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) < 0) {
  62                                 r = log_debug_errno(errno, "Failed to umount %s: %m", path);
  63                                 continue;
  64                         }
  65
  66                         log_debug("Successfully unmounted %s", path);
  67
  68                         again = true;
  69                         n++;
  70
  71                         break;
  72                 }
  73
  74         } while (again);
  75
  76         return n;
  77 }
  78
  79 static int get_mount_flags(
  80                 struct libmnt_table *table,
  81                 const char *path,
  82                 unsigned long *ret) {
  83         struct libmnt_fs *fs;
  84         struct statvfs buf;
  85         const char *opts;
  86         int r = 0;
  87
  88         /* Get the mount flags for the mountpoint at "path" from "table". We have a fallback using statvfs()
  89          * in place (which provides us with mostly the same info), but it's just a fallback, since using it
  90          * means triggering autofs or NFS mounts, which we'd rather avoid needlessly. */
  91
  92         fs = mnt_table_find_target(table, path, MNT_ITER_FORWARD);
  93         if (!fs) {
  94                 log_debug("Could not find '%s' in mount table, ignoring.", path);
  95                 goto fallback;
  96         }
  97
  98         opts = mnt_fs_get_vfs_options(fs);
  99         if (!opts) {
 100                 *ret = 0;
 101                 return 0;
 102         }
 103
 104         r = mnt_optstr_get_flags(opts, ret, mnt_get_builtin_optmap(MNT_LINUX_MAP));
 105         if (r != 0) {
 106                 log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path);
 107                 goto fallback;
 108         }
 109
 110         /* MS_RELATIME is default and trying to set it in an unprivileged container causes EPERM */
 111         *ret &= ~MS_RELATIME;
 112         return 0;
 113
 114 fallback:
 115         if (statvfs(path, &buf) < 0)
 116                 return -errno;
 117
 118         /* The statvfs() flags and the mount flags mostly have the same values, but for some cases do
 119          * not. Hence map the flags manually. (Strictly speaking, ST_RELATIME/MS_RELATIME is the most
 120          * prominent one that doesn't match, but that's the one we mask away anyway, see above.) */
 121
 122         *ret =
 123                 FLAGS_SET(buf.f_flag, ST_RDONLY) * MS_RDONLY |
 124                 FLAGS_SET(buf.f_flag, ST_NODEV) * MS_NODEV |
 125                 FLAGS_SET(buf.f_flag, ST_NOEXEC) * MS_NOEXEC |
 126                 FLAGS_SET(buf.f_flag, ST_NOSUID) * MS_NOSUID |
 127                 FLAGS_SET(buf.f_flag, ST_NOATIME) * MS_NOATIME |
 128                 FLAGS_SET(buf.f_flag, ST_NODIRATIME) * MS_NODIRATIME;
 129
 130         return 0;
 131 }
 132
 133 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
 134  * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
 135 int bind_remount_recursive_with_mountinfo(
 136                 const char *prefix,
 137                 unsigned long new_flags,
 138                 unsigned long flags_mask,
 139                 char **deny_list,
 140                 FILE *proc_self_mountinfo) {
 141
 142         _cleanup_set_free_free_ Set *done = NULL;
 143         _cleanup_free_ char *simplified = NULL;
 144         int r;
 145
 146         assert(prefix);
 147         assert(proc_self_mountinfo);
 148
 149         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 150          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 151          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 152          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 153          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 154          * do not have any effect on future submounts that might get propagated, they might be writable. This includes
 155          * future submounts that have been triggered via autofs.
 156          *
 157          * If the "deny_list" parameter is specified it may contain a list of subtrees to exclude from the
 158          * remount operation. Note that we'll ignore the deny list for the top-level path. */
 159
 160         simplified = strdup(prefix);
 161         if (!simplified)
 162                 return -ENOMEM;
 163
 164         path_simplify(simplified, false);
 165
 166         done = set_new(&path_hash_ops);
 167         if (!done)
 168                 return -ENOMEM;
 169
 170         for (;;) {
 171                 _cleanup_set_free_free_ Set *todo = NULL;
 172                 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
 173                 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
 174                 bool top_autofs = false;
 175                 char *x;
 176                 unsigned long orig_flags;
 177
 178                 todo = set_new(&path_hash_ops);
 179                 if (!todo)
 180                         return -ENOMEM;
 181
 182                 rewind(proc_self_mountinfo);
 183
 184                 r = libmount_parse("/proc/self/mountinfo", proc_self_mountinfo, &table, &iter);
 185                 if (r < 0)
 186                         return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
 187
 188                 for (;;) {
 189                         struct libmnt_fs *fs;
 190                         const char *path, *type;
 191
 192                         r = mnt_table_next_fs(table, iter, &fs);
 193                         if (r == 1)
 194                                 break;
 195                         if (r < 0)
 196                                 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
 197
 198                         path = mnt_fs_get_target(fs);
 199                         type = mnt_fs_get_fstype(fs);
 200                         if (!path || !type)
 201                                 continue;
 202
 203                         if (!path_startswith(path, simplified))
 204                                 continue;
 205
 206                         /* Ignore this mount if it is deny-listed, but only if it isn't the top-level mount
 207                          * we shall operate on. */
 208                         if (!path_equal(path, simplified)) {
 209                                 bool deny_listed = false;
 210                                 char **i;
 211
 212                                 STRV_FOREACH(i, deny_list) {
 213                                         if (path_equal(*i, simplified))
 214                                                 continue;
 215
 216                                         if (!path_startswith(*i, simplified))
 217                                                 continue;
 218
 219                                         if (path_startswith(path, *i)) {
 220                                                 deny_listed = true;
 221                                                 log_debug("Not remounting %s deny-listed by %s, called for %s",
 222                                                           path, *i, simplified);
 223                                                 break;
 224                                         }
 225                                 }
 226                                 if (deny_listed)
 227                                         continue;
 228                         }
 229
 230                         /* Let's ignore autofs mounts.  If they aren't
 231                          * triggered yet, we want to avoid triggering
 232                          * them, as we don't make any guarantees for
 233                          * future submounts anyway.  If they are
 234                          * already triggered, then we will find
 235                          * another entry for this. */
 236                         if (streq(type, "autofs")) {
 237                                 top_autofs = top_autofs || path_equal(path, simplified);
 238                                 continue;
 239                         }
 240
 241                         if (!set_contains(done, path)) {
 242                                 r = set_put_strdup(&todo, path);
 243                                 if (r < 0)
 244                                         return r;
 245                         }
 246                 }
 247
 248                 /* If we have no submounts to process anymore and if
 249                  * the root is either already done, or an autofs, we
 250                  * are done */
 251                 if (set_isempty(todo) &&
 252                     (top_autofs || set_contains(done, simplified)))
 253                         return 0;
 254
 255                 if (!set_contains(done, simplified) &&
 256                     !set_contains(todo, simplified)) {
 257                         /* The prefix directory itself is not yet a mount, make it one. */
 258                         if (mount(simplified, simplified, NULL, MS_BIND|MS_REC, NULL) < 0)
 259                                 return -errno;
 260
 261                         orig_flags = 0;
 262                         (void) get_mount_flags(table, simplified, &orig_flags);
 263
 264                         if (mount(NULL, simplified, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 265                                 return -errno;
 266
 267                         log_debug("Made top-level directory %s a mount point.", prefix);
 268
 269                         r = set_put_strdup(&done, simplified);
 270                         if (r < 0)
 271                                 return r;
 272                 }
 273
 274                 while ((x = set_steal_first(todo))) {
 275
 276                         r = set_consume(done, x);
 277                         if (IN_SET(r, 0, -EEXIST))
 278                                 continue;
 279                         if (r < 0)
 280                                 return r;
 281
 282                         /* Deal with mount points that are obstructed by a later mount */
 283                         r = path_is_mount_point(x, NULL, 0);
 284                         if (IN_SET(r, 0, -ENOENT))
 285                                 continue;
 286                         if (IN_SET(r, -EACCES, -EPERM)) {
 287                                 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
 288                                  * may not be acceessed. E.g.,
 289                                  *
 290                                  * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
 291                                  * $ bindfs --no-allow-other ~/mnt ~/mnt
 292                                  *
 293                                  * Then, root user cannot access the mount point ~/mnt/mnt.
 294                                  * In such cases, the submounts are ignored, as we have no way to manage them. */
 295                                 log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
 296                                 continue;
 297                         }
 298                         if (r < 0)
 299                                 return r;
 300
 301                         /* Try to reuse the original flag set */
 302                         orig_flags = 0;
 303                         (void) get_mount_flags(table, x, &orig_flags);
 304
 305                         if (mount(NULL, x, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 306                                 return -errno;
 307
 308                         log_debug("Remounted %s read-only.", x);
 309                 }
 310         }
 311 }
 312
 313 int bind_remount_recursive(
 314                 const char *prefix,
 315                 unsigned long new_flags,
 316                 unsigned long flags_mask,
 317                 char **deny_list) {
 318
 319         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 320         int r;
 321
 322         r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo);
 323         if (r < 0)
 324                 return r;
 325
 326         return bind_remount_recursive_with_mountinfo(prefix, new_flags, flags_mask, deny_list, proc_self_mountinfo);
 327 }
 328
 329 int bind_remount_one_with_mountinfo(
 330                 const char *path,
 331                 unsigned long new_flags,
 332                 unsigned long flags_mask,
 333                 FILE *proc_self_mountinfo) {
 334
 335         _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
 336         unsigned long orig_flags = 0;
 337         int r;
 338
 339         assert(path);
 340         assert(proc_self_mountinfo);
 341
 342         rewind(proc_self_mountinfo);
 343
 344         table = mnt_new_table();
 345         if (!table)
 346                 return -ENOMEM;
 347
 348         r = mnt_table_parse_stream(table, proc_self_mountinfo, "/proc/self/mountinfo");
 349         if (r < 0)
 350                 return r;
 351
 352         /* Try to reuse the original flag set */
 353         (void) get_mount_flags(table, path, &orig_flags);
 354
 355         if (mount(NULL, path, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 356                 return -errno;
 357
 358         return 0;
 359 }
 360
 361 int mount_move_root(const char *path) {
 362         assert(path);
 363
 364         if (chdir(path) < 0)
 365                 return -errno;
 366
 367         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 368                 return -errno;
 369
 370         if (chroot(".") < 0)
 371                 return -errno;
 372
 373         if (chdir("/") < 0)
 374                 return -errno;
 375
 376         return 0;
 377 }
 378
 379 int repeat_unmount(const char *path, int flags) {
 380         bool done = false;
 381
 382         assert(path);
 383
 384         /* If there are multiple mounts on a mount point, this
 385          * removes them all */
 386
 387         for (;;) {
 388                 if (umount2(path, flags) < 0) {
 389
 390                         if (errno == EINVAL)
 391                                 return done;
 392
 393                         return -errno;
 394                 }
 395
 396                 done = true;
 397         }
 398 }
 399
 400 int mode_to_inaccessible_node(
 401                 const char *runtime_dir,
 402                 mode_t mode,
 403                 char **ret) {
 404
 405         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created
 406          * during early boot by PID 1. In some cases we lacked the privs to create the character and block
 407          * devices (maybe because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a
 408          * devices policy that excludes device nodes with major and minor of 0), but that's fine, in that
 409          * case we use an AF_UNIX file node instead, which is not the same, but close enough for most
 410          * uses. And most importantly, the kernel allows bind mounts from socket nodes to any non-directory
 411          * file nodes, and that's the most important thing that matters.
 412          *
 413          * Note that the runtime directory argument shall be the top-level runtime directory, i.e. /run/ if
 414          * we operate in system context and $XDG_RUNTIME_DIR if we operate in user context. */
 415
 416         _cleanup_free_ char *d = NULL;
 417         const char *node = NULL;
 418         bool fallback = false;
 419
 420         assert(ret);
 421
 422         if (!runtime_dir)
 423                 runtime_dir = "/run";
 424
 425         switch(mode & S_IFMT) {
 426                 case S_IFREG:
 427                         node = "/systemd/inaccessible/reg";
 428                         break;
 429
 430                 case S_IFDIR:
 431                         node = "/systemd/inaccessible/dir";
 432                         break;
 433
 434                 case S_IFCHR:
 435                         node = "/systemd/inaccessible/chr";
 436                         fallback = true;
 437                         break;
 438
 439                 case S_IFBLK:
 440                         node = "/systemd/inaccessible/blk";
 441                         fallback = true;
 442                         break;
 443
 444                 case S_IFIFO:
 445                         node = "/systemd/inaccessible/fifo";
 446                         break;
 447
 448                 case S_IFSOCK:
 449                         node = "/systemd/inaccessible/sock";
 450                         break;
 451         }
 452         if (!node)
 453                 return -EINVAL;
 454
 455         d = path_join(runtime_dir, node);
 456         if (!d)
 457                 return -ENOMEM;
 458
 459         if (fallback && access(d, F_OK) < 0) {
 460                 free(d);
 461                 d = path_join(runtime_dir, "/systemd/inaccessible/sock");
 462                 if (!d)
 463                         return -ENOMEM;
 464         }
 465
 466         *ret = TAKE_PTR(d);
 467         return 0;
 468 }
 469
 470 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 471 static char* mount_flags_to_string(long unsigned flags) {
 472         char *x;
 473         _cleanup_free_ char *y = NULL;
 474         long unsigned overflow;
 475
 476         overflow = flags & ~(MS_RDONLY |
 477                              MS_NOSUID |
 478                              MS_NODEV |
 479                              MS_NOEXEC |
 480                              MS_SYNCHRONOUS |
 481                              MS_REMOUNT |
 482                              MS_MANDLOCK |
 483                              MS_DIRSYNC |
 484                              MS_NOATIME |
 485                              MS_NODIRATIME |
 486                              MS_BIND |
 487                              MS_MOVE |
 488                              MS_REC |
 489                              MS_SILENT |
 490                              MS_POSIXACL |
 491                              MS_UNBINDABLE |
 492                              MS_PRIVATE |
 493                              MS_SLAVE |
 494                              MS_SHARED |
 495                              MS_RELATIME |
 496                              MS_KERNMOUNT |
 497                              MS_I_VERSION |
 498                              MS_STRICTATIME |
 499                              MS_LAZYTIME);
 500
 501         if (flags == 0 || overflow != 0)
 502                 if (asprintf(&y, "%lx", overflow) < 0)
 503                         return NULL;
 504
 505         x = strjoin(FLAG(MS_RDONLY),
 506                     FLAG(MS_NOSUID),
 507                     FLAG(MS_NODEV),
 508                     FLAG(MS_NOEXEC),
 509                     FLAG(MS_SYNCHRONOUS),
 510                     FLAG(MS_REMOUNT),
 511                     FLAG(MS_MANDLOCK),
 512                     FLAG(MS_DIRSYNC),
 513                     FLAG(MS_NOATIME),
 514                     FLAG(MS_NODIRATIME),
 515                     FLAG(MS_BIND),
 516                     FLAG(MS_MOVE),
 517                     FLAG(MS_REC),
 518                     FLAG(MS_SILENT),
 519                     FLAG(MS_POSIXACL),
 520                     FLAG(MS_UNBINDABLE),
 521                     FLAG(MS_PRIVATE),
 522                     FLAG(MS_SLAVE),
 523                     FLAG(MS_SHARED),
 524                     FLAG(MS_RELATIME),
 525                     FLAG(MS_KERNMOUNT),
 526                     FLAG(MS_I_VERSION),
 527                     FLAG(MS_STRICTATIME),
 528                     FLAG(MS_LAZYTIME),
 529                     y);
 530         if (!x)
 531                 return NULL;
 532         if (!y)
 533                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 534         return x;
 535 }
 536
 537 int mount_verbose(
 538                 int error_log_level,
 539                 const char *what,
 540                 const char *where,
 541                 const char *type,
 542                 unsigned long flags,
 543                 const char *options) {
 544
 545         _cleanup_free_ char *fl = NULL, *o = NULL;
 546         unsigned long f;
 547         int r;
 548
 549         r = mount_option_mangle(options, flags, &f, &o);
 550         if (r < 0)
 551                 return log_full_errno(error_log_level, r,
 552                                       "Failed to mangle mount options %s: %m",
 553                                       strempty(options));
 554
 555         fl = mount_flags_to_string(f);
 556
 557         if ((f & MS_REMOUNT) && !what && !type)
 558                 log_debug("Remounting %s (%s \"%s\")...",
 559                           where, strnull(fl), strempty(o));
 560         else if (!what && !type)
 561                 log_debug("Mounting %s (%s \"%s\")...",
 562                           where, strnull(fl), strempty(o));
 563         else if ((f & MS_BIND) && !type)
 564                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 565                           what, where, strnull(fl), strempty(o));
 566         else if (f & MS_MOVE)
 567                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 568                           what, where, strnull(fl), strempty(o));
 569         else
 570                 log_debug("Mounting %s on %s (%s \"%s\")...",
 571                           strna(type), where, strnull(fl), strempty(o));
 572         if (mount(what, where, type, f, o) < 0)
 573                 return log_full_errno(error_log_level, errno,
 574                                       "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
 575                                       strna(what), strna(type), where, strnull(fl), strempty(o));
 576         return 0;
 577 }
 578
 579 int umount_verbose(const char *what) {
 580         log_debug("Umounting %s...", what);
 581         if (umount(what) < 0)
 582                 return log_error_errno(errno, "Failed to unmount %s: %m", what);
 583         return 0;
 584 }
 585
 586 int mount_option_mangle(
 587                 const char *options,
 588                 unsigned long mount_flags,
 589                 unsigned long *ret_mount_flags,
 590                 char **ret_remaining_options) {
 591
 592         const struct libmnt_optmap *map;
 593         _cleanup_free_ char *ret = NULL;
 594         const char *p;
 595         int r;
 596
 597         /* This extracts mount flags from the mount options, and store
 598          * non-mount-flag options to '*ret_remaining_options'.
 599          * E.g.,
 600          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 601          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 602          * "size=1630748k,mode=700,uid=1000,gid=1000".
 603          * See more examples in test-mount-utils.c.
 604          *
 605          * Note that if 'options' does not contain any non-mount-flag options,
 606          * then '*ret_remaining_options' is set to NULL instead of empty string.
 607          * Note that this does not check validity of options stored in
 608          * '*ret_remaining_options'.
 609          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 610          * to '*ret_mount_flags'. */
 611
 612         assert(ret_mount_flags);
 613         assert(ret_remaining_options);
 614
 615         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 616         if (!map)
 617                 return -EINVAL;
 618
 619         p = options;
 620         for (;;) {
 621                 _cleanup_free_ char *word = NULL;
 622                 const struct libmnt_optmap *ent;
 623
 624                 r = extract_first_word(&p, &word, ",", EXTRACT_UNQUOTE);
 625                 if (r < 0)
 626                         return r;
 627                 if (r == 0)
 628                         break;
 629
 630                 for (ent = map; ent->name; ent++) {
 631                         /* All entries in MNT_LINUX_MAP do not take any argument.
 632                          * Thus, ent->name does not contain "=" or "[=]". */
 633                         if (!streq(word, ent->name))
 634                                 continue;
 635
 636                         if (!(ent->mask & MNT_INVERT))
 637                                 mount_flags |= ent->id;
 638                         else if (mount_flags & ent->id)
 639                                 mount_flags ^= ent->id;
 640
 641                         break;
 642                 }
 643
 644                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 645                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 646                         return -ENOMEM;
 647         }
 648
 649         *ret_mount_flags = mount_flags;
 650         *ret_remaining_options = TAKE_PTR(ret);
 651
 652         return 0;
 653 }