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