src/shared/mount-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <stdlib.h>
   5 #include <sys/mount.h>
   6 #include <sys/stat.h>
   7 #include <sys/statvfs.h>
   8 #include <unistd.h>
   9
  10 #include "alloc-util.h"
  11 #include "extract-word.h"
  12 #include "fd-util.h"
  13 #include "fileio.h"
  14 #include "fs-util.h"
  15 #include "hashmap.h"
  16 #include "libmount-util.h"
  17 #include "mount-util.h"
  18 #include "mountpoint-util.h"
  19 #include "parse-util.h"
  20 #include "path-util.h"
  21 #include "set.h"
  22 #include "stdio-util.h"
  23 #include "string-util.h"
  24 #include "strv.h"
  25
  26 int umount_recursive(const char *prefix, int flags) {
  27         int n = 0, r;
  28         bool again;
  29
  30         /* Try to umount everything recursively below a
  31          * directory. Also, take care of stacked mounts, and keep
  32          * unmounting them until they are gone. */
  33
  34         do {
  35                 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
  36                 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
  37
  38                 again = false;
  39
  40                 r = libmount_parse("/proc/self/mountinfo", NULL, &table, &iter);
  41                 if (r < 0)
  42                         return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
  43
  44                 for (;;) {
  45                         struct libmnt_fs *fs;
  46                         const char *path;
  47
  48                         r = mnt_table_next_fs(table, iter, &fs);
  49                         if (r == 1)
  50                                 break;
  51                         if (r < 0)
  52                                 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
  53
  54                         path = mnt_fs_get_target(fs);
  55                         if (!path)
  56                                 continue;
  57
  58                         if (!path_startswith(path, prefix))
  59                                 continue;
  60
  61                         if (umount2(path, flags | UMOUNT_NOFOLLOW) < 0) {
  62                                 log_debug_errno(errno, "Failed to umount %s, ignoring: %m", path);
  63                                 continue;
  64                         }
  65
  66                         log_debug("Successfully unmounted %s", path);
  67
  68                         again = true;
  69                         n++;
  70
  71                         break;
  72                 }
  73         } while (again);
  74
  75         return n;
  76 }
  77
  78 static int get_mount_flags(
  79                 struct libmnt_table *table,
  80                 const char *path,
  81                 unsigned long *ret) {
  82         struct libmnt_fs *fs;
  83         struct statvfs buf;
  84         const char *opts;
  85         int r = 0;
  86
  87         /* Get the mount flags for the mountpoint at "path" from "table". We have a fallback using statvfs()
  88          * in place (which provides us with mostly the same info), but it's just a fallback, since using it
  89          * means triggering autofs or NFS mounts, which we'd rather avoid needlessly. */
  90
  91         fs = mnt_table_find_target(table, path, MNT_ITER_FORWARD);
  92         if (!fs) {
  93                 log_debug("Could not find '%s' in mount table, ignoring.", path);
  94                 goto fallback;
  95         }
  96
  97         opts = mnt_fs_get_vfs_options(fs);
  98         if (!opts) {
  99                 *ret = 0;
 100                 return 0;
 101         }
 102
 103         r = mnt_optstr_get_flags(opts, ret, mnt_get_builtin_optmap(MNT_LINUX_MAP));
 104         if (r != 0) {
 105                 log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path);
 106                 goto fallback;
 107         }
 108
 109         /* MS_RELATIME is default and trying to set it in an unprivileged container causes EPERM */
 110         *ret &= ~MS_RELATIME;
 111         return 0;
 112
 113 fallback:
 114         if (statvfs(path, &buf) < 0)
 115                 return -errno;
 116
 117         /* The statvfs() flags and the mount flags mostly have the same values, but for some cases do
 118          * not. Hence map the flags manually. (Strictly speaking, ST_RELATIME/MS_RELATIME is the most
 119          * prominent one that doesn't match, but that's the one we mask away anyway, see above.) */
 120
 121         *ret =
 122                 FLAGS_SET(buf.f_flag, ST_RDONLY) * MS_RDONLY |
 123                 FLAGS_SET(buf.f_flag, ST_NODEV) * MS_NODEV |
 124                 FLAGS_SET(buf.f_flag, ST_NOEXEC) * MS_NOEXEC |
 125                 FLAGS_SET(buf.f_flag, ST_NOSUID) * MS_NOSUID |
 126                 FLAGS_SET(buf.f_flag, ST_NOATIME) * MS_NOATIME |
 127                 FLAGS_SET(buf.f_flag, ST_NODIRATIME) * MS_NODIRATIME;
 128
 129         return 0;
 130 }
 131
 132 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
 133  * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
 134 int bind_remount_recursive_with_mountinfo(
 135                 const char *prefix,
 136                 unsigned long new_flags,
 137                 unsigned long flags_mask,
 138                 char **deny_list,
 139                 FILE *proc_self_mountinfo) {
 140
 141         _cleanup_set_free_free_ Set *done = NULL;
 142         _cleanup_free_ char *simplified = NULL;
 143         int r;
 144
 145         assert(prefix);
 146         assert(proc_self_mountinfo);
 147
 148         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 149          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 150          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 151          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 152          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 153          * do not have any effect on future submounts that might get propagated, they might be writable. This includes
 154          * future submounts that have been triggered via autofs.
 155          *
 156          * If the "deny_list" parameter is specified it may contain a list of subtrees to exclude from the
 157          * remount operation. Note that we'll ignore the deny list for the top-level path. */
 158
 159         simplified = strdup(prefix);
 160         if (!simplified)
 161                 return -ENOMEM;
 162
 163         path_simplify(simplified, false);
 164
 165         done = set_new(&path_hash_ops);
 166         if (!done)
 167                 return -ENOMEM;
 168
 169         for (;;) {
 170                 _cleanup_set_free_free_ Set *todo = NULL;
 171                 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
 172                 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
 173                 bool top_autofs = false;
 174                 char *x;
 175                 unsigned long orig_flags;
 176
 177                 todo = set_new(&path_hash_ops);
 178                 if (!todo)
 179                         return -ENOMEM;
 180
 181                 rewind(proc_self_mountinfo);
 182
 183                 r = libmount_parse("/proc/self/mountinfo", proc_self_mountinfo, &table, &iter);
 184                 if (r < 0)
 185                         return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
 186
 187                 for (;;) {
 188                         struct libmnt_fs *fs;
 189                         const char *path, *type;
 190
 191                         r = mnt_table_next_fs(table, iter, &fs);
 192                         if (r == 1)
 193                                 break;
 194                         if (r < 0)
 195                                 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
 196
 197                         path = mnt_fs_get_target(fs);
 198                         type = mnt_fs_get_fstype(fs);
 199                         if (!path || !type)
 200                                 continue;
 201
 202                         if (!path_startswith(path, simplified))
 203                                 continue;
 204
 205                         /* Ignore this mount if it is deny-listed, but only if it isn't the top-level mount
 206                          * we shall operate on. */
 207                         if (!path_equal(path, simplified)) {
 208                                 bool deny_listed = false;
 209                                 char **i;
 210
 211                                 STRV_FOREACH(i, deny_list) {
 212                                         if (path_equal(*i, simplified))
 213                                                 continue;
 214
 215                                         if (!path_startswith(*i, simplified))
 216                                                 continue;
 217
 218                                         if (path_startswith(path, *i)) {
 219                                                 deny_listed = true;
 220                                                 log_debug("Not remounting %s deny-listed by %s, called for %s",
 221                                                           path, *i, simplified);
 222                                                 break;
 223                                         }
 224                                 }
 225                                 if (deny_listed)
 226                                         continue;
 227                         }
 228
 229                         /* Let's ignore autofs mounts.  If they aren't
 230                          * triggered yet, we want to avoid triggering
 231                          * them, as we don't make any guarantees for
 232                          * future submounts anyway.  If they are
 233                          * already triggered, then we will find
 234                          * another entry for this. */
 235                         if (streq(type, "autofs")) {
 236                                 top_autofs = top_autofs || path_equal(path, simplified);
 237                                 continue;
 238                         }
 239
 240                         if (!set_contains(done, path)) {
 241                                 r = set_put_strdup(&todo, path);
 242                                 if (r < 0)
 243                                         return r;
 244                         }
 245                 }
 246
 247                 /* If we have no submounts to process anymore and if
 248                  * the root is either already done, or an autofs, we
 249                  * are done */
 250                 if (set_isempty(todo) &&
 251                     (top_autofs || set_contains(done, simplified)))
 252                         return 0;
 253
 254                 if (!set_contains(done, simplified) &&
 255                     !set_contains(todo, simplified)) {
 256                         /* The prefix directory itself is not yet a mount, make it one. */
 257                         if (mount(simplified, simplified, NULL, MS_BIND|MS_REC, NULL) < 0)
 258                                 return -errno;
 259
 260                         orig_flags = 0;
 261                         (void) get_mount_flags(table, simplified, &orig_flags);
 262
 263                         if (mount(NULL, simplified, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 264                                 return -errno;
 265
 266                         log_debug("Made top-level directory %s a mount point.", prefix);
 267
 268                         r = set_put_strdup(&done, simplified);
 269                         if (r < 0)
 270                                 return r;
 271                 }
 272
 273                 while ((x = set_steal_first(todo))) {
 274
 275                         r = set_consume(done, x);
 276                         if (IN_SET(r, 0, -EEXIST))
 277                                 continue;
 278                         if (r < 0)
 279                                 return r;
 280
 281                         /* Deal with mount points that are obstructed by a later mount */
 282                         r = path_is_mount_point(x, NULL, 0);
 283                         if (IN_SET(r, 0, -ENOENT))
 284                                 continue;
 285                         if (IN_SET(r, -EACCES, -EPERM)) {
 286                                 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
 287                                  * may not be acceessed. E.g.,
 288                                  *
 289                                  * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
 290                                  * $ bindfs --no-allow-other ~/mnt ~/mnt
 291                                  *
 292                                  * Then, root user cannot access the mount point ~/mnt/mnt.
 293                                  * In such cases, the submounts are ignored, as we have no way to manage them. */
 294                                 log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
 295                                 continue;
 296                         }
 297                         if (r < 0)
 298                                 return r;
 299
 300                         /* Try to reuse the original flag set */
 301                         orig_flags = 0;
 302                         (void) get_mount_flags(table, x, &orig_flags);
 303
 304                         if (mount(NULL, x, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 305                                 return -errno;
 306
 307                         log_debug("Remounted %s read-only.", x);
 308                 }
 309         }
 310 }
 311
 312 int bind_remount_recursive(
 313                 const char *prefix,
 314                 unsigned long new_flags,
 315                 unsigned long flags_mask,
 316                 char **deny_list) {
 317
 318         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 319         int r;
 320
 321         r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo);
 322         if (r < 0)
 323                 return r;
 324
 325         return bind_remount_recursive_with_mountinfo(prefix, new_flags, flags_mask, deny_list, proc_self_mountinfo);
 326 }
 327
 328 int bind_remount_one_with_mountinfo(
 329                 const char *path,
 330                 unsigned long new_flags,
 331                 unsigned long flags_mask,
 332                 FILE *proc_self_mountinfo) {
 333
 334         _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
 335         unsigned long orig_flags = 0;
 336         int r;
 337
 338         assert(path);
 339         assert(proc_self_mountinfo);
 340
 341         rewind(proc_self_mountinfo);
 342
 343         table = mnt_new_table();
 344         if (!table)
 345                 return -ENOMEM;
 346
 347         r = mnt_table_parse_stream(table, proc_self_mountinfo, "/proc/self/mountinfo");
 348         if (r < 0)
 349                 return r;
 350
 351         /* Try to reuse the original flag set */
 352         (void) get_mount_flags(table, path, &orig_flags);
 353
 354         if (mount(NULL, path, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 355                 return -errno;
 356
 357         return 0;
 358 }
 359
 360 int mount_move_root(const char *path) {
 361         assert(path);
 362
 363         if (chdir(path) < 0)
 364                 return -errno;
 365
 366         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 367                 return -errno;
 368
 369         if (chroot(".") < 0)
 370                 return -errno;
 371
 372         if (chdir("/") < 0)
 373                 return -errno;
 374
 375         return 0;
 376 }
 377
 378 int repeat_unmount(const char *path, int flags) {
 379         bool done = false;
 380
 381         assert(path);
 382
 383         /* If there are multiple mounts on a mount point, this
 384          * removes them all */
 385
 386         for (;;) {
 387                 if (umount2(path, flags) < 0) {
 388
 389                         if (errno == EINVAL)
 390                                 return done;
 391
 392                         return -errno;
 393                 }
 394
 395                 done = true;
 396         }
 397 }
 398
 399 int mode_to_inaccessible_node(
 400                 const char *runtime_dir,
 401                 mode_t mode,
 402                 char **ret) {
 403
 404         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created
 405          * during early boot by PID 1. In some cases we lacked the privs to create the character and block
 406          * devices (maybe because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a
 407          * devices policy that excludes device nodes with major and minor of 0), but that's fine, in that
 408          * case we use an AF_UNIX file node instead, which is not the same, but close enough for most
 409          * uses. And most importantly, the kernel allows bind mounts from socket nodes to any non-directory
 410          * file nodes, and that's the most important thing that matters.
 411          *
 412          * Note that the runtime directory argument shall be the top-level runtime directory, i.e. /run/ if
 413          * we operate in system context and $XDG_RUNTIME_DIR if we operate in user context. */
 414
 415         _cleanup_free_ char *d = NULL;
 416         const char *node = NULL;
 417         bool fallback = false;
 418
 419         assert(ret);
 420
 421         if (!runtime_dir)
 422                 runtime_dir = "/run";
 423
 424         switch(mode & S_IFMT) {
 425                 case S_IFREG:
 426                         node = "/systemd/inaccessible/reg";
 427                         break;
 428
 429                 case S_IFDIR:
 430                         node = "/systemd/inaccessible/dir";
 431                         break;
 432
 433                 case S_IFCHR:
 434                         node = "/systemd/inaccessible/chr";
 435                         fallback = true;
 436                         break;
 437
 438                 case S_IFBLK:
 439                         node = "/systemd/inaccessible/blk";
 440                         fallback = true;
 441                         break;
 442
 443                 case S_IFIFO:
 444                         node = "/systemd/inaccessible/fifo";
 445                         break;
 446
 447                 case S_IFSOCK:
 448                         node = "/systemd/inaccessible/sock";
 449                         break;
 450         }
 451         if (!node)
 452                 return -EINVAL;
 453
 454         d = path_join(runtime_dir, node);
 455         if (!d)
 456                 return -ENOMEM;
 457
 458         if (fallback && access(d, F_OK) < 0) {
 459                 free(d);
 460                 d = path_join(runtime_dir, "/systemd/inaccessible/sock");
 461                 if (!d)
 462                         return -ENOMEM;
 463         }
 464
 465         *ret = TAKE_PTR(d);
 466         return 0;
 467 }
 468
 469 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 470 static char* mount_flags_to_string(long unsigned flags) {
 471         char *x;
 472         _cleanup_free_ char *y = NULL;
 473         long unsigned overflow;
 474
 475         overflow = flags & ~(MS_RDONLY |
 476                              MS_NOSUID |
 477                              MS_NODEV |
 478                              MS_NOEXEC |
 479                              MS_SYNCHRONOUS |
 480                              MS_REMOUNT |
 481                              MS_MANDLOCK |
 482                              MS_DIRSYNC |
 483                              MS_NOATIME |
 484                              MS_NODIRATIME |
 485                              MS_BIND |
 486                              MS_MOVE |
 487                              MS_REC |
 488                              MS_SILENT |
 489                              MS_POSIXACL |
 490                              MS_UNBINDABLE |
 491                              MS_PRIVATE |
 492                              MS_SLAVE |
 493                              MS_SHARED |
 494                              MS_RELATIME |
 495                              MS_KERNMOUNT |
 496                              MS_I_VERSION |
 497                              MS_STRICTATIME |
 498                              MS_LAZYTIME);
 499
 500         if (flags == 0 || overflow != 0)
 501                 if (asprintf(&y, "%lx", overflow) < 0)
 502                         return NULL;
 503
 504         x = strjoin(FLAG(MS_RDONLY),
 505                     FLAG(MS_NOSUID),
 506                     FLAG(MS_NODEV),
 507                     FLAG(MS_NOEXEC),
 508                     FLAG(MS_SYNCHRONOUS),
 509                     FLAG(MS_REMOUNT),
 510                     FLAG(MS_MANDLOCK),
 511                     FLAG(MS_DIRSYNC),
 512                     FLAG(MS_NOATIME),
 513                     FLAG(MS_NODIRATIME),
 514                     FLAG(MS_BIND),
 515                     FLAG(MS_MOVE),
 516                     FLAG(MS_REC),
 517                     FLAG(MS_SILENT),
 518                     FLAG(MS_POSIXACL),
 519                     FLAG(MS_UNBINDABLE),
 520                     FLAG(MS_PRIVATE),
 521                     FLAG(MS_SLAVE),
 522                     FLAG(MS_SHARED),
 523                     FLAG(MS_RELATIME),
 524                     FLAG(MS_KERNMOUNT),
 525                     FLAG(MS_I_VERSION),
 526                     FLAG(MS_STRICTATIME),
 527                     FLAG(MS_LAZYTIME),
 528                     y);
 529         if (!x)
 530                 return NULL;
 531         if (!y)
 532                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 533         return x;
 534 }
 535
 536 int mount_verbose(
 537                 int error_log_level,
 538                 const char *what,
 539                 const char *where,
 540                 const char *type,
 541                 unsigned long flags,
 542                 const char *options) {
 543
 544         _cleanup_free_ char *fl = NULL, *o = NULL;
 545         unsigned long f;
 546         int r;
 547
 548         r = mount_option_mangle(options, flags, &f, &o);
 549         if (r < 0)
 550                 return log_full_errno(error_log_level, r,
 551                                       "Failed to mangle mount options %s: %m",
 552                                       strempty(options));
 553
 554         fl = mount_flags_to_string(f);
 555
 556         if ((f & MS_REMOUNT) && !what && !type)
 557                 log_debug("Remounting %s (%s \"%s\")...",
 558                           where, strnull(fl), strempty(o));
 559         else if (!what && !type)
 560                 log_debug("Mounting %s (%s \"%s\")...",
 561                           where, strnull(fl), strempty(o));
 562         else if ((f & MS_BIND) && !type)
 563                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 564                           what, where, strnull(fl), strempty(o));
 565         else if (f & MS_MOVE)
 566                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 567                           what, where, strnull(fl), strempty(o));
 568         else
 569                 log_debug("Mounting %s on %s (%s \"%s\")...",
 570                           strna(type), where, strnull(fl), strempty(o));
 571         if (mount(what, where, type, f, o) < 0)
 572                 return log_full_errno(error_log_level, errno,
 573                                       "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
 574                                       strna(what), strna(type), where, strnull(fl), strempty(o));
 575         return 0;
 576 }
 577
 578 int umount_verbose(const char *what) {
 579         log_debug("Umounting %s...", what);
 580         if (umount(what) < 0)
 581                 return log_error_errno(errno, "Failed to unmount %s: %m", what);
 582         return 0;
 583 }
 584
 585 int mount_option_mangle(
 586                 const char *options,
 587                 unsigned long mount_flags,
 588                 unsigned long *ret_mount_flags,
 589                 char **ret_remaining_options) {
 590
 591         const struct libmnt_optmap *map;
 592         _cleanup_free_ char *ret = NULL;
 593         const char *p;
 594         int r;
 595
 596         /* This extracts mount flags from the mount options, and store
 597          * non-mount-flag options to '*ret_remaining_options'.
 598          * E.g.,
 599          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 600          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 601          * "size=1630748k,mode=700,uid=1000,gid=1000".
 602          * See more examples in test-mount-utils.c.
 603          *
 604          * Note that if 'options' does not contain any non-mount-flag options,
 605          * then '*ret_remaining_options' is set to NULL instead of empty string.
 606          * Note that this does not check validity of options stored in
 607          * '*ret_remaining_options'.
 608          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 609          * to '*ret_mount_flags'. */
 610
 611         assert(ret_mount_flags);
 612         assert(ret_remaining_options);
 613
 614         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 615         if (!map)
 616                 return -EINVAL;
 617
 618         p = options;
 619         for (;;) {
 620                 _cleanup_free_ char *word = NULL;
 621                 const struct libmnt_optmap *ent;
 622
 623                 r = extract_first_word(&p, &word, ",", EXTRACT_UNQUOTE);
 624                 if (r < 0)
 625                         return r;
 626                 if (r == 0)
 627                         break;
 628
 629                 for (ent = map; ent->name; ent++) {
 630                         /* All entries in MNT_LINUX_MAP do not take any argument.
 631                          * Thus, ent->name does not contain "=" or "[=]". */
 632                         if (!streq(word, ent->name))
 633                                 continue;
 634
 635                         if (!(ent->mask & MNT_INVERT))
 636                                 mount_flags |= ent->id;
 637                         else if (mount_flags & ent->id)
 638                                 mount_flags ^= ent->id;
 639
 640                         break;
 641                 }
 642
 643                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 644                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 645                         return -ENOMEM;
 646         }
 647
 648         *ret_mount_flags = mount_flags;
 649         *ret_remaining_options = TAKE_PTR(ret);
 650
 651         return 0;
 652 }