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