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_multiply(*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, MountSettingsMask mount_settings) {
 229         const char *full, *top, *x;
 230         int r;
 231         unsigned long extra_flags = 0;
 232
 233         top = prefix_roota(dest, "/sys");
 234         r = path_check_fstype(top, SYSFS_MAGIC);
 235         if (r < 0)
 236                 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
 237         /* /sys might already be mounted as sysfs by the outer child in the
 238          * !netns case. In this case, it's all good. Don't touch it because we
 239          * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
 240          */
 241         if (r > 0)
 242                 return 0;
 243
 244         full = prefix_roota(top, "/full");
 245
 246         (void) mkdir(full, 0755);
 247
 248         if (mount_settings & MOUNT_APPLY_APIVFS_RO)
 249                 extra_flags |= MS_RDONLY;
 250
 251         r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs",
 252                           MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
 253         if (r < 0)
 254                 return r;
 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                 r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
 270                 if (r < 0)
 271                         return r;
 272
 273                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 274                                   MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 275                 if (r < 0)
 276                         return r;
 277         }
 278
 279         r = umount_verbose(full);
 280         if (r < 0)
 281                 return r;
 282
 283         if (rmdir(full) < 0)
 284                 return log_error_errno(errno, "Failed to remove %s: %m", full);
 285
 286         x = prefix_roota(top, "/fs/kdbus");
 287         (void) mkdir_p(x, 0755);
 288
 289         /* Create mountpoint for cgroups. Otherwise we are not allowed since we
 290          * remount /sys read-only.
 291          */
 292         if (cg_ns_supported()) {
 293                 x = prefix_roota(top, "/fs/cgroup");
 294                 (void) mkdir_p(x, 0755);
 295         }
 296
 297         return mount_verbose(LOG_ERR, NULL, top, NULL,
 298                              MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
 299 }
 300
 301 static int mkdir_userns(const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 302         int r;
 303
 304         assert(path);
 305
 306         r = mkdir(path, mode);
 307         if (r < 0 && errno != EEXIST)
 308                 return -errno;
 309
 310         if ((mask & MOUNT_USE_USERNS) == 0)
 311                 return 0;
 312
 313         if (mask & MOUNT_IN_USERNS)
 314                 return 0;
 315
 316         r = lchown(path, uid_shift, uid_shift);
 317         if (r < 0)
 318                 return -errno;
 319
 320         return 0;
 321 }
 322
 323 static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) {
 324         const char *p, *e;
 325         int r;
 326
 327         assert(path);
 328
 329         if (prefix && !path_startswith(path, prefix))
 330                 return -ENOTDIR;
 331
 332         /* create every parent directory in the path, except the last component */
 333         p = path + strspn(path, "/");
 334         for (;;) {
 335                 char t[strlen(path) + 1];
 336
 337                 e = p + strcspn(p, "/");
 338                 p = e + strspn(e, "/");
 339
 340                 /* Is this the last component? If so, then we're done */
 341                 if (*p == 0)
 342                         break;
 343
 344                 memcpy(t, path, e - path);
 345                 t[e-path] = 0;
 346
 347                 if (prefix && path_startswith(prefix, t))
 348                         continue;
 349
 350                 r = mkdir_userns(t, mode, mask, uid_shift);
 351                 if (r < 0)
 352                         return r;
 353         }
 354
 355         return mkdir_userns(path, mode, mask, uid_shift);
 356 }
 357
 358 int mount_all(const char *dest,
 359               MountSettingsMask mount_settings,
 360               uid_t uid_shift, uid_t uid_range,
 361               const char *selinux_apifs_context) {
 362
 363         typedef struct MountPoint {
 364                 const char *what;
 365                 const char *where;
 366                 const char *type;
 367                 const char *options;
 368                 unsigned long flags;
 369                 MountSettingsMask mount_settings;
 370         } MountPoint;
 371
 372         static const MountPoint mount_table[] = {
 373                 /* inner child mounts */
 374                 { "proc",                "/proc",               "proc",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_IN_USERNS },
 375                 { "/proc/sys",           "/proc/sys",           NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ...*/
 376                 { "/proc/sys/net",       "/proc/sys/net",       NULL,    NULL,        MS_BIND,                                                   MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
 377                 { NULL,                  "/proc/sys",           NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 378                 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND,                                                               MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* Bind mount first ...*/
 379                 { NULL,                  "/proc/sysrq-trigger", NULL,    NULL,        MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,             MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO },                          /* ... then, make it r/o */
 380                 { "tmpfs",               "/tmp",                "tmpfs", "mode=1777", MS_STRICTATIME,                                            MOUNT_FATAL|MOUNT_IN_USERNS },
 381
 382                 /* outer child mounts */
 383                 { "tmpfs",               "/sys",                "tmpfs", "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV,                              MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS },
 384                 { "sysfs",               "/sys",                "sysfs", NULL,        MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO },    /* skipped if above was mounted */
 385                 { "sysfs",               "/sys",                "sysfs", NULL,                  MS_NOSUID|MS_NOEXEC|MS_NODEV,                    MOUNT_FATAL },                          /* skipped if above was mounted */
 386
 387                 { "tmpfs",               "/dev",                "tmpfs", "mode=755",  MS_NOSUID|MS_STRICTATIME,                                  MOUNT_FATAL },
 388                 { "tmpfs",               "/dev/shm",            "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 389                 { "tmpfs",               "/run",                "tmpfs", "mode=755",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,                         MOUNT_FATAL },
 390 #ifdef HAVE_SELINUX
 391                 { "/sys/fs/selinux",     "/sys/fs/selinux",     NULL,     NULL,       MS_BIND,                                                   0 },  /* Bind mount first */
 392                 { NULL,                  "/sys/fs/selinux",     NULL,     NULL,       MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, 0 },  /* Then, make it r/o */
 393 #endif
 394         };
 395
 396         unsigned k;
 397         int r;
 398         bool use_userns = (mount_settings & MOUNT_USE_USERNS);
 399         bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS);
 400         bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO);
 401         bool in_userns = (mount_settings & MOUNT_IN_USERNS);
 402
 403         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 404                 _cleanup_free_ char *where = NULL, *options = NULL;
 405                 const char *o;
 406                 bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL);
 407
 408                 if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS))
 409                         continue;
 410
 411                 if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS))
 412                         continue;
 413
 414                 if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO))
 415                         continue;
 416
 417                 where = prefix_root(dest, mount_table[k].where);
 418                 if (!where)
 419                         return log_oom();
 420
 421                 r = path_is_mount_point(where, dest, AT_SYMLINK_FOLLOW);
 422                 if (r < 0 && r != -ENOENT)
 423                         return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
 424
 425                 /* Skip this entry if it is not a remount. */
 426                 if (mount_table[k].what && r > 0)
 427                         continue;
 428
 429                 r = mkdir_userns_p(dest, where, 0755, mount_settings, uid_shift);
 430                 if (r < 0 && r != -EEXIST) {
 431                         if (fatal)
 432                                 return log_error_errno(r, "Failed to create directory %s: %m", where);
 433
 434                         log_debug_errno(r, "Failed to create directory %s: %m", where);
 435                         continue;
 436                 }
 437
 438                 o = mount_table[k].options;
 439                 if (streq_ptr(mount_table[k].type, "tmpfs")) {
 440                         if (in_userns)
 441                                 r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options);
 442                         else
 443                                 r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 444                         if (r < 0)
 445                                 return log_oom();
 446                         if (r > 0)
 447                                 o = options;
 448                 }
 449
 450                 r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
 451                                   mount_table[k].what,
 452                                   where,
 453                                   mount_table[k].type,
 454                                   mount_table[k].flags,
 455                                   o);
 456                 if (r < 0 && fatal)
 457                         return r;
 458         }
 459
 460         return 0;
 461 }
 462
 463 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
 464         const char *p = options;
 465         unsigned long flags = *mount_flags;
 466         char *opts = NULL;
 467         int r;
 468
 469         assert(options);
 470
 471         for (;;) {
 472                 _cleanup_free_ char *word = NULL;
 473
 474                 r = extract_first_word(&p, &word, ",", 0);
 475                 if (r < 0)
 476                         return log_error_errno(r, "Failed to extract mount option: %m");
 477                 if (r == 0)
 478                         break;
 479
 480                 if (streq(word, "rbind"))
 481                         flags |= MS_REC;
 482                 else if (streq(word, "norbind"))
 483                         flags &= ~MS_REC;
 484                 else {
 485                         log_error("Invalid bind mount option: %s", word);
 486                         return -EINVAL;
 487                 }
 488         }
 489
 490         *mount_flags = flags;
 491         /* in the future mount_opts will hold string options for mount(2) */
 492         *mount_opts = opts;
 493
 494         return 0;
 495 }
 496
 497 static int mount_bind(const char *dest, CustomMount *m) {
 498         struct stat source_st, dest_st;
 499         const char *where;
 500         unsigned long mount_flags = MS_BIND | MS_REC;
 501         _cleanup_free_ char *mount_opts = NULL;
 502         int r;
 503
 504         assert(m);
 505
 506         if (m->options) {
 507                 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
 508                 if (r < 0)
 509                         return r;
 510         }
 511
 512         if (stat(m->source, &source_st) < 0)
 513                 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
 514
 515         where = prefix_roota(dest, m->destination);
 516
 517         if (stat(where, &dest_st) >= 0) {
 518                 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
 519                         log_error("Cannot bind mount directory %s on file %s.", m->source, where);
 520                         return -EINVAL;
 521                 }
 522
 523                 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
 524                         log_error("Cannot bind mount file %s on directory %s.", m->source, where);
 525                         return -EINVAL;
 526                 }
 527
 528         } else if (errno == ENOENT) {
 529                 r = mkdir_parents_label(where, 0755);
 530                 if (r < 0)
 531                         return log_error_errno(r, "Failed to make parents of %s: %m", where);
 532
 533                 /* Create the mount point. Any non-directory file can be
 534                 * mounted on any non-directory file (regular, fifo, socket,
 535                 * char, block).
 536                 */
 537                 if (S_ISDIR(source_st.st_mode))
 538                         r = mkdir_label(where, 0755);
 539                 else
 540                         r = touch(where);
 541                 if (r < 0)
 542                         return log_error_errno(r, "Failed to create mount point %s: %m", where);
 543
 544         } else
 545                 return log_error_errno(errno, "Failed to stat %s: %m", where);
 546
 547         r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
 548         if (r < 0)
 549                 return r;
 550
 551         if (m->read_only) {
 552                 r = bind_remount_recursive(where, true, NULL);
 553                 if (r < 0)
 554                         return log_error_errno(r, "Read-only bind mount failed: %m");
 555         }
 556
 557         return 0;
 558 }
 559
 560 static int mount_tmpfs(
 561                 const char *dest,
 562                 CustomMount *m,
 563                 bool userns, uid_t uid_shift, uid_t uid_range,
 564                 const char *selinux_apifs_context) {
 565
 566         const char *where, *options;
 567         _cleanup_free_ char *buf = NULL;
 568         int r;
 569
 570         assert(dest);
 571         assert(m);
 572
 573         where = prefix_roota(dest, m->destination);
 574
 575         r = mkdir_p_label(where, 0755);
 576         if (r < 0 && r != -EEXIST)
 577                 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
 578
 579         r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
 580         if (r < 0)
 581                 return log_oom();
 582         options = r > 0 ? buf : m->options;
 583
 584         return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
 585 }
 586
 587 static char *joined_and_escaped_lower_dirs(char * const *lower) {
 588         _cleanup_strv_free_ char **sv = NULL;
 589
 590         sv = strv_copy(lower);
 591         if (!sv)
 592                 return NULL;
 593
 594         strv_reverse(sv);
 595
 596         if (!strv_shell_escape(sv, ",:"))
 597                 return NULL;
 598
 599         return strv_join(sv, ":");
 600 }
 601
 602 static int mount_overlay(const char *dest, CustomMount *m) {
 603         _cleanup_free_ char *lower = NULL;
 604         const char *where, *options;
 605         int r;
 606
 607         assert(dest);
 608         assert(m);
 609
 610         where = prefix_roota(dest, m->destination);
 611
 612         r = mkdir_label(where, 0755);
 613         if (r < 0 && r != -EEXIST)
 614                 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
 615
 616         (void) mkdir_p_label(m->source, 0755);
 617
 618         lower = joined_and_escaped_lower_dirs(m->lower);
 619         if (!lower)
 620                 return log_oom();
 621
 622         if (m->read_only) {
 623                 _cleanup_free_ char *escaped_source = NULL;
 624
 625                 escaped_source = shell_escape(m->source, ",:");
 626                 if (!escaped_source)
 627                         return log_oom();
 628
 629                 options = strjoina("lowerdir=", escaped_source, ":", lower);
 630         } else {
 631                 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
 632
 633                 assert(m->work_dir);
 634                 (void) mkdir_label(m->work_dir, 0700);
 635
 636                 escaped_source = shell_escape(m->source, ",:");
 637                 if (!escaped_source)
 638                         return log_oom();
 639                 escaped_work_dir = shell_escape(m->work_dir, ",:");
 640                 if (!escaped_work_dir)
 641                         return log_oom();
 642
 643                 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
 644         }
 645
 646         return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
 647 }
 648
 649 int mount_custom(
 650                 const char *dest,
 651                 CustomMount *mounts, unsigned n,
 652                 bool userns, uid_t uid_shift, uid_t uid_range,
 653                 const char *selinux_apifs_context) {
 654
 655         unsigned i;
 656         int r;
 657
 658         assert(dest);
 659
 660         for (i = 0; i < n; i++) {
 661                 CustomMount *m = mounts + i;
 662
 663                 switch (m->type) {
 664
 665                 case CUSTOM_MOUNT_BIND:
 666                         r = mount_bind(dest, m);
 667                         break;
 668
 669                 case CUSTOM_MOUNT_TMPFS:
 670                         r = mount_tmpfs(dest, m, userns, uid_shift, uid_range, selinux_apifs_context);
 671                         break;
 672
 673                 case CUSTOM_MOUNT_OVERLAY:
 674                         r = mount_overlay(dest, m);
 675                         break;
 676
 677                 default:
 678                         assert_not_reached("Unknown custom mount type");
 679                 }
 680
 681                 if (r < 0)
 682                         return r;
 683         }
 684
 685         return 0;
 686 }
 687
 688 /* Retrieve existing subsystems. This function is called in a new cgroup
 689  * namespace.
 690  */
 691 static int get_controllers(Set *subsystems) {
 692         _cleanup_fclose_ FILE *f = NULL;
 693         char line[LINE_MAX];
 694
 695         assert(subsystems);
 696
 697         f = fopen("/proc/self/cgroup", "re");
 698         if (!f)
 699                 return errno == ENOENT ? -ESRCH : -errno;
 700
 701         FOREACH_LINE(line, f, return -errno) {
 702                 int r;
 703                 char *e, *l, *p;
 704
 705                 l = strchr(line, ':');
 706                 if (!l)
 707                         continue;
 708
 709                 l++;
 710                 e = strchr(l, ':');
 711                 if (!e)
 712                         continue;
 713
 714                 *e = 0;
 715
 716                 if (STR_IN_SET(l, "", "name=systemd"))
 717                         continue;
 718
 719                 p = strdup(l);
 720                 if (!p)
 721                         return -ENOMEM;
 722
 723                 r = set_consume(subsystems, p);
 724                 if (r < 0)
 725                         return r;
 726         }
 727
 728         return 0;
 729 }
 730
 731 static int mount_legacy_cgroup_hierarchy(
 732                 const char *dest,
 733                 const char *controller,
 734                 const char *hierarchy,
 735                 CGroupUnified unified_requested,
 736                 bool read_only) {
 737
 738         const char *to, *fstype, *opts;
 739         int r;
 740
 741         to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
 742
 743         r = path_is_mount_point(to, dest, 0);
 744         if (r < 0 && r != -ENOENT)
 745                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
 746         if (r > 0)
 747                 return 0;
 748
 749         mkdir_p(to, 0755);
 750
 751         /* The superblock mount options of the mount point need to be
 752          * identical to the hosts', and hence writable... */
 753         if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
 754                 if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
 755                         fstype = "cgroup2";
 756                         opts = NULL;
 757                 } else {
 758                         fstype = "cgroup";
 759                         opts = "none,name=systemd,xattr";
 760                 }
 761         } else {
 762                 fstype = "cgroup";
 763                 opts = controller;
 764         }
 765
 766         r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
 767         if (r < 0)
 768                 return r;
 769
 770         /* ... hence let's only make the bind mount read-only, not the superblock. */
 771         if (read_only) {
 772                 r = mount_verbose(LOG_ERR, NULL, to, NULL,
 773                                   MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
 774                 if (r < 0)
 775                         return r;
 776         }
 777
 778         return 1;
 779 }
 780
 781 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
 782 static int mount_legacy_cgns_supported(
 783                 const char *dest,
 784                 CGroupUnified unified_requested,
 785                 bool userns,
 786                 uid_t uid_shift,
 787                 uid_t uid_range,
 788                 const char *selinux_apifs_context) {
 789
 790         _cleanup_set_free_free_ Set *controllers = NULL;
 791         const char *cgroup_root = "/sys/fs/cgroup", *c;
 792         int r;
 793
 794         (void) mkdir_p(cgroup_root, 0755);
 795
 796         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 797         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 798         if (r < 0)
 799                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 800         if (r == 0) {
 801                 _cleanup_free_ char *options = NULL;
 802
 803                 /* When cgroup namespaces are enabled and user namespaces are
 804                  * used then the mount of the cgroupfs is done *inside* the new
 805                  * user namespace. We're root in the new user namespace and the
 806                  * kernel will happily translate our uid/gid to the correct
 807                  * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
 808                  * pass uid 0 and not uid_shift to tmpfs_patch_options().
 809                  */
 810                 r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options);
 811                 if (r < 0)
 812                         return log_oom();
 813
 814                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 815                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 816                 if (r < 0)
 817                         return r;
 818         }
 819
 820         if (cg_all_unified() > 0)
 821                 goto skip_controllers;
 822
 823         controllers = set_new(&string_hash_ops);
 824         if (!controllers)
 825                 return log_oom();
 826
 827         r = get_controllers(controllers);
 828         if (r < 0)
 829                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 830
 831         for (;;) {
 832                 _cleanup_free_ const char *controller = NULL;
 833
 834                 controller = set_steal_first(controllers);
 835                 if (!controller)
 836                         break;
 837
 838                 r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns);
 839                 if (r < 0)
 840                         return r;
 841
 842                 /* When multiple hierarchies are co-mounted, make their
 843                  * constituting individual hierarchies a symlink to the
 844                  * co-mount.
 845                  */
 846                 c = controller;
 847                 for (;;) {
 848                         _cleanup_free_ char *target = NULL, *tok = NULL;
 849
 850                         r = extract_first_word(&c, &tok, ",", 0);
 851                         if (r < 0)
 852                                 return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
 853                         if (r == 0)
 854                                 break;
 855
 856                         target = prefix_root("/sys/fs/cgroup", tok);
 857                         if (!target)
 858                                 return log_oom();
 859
 860                         if (streq(controller, tok))
 861                                 break;
 862
 863                         r = symlink_idempotent(controller, target);
 864                         if (r == -EINVAL)
 865                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 866                         if (r < 0)
 867                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 868                 }
 869         }
 870
 871 skip_controllers:
 872         r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 873         if (r < 0)
 874                 return r;
 875
 876         if (!userns)
 877                 return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 878                                      MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 879
 880         return 0;
 881 }
 882
 883 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
 884 static int mount_legacy_cgns_unsupported(
 885                 const char *dest,
 886                 CGroupUnified unified_requested,
 887                 bool userns,
 888                 uid_t uid_shift,
 889                 uid_t uid_range,
 890                 const char *selinux_apifs_context) {
 891
 892         _cleanup_set_free_free_ Set *controllers = NULL;
 893         const char *cgroup_root;
 894         int r;
 895
 896         cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
 897
 898         (void) mkdir_p(cgroup_root, 0755);
 899
 900         /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
 901         r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
 902         if (r < 0)
 903                 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
 904         if (r == 0) {
 905                 _cleanup_free_ char *options = NULL;
 906
 907                 r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options);
 908                 if (r < 0)
 909                         return log_oom();
 910
 911                 r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
 912                                   MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
 913                 if (r < 0)
 914                         return r;
 915         }
 916
 917         if (cg_all_unified() > 0)
 918                 goto skip_controllers;
 919
 920         controllers = set_new(&string_hash_ops);
 921         if (!controllers)
 922                 return log_oom();
 923
 924         r = cg_kernel_controllers(controllers);
 925         if (r < 0)
 926                 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
 927
 928         for (;;) {
 929                 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
 930
 931                 controller = set_steal_first(controllers);
 932                 if (!controller)
 933                         break;
 934
 935                 origin = prefix_root("/sys/fs/cgroup/", controller);
 936                 if (!origin)
 937                         return log_oom();
 938
 939                 r = readlink_malloc(origin, &combined);
 940                 if (r == -EINVAL) {
 941                         /* Not a symbolic link, but directly a single cgroup hierarchy */
 942
 943                         r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true);
 944                         if (r < 0)
 945                                 return r;
 946
 947                 } else if (r < 0)
 948                         return log_error_errno(r, "Failed to read link %s: %m", origin);
 949                 else {
 950                         _cleanup_free_ char *target = NULL;
 951
 952                         target = prefix_root(dest, origin);
 953                         if (!target)
 954                                 return log_oom();
 955
 956                         /* A symbolic link, a combination of controllers in one hierarchy */
 957
 958                         if (!filename_is_valid(combined)) {
 959                                 log_warning("Ignoring invalid combined hierarchy %s.", combined);
 960                                 continue;
 961                         }
 962
 963                         r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true);
 964                         if (r < 0)
 965                                 return r;
 966
 967                         r = symlink_idempotent(combined, target);
 968                         if (r == -EINVAL)
 969                                 return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
 970                         if (r < 0)
 971                                 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
 972                 }
 973         }
 974
 975 skip_controllers:
 976         r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false);
 977         if (r < 0)
 978                 return r;
 979
 980         return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
 981                              MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 982 }
 983
 984 static int mount_unified_cgroups(const char *dest) {
 985         const char *p;
 986         int r;
 987
 988         assert(dest);
 989
 990         p = prefix_roota(dest, "/sys/fs/cgroup");
 991
 992         (void) mkdir_p(p, 0755);
 993
 994         r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW);
 995         if (r < 0)
 996                 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
 997         if (r > 0) {
 998                 p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
 999                 if (access(p, F_OK) >= 0)
1000                         return 0;
1001                 if (errno != ENOENT)
1002                         return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
1003
1004                 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
1005                 return -EINVAL;
1006         }
1007
1008         return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
1009 }
1010
1011 int mount_cgroups(
1012                 const char *dest,
1013                 CGroupUnified unified_requested,
1014                 bool userns,
1015                 uid_t uid_shift,
1016                 uid_t uid_range,
1017                 const char *selinux_apifs_context,
1018                 bool use_cgns) {
1019
1020         if (unified_requested >= CGROUP_UNIFIED_ALL)
1021                 return mount_unified_cgroups(dest);
1022         else if (use_cgns)
1023                 return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1024
1025         return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
1026 }
1027
1028 int mount_systemd_cgroup_writable(
1029                 const char *dest,
1030                 CGroupUnified unified_requested) {
1031
1032         _cleanup_free_ char *own_cgroup_path = NULL;
1033         const char *systemd_root, *systemd_own;
1034         int r;
1035
1036         assert(dest);
1037
1038         r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1039         if (r < 0)
1040                 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1041
1042         /* If we are living in the top-level, then there's nothing to do... */
1043         if (path_equal(own_cgroup_path, "/"))
1044                 return 0;
1045
1046         if (unified_requested >= CGROUP_UNIFIED_ALL) {
1047                 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
1048                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
1049         } else {
1050                 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1051                 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
1052         }
1053
1054         /* Make our own cgroup a (writable) bind mount */
1055         r = mount_verbose(LOG_ERR, systemd_own, systemd_own,  NULL, MS_BIND, NULL);
1056         if (r < 0)
1057                 return r;
1058
1059         /* And then remount the systemd cgroup root read-only */
1060         return mount_verbose(LOG_ERR, NULL, systemd_root, NULL,
1061                              MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
1062 }
1063
1064 int setup_volatile_state(
1065                 const char *directory,
1066                 VolatileMode mode,
1067                 bool userns, uid_t uid_shift, uid_t uid_range,
1068                 const char *selinux_apifs_context) {
1069
1070         _cleanup_free_ char *buf = NULL;
1071         const char *p, *options;
1072         int r;
1073
1074         assert(directory);
1075
1076         if (mode != VOLATILE_STATE)
1077                 return 0;
1078
1079         /* --volatile=state means we simply overmount /var
1080            with a tmpfs, and the rest read-only. */
1081
1082         r = bind_remount_recursive(directory, true, NULL);
1083         if (r < 0)
1084                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1085
1086         p = prefix_roota(directory, "/var");
1087         r = mkdir(p, 0755);
1088         if (r < 0 && errno != EEXIST)
1089                 return log_error_errno(errno, "Failed to create %s: %m", directory);
1090
1091         options = "mode=755";
1092         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1093         if (r < 0)
1094                 return log_oom();
1095         if (r > 0)
1096                 options = buf;
1097
1098         return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1099 }
1100
1101 int setup_volatile(
1102                 const char *directory,
1103                 VolatileMode mode,
1104                 bool userns, uid_t uid_shift, uid_t uid_range,
1105                 const char *selinux_apifs_context) {
1106
1107         bool tmpfs_mounted = false, bind_mounted = false;
1108         char template[] = "/tmp/nspawn-volatile-XXXXXX";
1109         _cleanup_free_ char *buf = NULL;
1110         const char *f, *t, *options;
1111         int r;
1112
1113         assert(directory);
1114
1115         if (mode != VOLATILE_YES)
1116                 return 0;
1117
1118         /* --volatile=yes means we mount a tmpfs to the root dir, and
1119            the original /usr to use inside it, and that read-only. */
1120
1121         if (!mkdtemp(template))
1122                 return log_error_errno(errno, "Failed to create temporary directory: %m");
1123
1124         options = "mode=755";
1125         r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf);
1126         if (r < 0)
1127                 return log_oom();
1128         if (r > 0)
1129                 options = buf;
1130
1131         r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1132         if (r < 0)
1133                 goto fail;
1134
1135         tmpfs_mounted = true;
1136
1137         f = prefix_roota(directory, "/usr");
1138         t = prefix_roota(template, "/usr");
1139
1140         r = mkdir(t, 0755);
1141         if (r < 0 && errno != EEXIST) {
1142                 r = log_error_errno(errno, "Failed to create %s: %m", t);
1143                 goto fail;
1144         }
1145
1146         r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1147         if (r < 0)
1148                 goto fail;
1149
1150         bind_mounted = true;
1151
1152         r = bind_remount_recursive(t, true, NULL);
1153         if (r < 0) {
1154                 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1155                 goto fail;
1156         }
1157
1158         r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1159         if (r < 0)
1160                 goto fail;
1161
1162         (void) rmdir(template);
1163
1164         return 0;
1165
1166 fail:
1167         if (bind_mounted)
1168                 (void) umount_verbose(t);
1169
1170         if (tmpfs_mounted)
1171                 (void) umount_verbose(template);
1172         (void) rmdir(template);
1173         return r;
1174 }
1175
1176 VolatileMode volatile_mode_from_string(const char *s) {
1177         int b;
1178
1179         if (isempty(s))
1180                 return _VOLATILE_MODE_INVALID;
1181
1182         b = parse_boolean(s);
1183         if (b > 0)
1184                 return VOLATILE_YES;
1185         if (b == 0)
1186                 return VOLATILE_NO;
1187
1188         if (streq(s, "state"))
1189                 return VOLATILE_STATE;
1190
1191         return _VOLATILE_MODE_INVALID;
1192 }