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 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
  48
  49 typedef enum MountMode {
  50         /* This is ordered by priority! */
  51         INACCESSIBLE,
  52         READONLY,
  53         PRIVATE_TMP,
  54         PRIVATE_VAR_TMP,
  55         PRIVATE_DEV,
  56         READWRITE
  57 } MountMode;
  58
  59 typedef struct BindMount {
  60         const char *path;
  61         MountMode mode;
  62         bool done;
  63         bool ignore;
  64 } BindMount;
  65
  66 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
  67         char **i;
  68
  69         assert(p);
  70
  71         STRV_FOREACH(i, strv) {
  72
  73                 (*p)->ignore = false;
  74                 (*p)->done = false;
  75
  76                 if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
  77                         (*p)->ignore = true;
  78                         (*i)++;
  79                 }
  80
  81                 if (!path_is_absolute(*i))
  82                         return -EINVAL;
  83
  84                 (*p)->path = *i;
  85                 (*p)->mode = mode;
  86                 (*p)++;
  87         }
  88
  89         return 0;
  90 }
  91
  92 static int mount_path_compare(const void *a, const void *b) {
  93         const BindMount *p = a, *q = b;
  94         int d;
  95
  96         d = path_compare(p->path, q->path);
  97
  98         if (d == 0) {
  99                 /* If the paths are equal, check the mode */
 100                 if (p->mode < q->mode)
 101                         return -1;
 102
 103                 if (p->mode > q->mode)
 104                         return 1;
 105
 106                 return 0;
 107         }
 108
 109         /* If the paths are not equal, then order prefixes first */
 110         return d;
 111 }
 112
 113 static void drop_duplicates(BindMount *m, unsigned *n) {
 114         BindMount *f, *t, *previous;
 115
 116         assert(m);
 117         assert(n);
 118
 119         for (f = m, t = m, previous = NULL; f < m+*n; f++) {
 120
 121                 /* The first one wins */
 122                 if (previous && path_equal(f->path, previous->path))
 123                         continue;
 124
 125                 *t = *f;
 126
 127                 previous = t;
 128
 129                 t++;
 130         }
 131
 132         *n = t - m;
 133 }
 134
 135 static int mount_dev(BindMount *m) {
 136         static const char devnodes[] =
 137                 "/dev/null\0"
 138                 "/dev/zero\0"
 139                 "/dev/full\0"
 140                 "/dev/random\0"
 141                 "/dev/urandom\0"
 142                 "/dev/tty\0";
 143
 144         char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
 145         const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
 146         _cleanup_umask_ mode_t u;
 147         int r;
 148
 149         assert(m);
 150
 151         u = umask(0000);
 152
 153         if (!mkdtemp(temporary_mount))
 154                 return -errno;
 155
 156         dev = strjoina(temporary_mount, "/dev");
 157         (void) mkdir(dev, 0755);
 158         if (mount("tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=755") < 0) {
 159                 r = -errno;
 160                 goto fail;
 161         }
 162
 163         devpts = strjoina(temporary_mount, "/dev/pts");
 164         (void) mkdir(devpts, 0755);
 165         if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
 166                 r = -errno;
 167                 goto fail;
 168         }
 169
 170         devptmx = strjoina(temporary_mount, "/dev/ptmx");
 171         if (symlink("pts/ptmx", devptmx) < 0) {
 172                 r = -errno;
 173                 goto fail;
 174         }
 175
 176         devshm = strjoina(temporary_mount, "/dev/shm");
 177         (void) mkdir(devshm, 01777);
 178         r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
 179         if (r < 0) {
 180                 r = -errno;
 181                 goto fail;
 182         }
 183
 184         devmqueue = strjoina(temporary_mount, "/dev/mqueue");
 185         (void) mkdir(devmqueue, 0755);
 186         (void) mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
 187
 188         devhugepages = strjoina(temporary_mount, "/dev/hugepages");
 189         (void) mkdir(devhugepages, 0755);
 190         (void) mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
 191
 192         devlog = strjoina(temporary_mount, "/dev/log");
 193         (void) symlink("/run/systemd/journal/dev-log", devlog);
 194
 195         NULSTR_FOREACH(d, devnodes) {
 196                 _cleanup_free_ char *dn = NULL;
 197                 struct stat st;
 198
 199                 r = stat(d, &st);
 200                 if (r < 0) {
 201
 202                         if (errno == ENOENT)
 203                                 continue;
 204
 205                         r = -errno;
 206                         goto fail;
 207                 }
 208
 209                 if (!S_ISBLK(st.st_mode) &&
 210                     !S_ISCHR(st.st_mode)) {
 211                         r = -EINVAL;
 212                         goto fail;
 213                 }
 214
 215                 if (st.st_rdev == 0)
 216                         continue;
 217
 218                 dn = strappend(temporary_mount, d);
 219                 if (!dn) {
 220                         r = -ENOMEM;
 221                         goto fail;
 222                 }
 223
 224                 mac_selinux_create_file_prepare(d, st.st_mode);
 225                 r = mknod(dn, st.st_mode, st.st_rdev);
 226                 mac_selinux_create_file_clear();
 227
 228                 if (r < 0) {
 229                         r = -errno;
 230                         goto fail;
 231                 }
 232         }
 233
 234         dev_setup(temporary_mount, UID_INVALID, GID_INVALID);
 235
 236         /* Create the /dev directory if missing. It is more likely to be
 237          * missing when the service is started with RootDirectory. This is
 238          * consistent with mount units creating the mount points when missing.
 239          */
 240         (void) mkdir_p_label(m->path, 0755);
 241
 242         /* Unmount everything in old /dev */
 243         umount_recursive(m->path, 0);
 244         if (mount(dev, m->path, NULL, MS_MOVE, NULL) < 0) {
 245                 r = -errno;
 246                 goto fail;
 247         }
 248
 249         rmdir(dev);
 250         rmdir(temporary_mount);
 251
 252         return 0;
 253
 254 fail:
 255         if (devpts)
 256                 umount(devpts);
 257
 258         if (devshm)
 259                 umount(devshm);
 260
 261         if (devhugepages)
 262                 umount(devhugepages);
 263
 264         if (devmqueue)
 265                 umount(devmqueue);
 266
 267         umount(dev);
 268         rmdir(dev);
 269         rmdir(temporary_mount);
 270
 271         return r;
 272 }
 273
 274 static int apply_mount(
 275                 BindMount *m,
 276                 const char *tmp_dir,
 277                 const char *var_tmp_dir) {
 278
 279         const char *what;
 280         int r;
 281         struct stat target;
 282
 283         assert(m);
 284
 285         switch (m->mode) {
 286
 287         case INACCESSIBLE:
 288
 289                 /* First, get rid of everything that is below if there
 290                  * is anything... Then, overmount it with an
 291                  * inaccessible path. */
 292                 umount_recursive(m->path, 0);
 293
 294                 r = lstat(m->path, &target);
 295                 if (r != 0) {
 296                         if (m->ignore && errno == ENOENT)
 297                                 return 0;
 298                         return -errno;
 299                 }
 300
 301                 what = mode_to_inaccessible_node(target.st_mode);
 302                 if (what == NULL) {
 303                         log_debug("File type not supported. Note that symlinks are not allowed");
 304                         return -ELOOP;
 305                 }
 306                 break;
 307         case READONLY:
 308         case READWRITE:
 309                 /* Nothing to mount here, we just later toggle the
 310                  * MS_RDONLY bit for the mount point */
 311                 return 0;
 312
 313         case PRIVATE_TMP:
 314                 what = tmp_dir;
 315                 break;
 316
 317         case PRIVATE_VAR_TMP:
 318                 what = var_tmp_dir;
 319                 break;
 320
 321         case PRIVATE_DEV:
 322                 return mount_dev(m);
 323
 324         default:
 325                 assert_not_reached("Unknown mode");
 326         }
 327
 328         assert(what);
 329
 330         r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
 331         if (r >= 0) {
 332                 log_debug("Successfully mounted %s to %s", what, m->path);
 333                 return r;
 334         }
 335         else {
 336                 if (m->ignore && errno == ENOENT)
 337                         return 0;
 338                 log_debug("Failed mounting %s to %s: %s", what, m->path, strerror(errno));
 339                 return -errno;
 340         }
 341 }
 342
 343 static int make_read_only(BindMount *m) {
 344         int r;
 345
 346         assert(m);
 347
 348         if (IN_SET(m->mode, INACCESSIBLE, READONLY))
 349                 r = bind_remount_recursive(m->path, true);
 350         else if (IN_SET(m->mode, READWRITE, PRIVATE_TMP, PRIVATE_VAR_TMP, PRIVATE_DEV)) {
 351                 r = bind_remount_recursive(m->path, false);
 352                 if (r == 0 && m->mode == PRIVATE_DEV) /* can be readonly but the submounts can't*/
 353                         r = mount(NULL, m->path, NULL, MS_REMOUNT|DEV_MOUNT_OPTIONS|MS_RDONLY, NULL);
 354         } else
 355                 r = 0;
 356
 357         if (m->ignore && r == -ENOENT)
 358                 return 0;
 359
 360         return r;
 361 }
 362
 363 int setup_namespace(
 364                 const char* root_directory,
 365                 char** read_write_paths,
 366                 char** read_only_paths,
 367                 char** inaccessible_paths,
 368                 const char* tmp_dir,
 369                 const char* var_tmp_dir,
 370                 bool private_dev,
 371                 ProtectHome protect_home,
 372                 ProtectSystem protect_system,
 373                 unsigned long mount_flags) {
 374
 375         BindMount *m, *mounts = NULL;
 376         unsigned n;
 377         int r = 0;
 378
 379         if (mount_flags == 0)
 380                 mount_flags = MS_SHARED;
 381
 382         if (unshare(CLONE_NEWNS) < 0)
 383                 return -errno;
 384
 385         n = !!tmp_dir + !!var_tmp_dir +
 386                 strv_length(read_write_paths) +
 387                 strv_length(read_only_paths) +
 388                 strv_length(inaccessible_paths) +
 389                 private_dev +
 390                 (protect_home != PROTECT_HOME_NO ? 3 : 0) +
 391                 (protect_system != PROTECT_SYSTEM_NO ? 2 : 0) +
 392                 (protect_system == PROTECT_SYSTEM_FULL ? 1 : 0);
 393
 394         if (n > 0) {
 395                 m = mounts = (BindMount *) alloca0(n * sizeof(BindMount));
 396                 r = append_mounts(&m, read_write_paths, READWRITE);
 397                 if (r < 0)
 398                         return r;
 399
 400                 r = append_mounts(&m, read_only_paths, READONLY);
 401                 if (r < 0)
 402                         return r;
 403
 404                 r = append_mounts(&m, inaccessible_paths, INACCESSIBLE);
 405                 if (r < 0)
 406                         return r;
 407
 408                 if (tmp_dir) {
 409                         m->path = prefix_roota(root_directory, "/tmp");
 410                         m->mode = PRIVATE_TMP;
 411                         m++;
 412                 }
 413
 414                 if (var_tmp_dir) {
 415                         m->path = prefix_roota(root_directory, "/var/tmp");
 416                         m->mode = PRIVATE_VAR_TMP;
 417                         m++;
 418                 }
 419
 420                 if (private_dev) {
 421                         m->path = prefix_roota(root_directory, "/dev");
 422                         m->mode = PRIVATE_DEV;
 423                         m++;
 424                 }
 425
 426                 if (protect_home != PROTECT_HOME_NO) {
 427                         const char *home_dir, *run_user_dir, *root_dir;
 428
 429                         home_dir = prefix_roota(root_directory, "/home");
 430                         home_dir = strjoina("-", home_dir);
 431                         run_user_dir = prefix_roota(root_directory, "/run/user");
 432                         run_user_dir = strjoina("-", run_user_dir);
 433                         root_dir = prefix_roota(root_directory, "/root");
 434                         root_dir = strjoina("-", root_dir);
 435
 436                         r = append_mounts(&m, STRV_MAKE(home_dir, run_user_dir, root_dir),
 437                                 protect_home == PROTECT_HOME_READ_ONLY ? READONLY : INACCESSIBLE);
 438                         if (r < 0)
 439                                 return r;
 440                 }
 441
 442                 if (protect_system != PROTECT_SYSTEM_NO) {
 443                         const char *usr_dir, *boot_dir, *etc_dir;
 444
 445                         usr_dir = prefix_roota(root_directory, "/usr");
 446                         boot_dir = prefix_roota(root_directory, "/boot");
 447                         boot_dir = strjoina("-", boot_dir);
 448                         etc_dir = prefix_roota(root_directory, "/etc");
 449
 450                         r = append_mounts(&m, protect_system == PROTECT_SYSTEM_FULL
 451                                 ? STRV_MAKE(usr_dir, boot_dir, etc_dir)
 452                                 : STRV_MAKE(usr_dir, boot_dir), READONLY);
 453                         if (r < 0)
 454                                 return r;
 455                 }
 456
 457                 assert(mounts + n == m);
 458
 459                 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
 460                 drop_duplicates(mounts, &n);
 461         }
 462
 463         if (n > 0 || root_directory) {
 464                 /* Remount / as SLAVE so that nothing now mounted in the namespace
 465                    shows up in the parent */
 466                 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
 467                         return -errno;
 468         }
 469
 470         if (root_directory) {
 471                 /* Turn directory into bind mount */
 472                 if (mount(root_directory, root_directory, NULL, MS_BIND|MS_REC, NULL) < 0)
 473                         return -errno;
 474         }
 475
 476         if (n > 0) {
 477                 for (m = mounts; m < mounts + n; ++m) {
 478                         r = apply_mount(m, tmp_dir, var_tmp_dir);
 479                         if (r < 0)
 480                                 goto fail;
 481                 }
 482
 483                 for (m = mounts; m < mounts + n; ++m) {
 484                         r = make_read_only(m);
 485                         if (r < 0)
 486                                 goto fail;
 487                 }
 488         }
 489
 490         if (root_directory) {
 491                 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
 492                 r = mount_move_root(root_directory);
 493
 494                 /* at this point, we cannot rollback */
 495                 if (r < 0)
 496                         return r;
 497         }
 498
 499         /* Remount / as the desired mode. Not that this will not
 500          * reestablish propagation from our side to the host, since
 501          * what's disconnected is disconnected. */
 502         if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0)
 503                 /* at this point, we cannot rollback */
 504                 return -errno;
 505
 506         return 0;
 507
 508 fail:
 509         if (n > 0) {
 510                 for (m = mounts; m < mounts + n; ++m)
 511                         if (m->done)
 512                                 (void) umount2(m->path, MNT_DETACH);
 513         }
 514
 515         return r;
 516 }
 517
 518 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
 519         _cleanup_free_ char *x = NULL;
 520         char bid[SD_ID128_STRING_MAX];
 521         sd_id128_t boot_id;
 522         int r;
 523
 524         assert(id);
 525         assert(prefix);
 526         assert(path);
 527
 528         /* We include the boot id in the directory so that after a
 529          * reboot we can easily identify obsolete directories. */
 530
 531         r = sd_id128_get_boot(&boot_id);
 532         if (r < 0)
 533                 return r;
 534
 535         x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX", NULL);
 536         if (!x)
 537                 return -ENOMEM;
 538
 539         RUN_WITH_UMASK(0077)
 540                 if (!mkdtemp(x))
 541                         return -errno;
 542
 543         RUN_WITH_UMASK(0000) {
 544                 char *y;
 545
 546                 y = strjoina(x, "/tmp");
 547
 548                 if (mkdir(y, 0777 | S_ISVTX) < 0)
 549                         return -errno;
 550         }
 551
 552         *path = x;
 553         x = NULL;
 554
 555         return 0;
 556 }
 557
 558 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
 559         char *a, *b;
 560         int r;
 561
 562         assert(id);
 563         assert(tmp_dir);
 564         assert(var_tmp_dir);
 565
 566         r = setup_one_tmp_dir(id, "/tmp", &a);
 567         if (r < 0)
 568                 return r;
 569
 570         r = setup_one_tmp_dir(id, "/var/tmp", &b);
 571         if (r < 0) {
 572                 char *t;
 573
 574                 t = strjoina(a, "/tmp");
 575                 rmdir(t);
 576                 rmdir(a);
 577
 578                 free(a);
 579                 return r;
 580         }
 581
 582         *tmp_dir = a;
 583         *var_tmp_dir = b;
 584
 585         return 0;
 586 }
 587
 588 int setup_netns(int netns_storage_socket[2]) {
 589         _cleanup_close_ int netns = -1;
 590         int r, q;
 591
 592         assert(netns_storage_socket);
 593         assert(netns_storage_socket[0] >= 0);
 594         assert(netns_storage_socket[1] >= 0);
 595
 596         /* We use the passed socketpair as a storage buffer for our
 597          * namespace reference fd. Whatever process runs this first
 598          * shall create a new namespace, all others should just join
 599          * it. To serialize that we use a file lock on the socket
 600          * pair.
 601          *
 602          * It's a bit crazy, but hey, works great! */
 603
 604         if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
 605                 return -errno;
 606
 607         netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
 608         if (netns == -EAGAIN) {
 609                 /* Nothing stored yet, so let's create a new namespace */
 610
 611                 if (unshare(CLONE_NEWNET) < 0) {
 612                         r = -errno;
 613                         goto fail;
 614                 }
 615
 616                 loopback_setup();
 617
 618                 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
 619                 if (netns < 0) {
 620                         r = -errno;
 621                         goto fail;
 622                 }
 623
 624                 r = 1;
 625
 626         } else if (netns < 0) {
 627                 r = netns;
 628                 goto fail;
 629
 630         } else {
 631                 /* Yay, found something, so let's join the namespace */
 632                 if (setns(netns, CLONE_NEWNET) < 0) {
 633                         r = -errno;
 634                         goto fail;
 635                 }
 636
 637                 r = 0;
 638         }
 639
 640         q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
 641         if (q < 0) {
 642                 r = q;
 643                 goto fail;
 644         }
 645
 646 fail:
 647         lockf(netns_storage_socket[0], F_ULOCK, 0);
 648         return r;
 649 }
 650
 651 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
 652         [PROTECT_HOME_NO] = "no",
 653         [PROTECT_HOME_YES] = "yes",
 654         [PROTECT_HOME_READ_ONLY] = "read-only",
 655 };
 656
 657 DEFINE_STRING_TABLE_LOOKUP(protect_home, ProtectHome);
 658
 659 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
 660         [PROTECT_SYSTEM_NO] = "no",
 661         [PROTECT_SYSTEM_YES] = "yes",
 662         [PROTECT_SYSTEM_FULL] = "full",
 663 };
 664
 665 DEFINE_STRING_TABLE_LOOKUP(protect_system, ProtectSystem);