src/shared/mount-util.c

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