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 };
 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;
 149
 150         assert(p);
 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, true, true);
 158
 159         r = path_is_mount_point(p->where, AT_SYMLINK_FOLLOW);
 160         if (r < 0 && r != -ENOENT) {
 161                 log_full_errno((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, 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((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, 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, false, false);
 196
 197         return 1;
 198 }
 199
 200 static int mount_points_setup(unsigned n, bool loaded_policy) {
 201         unsigned i;
 202         int r = 0;
 203
 204         for (i = 0; i < n; i ++) {
 205                 int j;
 206
 207                 j = mount_one(mount_table + i, loaded_policy);
 208                 if (j != 0 && r >= 0)
 209                         r = j;
 210         }
 211
 212         return r;
 213 }
 214
 215 int mount_setup_early(void) {
 216         assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
 217
 218         /* Do a minimal mount of /proc and friends to enable the most
 219          * basic stuff, such as SELinux */
 220         return mount_points_setup(N_EARLY_MOUNT, false);
 221 }
 222
 223 int mount_cgroup_controllers(char ***join_controllers) {
 224         _cleanup_set_free_free_ Set *controllers = NULL;
 225         int r;
 226
 227         if (!cg_is_legacy_wanted())
 228                 return 0;
 229
 230         /* Mount all available cgroup controllers that are built into the kernel. */
 231
 232         controllers = set_new(&string_hash_ops);
 233         if (!controllers)
 234                 return log_oom();
 235
 236         r = cg_kernel_controllers(controllers);
 237         if (r < 0)
 238                 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
 239
 240         for (;;) {
 241                 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
 242                 MountPoint p = {
 243                         .what = "cgroup",
 244                         .type = "cgroup",
 245                         .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
 246                         .mode = MNT_IN_CONTAINER,
 247                 };
 248                 char ***k = NULL;
 249
 250                 controller = set_steal_first(controllers);
 251                 if (!controller)
 252                         break;
 253
 254                 if (join_controllers)
 255                         for (k = join_controllers; *k; k++)
 256                                 if (strv_find(*k, controller))
 257                                         break;
 258
 259                 if (k && *k) {
 260                         char **i, **j;
 261
 262                         for (i = *k, j = *k; *i; i++) {
 263
 264                                 if (!streq(*i, controller)) {
 265                                         _cleanup_free_ char *t;
 266
 267                                         t = set_remove(controllers, *i);
 268                                         if (!t) {
 269                                                 free(*i);
 270                                                 continue;
 271                                         }
 272                                 }
 273
 274                                 *(j++) = *i;
 275                         }
 276
 277                         *j = NULL;
 278
 279                         options = strv_join(*k, ",");
 280                         if (!options)
 281                                 return log_oom();
 282                 } else {
 283                         options = controller;
 284                         controller = NULL;
 285                 }
 286
 287                 where = strappend("/sys/fs/cgroup/", options);
 288                 if (!where)
 289                         return log_oom();
 290
 291                 p.where = where;
 292                 p.options = options;
 293
 294                 r = mount_one(&p, true);
 295                 if (r < 0)
 296                         return r;
 297
 298                 if (r > 0 && k && *k) {
 299                         char **i;
 300
 301                         for (i = *k; *i; i++) {
 302                                 _cleanup_free_ char *t = NULL;
 303
 304                                 t = strappend("/sys/fs/cgroup/", *i);
 305                                 if (!t)
 306                                         return log_oom();
 307
 308                                 r = symlink(options, t);
 309                                 if (r >= 0) {
 310 #ifdef SMACK_RUN_LABEL
 311                                         _cleanup_free_ char *src;
 312                                         src = strappend("/sys/fs/cgroup/", options);
 313                                         if (!src)
 314                                                 return log_oom();
 315                                         r = mac_smack_copy(t, src);
 316                                         if (r < 0 && r != -EOPNOTSUPP)
 317                                                 return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", src, t);
 318 #endif
 319                                 } else if (errno != EEXIST)
 320                                         return log_error_errno(errno, "Failed to create symlink %s: %m", t);
 321                         }
 322                 }
 323         }
 324
 325         /* Now that we mounted everything, let's make the tmpfs the
 326          * cgroup file systems are mounted into read-only. */
 327         (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 328
 329         return 0;
 330 }
 331
 332 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
 333 static int nftw_cb(
 334                 const char *fpath,
 335                 const struct stat *sb,
 336                 int tflag,
 337                 struct FTW *ftwbuf) {
 338
 339         /* No need to label /dev twice in a row... */
 340         if (_unlikely_(ftwbuf->level == 0))
 341                 return FTW_CONTINUE;
 342
 343         label_fix(fpath, false, false);
 344
 345         /* /run/initramfs is static data and big, no need to
 346          * dynamically relabel its contents at boot... */
 347         if (_unlikely_(ftwbuf->level == 1 &&
 348                       tflag == FTW_D &&
 349                       streq(fpath, "/run/initramfs")))
 350                 return FTW_SKIP_SUBTREE;
 351
 352         return FTW_CONTINUE;
 353 };
 354 #endif
 355
 356 int mount_setup(bool loaded_policy) {
 357         int r = 0;
 358
 359         r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
 360
 361         if (r < 0)
 362                 return r;
 363
 364 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
 365         /* Nodes in devtmpfs and /run need to be manually updated for
 366          * the appropriate labels, after mounting. The other virtual
 367          * API file systems like /sys and /proc do not need that, they
 368          * use the same label for all their files. */
 369         if (loaded_policy) {
 370                 usec_t before_relabel, after_relabel;
 371                 char timespan[FORMAT_TIMESPAN_MAX];
 372
 373                 before_relabel = now(CLOCK_MONOTONIC);
 374
 375                 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 376                 nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 377                 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 378
 379                 after_relabel = now(CLOCK_MONOTONIC);
 380
 381                 log_info("Relabelled /dev and /run in %s.",
 382                          format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
 383         }
 384 #endif
 385
 386         /* Create a few default symlinks, which are normally created
 387          * by udevd, but some scripts might need them before we start
 388          * udevd. */
 389         dev_setup(NULL, UID_INVALID, GID_INVALID);
 390
 391         /* Mark the root directory as shared in regards to mount
 392          * propagation. The kernel defaults to "private", but we think
 393          * it makes more sense to have a default of "shared" so that
 394          * nspawn and the container tools work out of the box. If
 395          * specific setups need other settings they can reset the
 396          * propagation mode to private if needed. */
 397         if (detect_container() <= 0)
 398                 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
 399                         log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
 400
 401         /* Create a few directories we always want around, Note that
 402          * sd_booted() checks for /run/systemd/system, so this mkdir
 403          * really needs to stay for good, otherwise software that
 404          * copied sd-daemon.c into their sources will misdetect
 405          * systemd. */
 406         mkdir_label("/run/systemd", 0755);
 407         mkdir_label("/run/systemd/system", 0755);
 408         mkdir_label("/run/systemd/inaccessible", 0000);
 409
 410         return 0;
 411 }