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