src/core/namespace.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 <errno.h>
  23 #include <sys/mount.h>
  24 #include <string.h>
  25 #include <stdio.h>
  26 #include <unistd.h>
  27 #include <sys/stat.h>
  28 #include <sys/types.h>
  29 #include <sched.h>
  30 #include <sys/syscall.h>
  31 #include <limits.h>
  32 #include <linux/fs.h>
  33 #include <sys/file.h>
  34
  35 #include "strv.h"
  36 #include "util.h"
  37 #include "path-util.h"
  38 #include "namespace.h"
  39 #include "missing.h"
  40 #include "execute.h"
  41 #include "loopback-setup.h"
  42 #include "mkdir.h"
  43 #include "dev-setup.h"
  44 #include "def.h"
  45 #include "label.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         PRIVATE_BUS_ENDPOINT,
  55         READWRITE
  56 } MountMode;
  57
  58 typedef struct BindMount {
  59         const char *path;
  60         MountMode mode;
  61         bool done;
  62         bool ignore;
  63 } BindMount;
  64
  65 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
  66         char **i;
  67
  68         assert(p);
  69
  70         STRV_FOREACH(i, strv) {
  71
  72                 (*p)->ignore = false;
  73                 (*p)->done = false;
  74
  75                 if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
  76                         (*p)->ignore = true;
  77                         (*i)++;
  78                 }
  79
  80                 if (!path_is_absolute(*i))
  81                         return -EINVAL;
  82
  83                 (*p)->path = *i;
  84                 (*p)->mode = mode;
  85                 (*p)++;
  86         }
  87
  88         return 0;
  89 }
  90
  91 static int mount_path_compare(const void *a, const void *b) {
  92         const BindMount *p = a, *q = b;
  93
  94         if (path_equal(p->path, q->path)) {
  95
  96                 /* If the paths are equal, check the mode */
  97                 if (p->mode < q->mode)
  98                         return -1;
  99
 100                 if (p->mode > q->mode)
 101                         return 1;
 102
 103                 return 0;
 104         }
 105
 106         /* If the paths are not equal, then order prefixes first */
 107         if (path_startswith(p->path, q->path))
 108                 return 1;
 109
 110         if (path_startswith(q->path, p->path))
 111                 return -1;
 112
 113         return 0;
 114 }
 115
 116 static void drop_duplicates(BindMount *m, unsigned *n) {
 117         BindMount *f, *t, *previous;
 118
 119         assert(m);
 120         assert(n);
 121
 122         for (f = m, t = m, previous = NULL; f < m+*n; f++) {
 123
 124                 /* The first one wins */
 125                 if (previous && path_equal(f->path, previous->path))
 126                         continue;
 127
 128                 *t = *f;
 129
 130                 previous = t;
 131
 132                 t++;
 133         }
 134
 135         *n = t - m;
 136 }
 137
 138 static int mount_dev(BindMount *m) {
 139         static const char devnodes[] =
 140                 "/dev/null\0"
 141                 "/dev/zero\0"
 142                 "/dev/full\0"
 143                 "/dev/random\0"
 144                 "/dev/urandom\0"
 145                 "/dev/tty\0";
 146
 147         char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
 148         const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devkdbus = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
 149         _cleanup_umask_ mode_t u;
 150         int r;
 151
 152         assert(m);
 153
 154         u = umask(0000);
 155
 156         if (!mkdtemp(temporary_mount))
 157                 return -errno;
 158
 159         dev = strappenda(temporary_mount, "/dev");
 160         mkdir(dev, 0755);
 161         if (mount("tmpfs", dev, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=755") < 0) {
 162                 r = -errno;
 163                 goto fail;
 164         }
 165
 166         devpts = strappenda(temporary_mount, "/dev/pts");
 167         mkdir(devpts, 0755);
 168         if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
 169                 r = -errno;
 170                 goto fail;
 171         }
 172
 173         devptmx = strappenda(temporary_mount, "/dev/ptmx");
 174         symlink("pts/ptmx", devptmx);
 175
 176         devshm = strappenda(temporary_mount, "/dev/shm");
 177         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 = strappenda(temporary_mount, "/dev/mqueue");
 185         mkdir(devmqueue, 0755);
 186         mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
 187
 188         devkdbus = strappenda(temporary_mount, "/dev/kdbus");
 189         mkdir(devkdbus, 0755);
 190         mount("/dev/kdbus", devkdbus, NULL, MS_BIND, NULL);
 191
 192         devhugepages = strappenda(temporary_mount, "/dev/hugepages");
 193         mkdir(devhugepages, 0755);
 194         mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
 195
 196         devlog = strappenda(temporary_mount, "/dev/log");
 197         symlink("/run/systemd/journal/dev-log", devlog);
 198
 199         NULSTR_FOREACH(d, devnodes) {
 200                 _cleanup_free_ char *dn = NULL;
 201                 struct stat st;
 202
 203                 r = stat(d, &st);
 204                 if (r < 0) {
 205
 206                         if (errno == ENOENT)
 207                                 continue;
 208
 209                         r = -errno;
 210                         goto fail;
 211                 }
 212
 213                 if (!S_ISBLK(st.st_mode) &&
 214                     !S_ISCHR(st.st_mode)) {
 215                         r = -EINVAL;
 216                         goto fail;
 217                 }
 218
 219                 if (st.st_rdev == 0)
 220                         continue;
 221
 222                 dn = strappend(temporary_mount, d);
 223                 if (!dn) {
 224                         r = -ENOMEM;
 225                         goto fail;
 226                 }
 227
 228                 label_context_set(d, st.st_mode);
 229                 r = mknod(dn, st.st_mode, st.st_rdev);
 230                 label_context_clear();
 231
 232                 if (r < 0) {
 233                         r = -errno;
 234                         goto fail;
 235                 }
 236         }
 237
 238         dev_setup(temporary_mount);
 239
 240         if (mount(dev, "/dev/", NULL, MS_MOVE, NULL) < 0) {
 241                 r = -errno;
 242                 goto fail;
 243         }
 244
 245         rmdir(dev);
 246         rmdir(temporary_mount);
 247
 248         return 0;
 249
 250 fail:
 251         if (devpts)
 252                 umount(devpts);
 253
 254         if (devshm)
 255                 umount(devshm);
 256
 257         if (devkdbus)
 258                 umount(devkdbus);
 259
 260         if (devhugepages)
 261                 umount(devhugepages);
 262
 263         if (devmqueue)
 264                 umount(devmqueue);
 265
 266         if (dev) {
 267                 umount(dev);
 268                 rmdir(dev);
 269         }
 270
 271         rmdir(temporary_mount);
 272
 273         return r;
 274 }
 275
 276 static int mount_kdbus(BindMount *m) {
 277
 278         char temporary_mount[] = "/tmp/kdbus-dev-XXXXXX";
 279         _cleanup_free_ char *basepath = NULL;
 280         _cleanup_umask_ mode_t u;
 281         char *busnode, *root;
 282         struct stat st;
 283         int r;
 284
 285         assert(m);
 286
 287         u = umask(0000);
 288
 289         if (!mkdtemp(temporary_mount)) {
 290                 log_error("Failed create temp dir: %m");
 291                 return -errno;
 292         }
 293
 294         root = strappenda(temporary_mount, "/kdbus");
 295         mkdir(root, 0755);
 296         if (mount("tmpfs", root, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=777") < 0) {
 297                 r = -errno;
 298                 goto fail;
 299         }
 300
 301         /* create a new /dev/null dev node copy so we have some fodder to
 302          * bind-mount the custom endpoint over. */
 303         if (stat("/dev/null", &st) < 0) {
 304                 log_error("Failed to stat /dev/null: %m");
 305                 r = -errno;
 306                 goto fail;
 307         }
 308
 309         busnode = strappenda(root, "/bus");
 310         if (mknod(busnode, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) {
 311                 log_error("mknod() for %s failed: %m", busnode);
 312                 r = -errno;
 313                 goto fail;
 314         }
 315
 316         r = mount(m->path, busnode, "bind", MS_BIND, NULL);
 317         if (r < 0) {
 318                 log_error("bind mount of %s failed: %m", m->path);
 319                 r = -errno;
 320                 goto fail;
 321         }
 322
 323         basepath = dirname_malloc(m->path);
 324         if (!basepath) {
 325                 r = -ENOMEM;
 326                 goto fail;
 327         }
 328
 329         if (mount(root, basepath, NULL, MS_MOVE, NULL) < 0) {
 330                 log_error("bind mount of %s failed: %m", basepath);
 331                 r = -errno;
 332                 goto fail;
 333         }
 334
 335         rmdir(temporary_mount);
 336         return 0;
 337
 338 fail:
 339         if (busnode) {
 340                 umount(busnode);
 341                 unlink(busnode);
 342         }
 343
 344         if (root) {
 345                 umount(root);
 346                 rmdir(root);
 347         }
 348
 349         rmdir(temporary_mount);
 350
 351         return r;
 352 }
 353
 354 static int apply_mount(
 355                 BindMount *m,
 356                 const char *tmp_dir,
 357                 const char *var_tmp_dir) {
 358
 359         const char *what;
 360         int r;
 361
 362         assert(m);
 363
 364         switch (m->mode) {
 365
 366         case INACCESSIBLE:
 367
 368                 /* First, get rid of everything that is below if there
 369                  * is anything... Then, overmount it with an
 370                  * inaccessible directory. */
 371                 umount_recursive(m->path, 0);
 372
 373                 what = "/run/systemd/inaccessible";
 374                 break;
 375
 376         case READONLY:
 377         case READWRITE:
 378                 /* Nothing to mount here, we just later toggle the
 379                  * MS_RDONLY bit for the mount point */
 380                 return 0;
 381
 382         case PRIVATE_TMP:
 383                 what = tmp_dir;
 384                 break;
 385
 386         case PRIVATE_VAR_TMP:
 387                 what = var_tmp_dir;
 388                 break;
 389
 390         case PRIVATE_DEV:
 391                 return mount_dev(m);
 392
 393         case PRIVATE_BUS_ENDPOINT:
 394                 return mount_kdbus(m);
 395
 396         default:
 397                 assert_not_reached("Unknown mode");
 398         }
 399
 400         assert(what);
 401
 402         r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
 403         if (r >= 0)
 404                 log_debug("Successfully mounted %s to %s", what, m->path);
 405         else if (m->ignore && errno == ENOENT)
 406                 return 0;
 407
 408         return r;
 409 }
 410
 411 static int make_read_only(BindMount *m) {
 412         int r;
 413
 414         assert(m);
 415
 416         if (IN_SET(m->mode, INACCESSIBLE, READONLY))
 417                 r = bind_remount_recursive(m->path, true);
 418         else if (IN_SET(m->mode, READWRITE, PRIVATE_TMP, PRIVATE_VAR_TMP, PRIVATE_DEV))
 419                 r = bind_remount_recursive(m->path, false);
 420         else
 421                 r = 0;
 422
 423         if (m->ignore && r == -ENOENT)
 424                 return 0;
 425
 426         return r;
 427 }
 428
 429 int setup_namespace(
 430                 char** read_write_dirs,
 431                 char** read_only_dirs,
 432                 char** inaccessible_dirs,
 433                 char* tmp_dir,
 434                 char* var_tmp_dir,
 435                 char* bus_endpoint_path,
 436                 bool private_dev,
 437                 ProtectHome protect_home,
 438                 ProtectSystem protect_system,
 439                 unsigned mount_flags) {
 440
 441         BindMount *m, *mounts = NULL;
 442         unsigned n;
 443         int r = 0;
 444
 445         if (mount_flags == 0)
 446                 mount_flags = MS_SHARED;
 447
 448         if (unshare(CLONE_NEWNS) < 0)
 449                 return -errno;
 450
 451         n = !!tmp_dir + !!var_tmp_dir + !!bus_endpoint_path +
 452                 strv_length(read_write_dirs) +
 453                 strv_length(read_only_dirs) +
 454                 strv_length(inaccessible_dirs) +
 455                 private_dev +
 456                 (protect_home != PROTECT_HOME_NO ? 3 : 0) +
 457                 (protect_system != PROTECT_SYSTEM_NO ? 2 : 0) +
 458                 (protect_system == PROTECT_SYSTEM_FULL ? 1 : 0);
 459
 460         if (n > 0) {
 461                 m = mounts = (BindMount *) alloca0(n * sizeof(BindMount));
 462                 r = append_mounts(&m, read_write_dirs, READWRITE);
 463                 if (r < 0)
 464                         return r;
 465
 466                 r = append_mounts(&m, read_only_dirs, READONLY);
 467                 if (r < 0)
 468                         return r;
 469
 470                 r = append_mounts(&m, inaccessible_dirs, INACCESSIBLE);
 471                 if (r < 0)
 472                         return r;
 473
 474                 if (tmp_dir) {
 475                         m->path = "/tmp";
 476                         m->mode = PRIVATE_TMP;
 477                         m++;
 478                 }
 479
 480                 if (var_tmp_dir) {
 481                         m->path = "/var/tmp";
 482                         m->mode = PRIVATE_VAR_TMP;
 483                         m++;
 484                 }
 485
 486                 if (private_dev) {
 487                         m->path = "/dev";
 488                         m->mode = PRIVATE_DEV;
 489                         m++;
 490                 }
 491
 492                 if (bus_endpoint_path) {
 493                         m->path = bus_endpoint_path;
 494                         m->mode = PRIVATE_BUS_ENDPOINT;
 495                         m++;
 496                 }
 497
 498                 if (protect_home != PROTECT_HOME_NO) {
 499                         r = append_mounts(&m, STRV_MAKE("-/home", "-/run/user", "-/root"), protect_home == PROTECT_HOME_READ_ONLY ? READONLY : INACCESSIBLE);
 500                         if (r < 0)
 501                                 return r;
 502                 }
 503
 504                 if (protect_system != PROTECT_SYSTEM_NO) {
 505                         r = append_mounts(&m, protect_system == PROTECT_SYSTEM_FULL ? STRV_MAKE("/usr", "-/boot", "/etc") : STRV_MAKE("/usr", "-/boot"), READONLY);
 506                         if (r < 0)
 507                                 return r;
 508                 }
 509
 510                 assert(mounts + n == m);
 511
 512                 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
 513                 drop_duplicates(mounts, &n);
 514         }
 515
 516         if (n > 0) {
 517                 /* Remount / as SLAVE so that nothing now mounted in the namespace
 518                    shows up in the parent */
 519                 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
 520                         return -errno;
 521
 522                 for (m = mounts; m < mounts + n; ++m) {
 523                         r = apply_mount(m, tmp_dir, var_tmp_dir);
 524                         if (r < 0)
 525                                 goto fail;
 526                 }
 527
 528                 for (m = mounts; m < mounts + n; ++m) {
 529                         r = make_read_only(m);
 530                         if (r < 0)
 531                                 goto fail;
 532                 }
 533         }
 534
 535         /* Remount / as the desired mode. Not that this will not
 536          * reestablish propagation from our side to the host, since
 537          * what's disconnected is disconnected. */
 538         if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
 539                 r = -errno;
 540                 goto fail;
 541         }
 542
 543         return 0;
 544
 545 fail:
 546         if (n > 0) {
 547                 for (m = mounts; m < mounts + n; ++m)
 548                         if (m->done)
 549                                 umount2(m->path, MNT_DETACH);
 550         }
 551
 552         return r;
 553 }
 554
 555 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
 556         _cleanup_free_ char *x = NULL;
 557         char bid[SD_ID128_STRING_MAX];
 558         sd_id128_t boot_id;
 559         int r;
 560
 561         assert(id);
 562         assert(prefix);
 563         assert(path);
 564
 565         /* We include the boot id in the directory so that after a
 566          * reboot we can easily identify obsolete directories. */
 567
 568         r = sd_id128_get_boot(&boot_id);
 569         if (r < 0)
 570                 return r;
 571
 572         x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX", NULL);
 573         if (!x)
 574                 return -ENOMEM;
 575
 576         RUN_WITH_UMASK(0077)
 577                 if (!mkdtemp(x))
 578                         return -errno;
 579
 580         RUN_WITH_UMASK(0000) {
 581                 char *y;
 582
 583                 y = strappenda(x, "/tmp");
 584
 585                 if (mkdir(y, 0777 | S_ISVTX) < 0)
 586                         return -errno;
 587         }
 588
 589         *path = x;
 590         x = NULL;
 591
 592         return 0;
 593 }
 594
 595 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
 596         char *a, *b;
 597         int r;
 598
 599         assert(id);
 600         assert(tmp_dir);
 601         assert(var_tmp_dir);
 602
 603         r = setup_one_tmp_dir(id, "/tmp", &a);
 604         if (r < 0)
 605                 return r;
 606
 607         r = setup_one_tmp_dir(id, "/var/tmp", &b);
 608         if (r < 0) {
 609                 char *t;
 610
 611                 t = strappenda(a, "/tmp");
 612                 rmdir(t);
 613                 rmdir(a);
 614
 615                 free(a);
 616                 return r;
 617         }
 618
 619         *tmp_dir = a;
 620         *var_tmp_dir = b;
 621
 622         return 0;
 623 }
 624
 625 int setup_netns(int netns_storage_socket[2]) {
 626         _cleanup_close_ int netns = -1;
 627         union {
 628                 struct cmsghdr cmsghdr;
 629                 uint8_t buf[CMSG_SPACE(sizeof(int))];
 630         } control = {};
 631         struct msghdr mh = {
 632                 .msg_control = &control,
 633                 .msg_controllen = sizeof(control),
 634         };
 635         struct cmsghdr *cmsg;
 636         int r;
 637
 638         assert(netns_storage_socket);
 639         assert(netns_storage_socket[0] >= 0);
 640         assert(netns_storage_socket[1] >= 0);
 641
 642         /* We use the passed socketpair as a storage buffer for our
 643          * namespace reference fd. Whatever process runs this first
 644          * shall create a new namespace, all others should just join
 645          * it. To serialize that we use a file lock on the socket
 646          * pair.
 647          *
 648          * It's a bit crazy, but hey, works great! */
 649
 650         if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
 651                 return -errno;
 652
 653         if (recvmsg(netns_storage_socket[0], &mh, MSG_DONTWAIT|MSG_CMSG_CLOEXEC) < 0) {
 654                 if (errno != EAGAIN) {
 655                         r = -errno;
 656                         goto fail;
 657                 }
 658
 659                 /* Nothing stored yet, so let's create a new namespace */
 660
 661                 if (unshare(CLONE_NEWNET) < 0) {
 662                         r = -errno;
 663                         goto fail;
 664                 }
 665
 666                 loopback_setup();
 667
 668                 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
 669                 if (netns < 0) {
 670                         r = -errno;
 671                         goto fail;
 672                 }
 673
 674                 r = 1;
 675         } else {
 676                 /* Yay, found something, so let's join the namespace */
 677
 678                 for (cmsg = CMSG_FIRSTHDR(&mh); cmsg; cmsg = CMSG_NXTHDR(&mh, cmsg)) {
 679                         if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
 680                                 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
 681                                 netns = *(int*) CMSG_DATA(cmsg);
 682                         }
 683                 }
 684
 685                 if (setns(netns, CLONE_NEWNET) < 0) {
 686                         r = -errno;
 687                         goto fail;
 688                 }
 689
 690                 r = 0;
 691         }
 692
 693         cmsg = CMSG_FIRSTHDR(&mh);
 694         cmsg->cmsg_level = SOL_SOCKET;
 695         cmsg->cmsg_type = SCM_RIGHTS;
 696         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
 697         memcpy(CMSG_DATA(cmsg), &netns, sizeof(int));
 698         mh.msg_controllen = cmsg->cmsg_len;
 699
 700         if (sendmsg(netns_storage_socket[1], &mh, MSG_DONTWAIT|MSG_NOSIGNAL) < 0) {
 701                 r = -errno;
 702                 goto fail;
 703         }
 704
 705 fail:
 706         lockf(netns_storage_socket[0], F_ULOCK, 0);
 707
 708         return r;
 709 }
 710
 711 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
 712         [PROTECT_HOME_NO] = "no",
 713         [PROTECT_HOME_YES] = "yes",
 714         [PROTECT_HOME_READ_ONLY] = "read-only",
 715 };
 716
 717 DEFINE_STRING_TABLE_LOOKUP(protect_home, ProtectHome);
 718
 719 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
 720         [PROTECT_SYSTEM_NO] = "no",
 721         [PROTECT_SYSTEM_YES] = "yes",
 722         [PROTECT_SYSTEM_FULL] = "full",
 723 };
 724
 725 DEFINE_STRING_TABLE_LOOKUP(protect_system, ProtectSystem);