src/core/namespace.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 <sched.h>
  22 #include <stdio.h>
  23 #include <string.h>
  24 #include <sys/mount.h>
  25 #include <sys/stat.h>
  26 #include <unistd.h>
  27 #include <linux/fs.h>
  28
  29 #include "alloc-util.h"
  30 #include "dev-setup.h"
  31 #include "fd-util.h"
  32 #include "loopback-setup.h"
  33 #include "missing.h"
  34 #include "mkdir.h"
  35 #include "mount-util.h"
  36 #include "namespace.h"
  37 #include "path-util.h"
  38 #include "selinux-util.h"
  39 #include "socket-util.h"
  40 #include "string-table.h"
  41 #include "string-util.h"
  42 #include "strv.h"
  43 #include "umask-util.h"
  44 #include "user-util.h"
  45 #include "util.h"
  46
  47 typedef enum MountMode {
  48         /* This is ordered by priority! */
  49         INACCESSIBLE,
  50         READONLY,
  51         PRIVATE_TMP,
  52         PRIVATE_VAR_TMP,
  53         PRIVATE_DEV,
  54         READWRITE
  55 } MountMode;
  56
  57 typedef struct BindMount {
  58         const char *path;
  59         MountMode mode;
  60         bool done;
  61         bool ignore;
  62 } BindMount;
  63
  64 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
  65         char **i;
  66
  67         assert(p);
  68
  69         STRV_FOREACH(i, strv) {
  70
  71                 (*p)->ignore = false;
  72                 (*p)->done = false;
  73
  74                 if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
  75                         (*p)->ignore = true;
  76                         (*i)++;
  77                 }
  78
  79                 if (!path_is_absolute(*i))
  80                         return -EINVAL;
  81
  82                 (*p)->path = *i;
  83                 (*p)->mode = mode;
  84                 (*p)++;
  85         }
  86
  87         return 0;
  88 }
  89
  90 static int mount_path_compare(const void *a, const void *b) {
  91         const BindMount *p = a, *q = b;
  92         int d;
  93
  94         d = path_compare(p->path, q->path);
  95
  96         if (d == 0) {
  97                 /* If the paths are equal, check the mode */
  98                 if (p->mode < q->mode)
  99                         return -1;
 100
 101                 if (p->mode > q->mode)
 102                         return 1;
 103
 104                 return 0;
 105         }
 106
 107         /* If the paths are not equal, then order prefixes first */
 108         return d;
 109 }
 110
 111 static void drop_duplicates(BindMount *m, unsigned *n) {
 112         BindMount *f, *t, *previous;
 113
 114         assert(m);
 115         assert(n);
 116
 117         for (f = m, t = m, previous = NULL; f < m+*n; f++) {
 118
 119                 /* The first one wins */
 120                 if (previous && path_equal(f->path, previous->path))
 121                         continue;
 122
 123                 *t = *f;
 124
 125                 previous = t;
 126
 127                 t++;
 128         }
 129
 130         *n = t - m;
 131 }
 132
 133 static int mount_dev(BindMount *m) {
 134         static const char devnodes[] =
 135                 "/dev/null\0"
 136                 "/dev/zero\0"
 137                 "/dev/full\0"
 138                 "/dev/random\0"
 139                 "/dev/urandom\0"
 140                 "/dev/tty\0";
 141
 142         char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
 143         const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
 144         _cleanup_umask_ mode_t u;
 145         int r;
 146
 147         assert(m);
 148
 149         u = umask(0000);
 150
 151         if (!mkdtemp(temporary_mount))
 152                 return -errno;
 153
 154         dev = strjoina(temporary_mount, "/dev");
 155         (void) mkdir(dev, 0755);
 156         if (mount("tmpfs", dev, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=755") < 0) {
 157                 r = -errno;
 158                 goto fail;
 159         }
 160
 161         devpts = strjoina(temporary_mount, "/dev/pts");
 162         (void) mkdir(devpts, 0755);
 163         if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
 164                 r = -errno;
 165                 goto fail;
 166         }
 167
 168         devptmx = strjoina(temporary_mount, "/dev/ptmx");
 169         if (symlink("pts/ptmx", devptmx) < 0) {
 170                 r = -errno;
 171                 goto fail;
 172         }
 173
 174         devshm = strjoina(temporary_mount, "/dev/shm");
 175         (void) mkdir(devshm, 01777);
 176         r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
 177         if (r < 0) {
 178                 r = -errno;
 179                 goto fail;
 180         }
 181
 182         devmqueue = strjoina(temporary_mount, "/dev/mqueue");
 183         (void) mkdir(devmqueue, 0755);
 184         (void) mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
 185
 186         devhugepages = strjoina(temporary_mount, "/dev/hugepages");
 187         (void) mkdir(devhugepages, 0755);
 188         (void) mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
 189
 190         devlog = strjoina(temporary_mount, "/dev/log");
 191         (void) symlink("/run/systemd/journal/dev-log", devlog);
 192
 193         NULSTR_FOREACH(d, devnodes) {
 194                 _cleanup_free_ char *dn = NULL;
 195                 struct stat st;
 196
 197                 r = stat(d, &st);
 198                 if (r < 0) {
 199
 200                         if (errno == ENOENT)
 201                                 continue;
 202
 203                         r = -errno;
 204                         goto fail;
 205                 }
 206
 207                 if (!S_ISBLK(st.st_mode) &&
 208                     !S_ISCHR(st.st_mode)) {
 209                         r = -EINVAL;
 210                         goto fail;
 211                 }
 212
 213                 if (st.st_rdev == 0)
 214                         continue;
 215
 216                 dn = strappend(temporary_mount, d);
 217                 if (!dn) {
 218                         r = -ENOMEM;
 219                         goto fail;
 220                 }
 221
 222                 mac_selinux_create_file_prepare(d, st.st_mode);
 223                 r = mknod(dn, st.st_mode, st.st_rdev);
 224                 mac_selinux_create_file_clear();
 225
 226                 if (r < 0) {
 227                         r = -errno;
 228                         goto fail;
 229                 }
 230         }
 231
 232         dev_setup(temporary_mount, UID_INVALID, GID_INVALID);
 233
 234         /* Create the /dev directory if missing. It is more likely to be
 235          * missing when the service is started with RootDirectory. This is
 236          * consistent with mount units creating the mount points when missing.
 237          */
 238         (void) mkdir_p_label(m->path, 0755);
 239
 240         /* Unmount everything in old /dev */
 241         umount_recursive(m->path, 0);
 242         if (mount(dev, m->path, NULL, MS_MOVE, NULL) < 0) {
 243                 r = -errno;
 244                 goto fail;
 245         }
 246
 247         rmdir(dev);
 248         rmdir(temporary_mount);
 249
 250         return 0;
 251
 252 fail:
 253         if (devpts)
 254                 umount(devpts);
 255
 256         if (devshm)
 257                 umount(devshm);
 258
 259         if (devhugepages)
 260                 umount(devhugepages);
 261
 262         if (devmqueue)
 263                 umount(devmqueue);
 264
 265         umount(dev);
 266         rmdir(dev);
 267         rmdir(temporary_mount);
 268
 269         return r;
 270 }
 271
 272 static int apply_mount(
 273                 BindMount *m,
 274                 const char *tmp_dir,
 275                 const char *var_tmp_dir) {
 276
 277         const char *what;
 278         int r;
 279
 280         assert(m);
 281
 282         switch (m->mode) {
 283
 284         case INACCESSIBLE:
 285
 286                 /* First, get rid of everything that is below if there
 287                  * is anything... Then, overmount it with an
 288                  * inaccessible directory. */
 289                 umount_recursive(m->path, 0);
 290
 291                 what = "/run/systemd/inaccessible";
 292                 break;
 293
 294         case READONLY:
 295         case READWRITE:
 296                 /* Nothing to mount here, we just later toggle the
 297                  * MS_RDONLY bit for the mount point */
 298                 return 0;
 299
 300         case PRIVATE_TMP:
 301                 what = tmp_dir;
 302                 break;
 303
 304         case PRIVATE_VAR_TMP:
 305                 what = var_tmp_dir;
 306                 break;
 307
 308         case PRIVATE_DEV:
 309                 return mount_dev(m);
 310
 311         default:
 312                 assert_not_reached("Unknown mode");
 313         }
 314
 315         assert(what);
 316
 317         r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
 318         if (r >= 0)
 319                 log_debug("Successfully mounted %s to %s", what, m->path);
 320         else if (m->ignore && errno == ENOENT)
 321                 return 0;
 322
 323         return r;
 324 }
 325
 326 static int make_read_only(BindMount *m) {
 327         int r;
 328
 329         assert(m);
 330
 331         if (IN_SET(m->mode, INACCESSIBLE, READONLY))
 332                 r = bind_remount_recursive(m->path, true);
 333         else if (IN_SET(m->mode, READWRITE, PRIVATE_TMP, PRIVATE_VAR_TMP, PRIVATE_DEV))
 334                 r = bind_remount_recursive(m->path, false);
 335         else
 336                 r = 0;
 337
 338         if (m->ignore && r == -ENOENT)
 339                 return 0;
 340
 341         return r;
 342 }
 343
 344 int setup_namespace(
 345                 const char* root_directory,
 346                 char** read_write_dirs,
 347                 char** read_only_dirs,
 348                 char** inaccessible_dirs,
 349                 const char* tmp_dir,
 350                 const char* var_tmp_dir,
 351                 bool private_dev,
 352                 ProtectHome protect_home,
 353                 ProtectSystem protect_system,
 354                 unsigned long mount_flags) {
 355
 356         BindMount *m, *mounts = NULL;
 357         unsigned n;
 358         int r = 0;
 359
 360         if (mount_flags == 0)
 361                 mount_flags = MS_SHARED;
 362
 363         if (unshare(CLONE_NEWNS) < 0)
 364                 return -errno;
 365
 366         n = !!tmp_dir + !!var_tmp_dir +
 367                 strv_length(read_write_dirs) +
 368                 strv_length(read_only_dirs) +
 369                 strv_length(inaccessible_dirs) +
 370                 private_dev +
 371                 (protect_home != PROTECT_HOME_NO ? 3 : 0) +
 372                 (protect_system != PROTECT_SYSTEM_NO ? 2 : 0) +
 373                 (protect_system == PROTECT_SYSTEM_FULL ? 1 : 0);
 374
 375         if (n > 0) {
 376                 m = mounts = (BindMount *) alloca0(n * sizeof(BindMount));
 377                 r = append_mounts(&m, read_write_dirs, READWRITE);
 378                 if (r < 0)
 379                         return r;
 380
 381                 r = append_mounts(&m, read_only_dirs, READONLY);
 382                 if (r < 0)
 383                         return r;
 384
 385                 r = append_mounts(&m, inaccessible_dirs, INACCESSIBLE);
 386                 if (r < 0)
 387                         return r;
 388
 389                 if (tmp_dir) {
 390                         m->path = prefix_roota(root_directory, "/tmp");
 391                         m->mode = PRIVATE_TMP;
 392                         m++;
 393                 }
 394
 395                 if (var_tmp_dir) {
 396                         m->path = prefix_roota(root_directory, "/var/tmp");
 397                         m->mode = PRIVATE_VAR_TMP;
 398                         m++;
 399                 }
 400
 401                 if (private_dev) {
 402                         m->path = prefix_roota(root_directory, "/dev");
 403                         m->mode = PRIVATE_DEV;
 404                         m++;
 405                 }
 406
 407                 if (protect_home != PROTECT_HOME_NO) {
 408                         const char *home_dir, *run_user_dir, *root_dir;
 409
 410                         home_dir = prefix_roota(root_directory, "/home");
 411                         home_dir = strjoina("-", home_dir);
 412                         run_user_dir = prefix_roota(root_directory, "/run/user");
 413                         run_user_dir = strjoina("-", run_user_dir);
 414                         root_dir = prefix_roota(root_directory, "/root");
 415                         root_dir = strjoina("-", root_dir);
 416
 417                         r = append_mounts(&m, STRV_MAKE(home_dir, run_user_dir, root_dir),
 418                                 protect_home == PROTECT_HOME_READ_ONLY ? READONLY : INACCESSIBLE);
 419                         if (r < 0)
 420                                 return r;
 421                 }
 422
 423                 if (protect_system != PROTECT_SYSTEM_NO) {
 424                         const char *usr_dir, *boot_dir, *etc_dir;
 425
 426                         usr_dir = prefix_roota(root_directory, "/usr");
 427                         boot_dir = prefix_roota(root_directory, "/boot");
 428                         boot_dir = strjoina("-", boot_dir);
 429                         etc_dir = prefix_roota(root_directory, "/etc");
 430
 431                         r = append_mounts(&m, protect_system == PROTECT_SYSTEM_FULL
 432                                 ? STRV_MAKE(usr_dir, boot_dir, etc_dir)
 433                                 : STRV_MAKE(usr_dir, boot_dir), READONLY);
 434                         if (r < 0)
 435                                 return r;
 436                 }
 437
 438                 assert(mounts + n == m);
 439
 440                 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
 441                 drop_duplicates(mounts, &n);
 442         }
 443
 444         if (n > 0 || root_directory) {
 445                 /* Remount / as SLAVE so that nothing now mounted in the namespace
 446                    shows up in the parent */
 447                 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
 448                         return -errno;
 449         }
 450
 451         if (root_directory) {
 452                 /* Turn directory into bind mount */
 453                 if (mount(root_directory, root_directory, NULL, MS_BIND|MS_REC, NULL) < 0)
 454                         return -errno;
 455         }
 456
 457         if (n > 0) {
 458                 for (m = mounts; m < mounts + n; ++m) {
 459                         r = apply_mount(m, tmp_dir, var_tmp_dir);
 460                         if (r < 0)
 461                                 goto fail;
 462                 }
 463
 464                 for (m = mounts; m < mounts + n; ++m) {
 465                         r = make_read_only(m);
 466                         if (r < 0)
 467                                 goto fail;
 468                 }
 469         }
 470
 471         if (root_directory) {
 472                 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
 473                 r = mount_move_root(root_directory);
 474
 475                 /* at this point, we cannot rollback */
 476                 if (r < 0)
 477                         return r;
 478         }
 479
 480         /* Remount / as the desired mode. Not that this will not
 481          * reestablish propagation from our side to the host, since
 482          * what's disconnected is disconnected. */
 483         if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0)
 484                 /* at this point, we cannot rollback */
 485                 return -errno;
 486
 487         return 0;
 488
 489 fail:
 490         if (n > 0) {
 491                 for (m = mounts; m < mounts + n; ++m)
 492                         if (m->done)
 493                                 (void) umount2(m->path, MNT_DETACH);
 494         }
 495
 496         return r;
 497 }
 498
 499 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
 500         _cleanup_free_ char *x = NULL;
 501         char bid[SD_ID128_STRING_MAX];
 502         sd_id128_t boot_id;
 503         int r;
 504
 505         assert(id);
 506         assert(prefix);
 507         assert(path);
 508
 509         /* We include the boot id in the directory so that after a
 510          * reboot we can easily identify obsolete directories. */
 511
 512         r = sd_id128_get_boot(&boot_id);
 513         if (r < 0)
 514                 return r;
 515
 516         x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX", NULL);
 517         if (!x)
 518                 return -ENOMEM;
 519
 520         RUN_WITH_UMASK(0077)
 521                 if (!mkdtemp(x))
 522                         return -errno;
 523
 524         RUN_WITH_UMASK(0000) {
 525                 char *y;
 526
 527                 y = strjoina(x, "/tmp");
 528
 529                 if (mkdir(y, 0777 | S_ISVTX) < 0)
 530                         return -errno;
 531         }
 532
 533         *path = x;
 534         x = NULL;
 535
 536         return 0;
 537 }
 538
 539 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
 540         char *a, *b;
 541         int r;
 542
 543         assert(id);
 544         assert(tmp_dir);
 545         assert(var_tmp_dir);
 546
 547         r = setup_one_tmp_dir(id, "/tmp", &a);
 548         if (r < 0)
 549                 return r;
 550
 551         r = setup_one_tmp_dir(id, "/var/tmp", &b);
 552         if (r < 0) {
 553                 char *t;
 554
 555                 t = strjoina(a, "/tmp");
 556                 rmdir(t);
 557                 rmdir(a);
 558
 559                 free(a);
 560                 return r;
 561         }
 562
 563         *tmp_dir = a;
 564         *var_tmp_dir = b;
 565
 566         return 0;
 567 }
 568
 569 int setup_netns(int netns_storage_socket[2]) {
 570         _cleanup_close_ int netns = -1;
 571         int r, q;
 572
 573         assert(netns_storage_socket);
 574         assert(netns_storage_socket[0] >= 0);
 575         assert(netns_storage_socket[1] >= 0);
 576
 577         /* We use the passed socketpair as a storage buffer for our
 578          * namespace reference fd. Whatever process runs this first
 579          * shall create a new namespace, all others should just join
 580          * it. To serialize that we use a file lock on the socket
 581          * pair.
 582          *
 583          * It's a bit crazy, but hey, works great! */
 584
 585         if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
 586                 return -errno;
 587
 588         netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
 589         if (netns == -EAGAIN) {
 590                 /* Nothing stored yet, so let's create a new namespace */
 591
 592                 if (unshare(CLONE_NEWNET) < 0) {
 593                         r = -errno;
 594                         goto fail;
 595                 }
 596
 597                 loopback_setup();
 598
 599                 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
 600                 if (netns < 0) {
 601                         r = -errno;
 602                         goto fail;
 603                 }
 604
 605                 r = 1;
 606
 607         } else if (netns < 0) {
 608                 r = netns;
 609                 goto fail;
 610
 611         } else {
 612                 /* Yay, found something, so let's join the namespace */
 613                 if (setns(netns, CLONE_NEWNET) < 0) {
 614                         r = -errno;
 615                         goto fail;
 616                 }
 617
 618                 r = 0;
 619         }
 620
 621         q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
 622         if (q < 0) {
 623                 r = q;
 624                 goto fail;
 625         }
 626
 627 fail:
 628         lockf(netns_storage_socket[0], F_ULOCK, 0);
 629         return r;
 630 }
 631
 632 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
 633         [PROTECT_HOME_NO] = "no",
 634         [PROTECT_HOME_YES] = "yes",
 635         [PROTECT_HOME_READ_ONLY] = "read-only",
 636 };
 637
 638 DEFINE_STRING_TABLE_LOOKUP(protect_home, ProtectHome);
 639
 640 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
 641         [PROTECT_SYSTEM_NO] = "no",
 642         [PROTECT_SYSTEM_YES] = "yes",
 643         [PROTECT_SYSTEM_FULL] = "full",
 644 };
 645
 646 DEFINE_STRING_TABLE_LOOKUP(protect_system, ProtectSystem);