src/core/mount-setup.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <ftw.h>
   5 #include <stdlib.h>
   6 #include <sys/mount.h>
   7 #include <sys/statvfs.h>
   8 #include <unistd.h>
   9
  10 #include "alloc-util.h"
  11 #include "bus-util.h"
  12 #include "cgroup-util.h"
  13 #include "conf-files.h"
  14 #include "cgroup-setup.h"
  15 #include "dev-setup.h"
  16 #include "dirent-util.h"
  17 #include "efi-loader.h"
  18 #include "fd-util.h"
  19 #include "fileio.h"
  20 #include "fs-util.h"
  21 #include "label.h"
  22 #include "log.h"
  23 #include "macro.h"
  24 #include "mkdir.h"
  25 #include "mount-setup.h"
  26 #include "mount-util.h"
  27 #include "mountpoint-util.h"
  28 #include "nulstr-util.h"
  29 #include "path-util.h"
  30 #include "set.h"
  31 #include "smack-util.h"
  32 #include "strv.h"
  33 #include "user-util.h"
  34 #include "virt.h"
  35
  36 typedef enum MountMode {
  37         MNT_NONE           = 0,
  38         MNT_FATAL          = 1 << 0,
  39         MNT_IN_CONTAINER   = 1 << 1,
  40         MNT_CHECK_WRITABLE = 1 << 2,
  41 } MountMode;
  42
  43 typedef struct MountPoint {
  44         const char *what;
  45         const char *where;
  46         const char *type;
  47         const char *options;
  48         unsigned long flags;
  49         bool (*condition_fn)(void);
  50         MountMode mode;
  51 } MountPoint;
  52
  53 /* The first three entries we might need before SELinux is up. The
  54  * fourth (securityfs) is needed by IMA to load a custom policy. The
  55  * other ones we can delay until SELinux and IMA are loaded. When
  56  * SMACK is enabled we need smackfs, too, so it's a fifth one. */
  57 #if ENABLE_SMACK
  58 #define N_EARLY_MOUNT 5
  59 #else
  60 #define N_EARLY_MOUNT 4
  61 #endif
  62
  63 static const MountPoint mount_table[] = {
  64         { "sysfs",       "/sys",                      "sysfs",      NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  65           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  66         { "proc",        "/proc",                     "proc",       NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  67           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  68         { "devtmpfs",    "/dev",                      "devtmpfs",   "mode=755" TMPFS_LIMITS_DEV,               MS_NOSUID|MS_NOEXEC|MS_STRICTATIME,
  69           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  70         { "securityfs",  "/sys/kernel/security",      "securityfs", NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  71           NULL,          MNT_NONE                   },
  72 #if ENABLE_SMACK
  73         { "smackfs",     "/sys/fs/smackfs",           "smackfs",    "smackfsdef=*",                            MS_NOSUID|MS_NOEXEC|MS_NODEV,
  74           mac_smack_use, MNT_FATAL                  },
  75         { "tmpfs",       "/dev/shm",                  "tmpfs",      "mode=1777,smackfsroot=*",                 MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  76           mac_smack_use, MNT_FATAL                  },
  77 #endif
  78         { "tmpfs",       "/dev/shm",                  "tmpfs",      "mode=1777",                               MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  79           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  80         { "devpts",      "/dev/pts",                  "devpts",     "mode=620,gid=" STRINGIFY(TTY_GID),        MS_NOSUID|MS_NOEXEC,
  81           NULL,          MNT_IN_CONTAINER           },
  82 #if ENABLE_SMACK
  83         { "tmpfs",       "/run",                      "tmpfs",      "mode=755,smackfsroot=*" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  84           mac_smack_use, MNT_FATAL                  },
  85 #endif
  86         { "tmpfs",       "/run",                      "tmpfs",      "mode=755" TMPFS_LIMITS_RUN,               MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  87           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  88         { "cgroup2",     "/sys/fs/cgroup",            "cgroup2",    "nsdelegate",                              MS_NOSUID|MS_NOEXEC|MS_NODEV,
  89           cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  90         { "cgroup2",     "/sys/fs/cgroup",            "cgroup2",    NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  91           cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  92         { "tmpfs",       "/sys/fs/cgroup",            "tmpfs",      "mode=755" TMPFS_LIMITS_SYS_FS_CGROUP,     MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
  93           cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
  94         { "cgroup2",     "/sys/fs/cgroup/unified",    "cgroup2",    "nsdelegate",                              MS_NOSUID|MS_NOEXEC|MS_NODEV,
  95           cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  96         { "cgroup2",     "/sys/fs/cgroup/unified",    "cgroup2",    NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  97           cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  98         { "cgroup",      "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd,xattr",                 MS_NOSUID|MS_NOEXEC|MS_NODEV,
  99           cg_is_legacy_wanted, MNT_IN_CONTAINER     },
 100         { "cgroup",      "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd",                       MS_NOSUID|MS_NOEXEC|MS_NODEV,
 101           cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
 102         { "pstore",      "/sys/fs/pstore",            "pstore",     NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 103           NULL,          MNT_NONE                   },
 104 #if ENABLE_EFI
 105         { "efivarfs",    "/sys/firmware/efi/efivars", "efivarfs",   NULL,                                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 106           is_efi_boot,   MNT_NONE                   },
 107 #endif
 108         { "bpf",         "/sys/fs/bpf",               "bpf",        "mode=700",                                MS_NOSUID|MS_NOEXEC|MS_NODEV,
 109           NULL,          MNT_NONE,                  },
 110 };
 111
 112 /* These are API file systems that might be mounted by other software,
 113  * we just list them here so that we know that we should ignore them */
 114
 115 static const char ignore_paths[] =
 116         /* SELinux file systems */
 117         "/sys/fs/selinux\0"
 118         /* Container bind mounts */
 119         "/proc/sys\0"
 120         "/dev/console\0"
 121         "/proc/kmsg\0";
 122
 123 bool mount_point_is_api(const char *path) {
 124         unsigned i;
 125
 126         /* Checks if this mount point is considered "API", and hence
 127          * should be ignored */
 128
 129         for (i = 0; i < ELEMENTSOF(mount_table); i ++)
 130                 if (path_equal(path, mount_table[i].where))
 131                         return true;
 132
 133         return path_startswith(path, "/sys/fs/cgroup/");
 134 }
 135
 136 bool mount_point_ignore(const char *path) {
 137         const char *i;
 138
 139         NULSTR_FOREACH(i, ignore_paths)
 140                 if (path_equal(path, i))
 141                         return true;
 142
 143         return false;
 144 }
 145
 146 static int mount_one(const MountPoint *p, bool relabel) {
 147         int r, priority;
 148
 149         assert(p);
 150
 151         priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
 152
 153         if (p->condition_fn && !p->condition_fn())
 154                 return 0;
 155
 156         /* Relabel first, just in case */
 157         if (relabel)
 158                 (void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
 159
 160         r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
 161         if (r < 0 && r != -ENOENT) {
 162                 log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
 163                 return (p->mode & MNT_FATAL) ? r : 0;
 164         }
 165         if (r > 0)
 166                 return 0;
 167
 168         /* Skip securityfs in a container */
 169         if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
 170                 return 0;
 171
 172         /* The access mode here doesn't really matter too much, since
 173          * the mounted file system will take precedence anyway. */
 174         if (relabel)
 175                 (void) mkdir_p_label(p->where, 0755);
 176         else
 177                 (void) mkdir_p(p->where, 0755);
 178
 179         log_debug("Mounting %s to %s of type %s with options %s.",
 180                   p->what,
 181                   p->where,
 182                   p->type,
 183                   strna(p->options));
 184
 185         if (mount(p->what,
 186                   p->where,
 187                   p->type,
 188                   p->flags,
 189                   p->options) < 0) {
 190                 log_full_errno(priority, errno, "Failed to mount %s at %s: %m", p->type, p->where);
 191                 return (p->mode & MNT_FATAL) ? -errno : 0;
 192         }
 193
 194         /* Relabel again, since we now mounted something fresh here */
 195         if (relabel)
 196                 (void) label_fix(p->where, 0);
 197
 198         if (p->mode & MNT_CHECK_WRITABLE) {
 199                 if (access(p->where, W_OK) < 0) {
 200                         r = -errno;
 201
 202                         (void) umount(p->where);
 203                         (void) rmdir(p->where);
 204
 205                         log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
 206                         return (p->mode & MNT_FATAL) ? r : 0;
 207                 }
 208         }
 209
 210         return 1;
 211 }
 212
 213 static int mount_points_setup(unsigned n, bool loaded_policy) {
 214         unsigned i;
 215         int r = 0;
 216
 217         for (i = 0; i < n; i ++) {
 218                 int j;
 219
 220                 j = mount_one(mount_table + i, loaded_policy);
 221                 if (j != 0 && r >= 0)
 222                         r = j;
 223         }
 224
 225         return r;
 226 }
 227
 228 int mount_setup_early(void) {
 229         assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
 230
 231         /* Do a minimal mount of /proc and friends to enable the most
 232          * basic stuff, such as SELinux */
 233         return mount_points_setup(N_EARLY_MOUNT, false);
 234 }
 235
 236 static const char *join_with(const char *controller) {
 237
 238         static const char* const pairs[] = {
 239                 "cpu", "cpuacct",
 240                 "net_cls", "net_prio",
 241                 NULL
 242         };
 243
 244         const char *const *x, *const *y;
 245
 246         assert(controller);
 247
 248         /* This will lookup which controller to mount another controller with. Input is a controller name, and output
 249          * is the other controller name. The function works both ways: you can input one and get the other, and input
 250          * the other to get the one. */
 251
 252         STRV_FOREACH_PAIR(x, y, pairs) {
 253                 if (streq(controller, *x))
 254                         return *y;
 255                 if (streq(controller, *y))
 256                         return *x;
 257         }
 258
 259         return NULL;
 260 }
 261
 262 static int symlink_controller(const char *target, const char *alias) {
 263         const char *a;
 264         int r;
 265
 266         assert(target);
 267         assert(alias);
 268
 269         a = strjoina("/sys/fs/cgroup/", alias);
 270
 271         r = symlink_idempotent(target, a, false);
 272         if (r < 0)
 273                 return log_error_errno(r, "Failed to create symlink %s: %m", a);
 274
 275 #ifdef SMACK_RUN_LABEL
 276         const char *p;
 277
 278         p = strjoina("/sys/fs/cgroup/", target);
 279
 280         r = mac_smack_copy(a, p);
 281         if (r < 0 && r != -EOPNOTSUPP)
 282                 return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", p, a);
 283 #endif
 284
 285         return 0;
 286 }
 287
 288 int mount_cgroup_controllers(void) {
 289         _cleanup_set_free_free_ Set *controllers = NULL;
 290         int r;
 291
 292         if (!cg_is_legacy_wanted())
 293                 return 0;
 294
 295         /* Mount all available cgroup controllers that are built into the kernel. */
 296         r = cg_kernel_controllers(&controllers);
 297         if (r < 0)
 298                 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
 299
 300         for (;;) {
 301                 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
 302                 const char *other_controller;
 303                 MountPoint p = {
 304                         .what = "cgroup",
 305                         .type = "cgroup",
 306                         .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
 307                         .mode = MNT_IN_CONTAINER,
 308                 };
 309
 310                 controller = set_steal_first(controllers);
 311                 if (!controller)
 312                         break;
 313
 314                 /* Check if we shall mount this together with another controller */
 315                 other_controller = join_with(controller);
 316                 if (other_controller) {
 317                         _cleanup_free_ char *c = NULL;
 318
 319                         /* Check if the other controller is actually available in the kernel too */
 320                         c = set_remove(controllers, other_controller);
 321                         if (c) {
 322
 323                                 /* Join the two controllers into one string, and maintain a stable ordering */
 324                                 if (strcmp(controller, other_controller) < 0)
 325                                         options = strjoin(controller, ",", other_controller);
 326                                 else
 327                                         options = strjoin(other_controller, ",", controller);
 328                                 if (!options)
 329                                         return log_oom();
 330                         }
 331                 }
 332
 333                 /* The simple case, where there's only one controller to mount together */
 334                 if (!options)
 335                         options = TAKE_PTR(controller);
 336
 337                 where = path_join("/sys/fs/cgroup", options);
 338                 if (!where)
 339                         return log_oom();
 340
 341                 p.where = where;
 342                 p.options = options;
 343
 344                 r = mount_one(&p, true);
 345                 if (r < 0)
 346                         return r;
 347
 348                 /* Create symlinks from the individual controller names, in case we have a joined mount */
 349                 if (controller)
 350                         (void) symlink_controller(options, controller);
 351                 if (other_controller)
 352                         (void) symlink_controller(options, other_controller);
 353         }
 354
 355         /* Now that we mounted everything, let's make the tmpfs the cgroup file systems are mounted into read-only. */
 356         (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755" TMPFS_LIMITS_SYS_FS_CGROUP);
 357
 358         return 0;
 359 }
 360
 361 #if HAVE_SELINUX || ENABLE_SMACK
 362 static int nftw_cb(
 363                 const char *fpath,
 364                 const struct stat *sb,
 365                 int tflag,
 366                 struct FTW *ftwbuf) {
 367
 368         /* No need to label /dev twice in a row... */
 369         if (_unlikely_(ftwbuf->level == 0))
 370                 return FTW_CONTINUE;
 371
 372         (void) label_fix(fpath, 0);
 373
 374         /* /run/initramfs is static data and big, no need to
 375          * dynamically relabel its contents at boot... */
 376         if (_unlikely_(ftwbuf->level == 1 &&
 377                       tflag == FTW_D &&
 378                       streq(fpath, "/run/initramfs")))
 379                 return FTW_SKIP_SUBTREE;
 380
 381         return FTW_CONTINUE;
 382 };
 383
 384 static int relabel_cgroup_filesystems(void) {
 385         int r;
 386         struct statfs st;
 387
 388         r = cg_all_unified();
 389         if (r == 0) {
 390                 /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
 391                    only when the filesystem has been already populated by a previous instance of systemd
 392                    running from initrd. Otherwise don't remount anything and leave the filesystem read-write
 393                    for the cgroup filesystems to be mounted inside. */
 394                 if (statfs("/sys/fs/cgroup", &st) < 0)
 395                         return log_error_errno(errno, "Failed to determine mount flags for /sys/fs/cgroup: %m");
 396
 397                 if (st.f_flags & ST_RDONLY)
 398                         (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
 399
 400                 (void) label_fix("/sys/fs/cgroup", 0);
 401                 (void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 402
 403                 if (st.f_flags & ST_RDONLY)
 404                         (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
 405
 406         } else if (r < 0)
 407                 return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
 408
 409         return 0;
 410 }
 411
 412 static int relabel_extra(void) {
 413         _cleanup_strv_free_ char **files = NULL;
 414         char **file;
 415         int r, c = 0;
 416
 417         /* Support for relabelling additional files or directories after loading the policy. For this, code in the
 418          * initrd simply has to drop in *.relabel files into /run/systemd/relabel-extra.d/. We'll read all such files
 419          * expecting one absolute path by line and will relabel each (and everyone below that in case the path refers
 420          * to a directory). These drop-in files are supposed to be absolutely minimal, and do not understand comments
 421          * and such. After the operation succeeded the files are removed, and the drop-in directory as well, if
 422          * possible.
 423          */
 424
 425         r = conf_files_list(&files, ".relabel", NULL,
 426                             CONF_FILES_FILTER_MASKED | CONF_FILES_REGULAR,
 427                             "/run/systemd/relabel-extra.d/");
 428         if (r < 0)
 429                 return log_error_errno(r, "Failed to enumerate /run/systemd/relabel-extra.d/, ignoring: %m");
 430
 431         STRV_FOREACH(file, files) {
 432                 _cleanup_fclose_ FILE *f = NULL;
 433
 434                 f = fopen(*file, "re");
 435                 if (!f) {
 436                         log_warning_errno(errno, "Failed to open %s, ignoring: %m", *file);
 437                         continue;
 438                 }
 439
 440                 for (;;) {
 441                         _cleanup_free_ char *line = NULL;
 442
 443                         r = read_line(f, LONG_LINE_MAX, &line);
 444                         if (r < 0) {
 445                                 log_warning_errno(r, "Failed to read %s, ignoring: %m", *file);
 446                                 break;
 447                         }
 448                         if (r == 0) /* EOF */
 449                                 break;
 450
 451                         path_simplify(line, true);
 452
 453                         if (!path_is_normalized(line)) {
 454                                 log_warning("Path to relabel is not normalized, ignoring: %s", line);
 455                                 continue;
 456                         }
 457
 458                         if (!path_is_absolute(line)) {
 459                                 log_warning("Path to relabel is not absolute, ignoring: %s", line);
 460                                 continue;
 461                         }
 462
 463                         log_debug("Relabelling additional file/directory '%s'.", line);
 464                         (void) label_fix(line, 0);
 465                         (void) nftw(line, nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 466                         c++;
 467                 }
 468
 469                 if (unlink(*file) < 0)
 470                         log_warning_errno(errno, "Failed to remove %s, ignoring: %m", *file);
 471         }
 472
 473         /* Remove when we complete things. */
 474         if (rmdir("/run/systemd/relabel-extra.d") < 0 &&
 475             errno != ENOENT)
 476                 log_warning_errno(errno, "Failed to remove /run/systemd/relabel-extra.d/ directory: %m");
 477
 478         return c;
 479 }
 480 #endif
 481
 482 int mount_setup(bool loaded_policy, bool leave_propagation) {
 483         int r = 0;
 484
 485         r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
 486         if (r < 0)
 487                 return r;
 488
 489 #if HAVE_SELINUX || ENABLE_SMACK
 490         /* Nodes in devtmpfs and /run need to be manually updated for
 491          * the appropriate labels, after mounting. The other virtual
 492          * API file systems like /sys and /proc do not need that, they
 493          * use the same label for all their files. */
 494         if (loaded_policy) {
 495                 usec_t before_relabel, after_relabel;
 496                 char timespan[FORMAT_TIMESPAN_MAX];
 497                 const char *i;
 498                 int n_extra;
 499
 500                 before_relabel = now(CLOCK_MONOTONIC);
 501
 502                 FOREACH_STRING(i, "/dev", "/dev/shm", "/run")
 503                         (void) nftw(i, nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 504
 505                 (void) relabel_cgroup_filesystems();
 506
 507                 n_extra = relabel_extra();
 508
 509                 after_relabel = now(CLOCK_MONOTONIC);
 510
 511                 log_info("Relabelled /dev, /dev/shm, /run, /sys/fs/cgroup%s in %s.",
 512                          n_extra > 0 ? ", additional files" : "",
 513                          format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
 514         }
 515 #endif
 516
 517         /* Create a few default symlinks, which are normally created
 518          * by udevd, but some scripts might need them before we start
 519          * udevd. */
 520         dev_setup(NULL, UID_INVALID, GID_INVALID);
 521
 522         /* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
 523          * think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
 524          * the box. If specific setups need other settings they can reset the propagation mode to private if
 525          * needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
 526          * container manager we assume the container manager knows what it is doing (for example, because it set up
 527          * some directories with different propagation modes). */
 528         if (detect_container() <= 0 && !leave_propagation)
 529                 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
 530                         log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
 531
 532         /* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
 533          * this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
 534          * misdetect systemd. */
 535         (void) mkdir_label("/run/systemd", 0755);
 536         (void) mkdir_label("/run/systemd/system", 0755);
 537
 538         /* Also create /run/systemd/inaccessible nodes, so that we always have something to mount
 539          * inaccessible nodes from. */
 540         (void) make_inaccessible_nodes(NULL, UID_INVALID, GID_INVALID);
 541
 542         return 0;
 543 }