE)
 994                 return 0;
 995
 996         /* --volatile=state means we simply overmount /var
 997            with a tmpfs, and the rest read-only. */
 998
 999         r = bind_remount_recursive(directory, true, NULL);
1000         if (r < 0)
1001                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1002
1003         p = prefix_roota(directory, "/var");
1004         r = mkdir(p, 0755);
1005         if (r < 0 && errno != EEXIST)
1006                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1007
1008         options = "mode=755";
1009         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1010         if (r < 0)
1011                 return log_oom();
1012         if (r > 0)
1013                 options = buf;
1014
1015         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1016 }
1017
1018 int setup_volatile(
1019                 const char *directory,
1020                 VolatileMode mode,
1021                 bool userns, uid_t uid_shift, uid_t uid_range,
1022                 const char *selinux_apifs_context) {
1023
1024         bool tmpfs_mounted = false, bind_mounted = false;
1025         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1026         _cleanup_free_ char *buf = NULL;
1027         const char *f, *t, *options;
1028         int r;
1029
1030         assert(directory);
1031
1032         if (mode != VOLATILE_YES)
1033                 return 0;
1034
1035         /* --volatile=yes means we mount a tmpfs to the root dir, and
1036            the original /usr to use inside it, and that read-only. */
1037
1038         if (!mkdtemp(template))
1039                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1040
1041         options = "mode=755";
1042         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1043         if (r < 0)
1044                 return log_oom();
1045         if (r > 0)
1046                 options = buf;
1047
1048         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1049         if (r < 0)
1050                 goto fail;
1051
1052         tmpfs_mounted = true;
1053
1054         f = prefix_roota(directory, "/usr");
1055         t = prefix_roota(template, "/usr");
1056
1057         r = mkdir(t, 0755);
1058         if (r < 0 && errno != EEXIST) {
1059                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1060                 goto fail;
1061         }
1062
1063         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1064         if (r < 0)
1065                 goto fail;
1066
1067         bind_mounted = true;
1068
1069         r = bind_remount_recursive(t, true, NULL);
1070         if (r < 0) {
1071                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1072                 goto fail;
1073         }
1074
1075         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1076         if (r < 0)
1077                 goto fail;
1078
1079         (void) rmdir(template);
1080
1081         return 0;
1082
1083 fail:
1084         if (bind_mounted)
1085                 (void) umount_verbose(t);
1086
1087         if (tmpfs_mounted)
1088                 (void) umount_verbose(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 }