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