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