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 (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT,
 197                              options ?: "", options ? "," : "",
 198                              uid_shift, uid_shift) < 0)
 199                         return -ENOMEM;
 200
 201                 options = buf;
 202         }
 203
 204 #ifdef HAVE_SELINUX
 205         if (selinux_apifs_context) {
 206                 char *t;
 207
 208                 t = strjoin(options ?: "", options ? "," : "",
 209                             "context=\"", selinux_apifs_context, "\"");
 210                 free(buf);
 211                 if (!t)
 212                         return -ENOMEM;
 213
 214                 buf = t;
 215         }
 216 #endif
 217
 218         if (!buf && options) {
 219                 buf = strdup(options);
 220                 if (!buf)
 221                         return -ENOMEM;
 222         }
 223         *ret = buf;
 224
 225         return !!buf;
 226 }
 227
 228 int mount_sysfs(const char *dest) {
 229         const char *full, *top, *x;
 230         int r;
 231
 232         top = prefix_roota(dest, "/sys");
 233         r = path_check_fstype(top, SYSFS_MAGIC);
 234         if (r < 0)
 235                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 236         /* /sys might already be mounted as sysfs by the outer child in the
 237          * !netns case. In this case, it's all good. Don't touch it because we
 238          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 239          */
 240         if (r > 0)
 241                 return 0;
 242
 243         full = prefix_roota(top, "/full");
 244
 245         (void) mkdir(full, 0755);
 246
 247         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 248                           MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 249         if (r < 0)
 250                 return r;
 251
 252         FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
 253                 _cleanup_free_ char *from = NULL, *to = NULL;
 254
 255                 from = prefix_root(full, x);
 256                 if (!from)
 257                         return log_oom();
 258
 259                 to = prefix_root(top, x);
 260                 if (!to)
 261                         return log_oom();
 262
 263                 (void) mkdir(to, 0755);
 264
 265                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 266                 if (r < 0)
 267                         return r;
 268
 269                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 270                                   MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL);
 271                 if (r < 0)
 272                         return r;
 273         }
 274
 275         r = umount_verbose(full);
 276         if (r < 0)
 277                 return r;
 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         return mount_verbose(LOG_ERR, NULL, top, NULL,
 294                              MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL);
 295 }
 296
 297 static int mkdir_userns(const char *path, mode_t mode, bool in_userns, uid_t uid_shift) {
 298         int r;
 299
 300         assert(path);
 301
 302         r = mkdir(path, mode);
 303         if (r < 0 && errno != EEXIST)
 304                 return -errno;
 305
 306         if (!in_userns) {
 307                 r = lchown(path, uid_shift, uid_shift);
 308                 if (r < 0)
 309                         return -errno;
 310         }
 311
 312         return 0;
 313 }
 314
 315 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, bool in_userns, uid_t uid_shift) {
 316         const char *p, *e;
 317         int r;
 318
 319         assert(path);
 320
 321         if (prefix && !path_startswith(path, prefix))
 322                 return -ENOTDIR;
 323
 324         /* create every parent directory in the path, except the last component */
 325         p = path + strspn(path, "/");
 326         for (;;) {
 327                 char t[strlen(path) + 1];
 328
 329                 e = p + strcspn(p, "/");
 330                 p = e + strspn(e, "/");
 331
 332                 /* Is this the last component? If so, then we're done */
 333                 if (*p == 0)
 334                         break;
 335
 336                 memcpy(t, path, e - path);
 337                 t[e-path] = 0;
 338
 339                 if (prefix && path_startswith(prefix, t))
 340                         continue;
 341
 342                 r = mkdir_userns(t, mode, in_userns, uid_shift);
 343                 if (r < 0)
 344                         return r;
 345         }
 346
 347         return mkdir_userns(path, mode, in_userns, uid_shift);
 348 }
 349
 350 int mount_all(const char *dest,
 351               bool use_userns, bool in_userns,
 352               bool use_netns,
 353               uid_t uid_shift, uid_t uid_range,
 354               const char *selinux_apifs_context) {
 355
 356         typedef struct MountPoint {
 357                 const char *what;
 358                 const char *where;
 359                 const char *type;
 360                 const char *options;
 361                 unsigned long flags;
 362                 bool fatal;
 363                 bool in_userns;
 364                 bool use_netns;
 365         } MountPoint;
 366
 367         static const MountPoint mount_table[] = {
 368                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  true,  false },
 369                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   true,  true,  false },   /* Bind mount first ...*/
 370                 { "/proc/sys/net",       "/proc/sys/net",       NULL,    NULL,        MS_BIND,                                                   true,  true,  true  },   /* (except for this) */
 371                 { 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 */
 372                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                   false, true,  false },   /* Bind mount first ...*/
 373                 { 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 */
 374                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              true,  false, true  },
 375                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    true,  false, false },
 376                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  true,  false, false },
 377                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 378                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         true,  false, false },
 379                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            true,  true,  false },
 380 #ifdef HAVE_SELINUX
 381                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   false, false, false },  /* Bind mount first */
 382                 { 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 */
 383 #endif
 384         };
 385
 386         unsigned k;
 387         int r;
 388
 389         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 390                 _cleanup_free_ char *where = NULL, *options = NULL;
 391                 const char *o;
 392
 393                 if (in_userns != mount_table[k].in_userns)
 394                         continue;
 395
 396                 if (!use_netns && mount_table[k].use_netns)
 397                         continue;
 398
 399                 where = prefix_root(dest, mount_table[k].where);
 400                 if (!where)
 401                         return log_oom();
 402
 403                 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
 404                 if (r < 0 && r != -ENOENT)
 405                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 406
 407                 /* Skip this entry if it is not a remount. */
 408                 if (mount_table[k].what && r > 0)
 409                         continue;
 410
 411                 r = mkdir_userns_p(dest, where, 0755, in_userns, uid_shift);
 412                 if (r < 0 && r != -EEXIST) {
 413                         if (mount_table[k].fatal)
 414                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 415
 416                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 417                         continue;
 418                 }
 419
 420                 o = mount_table[k].options;
 421                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 422                         if (in_userns)
 423                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 424                         else
 425                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 426                         if (r < 0)
 427                                 return log_oom();
 428                         if (r > 0)
 429                                 o = options;
 430                 }
 431
 432                 r = mount_verbose(mount_table[k].fatal ? LOG_ERR : LOG_DEBUG,
 433                                   mount_table[k].what,
 434                                   where,
 435                                   mount_table[k].type,
 436                                   mount_table[k].flags,
 437                                   o);
 438                 if (r < 0 && mount_table[k].fatal)
 439                         return r;
 440         }
 441
 442         return 0;
 443 }
 444
 445 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 446         const char *p = options;
 447         unsigned long flags = *mount_flags;
 448         char *opts = NULL;
 449
 450         assert(options);
 451
 452         for (;;) {
 453                 _cleanup_free_ char *word = NULL;
 454                 int r = extract_first_word(&p, &word, ",", 0);
 455                 if (r < 0)
 456                         return log_error_errno(r, "Failed to extract mount option: %m");
 457                 if (r == 0)
 458                         break;
 459
 460                 if (streq(word, "rbind"))
 461                         flags |= MS_REC;
 462                 else if (streq(word, "norbind"))
 463                         flags &= ~MS_REC;
 464                 else {
 465                         log_error("Invalid bind mount option: %s", word);
 466                         return -EINVAL;
 467                 }
 468         }
 469
 470         *mount_flags = flags;
 471         /* in the future mount_opts will hold string options for mount(2) */
 472         *mount_opts = opts;
 473
 474         return 0;
 475 }
 476
 477 static int mount_bind(const char *dest, CustomMount *m) {
 478         struct stat source_st, dest_st;
 479         const char *where;
 480         unsigned long mount_flags = MS_BIND | MS_REC;
 481         _cleanup_free_ char *mount_opts = NULL;
 482         int r;
 483
 484         assert(m);
 485
 486         if (m->options) {
 487                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 488                 if (r < 0)
 489                         return r;
 490         }
 491
 492         if (stat(m->source, &source_st) < 0)
 493                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 494
 495         where = prefix_roota(dest, m->destination);
 496
 497         if (stat(where, &dest_st) >= 0) {
 498                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 499                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 500                         return -EINVAL;
 501                 }
 502
 503                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 504                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 505                         return -EINVAL;
 506                 }
 507
 508         } else if (errno == ENOENT) {
 509                 r = mkdir_parents_label(where, 0755);
 510                 if (r < 0)
 511                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 512
 513                 /* Create the mount point. Any non-directory file can be
 514                 * mounted on any non-directory file (regular, fifo, socket,
 515                 * char, block).
 516                 */
 517                 if (S_ISDIR(source_st.st_mode))
 518                         r = mkdir_label(where, 0755);
 519                 else
 520                         r = touch(where);
 521                 if (r < 0)
 522                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 523
 524         } else
 525                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 526
 527         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 528         if (r < 0)
 529                 return r;
 530
 531         if (m->read_only) {
 532                 r = bind_remount_recursive(where, true, NULL);
 533                 if (r < 0)
 534                         return log_error_errno(r, "Read-only bind mount failed: %m");
 535         }
 536
 537         return 0;
 538 }
 539
 540 static int mount_tmpfs(
 541                 const char *dest,
 542                 CustomMount *m,
 543                 bool userns, uid_t uid_shift, uid_t uid_range,
 544                 const char *selinux_apifs_context) {
 545
 546         const char *where, *options;
 547         _cleanup_free_ char *buf = NULL;
 548         int r;
 549
 550         assert(dest);
 551         assert(m);
 552
 553         where = prefix_roota(dest, m->destination);
 554
 555         r = mkdir_p_label(where, 0755);
 556         if (r < 0 && r != -EEXIST)
 557                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 558
 559         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 560         if (r < 0)
 561                 return log_oom();
 562         options = r > 0 ? buf : m->options;
 563
 564         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 565 }
 566
 567 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 568         _cleanup_strv_free_ char **sv = NULL;
 569
 570         sv = strv_copy(lower);
 571         if (!sv)
 572                 return NULL;
 573
 574         strv_reverse(sv);
 575
 576         if (!strv_shell_escape(sv, ",:"))
 577                 return NULL;
 578
 579         return strv_join(sv, ":");
 580 }
 581
 582 static int mount_overlay(const char *dest, CustomMount *m) {
 583         _cleanup_free_ char *lower = NULL;
 584         const char *where, *options;
 585         int r;
 586
 587         assert(dest);
 588         assert(m);
 589
 590         where = prefix_roota(dest, m->destination);
 591
 592         r = mkdir_label(where, 0755);
 593         if (r < 0 && r != -EEXIST)
 594                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 595
 596         (void) mkdir_p_label(m->source, 0755);
 597
 598         lower = joined_and_escaped_lower_dirs(m->lower);
 599         if (!lower)
 600                 return log_oom();
 601
 602         if (m->read_only) {
 603                 _cleanup_free_ char *escaped_source = NULL;
 604
 605                 escaped_source = shell_escape(m->source, ",:");
 606                 if (!escaped_source)
 607                         return log_oom();
 608
 609                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 610         } else {
 611                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 612
 613                 assert(m->work_dir);
 614                 (void) mkdir_label(m->work_dir, 0700);
 615
 616                 escaped_source = shell_escape(m->source, ",:");
 617                 if (!escaped_source)
 618                         return log_oom();
 619                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 620                 if (!escaped_work_dir)
 621                         return log_oom();
 622
 623                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 624         }
 625
 626         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 627 }
 628
 629 int mount_custom(
 630                 const char *dest,
 631                 CustomMount *mounts, unsigned n,
 632                 bool userns, uid_t uid_shift, uid_t uid_range,
 633                 const char *selinux_apifs_context) {
 634
 635         unsigned i;
 636         int r;
 637
 638         assert(dest);
 639
 640         for (i = 0; i < n; i++) {
 641                 CustomMount *m = mounts + i;
 642
 643                 switch (m->type) {
 644
 645                 case CUSTOM_MOUNT_BIND:
 646                         r = mount_bind(dest, m);
 647                         break;
 648
 649                 case CUSTOM_MOUNT_TMPFS:
 650                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 651                         break;
 652
 653                 case CUSTOM_MOUNT_OVERLAY:
 654                         r = mount_overlay(dest, m);
 655                         break;
 656
 657                 default:
 658                         assert_not_reached("Unknown custom mount type");
 659                 }
 660
 661                 if (r < 0)
 662                         return r;
 663         }
 664
 665         return 0;
 666 }
 667
 668 /* Retrieve existing subsystems. This function is called in a new cgroup
 669  * namespace.
 670  */
 671 static int get_controllers(Set *subsystems) {
 672         _cleanup_fclose_ FILE *f = NULL;
 673         char line[LINE_MAX];
 674
 675         assert(subsystems);
 676
 677         f = fopen("/proc/self/cgroup", "re");
 678         if (!f)
 679                 return errno == ENOENT ? -ESRCH : -errno;
 680
 681         FOREACH_LINE(line, f, return -errno) {
 682                 int r;
 683                 char *e, *l, *p;
 684
 685                 l = strchr(line, ':');
 686                 if (!l)
 687                         continue;
 688
 689                 l++;
 690                 e = strchr(l, ':');
 691                 if (!e)
 692                         continue;
 693
 694                 *e = 0;
 695
 696                 if (STR_IN_SET(l, "", "name=systemd"))
 697                         continue;
 698
 699                 p = strdup(l);
 700                 if (!p)
 701                         return -ENOMEM;
 702
 703                 r = set_consume(subsystems, p);
 704                 if (r < 0)
 705                         return r;
 706         }
 707
 708         return 0;
 709 }
 710
 711 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy,
 712                                          CGroupUnified unified_requested, bool read_only) {
 713         const char *to, *fstype, *opts;
 714         int r;
 715
 716         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 717
 718         r = path_is_mount_point(to, 0);
 719         if (r < 0 && r != -ENOENT)
 720                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 721         if (r > 0)
 722                 return 0;
 723
 724         mkdir_p(to, 0755);
 725
 726         /* The superblock mount options of the mount point need to be
 727          * identical to the hosts', and hence writable... */
 728         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 729                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
 730                         fstype = "cgroup2";
 731                         opts = NULL;
 732                 } else {
 733                         fstype = "cgroup";
 734                         opts = "none,name=systemd,xattr";
 735                 }
 736         } else {
 737                 fstype = "cgroup";
 738                 opts = controller;
 739         }
 740
 741         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 742         if (r < 0)
 743                 return r;
 744
 745         /* ... hence let's only make the bind mount read-only, not the superblock. */
 746         if (read_only) {
 747                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 748                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 749                 if (r < 0)
 750                         return r;
 751         }
 752
 753         return 1;
 754 }
 755
 756 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 757 static int mount_legacy_cgns_supported(
 758                 CGroupUnified unified_requested, bool userns, uid_t uid_shift,
 759                 uid_t uid_range, const char *selinux_apifs_context) {
 760         _cleanup_set_free_free_ Set *controllers = NULL;
 761         const char *cgroup_root = "/sys/fs/cgroup", *c;
 762         int r;
 763
 764         (void) mkdir_p(cgroup_root, 0755);
 765
 766         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 767         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 768         if (r < 0)
 769                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 770         if (r == 0) {
 771                 _cleanup_free_ char *options = NULL;
 772
 773                 /* When cgroup namespaces are enabled and user namespaces are
 774                  * used then the mount of the cgroupfs is done *inside* the new
 775                  * user namespace. We're root in the new user namespace and the
 776                  * kernel will happily translate our uid/gid to the correct
 777                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 778                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 779                  */
 780                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 781                 if (r < 0)
 782                         return log_oom();
 783
 784                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 785                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 786                 if (r < 0)
 787                         return r;
 788         }
 789
 790         if (cg_all_unified() > 0)
 791                 goto skip_controllers;
 792
 793         controllers = set_new(&string_hash_ops);
 794         if (!controllers)
 795                 return log_oom();
 796
 797         r = get_controllers(controllers);
 798         if (r < 0)
 799                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 800
 801         for (;;) {
 802                 _cleanup_free_ const char *controller = NULL;
 803
 804                 controller = set_steal_first(controllers);
 805                 if (!controller)
 806                         break;
 807
 808                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 809                 if (r < 0)
 810                         return r;
 811
 812                 /* When multiple hierarchies are co-mounted, make their
 813                  * constituting individual hierarchies a symlink to the
 814                  * co-mount.
 815                  */
 816                 c = controller;
 817                 for (;;) {
 818                         _cleanup_free_ char *target = NULL, *tok = NULL;
 819
 820                         r = extract_first_word(&c, &tok, ",", 0);
 821                         if (r < 0)
 822                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 823                         if (r == 0)
 824                                 break;
 825
 826                         target = prefix_root("/sys/fs/cgroup", tok);
 827                         if (!target)
 828                                 return log_oom();
 829
 830                         if (streq(controller, tok))
 831                                 break;
 832
 833                         r = symlink_idempotent(controller, target);
 834                         if (r == -EINVAL)
 835                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 836                         if (r < 0)
 837                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 838                 }
 839         }
 840
 841 skip_controllers:
 842         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 843         if (r < 0)
 844                 return r;
 845
 846         if (!userns)
 847                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 848                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 849
 850         return 0;
 851 }
 852
 853 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 854 static int mount_legacy_cgns_unsupported(
 855                 const char *dest,
 856                 CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range,
 857                 const char *selinux_apifs_context) {
 858         _cleanup_set_free_free_ Set *controllers = NULL;
 859         const char *cgroup_root;
 860         int r;
 861
 862         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 863
 864         (void) mkdir_p(cgroup_root, 0755);
 865
 866         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 867         r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
 868         if (r < 0)
 869                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 870         if (r == 0) {
 871                 _cleanup_free_ char *options = NULL;
 872
 873                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 874                 if (r < 0)
 875                         return log_oom();
 876
 877                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 878                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 879                 if (r < 0)
 880                         return r;
 881         }
 882
 883         if (cg_all_unified() > 0)
 884                 goto skip_controllers;
 885
 886         controllers = set_new(&string_hash_ops);
 887         if (!controllers)
 888                 return log_oom();
 889
 890         r = cg_kernel_controllers(controllers);
 891         if (r < 0)
 892                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 893
 894         for (;;) {
 895                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 896
 897                 controller = set_steal_first(controllers);
 898                 if (!controller)
 899                         break;
 900
 901                 origin = prefix_root("/sys/fs/cgroup/", controller);
 902                 if (!origin)
 903                         return log_oom();
 904
 905                 r = readlink_malloc(origin, &combined);
 906                 if (r == -EINVAL) {
 907                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 908
 909                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
 910                         if (r < 0)
 911                                 return r;
 912
 913                 } else if (r < 0)
 914                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 915                 else {
 916                         _cleanup_free_ char *target = NULL;
 917
 918                         target = prefix_root(dest, origin);
 919                         if (!target)
 920                                 return log_oom();
 921
 922                         /* A symbolic link, a combination of controllers in one hierarchy */
 923
 924                         if (!filename_is_valid(combined)) {
 925                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 926                                 continue;
 927                         }
 928
 929                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
 930                         if (r < 0)
 931                                 return r;
 932
 933                         r = symlink_idempotent(combined, target);
 934                         if (r == -EINVAL)
 935                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 936                         if (r < 0)
 937                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 938                 }
 939         }
 940
 941 skip_controllers:
 942         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 943         if (r < 0)
 944                 return r;
 945
 946         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 947                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 948 }
 949
 950 static int mount_unified_cgroups(const char *dest) {
 951         const char *p;
 952         int r;
 953
 954         assert(dest);
 955
 956         p = prefix_roota(dest, "/sys/fs/cgroup");
 957
 958         (void) mkdir_p(p, 0755);
 959
 960         r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
 961         if (r < 0)
 962                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 963         if (r > 0) {
 964                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 965                 if (access(p, F_OK) >= 0)
 966                         return 0;
 967                 if (errno != ENOENT)
 968                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
 969
 970                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
 971                 return -EINVAL;
 972         }
 973
 974         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
 975 }
 976
 977 int mount_cgroups(
 978                 const char *dest,
 979                 CGroupUnified unified_requested,
 980                 bool userns, uid_t uid_shift, uid_t uid_range,
 981                 const char *selinux_apifs_context,
 982                 bool use_cgns) {
 983
 984         if (unified_requested >= CGROUP_UNIFIED_ALL)
 985                 return mount_unified_cgroups(dest);
 986         else if (use_cgns)
 987                 return mount_legacy_cgns_supported(unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 988
 989         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
 990 }
 991
 992 int mount_systemd_cgroup_writable(
 993                 const char *dest,
 994                 CGroupUnified unified_requested) {
 995
 996         _cleanup_free_ char *own_cgroup_path = NULL;
 997         const char *systemd_root, *systemd_own;
 998         int r;
 999
1000         assert(dest);
1001
1002         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1003         if (r < 0)
1004                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1005
1006         /* If we are living in the top-level, then there's nothing to do... */
1007         if (path_equal(own_cgroup_path, "/"))
1008                 return 0;
1009
1010         if (unified_requested >= CGROUP_UNIFIED_ALL) {
1011                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
1012                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
1013         } else {
1014                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1015                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
1016         }
1017
1018         /* Make our own cgroup a (writable) bind mount */
1019         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1020         if (r < 0)
1021                 return r;
1022
1023         /* And then remount the systemd cgroup root read-only */
1024         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1025                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1026 }
1027
1028 int setup_volatile_state(
1029                 const char *directory,
1030                 VolatileMode mode,
1031                 bool userns, uid_t uid_shift, uid_t uid_range,
1032                 const char *selinux_apifs_context) {
1033
1034         _cleanup_free_ char *buf = NULL;
1035         const char *p, *options;
1036         int r;
1037
1038         assert(directory);
1039
1040         if (mode != VOLATILE_STATE)
1041                 return 0;
1042
1043         /* --volatile=state means we simply overmount /var
1044            with a tmpfs, and the rest read-only. */
1045
1046         r = bind_remount_recursive(directory, true, NULL);
1047         if (r < 0)
1048                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1049
1050         p = prefix_roota(directory, "/var");
1051         r = mkdir(p, 0755);
1052         if (r < 0 && errno != EEXIST)
1053                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1054
1055         options = "mode=755";
1056         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1057         if (r < 0)
1058                 return log_oom();
1059         if (r > 0)
1060                 options = buf;
1061
1062         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1063 }
1064
1065 int setup_volatile(
1066                 const char *directory,
1067                 VolatileMode mode,
1068                 bool userns, uid_t uid_shift, uid_t uid_range,
1069                 const char *selinux_apifs_context) {
1070
1071         bool tmpfs_mounted = false, bind_mounted = false;
1072         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1073         _cleanup_free_ char *buf = NULL;
1074         const char *f, *t, *options;
1075         int r;
1076
1077         assert(directory);
1078
1079         if (mode != VOLATILE_YES)
1080                 return 0;
1081
1082         /* --volatile=yes means we mount a tmpfs to the root dir, and
1083            the original /usr to use inside it, and that read-only. */
1084
1085         if (!mkdtemp(template))
1086                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1087
1088         options = "mode=755";
1089         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1090         if (r < 0)
1091                 return log_oom();
1092         if (r > 0)
1093                 options = buf;
1094
1095         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1096         if (r < 0)
1097                 goto fail;
1098
1099         tmpfs_mounted = true;
1100
1101         f = prefix_roota(directory, "/usr");
1102         t = prefix_roota(template, "/usr");
1103
1104         r = mkdir(t, 0755);
1105         if (r < 0 && errno != EEXIST) {
1106                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1107                 goto fail;
1108         }
1109
1110         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1111         if (r < 0)
1112                 goto fail;
1113
1114         bind_mounted = true;
1115
1116         r = bind_remount_recursive(t, true, NULL);
1117         if (r < 0) {
1118                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1119                 goto fail;
1120         }
1121
1122         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1123         if (r < 0)
1124                 goto fail;
1125
1126         (void) rmdir(template);
1127
1128         return 0;
1129
1130 fail:
1131         if (bind_mounted)
1132                 (void) umount_verbose(t);
1133
1134         if (tmpfs_mounted)
1135                 (void) umount_verbose(template);
1136         (void) rmdir(template);
1137         return r;
1138 }
1139
1140 VolatileMode volatile_mode_from_string(const char *s) {
1141         int b;
1142
1143         if (isempty(s))
1144                 return _VOLATILE_MODE_INVALID;
1145
1146         b = parse_boolean(s);
1147         if (b > 0)
1148                 return VOLATILE_YES;
1149         if (b == 0)
1150                 return VOLATILE_NO;
1151
1152         if (streq(s, "state"))
1153                 return VOLATILE_STATE;
1154
1155         return _VOLATILE_MODE_INVALID;
1156 }