src/shared/mount-util.c

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