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