src/core/mount-setup.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2010 Lennart Poettering
   6 ***/
   7
   8 #include <errno.h>
   9 #include <ftw.h>
  10 #include <stdlib.h>
  11 #include <sys/mount.h>
  12 #include <sys/statvfs.h>
  13 #include <unistd.h>
  14
  15 #include "alloc-util.h"
  16 #include "bus-util.h"
  17 #include "cgroup-util.h"
  18 #include "dev-setup.h"
  19 #include "efivars.h"
  20 #include "fileio.h"
  21 #include "fs-util.h"
  22 #include "label.h"
  23 #include "log.h"
  24 #include "macro.h"
  25 #include "missing.h"
  26 #include "mkdir.h"
  27 #include "mount-setup.h"
  28 #include "mount-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 "util.h"
  35 #include "virt.h"
  36
  37 typedef enum MountMode {
  38         MNT_NONE  =           0,
  39         MNT_FATAL =           1 <<  0,
  40         MNT_IN_CONTAINER =    1 <<  1,
  41         MNT_CHECK_WRITABLE  = 1 <<  2,
  42 } MountMode;
  43
  44 typedef struct MountPoint {
  45         const char *what;
  46         const char *where;
  47         const char *type;
  48         const char *options;
  49         unsigned long flags;
  50         bool (*condition_fn)(void);
  51         MountMode mode;
  52 } MountPoint;
  53
  54 /* The first three entries we might need before SELinux is up. The
  55  * fourth (securityfs) is needed by IMA to load a custom policy. The
  56  * other ones we can delay until SELinux and IMA are loaded. When
  57  * SMACK is enabled we need smackfs, too, so it's a fifth one. */
  58 #if ENABLE_SMACK
  59 #define N_EARLY_MOUNT 5
  60 #else
  61 #define N_EARLY_MOUNT 4
  62 #endif
  63
  64 static const MountPoint mount_table[] = {
  65         { "sysfs",       "/sys",                      "sysfs",      NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  66           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  67         { "proc",        "/proc",                     "proc",       NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  68           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  69         { "devtmpfs",    "/dev",                      "devtmpfs",   "mode=755",                MS_NOSUID|MS_STRICTATIME,
  70           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  71         { "securityfs",  "/sys/kernel/security",      "securityfs", NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  72           NULL,          MNT_NONE                   },
  73 #if ENABLE_SMACK
  74         { "smackfs",     "/sys/fs/smackfs",           "smackfs",    "smackfsdef=*",            MS_NOSUID|MS_NOEXEC|MS_NODEV,
  75           mac_smack_use, MNT_FATAL                  },
  76         { "tmpfs",       "/dev/shm",                  "tmpfs",      "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  77           mac_smack_use, MNT_FATAL                  },
  78 #endif
  79         { "tmpfs",       "/dev/shm",                  "tmpfs",      "mode=1777",               MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  80           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  81         { "devpts",      "/dev/pts",                  "devpts",     "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
  82           NULL,          MNT_IN_CONTAINER           },
  83 #if ENABLE_SMACK
  84         { "tmpfs",       "/run",                      "tmpfs",      "mode=755,smackfsroot=*",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  85           mac_smack_use, MNT_FATAL                  },
  86 #endif
  87         { "tmpfs",       "/run",                      "tmpfs",      "mode=755",                MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  88           NULL,          MNT_FATAL|MNT_IN_CONTAINER },
  89         { "cgroup2",     "/sys/fs/cgroup",            "cgroup2",    "nsdelegate",              MS_NOSUID|MS_NOEXEC|MS_NODEV,
  90           cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  91         { "cgroup2",     "/sys/fs/cgroup",            "cgroup2",    NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  92           cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  93         { "tmpfs",       "/sys/fs/cgroup",            "tmpfs",      "mode=755",                MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
  94           cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
  95         { "cgroup2",     "/sys/fs/cgroup/unified",    "cgroup2",    "nsdelegate",              MS_NOSUID|MS_NOEXEC|MS_NODEV,
  96           cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  97         { "cgroup2",     "/sys/fs/cgroup/unified",    "cgroup2",    NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
  98           cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
  99         { "cgroup",      "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
 100           cg_is_legacy_wanted, MNT_IN_CONTAINER     },
 101         { "cgroup",      "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd",       MS_NOSUID|MS_NOEXEC|MS_NODEV,
 102           cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
 103         { "pstore",      "/sys/fs/pstore",            "pstore",     NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 104           NULL,          MNT_NONE                   },
 105 #if ENABLE_EFI
 106         { "efivarfs",    "/sys/firmware/efi/efivars", "efivarfs",   NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
 107           is_efi_boot,   MNT_NONE                   },
 108 #endif
 109         { "bpf",         "/sys/fs/bpf",               "bpf",        "mode=700",                MS_NOSUID|MS_NOEXEC|MS_NODEV,
 110           NULL,          MNT_NONE,                  },
 111 };
 112
 113 /* These are API file systems that might be mounted by other software,
 114  * we just list them here so that we know that we should ignore them */
 115
 116 static const char ignore_paths[] =
 117         /* SELinux file systems */
 118         "/sys/fs/selinux\0"
 119         /* Container bind mounts */
 120         "/proc/sys\0"
 121         "/dev/console\0"
 122         "/proc/kmsg\0";
 123
 124 bool mount_point_is_api(const char *path) {
 125         unsigned i;
 126
 127         /* Checks if this mount point is considered "API", and hence
 128          * should be ignored */
 129
 130         for (i = 0; i < ELEMENTSOF(mount_table); i ++)
 131                 if (path_equal(path, mount_table[i].where))
 132                         return true;
 133
 134         return path_startswith(path, "/sys/fs/cgroup/");
 135 }
 136
 137 bool mount_point_ignore(const char *path) {
 138         const char *i;
 139
 140         NULSTR_FOREACH(i, ignore_paths)
 141                 if (path_equal(path, i))
 142                         return true;
 143
 144         return false;
 145 }
 146
 147 static int mount_one(const MountPoint *p, bool relabel) {
 148         int r, priority;
 149
 150         assert(p);
 151
 152         priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
 153
 154         if (p->condition_fn && !p->condition_fn())
 155                 return 0;
 156
 157         /* Relabel first, just in case */
 158         if (relabel)
 159                 (void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
 160
 161         r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
 162         if (r < 0 && r != -ENOENT) {
 163                 log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
 164                 return (p->mode & MNT_FATAL) ? r : 0;
 165         }
 166         if (r > 0)
 167                 return 0;
 168
 169         /* Skip securityfs in a container */
 170         if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
 171                 return 0;
 172
 173         /* The access mode here doesn't really matter too much, since
 174          * the mounted file system will take precedence anyway. */
 175         if (relabel)
 176                 (void) mkdir_p_label(p->where, 0755);
 177         else
 178                 (void) mkdir_p(p->where, 0755);
 179
 180         log_debug("Mounting %s to %s of type %s with options %s.",
 181                   p->what,
 182                   p->where,
 183                   p->type,
 184                   strna(p->options));
 185
 186         if (mount(p->what,
 187                   p->where,
 188                   p->type,
 189                   p->flags,
 190                   p->options) < 0) {
 191                 log_full_errno(priority, errno, "Failed to mount %s at %s: %m", p->type, p->where);
 192                 return (p->mode & MNT_FATAL) ? -errno : 0;
 193         }
 194
 195         /* Relabel again, since we now mounted something fresh here */
 196         if (relabel)
 197                 (void) label_fix(p->where, 0);
 198
 199         if (p->mode & MNT_CHECK_WRITABLE) {
 200                 if (access(p->where, W_OK) < 0) {
 201                         r = -errno;
 202
 203                         (void) umount(p->where);
 204                         (void) rmdir(p->where);
 205
 206                         log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
 207                         return (p->mode & MNT_FATAL) ? r : 0;
 208                 }
 209         }
 210
 211         return 1;
 212 }
 213
 214 static int mount_points_setup(unsigned n, bool loaded_policy) {
 215         unsigned i;
 216         int r = 0;
 217
 218         for (i = 0; i < n; i ++) {
 219                 int j;
 220
 221                 j = mount_one(mount_table + i, loaded_policy);
 222                 if (j != 0 && r >= 0)
 223                         r = j;
 224         }
 225
 226         return r;
 227 }
 228
 229 int mount_setup_early(void) {
 230         assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
 231
 232         /* Do a minimal mount of /proc and friends to enable the most
 233          * basic stuff, such as SELinux */
 234         return mount_points_setup(N_EARLY_MOUNT, false);
 235 }
 236
 237 int mount_cgroup_controllers(char ***join_controllers) {
 238         _cleanup_set_free_free_ Set *controllers = NULL;
 239         bool has_argument = !!join_controllers;
 240         int r;
 241
 242         if (!cg_is_legacy_wanted())
 243                 return 0;
 244
 245         /* Mount all available cgroup controllers that are built into the kernel. */
 246
 247         if (!has_argument)
 248                 /* The defaults:
 249                  * mount "cpu" + "cpuacct" together, and "net_cls" + "net_prio".
 250                  *
 251                  * We'd like to add "cpuset" to the mix, but "cpuset" doesn't really
 252                  * work for groups with no initialized attributes.
 253                  */
 254                 join_controllers = (char**[]) {
 255                         STRV_MAKE("cpu", "cpuacct"),
 256                         STRV_MAKE("net_cls", "net_prio"),
 257                         NULL,
 258                 };
 259
 260         r = cg_kernel_controllers(&controllers);
 261         if (r < 0)
 262                 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
 263
 264         for (;;) {
 265                 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
 266                 MountPoint p = {
 267                         .what = "cgroup",
 268                         .type = "cgroup",
 269                         .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
 270                         .mode = MNT_IN_CONTAINER,
 271                 };
 272                 char ***k = NULL;
 273
 274                 controller = set_steal_first(controllers);
 275                 if (!controller)
 276                         break;
 277
 278                 for (k = join_controllers; *k; k++)
 279                         if (strv_find(*k, controller))
 280                                 break;
 281
 282                 if (*k) {
 283                         char **i, **j;
 284
 285                         for (i = *k, j = *k; *i; i++) {
 286
 287                                 if (!streq(*i, controller)) {
 288                                         _cleanup_free_ char *t;
 289
 290                                         t = set_remove(controllers, *i);
 291                                         if (!t) {
 292                                                 if (has_argument)
 293                                                         free(*i);
 294                                                 continue;
 295                                         }
 296                                 }
 297
 298                                 *(j++) = *i;
 299                         }
 300
 301                         *j = NULL;
 302
 303                         options = strv_join(*k, ",");
 304                         if (!options)
 305                                 return log_oom();
 306                 } else
 307                         options = TAKE_PTR(controller);
 308
 309                 where = strappend("/sys/fs/cgroup/", options);
 310                 if (!where)
 311                         return log_oom();
 312
 313                 p.where = where;
 314                 p.options = options;
 315
 316                 r = mount_one(&p, true);
 317                 if (r < 0)
 318                         return r;
 319
 320                 if (r > 0 && *k) {
 321                         char **i;
 322
 323                         for (i = *k; *i; i++) {
 324                                 _cleanup_free_ char *t = NULL;
 325
 326                                 t = strappend("/sys/fs/cgroup/", *i);
 327                                 if (!t)
 328                                         return log_oom();
 329
 330                                 r = symlink(options, t);
 331                                 if (r >= 0) {
 332 #ifdef SMACK_RUN_LABEL
 333                                         _cleanup_free_ char *src;
 334                                         src = strappend("/sys/fs/cgroup/", options);
 335                                         if (!src)
 336                                                 return log_oom();
 337                                         r = mac_smack_copy(t, src);
 338                                         if (r < 0 && r != -EOPNOTSUPP)
 339                                                 return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", src, t);
 340 #endif
 341                                 } else if (errno != EEXIST)
 342                                         return log_error_errno(errno, "Failed to create symlink %s: %m", t);
 343                         }
 344                 }
 345         }
 346
 347         /* Now that we mounted everything, let's make the tmpfs the
 348          * cgroup file systems are mounted into read-only. */
 349         (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 350
 351         return 0;
 352 }
 353
 354 #if HAVE_SELINUX || ENABLE_SMACK
 355 static int nftw_cb(
 356                 const char *fpath,
 357                 const struct stat *sb,
 358                 int tflag,
 359                 struct FTW *ftwbuf) {
 360
 361         /* No need to label /dev twice in a row... */
 362         if (_unlikely_(ftwbuf->level == 0))
 363                 return FTW_CONTINUE;
 364
 365         (void) label_fix(fpath, 0);
 366
 367         /* /run/initramfs is static data and big, no need to
 368          * dynamically relabel its contents at boot... */
 369         if (_unlikely_(ftwbuf->level == 1 &&
 370                       tflag == FTW_D &&
 371                       streq(fpath, "/run/initramfs")))
 372                 return FTW_SKIP_SUBTREE;
 373
 374         return FTW_CONTINUE;
 375 };
 376
 377 static int relabel_cgroup_filesystems(void) {
 378         int r;
 379         struct statfs st;
 380
 381         r = cg_all_unified();
 382         if (r == 0) {
 383                 /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
 384                    only when the filesystem has been already populated by a previous instance of systemd
 385                    running from initrd. Otherwise don't remount anything and leave the filesystem read-write
 386                    for the cgroup filesystems to be mounted inside. */
 387                 if (statfs("/sys/fs/cgroup", &st) < 0)
 388                         return log_error_errno(errno, "Failed to determine mount flags for /sys/fs/cgroup: %m");
 389
 390                 if (st.f_flags & ST_RDONLY)
 391                         (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
 392
 393                 (void) label_fix("/sys/fs/cgroup", 0);
 394                 (void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 395
 396                 if (st.f_flags & ST_RDONLY)
 397                         (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
 398
 399         } else if (r < 0)
 400                 return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
 401
 402         return 0;
 403 }
 404 #endif
 405
 406 int mount_setup(bool loaded_policy) {
 407         int r = 0;
 408
 409         r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
 410         if (r < 0)
 411                 return r;
 412
 413 #if HAVE_SELINUX || ENABLE_SMACK
 414         /* Nodes in devtmpfs and /run need to be manually updated for
 415          * the appropriate labels, after mounting. The other virtual
 416          * API file systems like /sys and /proc do not need that, they
 417          * use the same label for all their files. */
 418         if (loaded_policy) {
 419                 usec_t before_relabel, after_relabel;
 420                 char timespan[FORMAT_TIMESPAN_MAX];
 421
 422                 before_relabel = now(CLOCK_MONOTONIC);
 423
 424                 (void) nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 425                 (void) nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 426                 (void) nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 427
 428                 r = relabel_cgroup_filesystems();
 429                 if (r < 0)
 430                         return r;
 431
 432                 after_relabel = now(CLOCK_MONOTONIC);
 433
 434                 log_info("Relabelled /dev, /run and /sys/fs/cgroup in %s.",
 435                          format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
 436         }
 437 #endif
 438
 439         /* Create a few default symlinks, which are normally created
 440          * by udevd, but some scripts might need them before we start
 441          * udevd. */
 442         dev_setup(NULL, UID_INVALID, GID_INVALID);
 443
 444         /* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
 445          * think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
 446          * the box. If specific setups need other settings they can reset the propagation mode to private if
 447          * needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
 448          * container manager we assume the container manager knows what it is doing (for example, because it set up
 449          * some directories with different propagation modes). */
 450         if (detect_container() <= 0)
 451                 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
 452                         log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
 453
 454         /* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
 455          * this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
 456          * misdetect systemd. */
 457         (void) mkdir_label("/run/systemd", 0755);
 458         (void) mkdir_label("/run/systemd/system", 0755);
 459
 460         /* Set up inaccessible (and empty) file nodes of all types */
 461         (void) mkdir_label("/run/systemd/inaccessible", 0000);
 462         (void) mknod("/run/systemd/inaccessible/reg", S_IFREG | 0000, 0);
 463         (void) mkdir_label("/run/systemd/inaccessible/dir", 0000);
 464         (void) mkfifo("/run/systemd/inaccessible/fifo", 0000);
 465         (void) mknod("/run/systemd/inaccessible/sock", S_IFSOCK | 0000, 0);
 466
 467         /* The following two are likely to fail if we lack the privs for it (for example in an userns environment, if
 468          * CAP_SYS_MKNOD is missing, or if a device node policy prohibit major/minor of 0 device nodes to be
 469          * created). But that's entirely fine. Consumers of these files should carry fallback to use a different node
 470          * then, for example /run/systemd/inaccessible/sock, which is close enough in behaviour and semantics for most
 471          * uses. */
 472         (void) mknod("/run/systemd/inaccessible/chr", S_IFCHR | 0000, makedev(0, 0));
 473         (void) mknod("/run/systemd/inaccessible/blk", S_IFBLK | 0000, makedev(0, 0));
 474
 475         return 0;
 476 }