src/basic/mount-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <stdio_ext.h>
   5 #include <stdlib.h>
   6 #include <string.h>
   7 #include <sys/mount.h>
   8 #include <sys/stat.h>
   9 #include <sys/statvfs.h>
  10 #include <unistd.h>
  11
  12 /* Include later */
  13 #include <libmount.h>
  14
  15 #include "alloc-util.h"
  16 #include "escape.h"
  17 #include "extract-word.h"
  18 #include "fd-util.h"
  19 #include "fileio.h"
  20 #include "fs-util.h"
  21 #include "hashmap.h"
  22 #include "mount-util.h"
  23 #include "parse-util.h"
  24 #include "path-util.h"
  25 #include "set.h"
  26 #include "stdio-util.h"
  27 #include "string-util.h"
  28 #include "strv.h"
  29
  30 /* This is the original MAX_HANDLE_SZ definition from the kernel, when the API was introduced. We use that in place of
  31  * any more currently defined value to future-proof things: if the size is increased in the API headers, and our code
  32  * is recompiled then it would cease working on old kernels, as those refuse any sizes larger than this value with
  33  * EINVAL right-away. Hence, let's disconnect ourselves from any such API changes, and stick to the original definition
  34  * from when it was introduced. We use it as a start value only anyway (see below), and hence should be able to deal
  35  * with large file handles anyway. */
  36 #define ORIGINAL_MAX_HANDLE_SZ 128
  37
  38 int name_to_handle_at_loop(
  39                 int fd,
  40                 const char *path,
  41                 struct file_handle **ret_handle,
  42                 int *ret_mnt_id,
  43                 int flags) {
  44
  45         _cleanup_free_ struct file_handle *h = NULL;
  46         size_t n = ORIGINAL_MAX_HANDLE_SZ;
  47
  48         /* We need to invoke name_to_handle_at() in a loop, given that it might return EOVERFLOW when the specified
  49          * buffer is too small. Note that in contrast to what the docs might suggest, MAX_HANDLE_SZ is only good as a
  50          * start value, it is not an upper bound on the buffer size required.
  51          *
  52          * This improves on raw name_to_handle_at() also in one other regard: ret_handle and ret_mnt_id can be passed
  53          * as NULL if there's no interest in either. */
  54
  55         for (;;) {
  56                 int mnt_id = -1;
  57
  58                 h = malloc0(offsetof(struct file_handle, f_handle) + n);
  59                 if (!h)
  60                         return -ENOMEM;
  61
  62                 h->handle_bytes = n;
  63
  64                 if (name_to_handle_at(fd, path, h, &mnt_id, flags) >= 0) {
  65
  66                         if (ret_handle)
  67                                 *ret_handle = TAKE_PTR(h);
  68
  69                         if (ret_mnt_id)
  70                                 *ret_mnt_id = mnt_id;
  71
  72                         return 0;
  73                 }
  74                 if (errno != EOVERFLOW)
  75                         return -errno;
  76
  77                 if (!ret_handle && ret_mnt_id && mnt_id >= 0) {
  78
  79                         /* As it appears, name_to_handle_at() fills in mnt_id even when it returns EOVERFLOW when the
  80                          * buffer is too small, but that's undocumented. Hence, let's make use of this if it appears to
  81                          * be filled in, and the caller was interested in only the mount ID an nothing else. */
  82
  83                         *ret_mnt_id = mnt_id;
  84                         return 0;
  85                 }
  86
  87                 /* If name_to_handle_at() didn't increase the byte size, then this EOVERFLOW is caused by something
  88                  * else (apparently EOVERFLOW is returned for untriggered nfs4 mounts sometimes), not by the too small
  89                  * buffer. In that case propagate EOVERFLOW */
  90                 if (h->handle_bytes <= n)
  91                         return -EOVERFLOW;
  92
  93                 /* The buffer was too small. Size the new buffer by what name_to_handle_at() returned. */
  94                 n = h->handle_bytes;
  95                 if (offsetof(struct file_handle, f_handle) + n < n) /* check for addition overflow */
  96                         return -EOVERFLOW;
  97
  98                 h = mfree(h);
  99         }
 100 }
 101
 102 static int fd_fdinfo_mnt_id(int fd, const char *filename, int flags, int *mnt_id) {
 103         char path[STRLEN("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)];
 104         _cleanup_free_ char *fdinfo = NULL;
 105         _cleanup_close_ int subfd = -1;
 106         char *p;
 107         int r;
 108
 109         if ((flags & AT_EMPTY_PATH) && isempty(filename))
 110                 xsprintf(path, "/proc/self/fdinfo/%i", fd);
 111         else {
 112                 subfd = openat(fd, filename, O_CLOEXEC|O_PATH);
 113                 if (subfd < 0)
 114                         return -errno;
 115
 116                 xsprintf(path, "/proc/self/fdinfo/%i", subfd);
 117         }
 118
 119         r = read_full_file(path, &fdinfo, NULL);
 120         if (r == -ENOENT) /* The fdinfo directory is a relatively new addition */
 121                 return -EOPNOTSUPP;
 122         if (r < 0)
 123                 return r;
 124
 125         p = startswith(fdinfo, "mnt_id:");
 126         if (!p) {
 127                 p = strstr(fdinfo, "\nmnt_id:");
 128                 if (!p) /* The mnt_id field is a relatively new addition */
 129                         return -EOPNOTSUPP;
 130
 131                 p += 8;
 132         }
 133
 134         p += strspn(p, WHITESPACE);
 135         p[strcspn(p, WHITESPACE)] = 0;
 136
 137         return safe_atoi(p, mnt_id);
 138 }
 139
 140 int fd_is_mount_point(int fd, const char *filename, int flags) {
 141         _cleanup_free_ struct file_handle *h = NULL, *h_parent = NULL;
 142         int mount_id = -1, mount_id_parent = -1;
 143         bool nosupp = false, check_st_dev = true;
 144         struct stat a, b;
 145         int r;
 146
 147         assert(fd >= 0);
 148         assert(filename);
 149
 150         /* First we will try the name_to_handle_at() syscall, which
 151          * tells us the mount id and an opaque file "handle". It is
 152          * not supported everywhere though (kernel compile-time
 153          * option, not all file systems are hooked up). If it works
 154          * the mount id is usually good enough to tell us whether
 155          * something is a mount point.
 156          *
 157          * If that didn't work we will try to read the mount id from
 158          * /proc/self/fdinfo/<fd>. This is almost as good as
 159          * name_to_handle_at(), however, does not return the
 160          * opaque file handle. The opaque file handle is pretty useful
 161          * to detect the root directory, which we should always
 162          * consider a mount point. Hence we use this only as
 163          * fallback. Exporting the mnt_id in fdinfo is a pretty recent
 164          * kernel addition.
 165          *
 166          * As last fallback we do traditional fstat() based st_dev
 167          * comparisons. This is how things were traditionally done,
 168          * but unionfs breaks this since it exposes file
 169          * systems with a variety of st_dev reported. Also, btrfs
 170          * subvolumes have different st_dev, even though they aren't
 171          * real mounts of their own. */
 172
 173         r = name_to_handle_at_loop(fd, filename, &h, &mount_id, flags);
 174         if (IN_SET(r, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL))
 175                 /* This kernel does not support name_to_handle_at() at all (ENOSYS), or the syscall was blocked
 176                  * (EACCES/EPERM; maybe through seccomp, because we are running inside of a container?), or the mount
 177                  * point is not triggered yet (EOVERFLOW, think nfs4), or some general name_to_handle_at() flakiness
 178                  * (EINVAL): fall back to simpler logic. */
 179                 goto fallback_fdinfo;
 180         else if (r == -EOPNOTSUPP)
 181                 /* This kernel or file system does not support name_to_handle_at(), hence let's see if the upper fs
 182                  * supports it (in which case it is a mount point), otherwise fallback to the traditional stat()
 183                  * logic */
 184                 nosupp = true;
 185         else if (r < 0)
 186                 return r;
 187
 188         r = name_to_handle_at_loop(fd, "", &h_parent, &mount_id_parent, AT_EMPTY_PATH);
 189         if (r == -EOPNOTSUPP) {
 190                 if (nosupp)
 191                         /* Neither parent nor child do name_to_handle_at()?  We have no choice but to fall back. */
 192                         goto fallback_fdinfo;
 193                 else
 194                         /* The parent can't do name_to_handle_at() but the directory we are interested in can?  If so,
 195                          * it must be a mount point. */
 196                         return 1;
 197         } else if (r < 0)
 198                 return r;
 199
 200         /* The parent can do name_to_handle_at() but the
 201          * directory we are interested in can't? If so, it
 202          * must be a mount point. */
 203         if (nosupp)
 204                 return 1;
 205
 206         /* If the file handle for the directory we are
 207          * interested in and its parent are identical, we
 208          * assume this is the root directory, which is a mount
 209          * point. */
 210
 211         if (h->handle_bytes == h_parent->handle_bytes &&
 212             h->handle_type == h_parent->handle_type &&
 213             memcmp(h->f_handle, h_parent->f_handle, h->handle_bytes) == 0)
 214                 return 1;
 215
 216         return mount_id != mount_id_parent;
 217
 218 fallback_fdinfo:
 219         r = fd_fdinfo_mnt_id(fd, filename, flags, &mount_id);
 220         if (IN_SET(r, -EOPNOTSUPP, -EACCES, -EPERM))
 221                 goto fallback_fstat;
 222         if (r < 0)
 223                 return r;
 224
 225         r = fd_fdinfo_mnt_id(fd, "", AT_EMPTY_PATH, &mount_id_parent);
 226         if (r < 0)
 227                 return r;
 228
 229         if (mount_id != mount_id_parent)
 230                 return 1;
 231
 232         /* Hmm, so, the mount ids are the same. This leaves one
 233          * special case though for the root file system. For that,
 234          * let's see if the parent directory has the same inode as we
 235          * are interested in. Hence, let's also do fstat() checks now,
 236          * too, but avoid the st_dev comparisons, since they aren't
 237          * that useful on unionfs mounts. */
 238         check_st_dev = false;
 239
 240 fallback_fstat:
 241         /* yay for fstatat() taking a different set of flags than the other
 242          * _at() above */
 243         if (flags & AT_SYMLINK_FOLLOW)
 244                 flags &= ~AT_SYMLINK_FOLLOW;
 245         else
 246                 flags |= AT_SYMLINK_NOFOLLOW;
 247         if (fstatat(fd, filename, &a, flags) < 0)
 248                 return -errno;
 249
 250         if (fstatat(fd, "", &b, AT_EMPTY_PATH) < 0)
 251                 return -errno;
 252
 253         /* A directory with same device and inode as its parent? Must
 254          * be the root directory */
 255         if (a.st_dev == b.st_dev &&
 256             a.st_ino == b.st_ino)
 257                 return 1;
 258
 259         return check_st_dev && (a.st_dev != b.st_dev);
 260 }
 261
 262 /* flags can be AT_SYMLINK_FOLLOW or 0 */
 263 int path_is_mount_point(const char *t, const char *root, int flags) {
 264         _cleanup_free_ char *canonical = NULL, *parent = NULL;
 265         _cleanup_close_ int fd = -1;
 266         int r;
 267
 268         assert(t);
 269         assert((flags & ~AT_SYMLINK_FOLLOW) == 0);
 270
 271         if (path_equal(t, "/"))
 272                 return 1;
 273
 274         /* we need to resolve symlinks manually, we can't just rely on
 275          * fd_is_mount_point() to do that for us; if we have a structure like
 276          * /bin -> /usr/bin/ and /usr is a mount point, then the parent that we
 277          * look at needs to be /usr, not /. */
 278         if (flags & AT_SYMLINK_FOLLOW) {
 279                 r = chase_symlinks(t, root, CHASE_TRAIL_SLASH, &canonical);
 280                 if (r < 0)
 281                         return r;
 282
 283                 t = canonical;
 284         }
 285
 286         parent = dirname_malloc(t);
 287         if (!parent)
 288                 return -ENOMEM;
 289
 290         fd = openat(AT_FDCWD, parent, O_DIRECTORY|O_CLOEXEC|O_PATH);
 291         if (fd < 0)
 292                 return -errno;
 293
 294         return fd_is_mount_point(fd, last_path_component(t), flags);
 295 }
 296
 297 int path_get_mnt_id(const char *path, int *ret) {
 298         int r;
 299
 300         r = name_to_handle_at_loop(AT_FDCWD, path, NULL, ret, 0);
 301         if (IN_SET(r, -EOPNOTSUPP, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL)) /* kernel/fs don't support this, or seccomp blocks access, or untriggered mount, or name_to_handle_at() is flaky */
 302                 return fd_fdinfo_mnt_id(AT_FDCWD, path, 0, ret);
 303
 304         return r;
 305 }
 306
 307 int umount_recursive(const char *prefix, int flags) {
 308         bool again;
 309         int n = 0, r;
 310
 311         /* Try to umount everything recursively below a
 312          * directory. Also, take care of stacked mounts, and keep
 313          * unmounting them until they are gone. */
 314
 315         do {
 316                 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 317
 318                 again = false;
 319                 r = 0;
 320
 321                 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
 322                 if (!proc_self_mountinfo)
 323                         return -errno;
 324
 325                 (void) __fsetlocking(proc_self_mountinfo, FSETLOCKING_BYCALLER);
 326
 327                 for (;;) {
 328                         _cleanup_free_ char *path = NULL, *p = NULL;
 329                         int k;
 330
 331                         k = fscanf(proc_self_mountinfo,
 332                                    "%*s "       /* (1) mount id */
 333                                    "%*s "       /* (2) parent id */
 334                                    "%*s "       /* (3) major:minor */
 335                                    "%*s "       /* (4) root */
 336                                    "%ms "       /* (5) mount point */
 337                                    "%*s"        /* (6) mount options */
 338                                    "%*[^-]"     /* (7) optional fields */
 339                                    "- "         /* (8) separator */
 340                                    "%*s "       /* (9) file system type */
 341                                    "%*s"        /* (10) mount source */
 342                                    "%*s"        /* (11) mount options 2 */
 343                                    "%*[^\n]",   /* some rubbish at the end */
 344                                    &path);
 345                         if (k != 1) {
 346                                 if (k == EOF)
 347                                         break;
 348
 349                                 continue;
 350                         }
 351
 352                         r = cunescape(path, UNESCAPE_RELAX, &p);
 353                         if (r < 0)
 354                                 return r;
 355
 356                         if (!path_startswith(p, prefix))
 357                                 continue;
 358
 359                         if (umount2(p, flags) < 0) {
 360                                 r = log_debug_errno(errno, "Failed to umount %s: %m", p);
 361                                 continue;
 362                         }
 363
 364                         log_debug("Successfully unmounted %s", p);
 365
 366                         again = true;
 367                         n++;
 368
 369                         break;
 370                 }
 371
 372         } while (again);
 373
 374         return r ? r : n;
 375 }
 376
 377 static int get_mount_flags(const char *path, unsigned long *flags) {
 378         struct statvfs buf;
 379
 380         if (statvfs(path, &buf) < 0)
 381                 return -errno;
 382         *flags = buf.f_flag;
 383         return 0;
 384 }
 385
 386 /* Use this function only if do you have direct access to /proc/self/mountinfo
 387  * and need the caller to open it for you. This is the case when /proc is
 388  * masked or not mounted. Otherwise, use bind_remount_recursive. */
 389 int bind_remount_recursive_with_mountinfo(const char *prefix, bool ro, char **blacklist, FILE *proc_self_mountinfo) {
 390         _cleanup_set_free_free_ Set *done = NULL;
 391         _cleanup_free_ char *cleaned = NULL;
 392         int r;
 393
 394         assert(proc_self_mountinfo);
 395
 396         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 397          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 398          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 399          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 400          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 401          * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
 402          * future submounts that have been triggered via autofs.
 403          *
 404          * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
 405          * remount operation. Note that we'll ignore the blacklist for the top-level path. */
 406
 407         cleaned = strdup(prefix);
 408         if (!cleaned)
 409                 return -ENOMEM;
 410
 411         path_simplify(cleaned, false);
 412
 413         done = set_new(&path_hash_ops);
 414         if (!done)
 415                 return -ENOMEM;
 416
 417         for (;;) {
 418                 _cleanup_set_free_free_ Set *todo = NULL;
 419                 bool top_autofs = false;
 420                 char *x;
 421                 unsigned long orig_flags;
 422
 423                 todo = set_new(&path_hash_ops);
 424                 if (!todo)
 425                         return -ENOMEM;
 426
 427                 rewind(proc_self_mountinfo);
 428
 429                 for (;;) {
 430                         _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
 431                         int k;
 432
 433                         k = fscanf(proc_self_mountinfo,
 434                                    "%*s "       /* (1) mount id */
 435                                    "%*s "       /* (2) parent id */
 436                                    "%*s "       /* (3) major:minor */
 437                                    "%*s "       /* (4) root */
 438                                    "%ms "       /* (5) mount point */
 439                                    "%*s"        /* (6) mount options (superblock) */
 440                                    "%*[^-]"     /* (7) optional fields */
 441                                    "- "         /* (8) separator */
 442                                    "%ms "       /* (9) file system type */
 443                                    "%*s"        /* (10) mount source */
 444                                    "%*s"        /* (11) mount options (bind mount) */
 445                                    "%*[^\n]",   /* some rubbish at the end */
 446                                    &path,
 447                                    &type);
 448                         if (k != 2) {
 449                                 if (k == EOF)
 450                                         break;
 451
 452                                 continue;
 453                         }
 454
 455                         r = cunescape(path, UNESCAPE_RELAX, &p);
 456                         if (r < 0)
 457                                 return r;
 458
 459                         if (!path_startswith(p, cleaned))
 460                                 continue;
 461
 462                         /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
 463                          * operate on. */
 464                         if (!path_equal(cleaned, p)) {
 465                                 bool blacklisted = false;
 466                                 char **i;
 467
 468                                 STRV_FOREACH(i, blacklist) {
 469
 470                                         if (path_equal(*i, cleaned))
 471                                                 continue;
 472
 473                                         if (!path_startswith(*i, cleaned))
 474                                                 continue;
 475
 476                                         if (path_startswith(p, *i)) {
 477                                                 blacklisted = true;
 478                                                 log_debug("Not remounting %s, because blacklisted by %s, called for %s", p, *i, cleaned);
 479                                                 break;
 480                                         }
 481                                 }
 482                                 if (blacklisted)
 483                                         continue;
 484                         }
 485
 486                         /* Let's ignore autofs mounts.  If they aren't
 487                          * triggered yet, we want to avoid triggering
 488                          * them, as we don't make any guarantees for
 489                          * future submounts anyway.  If they are
 490                          * already triggered, then we will find
 491                          * another entry for this. */
 492                         if (streq(type, "autofs")) {
 493                                 top_autofs = top_autofs || path_equal(cleaned, p);
 494                                 continue;
 495                         }
 496
 497                         if (!set_contains(done, p)) {
 498                                 r = set_consume(todo, p);
 499                                 p = NULL;
 500                                 if (r == -EEXIST)
 501                                         continue;
 502                                 if (r < 0)
 503                                         return r;
 504                         }
 505                 }
 506
 507                 /* If we have no submounts to process anymore and if
 508                  * the root is either already done, or an autofs, we
 509                  * are done */
 510                 if (set_isempty(todo) &&
 511                     (top_autofs || set_contains(done, cleaned)))
 512                         return 0;
 513
 514                 if (!set_contains(done, cleaned) &&
 515                     !set_contains(todo, cleaned)) {
 516                         /* The prefix directory itself is not yet a mount, make it one. */
 517                         if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
 518                                 return -errno;
 519
 520                         orig_flags = 0;
 521                         (void) get_mount_flags(cleaned, &orig_flags);
 522                         orig_flags &= ~MS_RDONLY;
 523
 524                         if (mount(NULL, prefix, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
 525                                 return -errno;
 526
 527                         log_debug("Made top-level directory %s a mount point.", prefix);
 528
 529                         x = strdup(cleaned);
 530                         if (!x)
 531                                 return -ENOMEM;
 532
 533                         r = set_consume(done, x);
 534                         if (r < 0)
 535                                 return r;
 536                 }
 537
 538                 while ((x = set_steal_first(todo))) {
 539
 540                         r = set_consume(done, x);
 541                         if (IN_SET(r, 0, -EEXIST))
 542                                 continue;
 543                         if (r < 0)
 544                                 return r;
 545
 546                         /* Deal with mount points that are obstructed by a later mount */
 547                         r = path_is_mount_point(x, NULL, 0);
 548                         if (IN_SET(r, 0, -ENOENT))
 549                                 continue;
 550                         if (r < 0)
 551                                 return r;
 552
 553                         /* Try to reuse the original flag set */
 554                         orig_flags = 0;
 555                         (void) get_mount_flags(x, &orig_flags);
 556                         orig_flags &= ~MS_RDONLY;
 557
 558                         if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
 559                                 return -errno;
 560
 561                         log_debug("Remounted %s read-only.", x);
 562                 }
 563         }
 564 }
 565
 566 int bind_remount_recursive(const char *prefix, bool ro, char **blacklist) {
 567         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 568
 569         proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
 570         if (!proc_self_mountinfo)
 571                 return -errno;
 572
 573         (void) __fsetlocking(proc_self_mountinfo, FSETLOCKING_BYCALLER);
 574
 575         return bind_remount_recursive_with_mountinfo(prefix, ro, blacklist, proc_self_mountinfo);
 576 }
 577
 578 int mount_move_root(const char *path) {
 579         assert(path);
 580
 581         if (chdir(path) < 0)
 582                 return -errno;
 583
 584         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 585                 return -errno;
 586
 587         if (chroot(".") < 0)
 588                 return -errno;
 589
 590         if (chdir("/") < 0)
 591                 return -errno;
 592
 593         return 0;
 594 }
 595
 596 bool fstype_is_network(const char *fstype) {
 597         const char *x;
 598
 599         x = startswith(fstype, "fuse.");
 600         if (x)
 601                 fstype = x;
 602
 603         return STR_IN_SET(fstype,
 604                           "afs",
 605                           "cifs",
 606                           "smbfs",
 607                           "sshfs",
 608                           "ncpfs",
 609                           "ncp",
 610                           "nfs",
 611                           "nfs4",
 612                           "gfs",
 613                           "gfs2",
 614                           "glusterfs",
 615                           "pvfs2", /* OrangeFS */
 616                           "ocfs2",
 617                           "lustre");
 618 }
 619
 620 bool fstype_is_api_vfs(const char *fstype) {
 621         return STR_IN_SET(fstype,
 622                           "autofs",
 623                           "bpf",
 624                           "cgroup",
 625                           "cgroup2",
 626                           "configfs",
 627                           "cpuset",
 628                           "debugfs",
 629                           "devpts",
 630                           "devtmpfs",
 631                           "efivarfs",
 632                           "fusectl",
 633                           "hugetlbfs",
 634                           "mqueue",
 635                           "proc",
 636                           "pstore",
 637                           "ramfs",
 638                           "securityfs",
 639                           "sysfs",
 640                           "tmpfs",
 641                           "tracefs");
 642 }
 643
 644 bool fstype_is_ro(const char *fstype) {
 645         /* All Linux file systems that are necessarily read-only */
 646         return STR_IN_SET(fstype,
 647                           "DM_verity_hash",
 648                           "iso9660",
 649                           "squashfs");
 650 }
 651
 652 bool fstype_can_discard(const char *fstype) {
 653         return STR_IN_SET(fstype,
 654                           "btrfs",
 655                           "ext4",
 656                           "vfat",
 657                           "xfs");
 658 }
 659
 660 bool fstype_can_uid_gid(const char *fstype) {
 661
 662         /* All file systems that have a uid=/gid= mount option that fixates the owners of all files and directories,
 663          * current and future. */
 664
 665         return STR_IN_SET(fstype,
 666                           "adfs",
 667                           "fat",
 668                           "hfs",
 669                           "hpfs",
 670                           "iso9660",
 671                           "msdos",
 672                           "ntfs",
 673                           "vfat");
 674 }
 675
 676 int repeat_unmount(const char *path, int flags) {
 677         bool done = false;
 678
 679         assert(path);
 680
 681         /* If there are multiple mounts on a mount point, this
 682          * removes them all */
 683
 684         for (;;) {
 685                 if (umount2(path, flags) < 0) {
 686
 687                         if (errno == EINVAL)
 688                                 return done;
 689
 690                         return -errno;
 691                 }
 692
 693                 done = true;
 694         }
 695 }
 696
 697 const char* mode_to_inaccessible_node(mode_t mode) {
 698         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created during
 699          * early boot by PID 1. In some cases we lacked the privs to create the character and block devices (maybe
 700          * because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a devices policy that excludes
 701          * device nodes with major and minor of 0), but that's fine, in that case we use an AF_UNIX file node instead,
 702          * which is not the same, but close enough for most uses. And most importantly, the kernel allows bind mounts
 703          * from socket nodes to any non-directory file nodes, and that's the most important thing that matters. */
 704
 705         switch(mode & S_IFMT) {
 706                 case S_IFREG:
 707                         return "/run/systemd/inaccessible/reg";
 708
 709                 case S_IFDIR:
 710                         return "/run/systemd/inaccessible/dir";
 711
 712                 case S_IFCHR:
 713                         if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
 714                                 return "/run/systemd/inaccessible/chr";
 715                         return "/run/systemd/inaccessible/sock";
 716
 717                 case S_IFBLK:
 718                         if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
 719                                 return "/run/systemd/inaccessible/blk";
 720                         return "/run/systemd/inaccessible/sock";
 721
 722                 case S_IFIFO:
 723                         return "/run/systemd/inaccessible/fifo";
 724
 725                 case S_IFSOCK:
 726                         return "/run/systemd/inaccessible/sock";
 727         }
 728         return NULL;
 729 }
 730
 731 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 732 static char* mount_flags_to_string(long unsigned flags) {
 733         char *x;
 734         _cleanup_free_ char *y = NULL;
 735         long unsigned overflow;
 736
 737         overflow = flags & ~(MS_RDONLY |
 738                              MS_NOSUID |
 739                              MS_NODEV |
 740                              MS_NOEXEC |
 741                              MS_SYNCHRONOUS |
 742                              MS_REMOUNT |
 743                              MS_MANDLOCK |
 744                              MS_DIRSYNC |
 745                              MS_NOATIME |
 746                              MS_NODIRATIME |
 747                              MS_BIND |
 748                              MS_MOVE |
 749                              MS_REC |
 750                              MS_SILENT |
 751                              MS_POSIXACL |
 752                              MS_UNBINDABLE |
 753                              MS_PRIVATE |
 754                              MS_SLAVE |
 755                              MS_SHARED |
 756                              MS_RELATIME |
 757                              MS_KERNMOUNT |
 758                              MS_I_VERSION |
 759                              MS_STRICTATIME |
 760                              MS_LAZYTIME);
 761
 762         if (flags == 0 || overflow != 0)
 763                 if (asprintf(&y, "%lx", overflow) < 0)
 764                         return NULL;
 765
 766         x = strjoin(FLAG(MS_RDONLY),
 767                     FLAG(MS_NOSUID),
 768                     FLAG(MS_NODEV),
 769                     FLAG(MS_NOEXEC),
 770                     FLAG(MS_SYNCHRONOUS),
 771                     FLAG(MS_REMOUNT),
 772                     FLAG(MS_MANDLOCK),
 773                     FLAG(MS_DIRSYNC),
 774                     FLAG(MS_NOATIME),
 775                     FLAG(MS_NODIRATIME),
 776                     FLAG(MS_BIND),
 777                     FLAG(MS_MOVE),
 778                     FLAG(MS_REC),
 779                     FLAG(MS_SILENT),
 780                     FLAG(MS_POSIXACL),
 781                     FLAG(MS_UNBINDABLE),
 782                     FLAG(MS_PRIVATE),
 783                     FLAG(MS_SLAVE),
 784                     FLAG(MS_SHARED),
 785                     FLAG(MS_RELATIME),
 786                     FLAG(MS_KERNMOUNT),
 787                     FLAG(MS_I_VERSION),
 788                     FLAG(MS_STRICTATIME),
 789                     FLAG(MS_LAZYTIME),
 790                     y);
 791         if (!x)
 792                 return NULL;
 793         if (!y)
 794                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 795         return x;
 796 }
 797
 798 int mount_verbose(
 799                 int error_log_level,
 800                 const char *what,
 801                 const char *where,
 802                 const char *type,
 803                 unsigned long flags,
 804                 const char *options) {
 805
 806         _cleanup_free_ char *fl = NULL, *o = NULL;
 807         unsigned long f;
 808         int r;
 809
 810         r = mount_option_mangle(options, flags, &f, &o);
 811         if (r < 0)
 812                 return log_full_errno(error_log_level, r,
 813                                       "Failed to mangle mount options %s: %m",
 814                                       strempty(options));
 815
 816         fl = mount_flags_to_string(f);
 817
 818         if ((f & MS_REMOUNT) && !what && !type)
 819                 log_debug("Remounting %s (%s \"%s\")...",
 820                           where, strnull(fl), strempty(o));
 821         else if (!what && !type)
 822                 log_debug("Mounting %s (%s \"%s\")...",
 823                           where, strnull(fl), strempty(o));
 824         else if ((f & MS_BIND) && !type)
 825                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 826                           what, where, strnull(fl), strempty(o));
 827         else if (f & MS_MOVE)
 828                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 829                           what, where, strnull(fl), strempty(o));
 830         else
 831                 log_debug("Mounting %s on %s (%s \"%s\")...",
 832                           strna(type), where, strnull(fl), strempty(o));
 833         if (mount(what, where, type, f, o) < 0)
 834                 return log_full_errno(error_log_level, errno,
 835                                       "Failed to mount %s on %s (%s \"%s\"): %m",
 836                                       strna(type), where, strnull(fl), strempty(o));
 837         return 0;
 838 }
 839
 840 int umount_verbose(const char *what) {
 841         log_debug("Umounting %s...", what);
 842         if (umount(what) < 0)
 843                 return log_error_errno(errno, "Failed to unmount %s: %m", what);
 844         return 0;
 845 }
 846
 847 const char *mount_propagation_flags_to_string(unsigned long flags) {
 848
 849         switch (flags & (MS_SHARED|MS_SLAVE|MS_PRIVATE)) {
 850         case 0:
 851                 return "";
 852         case MS_SHARED:
 853                 return "shared";
 854         case MS_SLAVE:
 855                 return "slave";
 856         case MS_PRIVATE:
 857                 return "private";
 858         }
 859
 860         return NULL;
 861 }
 862
 863 int mount_propagation_flags_from_string(const char *name, unsigned long *ret) {
 864
 865         if (isempty(name))
 866                 *ret = 0;
 867         else if (streq(name, "shared"))
 868                 *ret = MS_SHARED;
 869         else if (streq(name, "slave"))
 870                 *ret = MS_SLAVE;
 871         else if (streq(name, "private"))
 872                 *ret = MS_PRIVATE;
 873         else
 874                 return -EINVAL;
 875         return 0;
 876 }
 877
 878 int mount_option_mangle(
 879                 const char *options,
 880                 unsigned long mount_flags,
 881                 unsigned long *ret_mount_flags,
 882                 char **ret_remaining_options) {
 883
 884         const struct libmnt_optmap *map;
 885         _cleanup_free_ char *ret = NULL;
 886         const char *p;
 887         int r;
 888
 889         /* This extracts mount flags from the mount options, and store
 890          * non-mount-flag options to '*ret_remaining_options'.
 891          * E.g.,
 892          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 893          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 894          * "size=1630748k,mode=700,uid=1000,gid=1000".
 895          * See more examples in test-mount-utils.c.
 896          *
 897          * Note that if 'options' does not contain any non-mount-flag options,
 898          * then '*ret_remaining_options' is set to NULL instread of empty string.
 899          * Note that this does not check validity of options stored in
 900          * '*ret_remaining_options'.
 901          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 902          * to '*ret_mount_flags'. */
 903
 904         assert(ret_mount_flags);
 905         assert(ret_remaining_options);
 906
 907         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 908         if (!map)
 909                 return -EINVAL;
 910
 911         p = options;
 912         for (;;) {
 913                 _cleanup_free_ char *word = NULL;
 914                 const struct libmnt_optmap *ent;
 915
 916                 r = extract_first_word(&p, &word, ",", EXTRACT_QUOTES);
 917                 if (r < 0)
 918                         return r;
 919                 if (r == 0)
 920                         break;
 921
 922                 for (ent = map; ent->name; ent++) {
 923                         /* All entries in MNT_LINUX_MAP do not take any argument.
 924                          * Thus, ent->name does not contain "=" or "[=]". */
 925                         if (!streq(word, ent->name))
 926                                 continue;
 927
 928                         if (!(ent->mask & MNT_INVERT))
 929                                 mount_flags |= ent->id;
 930                         else if (mount_flags & ent->id)
 931                                 mount_flags ^= ent->id;
 932
 933                         break;
 934                 }
 935
 936                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 937                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 938                         return -ENOMEM;
 939         }
 940
 941         *ret_mount_flags = mount_flags;
 942         *ret_remaining_options = TAKE_PTR(ret);
 943
 944         return 0;
 945 }