src/core/mount-setup.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <sys/mount.h>
  23 #include <errno.h>
  24 #include <sys/stat.h>
  25 #include <stdlib.h>
  26 #include <string.h>
  27 #include <libgen.h>
  28 #include <assert.h>
  29 #include <unistd.h>
  30 #include <ftw.h>
  31
  32 #include "mount-setup.h"
  33 #include "dev-setup.h"
  34 #include "log.h"
  35 #include "macro.h"
  36 #include "util.h"
  37 #include "label.h"
  38 #include "set.h"
  39 #include "strv.h"
  40 #include "mkdir.h"
  41 #include "path-util.h"
  42 #include "missing.h"
  43 #include "virt.h"
  44 #include "efivars.h"
  45 #include "smack-util.h"
  46
  47 #ifndef TTY_GID
  48 #define TTY_GID 5
  49 #endif
  50
  51 typedef enum MountMode {
  52         MNT_NONE  =        0,
  53         MNT_FATAL =        1 <<  0,
  54         MNT_IN_CONTAINER = 1 <<  1,
  55 } MountMode;
  56
  57 typedef struct MountPoint {
  58         const char *what;
  59         const char *where;
  60         const char *type;
  61         const char *options;
  62         unsigned long flags;
  63         bool (*condition_fn)(void);
  64         MountMode mode;
  65 } MountPoint;
  66
  67 /* The first three entries we might need before SELinux is up. The
  68  * fourth (securityfs) is needed by IMA to load a custom policy. The
  69  * other ones we can delay until SELinux and IMA are loaded. */
  70 #define N_EARLY_MOUNT 5
  71
  72 static const MountPoint mount_table[] = {
  73         { "proc",       "/proc",                     "proc",       NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
  74           NULL,       MNT_FATAL|MNT_IN_CONTAINER },
  75         { "sysfs",      "/sys",                      "sysfs",      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|MS_STRICTATIME,
  83           use_smack,  MNT_FATAL },
  84         { "tmpfs",      "/dev/shm",                  "tmpfs",      "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  85           use_smack,  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           use_smack,  MNT_FATAL },
  94 #endif
  95         { "tmpfs",      "/run",                      "tmpfs",      "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
  96           NULL,       MNT_FATAL|MNT_IN_CONTAINER },
  97         { "tmpfs",      "/sys/fs/cgroup",            "tmpfs",      "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
  98           NULL,       MNT_IN_CONTAINER },
  99 #ifdef HAVE_XATTR
 100         { "cgroup",     "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
 101           NULL,       MNT_IN_CONTAINER },
 102 #endif
 103         { "cgroup",     "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
 104           NULL,       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         "/selinux\0"
 120         /* Legacy cgroup mount points */
 121         "/dev/cgroup\0"
 122         "/cgroup\0"
 123         /* Legacy kernel file system */
 124         "/proc/bus/usb\0"
 125         /* Container bind mounts */
 126         "/proc/sys\0"
 127         "/dev/console\0"
 128         "/proc/kmsg\0";
 129
 130 bool mount_point_is_api(const char *path) {
 131         unsigned i;
 132
 133         /* Checks if this mount point is considered "API", and hence
 134          * should be ignored */
 135
 136         for (i = 0; i < ELEMENTSOF(mount_table); i ++)
 137                 if (path_equal(path, mount_table[i].where))
 138                         return true;
 139
 140         return path_startswith(path, "/sys/fs/cgroup/");
 141 }
 142
 143 bool mount_point_ignore(const char *path) {
 144         const char *i;
 145
 146         NULSTR_FOREACH(i, ignore_paths)
 147                 if (path_equal(path, i))
 148                         return true;
 149
 150         return false;
 151 }
 152
 153 static int mount_one(const MountPoint *p, bool relabel) {
 154         int r;
 155
 156         assert(p);
 157
 158         if (p->condition_fn && !p->condition_fn())
 159                 return 0;
 160
 161         /* Relabel first, just in case */
 162         if (relabel)
 163                 label_fix(p->where, true, true);
 164
 165         r = path_is_mount_point(p->where, true);
 166         if (r < 0)
 167                 return r;
 168
 169         if (r > 0)
 170                 return 0;
 171
 172         /* Skip securityfs in a container */
 173         if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0)
 174                 return 0;
 175
 176         /* The access mode here doesn't really matter too much, since
 177          * the mounted file system will take precedence anyway. */
 178         mkdir_p_label(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((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s: %m", 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                 label_fix(p->where, false, false);
 198
 199         return 1;
 200 }
 201
 202 int mount_setup_early(void) {
 203         unsigned i;
 204         int r = 0;
 205
 206         assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
 207
 208         /* Do a minimal mount of /proc and friends to enable the most
 209          * basic stuff, such as SELinux */
 210         for (i = 0; i < N_EARLY_MOUNT; i ++)  {
 211                 int j;
 212
 213                 j = mount_one(mount_table + i, false);
 214                 if (r == 0)
 215                         r = j;
 216         }
 217
 218         return r;
 219 }
 220
 221 int mount_cgroup_controllers(char ***join_controllers) {
 222         int r;
 223         char buf[LINE_MAX];
 224         _cleanup_set_free_free_ Set *controllers = NULL;
 225         _cleanup_fclose_ FILE *f;
 226
 227         /* Mount all available cgroup controllers that are built into the kernel. */
 228
 229         f = fopen("/proc/cgroups", "re");
 230         if (!f) {
 231                 log_error("Failed to enumerate cgroup controllers: %m");
 232                 return 0;
 233         }
 234
 235         controllers = set_new(string_hash_func, string_compare_func);
 236         if (!controllers)
 237                 return log_oom();
 238
 239         /* Ignore the header line */
 240         (void) fgets(buf, sizeof(buf), f);
 241
 242         for (;;) {
 243                 char *controller;
 244                 int enabled = 0;
 245
 246                 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
 247
 248                         if (feof(f))
 249                                 break;
 250
 251                         log_error("Failed to parse /proc/cgroups.");
 252                         return -EIO;
 253                 }
 254
 255                 if (!enabled) {
 256                         free(controller);
 257                         continue;
 258                 }
 259
 260                 r = set_consume(controllers, controller);
 261                 if (r < 0) {
 262                         log_error("Failed to add controller to set.");
 263                         return r;
 264                 }
 265         }
 266
 267         for (;;) {
 268                 MountPoint p = {
 269                         .what = "cgroup",
 270                         .type = "cgroup",
 271                         .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
 272                         .mode = MNT_IN_CONTAINER,
 273                 };
 274                 char ***k = NULL;
 275                 _cleanup_free_ char *options = NULL, *controller;
 276
 277                 controller = set_steal_first(controllers);
 278                 if (!controller)
 279                         break;
 280
 281                 if (join_controllers)
 282                         for (k = join_controllers; *k; k++)
 283                                 if (strv_find(*k, controller))
 284                                         break;
 285
 286                 if (k && *k) {
 287                         char **i, **j;
 288
 289                         for (i = *k, j = *k; *i; i++) {
 290
 291                                 if (!streq(*i, controller)) {
 292                                         char _cleanup_free_ *t;
 293
 294                                         t = set_remove(controllers, *i);
 295                                         if (!t) {
 296                                                 free(*i);
 297                                                 continue;
 298                                         }
 299                                 }
 300
 301                                 *(j++) = *i;
 302                         }
 303
 304                         *j = NULL;
 305
 306                         options = strv_join(*k, ",");
 307                         if (!options)
 308                                 return log_oom();
 309                 } else {
 310                         options = controller;
 311                         controller = NULL;
 312                 }
 313
 314                 p.where = strappenda("/sys/fs/cgroup/", options);
 315                 p.options = options;
 316
 317                 r = mount_one(&p, true);
 318                 if (r < 0)
 319                         return r;
 320
 321                 if (r > 0 && k && *k) {
 322                         char **i;
 323
 324                         for (i = *k; *i; i++) {
 325                                 char *t = strappenda("/sys/fs/cgroup/", *i);
 326
 327                                 r = symlink(options, t);
 328                                 if (r < 0 && errno != EEXIST) {
 329                                         log_error("Failed to create symlink %s: %m", t);
 330                                         return -errno;
 331                                 }
 332                         }
 333                 }
 334         }
 335
 336         return 0;
 337 }
 338
 339 static int nftw_cb(
 340                 const char *fpath,
 341                 const struct stat *sb,
 342                 int tflag,
 343                 struct FTW *ftwbuf) {
 344
 345         /* No need to label /dev twice in a row... */
 346         if (_unlikely_(ftwbuf->level == 0))
 347                 return FTW_CONTINUE;
 348
 349         label_fix(fpath, false, false);
 350
 351         /* /run/initramfs is static data and big, no need to
 352          * dynamically relabel its contents at boot... */
 353         if (_unlikely_(ftwbuf->level == 1 &&
 354                       tflag == FTW_D &&
 355                       streq(fpath, "/run/initramfs")))
 356                 return FTW_SKIP_SUBTREE;
 357
 358         return FTW_CONTINUE;
 359 };
 360
 361 int mount_setup(bool loaded_policy) {
 362         int r;
 363         unsigned i;
 364
 365         for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
 366                 r = mount_one(mount_table + i, true);
 367
 368                 if (r < 0)
 369                         return r;
 370         }
 371
 372         /* Nodes in devtmpfs and /run need to be manually updated for
 373          * the appropriate labels, after mounting. The other virtual
 374          * API file systems like /sys and /proc do not need that, they
 375          * use the same label for all their files. */
 376         if (loaded_policy) {
 377                 usec_t before_relabel, after_relabel;
 378                 char timespan[FORMAT_TIMESPAN_MAX];
 379
 380                 before_relabel = now(CLOCK_MONOTONIC);
 381
 382                 nftw("/dev", 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
 391         /* Create a few default symlinks, which are normally created
 392          * by udevd, but some scripts might need them before we start
 393          * udevd. */
 394         dev_setup(NULL);
 395
 396         /* Mark the root directory as shared in regards to mount
 397          * propagation. The kernel defaults to "private", but we think
 398          * it makes more sense to have a default of "shared" so that
 399          * nspawn and the container tools work out of the box. If
 400          * specific setups need other settings they can reset the
 401          * propagation mode to private if needed. */
 402         if (detect_container(NULL) <= 0)
 403                 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
 404                         log_warning("Failed to set up the root directory for shared mount propagation: %m");
 405
 406         /* Create a few directories we always want around, Note that
 407          * sd_booted() checks for /run/systemd/system, so this mkdir
 408          * really needs to stay for good, otherwise software that
 409          * copied sd-daemon.c into their sources will misdetect
 410          * systemd. */
 411         mkdir_label("/run/systemd", 0755);
 412         mkdir_label("/run/systemd/system", 0755);
 413         mkdir_label("/run/systemd/inaccessible", 0000);
 414
 415         return 0;
 416 }