src/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 General Public License as published by
  10   the Free Software Foundation; either version 2 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   General Public License for more details.
  17
  18   You should have received a copy of the GNU 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 "log.h"
  34 #include "macro.h"
  35 #include "util.h"
  36 #include "label.h"
  37 #include "set.h"
  38 #include "strv.h"
  39
  40 #ifndef TTY_GID
  41 #define TTY_GID 5
  42 #endif
  43
  44 typedef struct MountPoint {
  45         const char *what;
  46         const char *where;
  47         const char *type;
  48         const char *options;
  49         unsigned long flags;
  50         bool fatal;
  51 } MountPoint;
  52
  53 /* The first three entries we might need before SELinux is up. The
  54  * other ones we can delay until SELinux is loaded. */
  55 #define N_EARLY_MOUNT 3
  56
  57 static const MountPoint mount_table[] = {
  58         { "proc",     "/proc",                  "proc",     NULL,                MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
  59         { "sysfs",    "/sys",                   "sysfs",    NULL,                MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
  60         { "devtmpfs", "/dev",                   "devtmpfs", "mode=755",          MS_NOSUID,                    true },
  61         { "tmpfs",    "/dev/shm",               "tmpfs",    "mode=1777",         MS_NOSUID|MS_NODEV,           true },
  62         { "devpts",   "/dev/pts",               "devpts",   "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC, false },
  63         { "tmpfs",    "/run",                   "tmpfs",    "mode=755",          MS_NOSUID|MS_NODEV, true },
  64         { "tmpfs",    "/sys/fs/cgroup",         "tmpfs",    "mode=755",          MS_NOSUID|MS_NOEXEC|MS_NODEV, false },
  65         { "cgroup",   "/sys/fs/cgroup/systemd", "cgroup",   "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV, false },
  66 };
  67
  68 /* These are API file systems that might be mounted by other software,
  69  * we just list them here so that we know that we should ignore them */
  70
  71 static const char * const ignore_paths[] = {
  72         "/sys/fs/selinux",
  73         "/selinux",
  74         "/proc/bus/usb"
  75 };
  76
  77 bool mount_point_is_api(const char *path) {
  78         unsigned i;
  79
  80         /* Checks if this mount point is considered "API", and hence
  81          * should be ignored */
  82
  83         for (i = 0; i < ELEMENTSOF(mount_table); i ++)
  84                 if (path_equal(path, mount_table[i].where))
  85                         return true;
  86
  87         return path_startswith(path, "/sys/fs/cgroup/");
  88 }
  89
  90 bool mount_point_ignore(const char *path) {
  91         unsigned i;
  92
  93         for (i = 0; i < ELEMENTSOF(ignore_paths); i++)
  94                 if (path_equal(path, ignore_paths[i]))
  95                         return true;
  96
  97         return false;
  98 }
  99
 100 static int mount_one(const MountPoint *p, bool relabel) {
 101         int r;
 102
 103         assert(p);
 104
 105         /* Relabel first, just in case */
 106         if (relabel)
 107                 label_fix(p->where, true);
 108
 109         if ((r = path_is_mount_point(p->where, true)) < 0)
 110                 return r;
 111
 112         if (r > 0)
 113                 return 0;
 114
 115         /* The access mode here doesn't really matter too much, since
 116          * the mounted file system will take precedence anyway. */
 117         mkdir_p(p->where, 0755);
 118
 119         log_debug("Mounting %s to %s of type %s with options %s.",
 120                   p->what,
 121                   p->where,
 122                   p->type,
 123                   strna(p->options));
 124
 125         if (mount(p->what,
 126                   p->where,
 127                   p->type,
 128                   p->flags,
 129                   p->options) < 0) {
 130                 log_error("Failed to mount %s: %s", p->where, strerror(errno));
 131                 return p->fatal ? -errno : 0;
 132         }
 133
 134         /* Relabel again, since we now mounted something fresh here */
 135         if (relabel)
 136                 label_fix(p->where, false);
 137
 138         return 1;
 139 }
 140
 141 int mount_setup_early(void) {
 142         unsigned i;
 143         int r = 0;
 144
 145         assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
 146
 147         /* Do a minimal mount of /proc and friends to enable the most
 148          * basic stuff, such as SELinux */
 149         for (i = 0; i < N_EARLY_MOUNT; i ++)  {
 150                 int j;
 151
 152                 j = mount_one(mount_table + i, false);
 153                 if (r == 0)
 154                         r = j;
 155         }
 156
 157         return r;
 158 }
 159
 160 int mount_cgroup_controllers(char ***join_controllers) {
 161         int r;
 162         FILE *f;
 163         char buf[LINE_MAX];
 164         Set *controllers;
 165
 166         /* Mount all available cgroup controllers that are built into the kernel. */
 167
 168         f = fopen("/proc/cgroups", "re");
 169         if (!f) {
 170                 log_error("Failed to enumerate cgroup controllers: %m");
 171                 return 0;
 172         }
 173
 174         controllers = set_new(string_hash_func, string_compare_func);
 175         if (!controllers) {
 176                 r = -ENOMEM;
 177                 log_error("Failed to allocate controller set.");
 178                 goto finish;
 179         }
 180
 181         /* Ignore the header line */
 182         (void) fgets(buf, sizeof(buf), f);
 183
 184         for (;;) {
 185                 char *controller;
 186                 int enabled = 0;
 187
 188                 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
 189
 190                         if (feof(f))
 191                                 break;
 192
 193                         log_error("Failed to parse /proc/cgroups.");
 194                         r = -EIO;
 195                         goto finish;
 196                 }
 197
 198                 if (!enabled) {
 199                         free(controller);
 200                         continue;
 201                 }
 202
 203                 r = set_put(controllers, controller);
 204                 if (r < 0) {
 205                         log_error("Failed to add controller to set.");
 206                         free(controller);
 207                         goto finish;
 208                 }
 209         }
 210
 211         for (;;) {
 212                 MountPoint p;
 213                 char *controller, *where, *options;
 214                 char ***k = NULL;
 215
 216                 controller = set_steal_first(controllers);
 217                 if (!controller)
 218                         break;
 219
 220                 if (join_controllers)
 221                         for (k = join_controllers; *k; k++)
 222                                 if (strv_find(*k, controller))
 223                                         break;
 224
 225                 if (k && *k) {
 226                         char **i, **j;
 227
 228                         for (i = *k, j = *k; *i; i++) {
 229
 230                                 if (!streq(*i, controller)) {
 231                                         char *t;
 232
 233                                         t = set_remove(controllers, *i);
 234                                         if (!t) {
 235                                                 free(*i);
 236                                                 continue;
 237                                         }
 238                                         free(t);
 239                                 }
 240
 241                                 *(j++) = *i;
 242                         }
 243
 244                         *j = NULL;
 245
 246                         options = strv_join(*k, ",");
 247                         if (!options) {
 248                                 log_error("Failed to join options");
 249                                 free(controller);
 250                                 r = -ENOMEM;
 251                                 goto finish;
 252                         }
 253
 254                 } else {
 255                         options = controller;
 256                         controller = NULL;
 257                 }
 258
 259                 where = strappend("/sys/fs/cgroup/", options);
 260                 if (!where) {
 261                         log_error("Failed to build path");
 262                         free(options);
 263                         r = -ENOMEM;
 264                         goto finish;
 265                 }
 266
 267                 zero(p);
 268                 p.what = "cgroup";
 269                 p.where = where;
 270                 p.type = "cgroup";
 271                 p.options = options;
 272                 p.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV;
 273                 p.fatal = false;
 274
 275                 r = mount_one(&p, true);
 276                 free(controller);
 277                 free(where);
 278
 279                 if (r < 0) {
 280                         free(options);
 281                         goto finish;
 282                 }
 283
 284                 if (r > 0 && k && *k) {
 285                         char **i;
 286
 287                         for (i = *k; *i; i++) {
 288                                 char *t;
 289
 290                                 t = strappend("/sys/fs/cgroup/", *i);
 291                                 if (!t) {
 292                                         log_error("Failed to build path");
 293                                         r = -ENOMEM;
 294                                         free(options);
 295                                         goto finish;
 296                                 }
 297
 298                                 r = symlink(options, t);
 299                                 free(t);
 300
 301                                 if (r < 0 && errno != EEXIST) {
 302                                         log_error("Failed to create symlink: %m");
 303                                         r = -errno;
 304                                         free(options);
 305                                         goto finish;
 306                                 }
 307                         }
 308                 }
 309
 310                 free(options);
 311         }
 312
 313         r = 0;
 314
 315 finish:
 316         set_free_free(controllers);
 317
 318         fclose(f);
 319
 320         return r;
 321 }
 322
 323 static int symlink_and_label(const char *old_path, const char *new_path) {
 324         int r;
 325
 326         assert(old_path);
 327         assert(new_path);
 328
 329         if ((r = label_symlinkfile_set(new_path)) < 0)
 330                 return r;
 331
 332         if (symlink(old_path, new_path) < 0)
 333                 r = -errno;
 334
 335         label_file_clear();
 336
 337         return r;
 338 }
 339
 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 (ftwbuf->level == 0)
 348                 return 0;
 349
 350         label_fix(fpath, true);
 351         return 0;
 352 };
 353
 354 int mount_setup(bool loaded_policy) {
 355
 356         const char symlinks[] =
 357                 "/proc/kcore\0"      "/dev/core\0"
 358                 "/proc/self/fd\0"    "/dev/fd\0"
 359                 "/proc/self/fd/0\0"  "/dev/stdin\0"
 360                 "/proc/self/fd/1\0"  "/dev/stdout\0"
 361                 "/proc/self/fd/2\0"  "/dev/stderr\0";
 362
 363         int r;
 364         unsigned i;
 365         const char *j, *k;
 366
 367         for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
 368                 r = mount_one(mount_table + i, true);
 369
 370                 if (r < 0)
 371                         return r;
 372         }
 373
 374         /* Nodes in devtmpfs and /run need to be manually updated for
 375          * the appropriate labels, after mounting. The other virtual
 376          * API file systems like /sys and /proc do not need that, they
 377          * use the same label for all their files. */
 378         if (loaded_policy) {
 379                 usec_t before_relabel, after_relabel;
 380                 char timespan[FORMAT_TIMESPAN_MAX];
 381
 382                 before_relabel = now(CLOCK_MONOTONIC);
 383
 384                 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS);
 385                 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS);
 386
 387                 after_relabel = now(CLOCK_MONOTONIC);
 388
 389                 log_info("Relabelled /dev and /run in %s.",
 390                          format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel));
 391
 392         }
 393
 394         /* Create a few default symlinks, which are normally created
 395          * by udevd, but some scripts might need them before we start
 396          * udevd. */
 397         NULSTR_FOREACH_PAIR(j, k, symlinks)
 398                 symlink_and_label(j, k);
 399
 400         /* Create a few directories we always want around */
 401         mkdir("/run/systemd", 0755);
 402         mkdir("/run/systemd/system", 0755);
 403
 404         return 0;
 405 }