src/core/mount-setup.c

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