src/shared/mount-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   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 "stat-util.h"
  23 #include "stdio-util.h"
  24 #include "string-util.h"
  25 #include "strv.h"
  26
  27 int mount_fd(const char *source,
  28              int target_fd,
  29              const char *filesystemtype,
  30              unsigned long mountflags,
  31              const void *data) {
  32
  33         char path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)];
  34
  35         xsprintf(path, "/proc/self/fd/%i", target_fd);
  36         if (mount(source, path, filesystemtype, mountflags, data) < 0) {
  37                 if (errno != ENOENT)
  38                         return -errno;
  39
  40                 /* ENOENT can mean two things: either that the source is missing, or that /proc/ isn't
  41                  * mounted. Check for the latter to generate better error messages. */
  42                 if (proc_mounted() == 0)
  43                         return -ENOSYS;
  44
  45                 return -ENOENT;
  46         }
  47
  48         return 0;
  49 }
  50
  51 int mount_nofollow(
  52                 const char *source,
  53                 const char *target,
  54                 const char *filesystemtype,
  55                 unsigned long mountflags,
  56                 const void *data) {
  57
  58         _cleanup_close_ int fd = -1;
  59
  60         /* In almost all cases we want to manipulate the mount table without following symlinks, hence
  61          * mount_nofollow() is usually the way to go. The only exceptions are environments where /proc/ is
  62          * not available yet, since we need /proc/self/fd/ for this logic to work. i.e. during the early
  63          * initialization of namespacing/container stuff where /proc is not yet mounted (and maybe even the
  64          * fs to mount) we can only use traditional mount() directly.
  65          *
  66          * Note that this disables following only for the final component of the target, i.e symlinks within
  67          * the path of the target are honoured, as are symlinks in the source path everywhere. */
  68
  69         fd = open(target, O_PATH|O_CLOEXEC|O_NOFOLLOW);
  70         if (fd < 0)
  71                 return -errno;
  72
  73         return mount_fd(source, fd, filesystemtype, mountflags, data);
  74 }
  75
  76 int umount_recursive(const char *prefix, int flags) {
  77         int n = 0, r;
  78         bool again;
  79
  80         /* Try to umount everything recursively below a
  81          * directory. Also, take care of stacked mounts, and keep
  82          * unmounting them until they are gone. */
  83
  84         do {
  85                 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
  86                 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
  87
  88                 again = false;
  89
  90                 r = libmount_parse("/proc/self/mountinfo", NULL, &table, &iter);
  91                 if (r < 0)
  92                         return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
  93
  94                 for (;;) {
  95                         struct libmnt_fs *fs;
  96                         const char *path;
  97
  98                         r = mnt_table_next_fs(table, iter, &fs);
  99                         if (r == 1)
 100                                 break;
 101                         if (r < 0)
 102                                 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
 103
 104                         path = mnt_fs_get_target(fs);
 105                         if (!path)
 106                                 continue;
 107
 108                         if (!path_startswith(path, prefix))
 109                                 continue;
 110
 111                         if (umount2(path, flags | UMOUNT_NOFOLLOW) < 0) {
 112                                 log_debug_errno(errno, "Failed to umount %s, ignoring: %m", path);
 113                                 continue;
 114                         }
 115
 116                         log_debug("Successfully unmounted %s", path);
 117
 118                         again = true;
 119                         n++;
 120
 121                         break;
 122                 }
 123         } while (again);
 124
 125         return n;
 126 }
 127
 128 static int get_mount_flags(
 129                 struct libmnt_table *table,
 130                 const char *path,
 131                 unsigned long *ret) {
 132
 133         _cleanup_close_ int fd = -1;
 134         struct libmnt_fs *fs;
 135         struct statvfs buf;
 136         const char *opts;
 137         int r;
 138
 139         /* Get the mount flags for the mountpoint at "path" from "table". We have a fallback using statvfs()
 140          * in place (which provides us with mostly the same info), but it's just a fallback, since using it
 141          * means triggering autofs or NFS mounts, which we'd rather avoid needlessly.
 142          *
 143          * This generally doesn't follow symlinks. */
 144
 145         fs = mnt_table_find_target(table, path, MNT_ITER_FORWARD);
 146         if (!fs) {
 147                 log_debug("Could not find '%s' in mount table, ignoring.", path);
 148                 goto fallback;
 149         }
 150
 151         opts = mnt_fs_get_vfs_options(fs);
 152         if (!opts) {
 153                 *ret = 0;
 154                 return 0;
 155         }
 156
 157         r = mnt_optstr_get_flags(opts, ret, mnt_get_builtin_optmap(MNT_LINUX_MAP));
 158         if (r != 0) {
 159                 log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path);
 160                 goto fallback;
 161         }
 162
 163         /* MS_RELATIME is default and trying to set it in an unprivileged container causes EPERM */
 164         *ret &= ~MS_RELATIME;
 165         return 0;
 166
 167 fallback:
 168         fd = open(path, O_PATH|O_CLOEXEC|O_NOFOLLOW);
 169         if (fd < 0)
 170                 return -errno;
 171
 172         if (fstatvfs(fd, &buf) < 0)
 173                 return -errno;
 174
 175         /* The statvfs() flags and the mount flags mostly have the same values, but for some cases do
 176          * not. Hence map the flags manually. (Strictly speaking, ST_RELATIME/MS_RELATIME is the most
 177          * prominent one that doesn't match, but that's the one we mask away anyway, see above.) */
 178
 179         *ret =
 180                 FLAGS_SET(buf.f_flag, ST_RDONLY) * MS_RDONLY |
 181                 FLAGS_SET(buf.f_flag, ST_NODEV) * MS_NODEV |
 182                 FLAGS_SET(buf.f_flag, ST_NOEXEC) * MS_NOEXEC |
 183                 FLAGS_SET(buf.f_flag, ST_NOSUID) * MS_NOSUID |
 184                 FLAGS_SET(buf.f_flag, ST_NOATIME) * MS_NOATIME |
 185                 FLAGS_SET(buf.f_flag, ST_NODIRATIME) * MS_NODIRATIME;
 186
 187         return 0;
 188 }
 189
 190 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
 191  * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
 192 int bind_remount_recursive_with_mountinfo(
 193                 const char *prefix,
 194                 unsigned long new_flags,
 195                 unsigned long flags_mask,
 196                 char **deny_list,
 197                 FILE *proc_self_mountinfo) {
 198
 199         _cleanup_set_free_free_ Set *done = NULL;
 200         _cleanup_free_ char *simplified = NULL;
 201         int r;
 202
 203         assert(prefix);
 204         assert(proc_self_mountinfo);
 205
 206         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 207          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 208          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 209          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 210          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 211          * do not have any effect on future submounts that might get propagated, they might be writable. This includes
 212          * future submounts that have been triggered via autofs.
 213          *
 214          * If the "deny_list" parameter is specified it may contain a list of subtrees to exclude from the
 215          * remount operation. Note that we'll ignore the deny list for the top-level path. */
 216
 217         simplified = strdup(prefix);
 218         if (!simplified)
 219                 return -ENOMEM;
 220
 221         path_simplify(simplified, false);
 222
 223         done = set_new(&path_hash_ops);
 224         if (!done)
 225                 return -ENOMEM;
 226
 227         for (;;) {
 228                 _cleanup_set_free_free_ Set *todo = NULL;
 229                 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
 230                 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
 231                 bool top_autofs = false;
 232                 char *x;
 233                 unsigned long orig_flags;
 234
 235                 todo = set_new(&path_hash_ops);
 236                 if (!todo)
 237                         return -ENOMEM;
 238
 239                 rewind(proc_self_mountinfo);
 240
 241                 r = libmount_parse("/proc/self/mountinfo", proc_self_mountinfo, &table, &iter);
 242                 if (r < 0)
 243                         return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
 244
 245                 for (;;) {
 246                         struct libmnt_fs *fs;
 247                         const char *path, *type;
 248
 249                         r = mnt_table_next_fs(table, iter, &fs);
 250                         if (r == 1)
 251                                 break;
 252                         if (r < 0)
 253                                 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
 254
 255                         path = mnt_fs_get_target(fs);
 256                         type = mnt_fs_get_fstype(fs);
 257                         if (!path || !type)
 258                                 continue;
 259
 260                         if (!path_startswith(path, simplified))
 261                                 continue;
 262
 263                         /* Ignore this mount if it is deny-listed, but only if it isn't the top-level mount
 264                          * we shall operate on. */
 265                         if (!path_equal(path, simplified)) {
 266                                 bool deny_listed = false;
 267                                 char **i;
 268
 269                                 STRV_FOREACH(i, deny_list) {
 270                                         if (path_equal(*i, simplified))
 271                                                 continue;
 272
 273                                         if (!path_startswith(*i, simplified))
 274                                                 continue;
 275
 276                                         if (path_startswith(path, *i)) {
 277                                                 deny_listed = true;
 278                                                 log_debug("Not remounting %s deny-listed by %s, called for %s",
 279                                                           path, *i, simplified);
 280                                                 break;
 281                                         }
 282                                 }
 283                                 if (deny_listed)
 284                                         continue;
 285                         }
 286
 287                         /* Let's ignore autofs mounts.  If they aren't
 288                          * triggered yet, we want to avoid triggering
 289                          * them, as we don't make any guarantees for
 290                          * future submounts anyway.  If they are
 291                          * already triggered, then we will find
 292                          * another entry for this. */
 293                         if (streq(type, "autofs")) {
 294                                 top_autofs = top_autofs || path_equal(path, simplified);
 295                                 continue;
 296                         }
 297
 298                         if (!set_contains(done, path)) {
 299                                 r = set_put_strdup(&todo, path);
 300                                 if (r < 0)
 301                                         return r;
 302                         }
 303                 }
 304
 305                 /* If we have no submounts to process anymore and if
 306                  * the root is either already done, or an autofs, we
 307                  * are done */
 308                 if (set_isempty(todo) &&
 309                     (top_autofs || set_contains(done, simplified)))
 310                         return 0;
 311
 312                 if (!set_contains(done, simplified) &&
 313                     !set_contains(todo, simplified)) {
 314                         /* The prefix directory itself is not yet a mount, make it one. */
 315                         r = mount_nofollow(simplified, simplified, NULL, MS_BIND|MS_REC, NULL);
 316                         if (r < 0)
 317                                 return r;
 318
 319                         orig_flags = 0;
 320                         (void) get_mount_flags(table, simplified, &orig_flags);
 321
 322                         r = mount_nofollow(NULL, simplified, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL);
 323                         if (r < 0)
 324                                 return r;
 325
 326                         log_debug("Made top-level directory %s a mount point.", prefix);
 327
 328                         r = set_put_strdup(&done, simplified);
 329                         if (r < 0)
 330                                 return r;
 331                 }
 332
 333                 while ((x = set_steal_first(todo))) {
 334
 335                         r = set_consume(done, x);
 336                         if (IN_SET(r, 0, -EEXIST))
 337                                 continue;
 338                         if (r < 0)
 339                                 return r;
 340
 341                         /* Deal with mount points that are obstructed by a later mount */
 342                         r = path_is_mount_point(x, NULL, 0);
 343                         if (IN_SET(r, 0, -ENOENT))
 344                                 continue;
 345                         if (r < 0) {
 346                                 if (!ERRNO_IS_PRIVILEGE(r))
 347                                         return r;
 348
 349                                 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
 350                                  * may not be acceessed. E.g.,
 351                                  *
 352                                  * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
 353                                  * $ bindfs --no-allow-other ~/mnt ~/mnt
 354                                  *
 355                                  * Then, root user cannot access the mount point ~/mnt/mnt.
 356                                  * In such cases, the submounts are ignored, as we have no way to manage them. */
 357                                 log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
 358                                 continue;
 359                         }
 360
 361                         /* Try to reuse the original flag set */
 362                         orig_flags = 0;
 363                         (void) get_mount_flags(table, x, &orig_flags);
 364
 365                         r = mount_nofollow(NULL, x, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL);
 366                         if (r < 0)
 367                                 return r;
 368
 369                         log_debug("Remounted %s read-only.", x);
 370                 }
 371         }
 372 }
 373
 374 int bind_remount_recursive(
 375                 const char *prefix,
 376                 unsigned long new_flags,
 377                 unsigned long flags_mask,
 378                 char **deny_list) {
 379
 380         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 381         int r;
 382
 383         r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo);
 384         if (r < 0)
 385                 return r;
 386
 387         return bind_remount_recursive_with_mountinfo(prefix, new_flags, flags_mask, deny_list, proc_self_mountinfo);
 388 }
 389
 390 int bind_remount_one_with_mountinfo(
 391                 const char *path,
 392                 unsigned long new_flags,
 393                 unsigned long flags_mask,
 394                 FILE *proc_self_mountinfo) {
 395
 396         _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
 397         unsigned long orig_flags = 0;
 398         int r;
 399
 400         assert(path);
 401         assert(proc_self_mountinfo);
 402
 403         rewind(proc_self_mountinfo);
 404
 405         table = mnt_new_table();
 406         if (!table)
 407                 return -ENOMEM;
 408
 409         r = mnt_table_parse_stream(table, proc_self_mountinfo, "/proc/self/mountinfo");
 410         if (r < 0)
 411                 return r;
 412
 413         /* Try to reuse the original flag set */
 414         (void) get_mount_flags(table, path, &orig_flags);
 415
 416         r = mount_nofollow(NULL, path, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL);
 417         if (r < 0)
 418                 return r;
 419
 420         return 0;
 421 }
 422
 423 int mount_move_root(const char *path) {
 424         assert(path);
 425
 426         if (chdir(path) < 0)
 427                 return -errno;
 428
 429         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 430                 return -errno;
 431
 432         if (chroot(".") < 0)
 433                 return -errno;
 434
 435         if (chdir("/") < 0)
 436                 return -errno;
 437
 438         return 0;
 439 }
 440
 441 int repeat_unmount(const char *path, int flags) {
 442         bool done = false;
 443
 444         assert(path);
 445
 446         /* If there are multiple mounts on a mount point, this
 447          * removes them all */
 448
 449         for (;;) {
 450                 if (umount2(path, flags) < 0) {
 451
 452                         if (errno == EINVAL)
 453                                 return done;
 454
 455                         return -errno;
 456                 }
 457
 458                 done = true;
 459         }
 460 }
 461
 462 int mode_to_inaccessible_node(
 463                 const char *runtime_dir,
 464                 mode_t mode,
 465                 char **ret) {
 466
 467         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created
 468          * during early boot by PID 1. In some cases we lacked the privs to create the character and block
 469          * devices (maybe because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a
 470          * devices policy that excludes device nodes with major and minor of 0), but that's fine, in that
 471          * case we use an AF_UNIX file node instead, which is not the same, but close enough for most
 472          * uses. And most importantly, the kernel allows bind mounts from socket nodes to any non-directory
 473          * file nodes, and that's the most important thing that matters.
 474          *
 475          * Note that the runtime directory argument shall be the top-level runtime directory, i.e. /run/ if
 476          * we operate in system context and $XDG_RUNTIME_DIR if we operate in user context. */
 477
 478         _cleanup_free_ char *d = NULL;
 479         const char *node = NULL;
 480
 481         assert(ret);
 482
 483         if (!runtime_dir)
 484                 runtime_dir = "/run";
 485
 486         switch(mode & S_IFMT) {
 487                 case S_IFREG:
 488                         node = "/systemd/inaccessible/reg";
 489                         break;
 490
 491                 case S_IFDIR:
 492                         node = "/systemd/inaccessible/dir";
 493                         break;
 494
 495                 case S_IFCHR:
 496                         node = "/systemd/inaccessible/chr";
 497                         break;
 498
 499                 case S_IFBLK:
 500                         node = "/systemd/inaccessible/blk";
 501                         break;
 502
 503                 case S_IFIFO:
 504                         node = "/systemd/inaccessible/fifo";
 505                         break;
 506
 507                 case S_IFSOCK:
 508                         node = "/systemd/inaccessible/sock";
 509                         break;
 510         }
 511         if (!node)
 512                 return -EINVAL;
 513
 514         d = path_join(runtime_dir, node);
 515         if (!d)
 516                 return -ENOMEM;
 517
 518         /* On new kernels unprivileged users are permitted to create 0:0 char device nodes (because they also
 519          * act as whiteout inode for overlayfs), but no other char or block device nodes. On old kernels no
 520          * device node whatsoever may be created by unprivileged processes. Hence, if the caller asks for the
 521          * inaccessible block device node let's see if the block device node actually exists, and if not,
 522          * fall back to the character device node. From there fall back to the socket device node. This means
 523          * in the best case we'll get the right device node type — but if not we'll hopefully at least get a
 524          * device node at all. */
 525
 526         if (S_ISBLK(mode) &&
 527             access(d, F_OK) < 0 && errno == ENOENT) {
 528                 free(d);
 529                 d = path_join(runtime_dir, "/systemd/inaccessible/chr");
 530                 if (!d)
 531                         return -ENOMEM;
 532         }
 533
 534         if (IN_SET(mode & S_IFMT, S_IFBLK, S_IFCHR) &&
 535             access(d, F_OK) < 0 && errno == ENOENT) {
 536                 free(d);
 537                 d = path_join(runtime_dir, "/systemd/inaccessible/sock");
 538                 if (!d)
 539                         return -ENOMEM;
 540         }
 541
 542         *ret = TAKE_PTR(d);
 543         return 0;
 544 }
 545
 546 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 547 static char* mount_flags_to_string(long unsigned flags) {
 548         char *x;
 549         _cleanup_free_ char *y = NULL;
 550         long unsigned overflow;
 551
 552         overflow = flags & ~(MS_RDONLY |
 553                              MS_NOSUID |
 554                              MS_NODEV |
 555                              MS_NOEXEC |
 556                              MS_SYNCHRONOUS |
 557                              MS_REMOUNT |
 558                              MS_MANDLOCK |
 559                              MS_DIRSYNC |
 560                              MS_NOATIME |
 561                              MS_NODIRATIME |
 562                              MS_BIND |
 563                              MS_MOVE |
 564                              MS_REC |
 565                              MS_SILENT |
 566                              MS_POSIXACL |
 567                              MS_UNBINDABLE |
 568                              MS_PRIVATE |
 569                              MS_SLAVE |
 570                              MS_SHARED |
 571                              MS_RELATIME |
 572                              MS_KERNMOUNT |
 573                              MS_I_VERSION |
 574                              MS_STRICTATIME |
 575                              MS_LAZYTIME);
 576
 577         if (flags == 0 || overflow != 0)
 578                 if (asprintf(&y, "%lx", overflow) < 0)
 579                         return NULL;
 580
 581         x = strjoin(FLAG(MS_RDONLY),
 582                     FLAG(MS_NOSUID),
 583                     FLAG(MS_NODEV),
 584                     FLAG(MS_NOEXEC),
 585                     FLAG(MS_SYNCHRONOUS),
 586                     FLAG(MS_REMOUNT),
 587                     FLAG(MS_MANDLOCK),
 588                     FLAG(MS_DIRSYNC),
 589                     FLAG(MS_NOATIME),
 590                     FLAG(MS_NODIRATIME),
 591                     FLAG(MS_BIND),
 592                     FLAG(MS_MOVE),
 593                     FLAG(MS_REC),
 594                     FLAG(MS_SILENT),
 595                     FLAG(MS_POSIXACL),
 596                     FLAG(MS_UNBINDABLE),
 597                     FLAG(MS_PRIVATE),
 598                     FLAG(MS_SLAVE),
 599                     FLAG(MS_SHARED),
 600                     FLAG(MS_RELATIME),
 601                     FLAG(MS_KERNMOUNT),
 602                     FLAG(MS_I_VERSION),
 603                     FLAG(MS_STRICTATIME),
 604                     FLAG(MS_LAZYTIME),
 605                     y);
 606         if (!x)
 607                 return NULL;
 608         if (!y)
 609                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 610         return x;
 611 }
 612
 613 int mount_verbose_full(
 614                 int error_log_level,
 615                 const char *what,
 616                 const char *where,
 617                 const char *type,
 618                 unsigned long flags,
 619                 const char *options,
 620                 bool follow_symlink) {
 621
 622         _cleanup_free_ char *fl = NULL, *o = NULL;
 623         unsigned long f;
 624         int r;
 625
 626         r = mount_option_mangle(options, flags, &f, &o);
 627         if (r < 0)
 628                 return log_full_errno(error_log_level, r,
 629                                       "Failed to mangle mount options %s: %m",
 630                                       strempty(options));
 631
 632         fl = mount_flags_to_string(f);
 633
 634         if ((f & MS_REMOUNT) && !what && !type)
 635                 log_debug("Remounting %s (%s \"%s\")...",
 636                           where, strnull(fl), strempty(o));
 637         else if (!what && !type)
 638                 log_debug("Mounting %s (%s \"%s\")...",
 639                           where, strnull(fl), strempty(o));
 640         else if ((f & MS_BIND) && !type)
 641                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 642                           what, where, strnull(fl), strempty(o));
 643         else if (f & MS_MOVE)
 644                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 645                           what, where, strnull(fl), strempty(o));
 646         else
 647                 log_debug("Mounting %s (%s) on %s (%s \"%s\")...",
 648                           strna(what), strna(type), where, strnull(fl), strempty(o));
 649
 650         if (follow_symlink)
 651                 r = mount(what, where, type, f, o) < 0 ? -errno : 0;
 652         else
 653                 r = mount_nofollow(what, where, type, f, o);
 654         if (r < 0)
 655                 return log_full_errno(error_log_level, r,
 656                                       "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
 657                                       strna(what), strna(type), where, strnull(fl), strempty(o));
 658         return 0;
 659 }
 660
 661 int umount_verbose(
 662                 int error_log_level,
 663                 const char *what,
 664                 int flags) {
 665
 666         assert(what);
 667
 668         log_debug("Umounting %s...", what);
 669
 670         if (umount2(what, flags) < 0)
 671                 return log_full_errno(error_log_level, errno,
 672                                       "Failed to unmount %s: %m", what);
 673
 674         return 0;
 675 }
 676
 677 int mount_option_mangle(
 678                 const char *options,
 679                 unsigned long mount_flags,
 680                 unsigned long *ret_mount_flags,
 681                 char **ret_remaining_options) {
 682
 683         const struct libmnt_optmap *map;
 684         _cleanup_free_ char *ret = NULL;
 685         const char *p;
 686         int r;
 687
 688         /* This extracts mount flags from the mount options, and store
 689          * non-mount-flag options to '*ret_remaining_options'.
 690          * E.g.,
 691          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 692          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 693          * "size=1630748k,mode=700,uid=1000,gid=1000".
 694          * See more examples in test-mount-utils.c.
 695          *
 696          * Note that if 'options' does not contain any non-mount-flag options,
 697          * then '*ret_remaining_options' is set to NULL instead of empty string.
 698          * Note that this does not check validity of options stored in
 699          * '*ret_remaining_options'.
 700          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 701          * to '*ret_mount_flags'. */
 702
 703         assert(ret_mount_flags);
 704         assert(ret_remaining_options);
 705
 706         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 707         if (!map)
 708                 return -EINVAL;
 709
 710         p = options;
 711         for (;;) {
 712                 _cleanup_free_ char *word = NULL;
 713                 const struct libmnt_optmap *ent;
 714
 715                 r = extract_first_word(&p, &word, ",", EXTRACT_UNQUOTE);
 716                 if (r < 0)
 717                         return r;
 718                 if (r == 0)
 719                         break;
 720
 721                 for (ent = map; ent->name; ent++) {
 722                         /* All entries in MNT_LINUX_MAP do not take any argument.
 723                          * Thus, ent->name does not contain "=" or "[=]". */
 724                         if (!streq(word, ent->name))
 725                                 continue;
 726
 727                         if (!(ent->mask & MNT_INVERT))
 728                                 mount_flags |= ent->id;
 729                         else if (mount_flags & ent->id)
 730                                 mount_flags ^= ent->id;
 731
 732                         break;
 733                 }
 734
 735                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 736                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 737                         return -ENOMEM;
 738         }
 739
 740         *ret_mount_flags = mount_flags;
 741         *ret_remaining_options = TAKE_PTR(ret);
 742
 743         return 0;
 744 }