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         bool again;
  33         int n = 0, r;
  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                 r = 0;
  44
  45                 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
  46                 if (!proc_self_mountinfo)
  47                         return -errno;
  48
  49                 (void) __fsetlocking(proc_self_mountinfo, FSETLOCKING_BYCALLER);
  50
  51                 for (;;) {
  52                         _cleanup_free_ char *path = NULL, *p = NULL;
  53                         int k;
  54
  55                         k = fscanf(proc_self_mountinfo,
  56                                    "%*s "       /* (1) mount id */
  57                                    "%*s "       /* (2) parent id */
  58                                    "%*s "       /* (3) major:minor */
  59                                    "%*s "       /* (4) root */
  60                                    "%ms "       /* (5) mount point */
  61                                    "%*s"        /* (6) mount options */
  62                                    "%*[^-]"     /* (7) optional fields */
  63                                    "- "         /* (8) separator */
  64                                    "%*s "       /* (9) file system type */
  65                                    "%*s"        /* (10) mount source */
  66                                    "%*s"        /* (11) mount options 2 */
  67                                    "%*[^\n]",   /* some rubbish at the end */
  68                                    &path);
  69                         if (k != 1) {
  70                                 if (k == EOF)
  71                                         break;
  72
  73                                 continue;
  74                         }
  75
  76                         r = cunescape(path, UNESCAPE_RELAX, &p);
  77                         if (r < 0)
  78                                 return r;
  79
  80                         if (!path_startswith(p, prefix))
  81                                 continue;
  82
  83                         if (umount2(p, flags) < 0) {
  84                                 r = log_debug_errno(errno, "Failed to umount %s: %m", p);
  85                                 continue;
  86                         }
  87
  88                         log_debug("Successfully unmounted %s", p);
  89
  90                         again = true;
  91                         n++;
  92
  93                         break;
  94                 }
  95
  96         } while (again);
  97
  98         return r ? r : n;
  99 }
 100
 101 static int get_mount_flags(const char *path, unsigned long *flags) {
 102         struct statvfs buf;
 103
 104         if (statvfs(path, &buf) < 0)
 105                 return -errno;
 106         *flags = buf.f_flag;
 107         return 0;
 108 }
 109
 110 /* Use this function only if do you have direct access to /proc/self/mountinfo
 111  * and need the caller to open it for you. This is the case when /proc is
 112  * masked or not mounted. Otherwise, use bind_remount_recursive. */
 113 int bind_remount_recursive_with_mountinfo(const char *prefix, bool ro, char **blacklist, FILE *proc_self_mountinfo) {
 114         _cleanup_set_free_free_ Set *done = NULL;
 115         _cleanup_free_ char *cleaned = NULL;
 116         int r;
 117
 118         assert(proc_self_mountinfo);
 119
 120         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 121          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 122          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 123          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 124          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 125          * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
 126          * future submounts that have been triggered via autofs.
 127          *
 128          * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
 129          * remount operation. Note that we'll ignore the blacklist for the top-level path. */
 130
 131         cleaned = strdup(prefix);
 132         if (!cleaned)
 133                 return -ENOMEM;
 134
 135         path_simplify(cleaned, false);
 136
 137         done = set_new(&path_hash_ops);
 138         if (!done)
 139                 return -ENOMEM;
 140
 141         for (;;) {
 142                 _cleanup_set_free_free_ Set *todo = NULL;
 143                 bool top_autofs = false;
 144                 char *x;
 145                 unsigned long orig_flags;
 146
 147                 todo = set_new(&path_hash_ops);
 148                 if (!todo)
 149                         return -ENOMEM;
 150
 151                 rewind(proc_self_mountinfo);
 152
 153                 for (;;) {
 154                         _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
 155                         int k;
 156
 157                         k = fscanf(proc_self_mountinfo,
 158                                    "%*s "       /* (1) mount id */
 159                                    "%*s "       /* (2) parent id */
 160                                    "%*s "       /* (3) major:minor */
 161                                    "%*s "       /* (4) root */
 162                                    "%ms "       /* (5) mount point */
 163                                    "%*s"        /* (6) mount options (superblock) */
 164                                    "%*[^-]"     /* (7) optional fields */
 165                                    "- "         /* (8) separator */
 166                                    "%ms "       /* (9) file system type */
 167                                    "%*s"        /* (10) mount source */
 168                                    "%*s"        /* (11) mount options (bind mount) */
 169                                    "%*[^\n]",   /* some rubbish at the end */
 170                                    &path,
 171                                    &type);
 172                         if (k != 2) {
 173                                 if (k == EOF)
 174                                         break;
 175
 176                                 continue;
 177                         }
 178
 179                         r = cunescape(path, UNESCAPE_RELAX, &p);
 180                         if (r < 0)
 181                                 return r;
 182
 183                         if (!path_startswith(p, cleaned))
 184                                 continue;
 185
 186                         /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
 187                          * operate on. */
 188                         if (!path_equal(cleaned, p)) {
 189                                 bool blacklisted = false;
 190                                 char **i;
 191
 192                                 STRV_FOREACH(i, blacklist) {
 193
 194                                         if (path_equal(*i, cleaned))
 195                                                 continue;
 196
 197                                         if (!path_startswith(*i, cleaned))
 198                                                 continue;
 199
 200                                         if (path_startswith(p, *i)) {
 201                                                 blacklisted = true;
 202                                                 log_debug("Not remounting %s blacklisted by %s, called for %s", p, *i, cleaned);
 203                                                 break;
 204                                         }
 205                                 }
 206                                 if (blacklisted)
 207                                         continue;
 208                         }
 209
 210                         /* Let's ignore autofs mounts.  If they aren't
 211                          * triggered yet, we want to avoid triggering
 212                          * them, as we don't make any guarantees for
 213                          * future submounts anyway.  If they are
 214                          * already triggered, then we will find
 215                          * another entry for this. */
 216                         if (streq(type, "autofs")) {
 217                                 top_autofs = top_autofs || path_equal(cleaned, p);
 218                                 continue;
 219                         }
 220
 221                         if (!set_contains(done, p)) {
 222                                 r = set_consume(todo, p);
 223                                 p = NULL;
 224                                 if (r == -EEXIST)
 225                                         continue;
 226                                 if (r < 0)
 227                                         return r;
 228                         }
 229                 }
 230
 231                 /* If we have no submounts to process anymore and if
 232                  * the root is either already done, or an autofs, we
 233                  * are done */
 234                 if (set_isempty(todo) &&
 235                     (top_autofs || set_contains(done, cleaned)))
 236                         return 0;
 237
 238                 if (!set_contains(done, cleaned) &&
 239                     !set_contains(todo, cleaned)) {
 240                         /* The prefix directory itself is not yet a mount, make it one. */
 241                         if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
 242                                 return -errno;
 243
 244                         orig_flags = 0;
 245                         (void) get_mount_flags(cleaned, &orig_flags);
 246                         orig_flags &= ~MS_RDONLY;
 247
 248                         if (mount(NULL, cleaned, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
 249                                 return -errno;
 250
 251                         log_debug("Made top-level directory %s a mount point.", prefix);
 252
 253                         x = strdup(cleaned);
 254                         if (!x)
 255                                 return -ENOMEM;
 256
 257                         r = set_consume(done, x);
 258                         if (r < 0)
 259                                 return r;
 260                 }
 261
 262                 while ((x = set_steal_first(todo))) {
 263
 264                         r = set_consume(done, x);
 265                         if (IN_SET(r, 0, -EEXIST))
 266                                 continue;
 267                         if (r < 0)
 268                                 return r;
 269
 270                         /* Deal with mount points that are obstructed by a later mount */
 271                         r = path_is_mount_point(x, NULL, 0);
 272                         if (IN_SET(r, 0, -ENOENT))
 273                                 continue;
 274                         if (IN_SET(r, -EACCES, -EPERM)) {
 275                                 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
 276                                  * may not be acceessed. E.g.,
 277                                  *
 278                                  * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
 279                                  * $ bindfs --no-allow-other ~/mnt ~/mnt
 280                                  *
 281                                  * Then, root user cannot access the mount point ~/mnt/mnt.
 282                                  * In such cases, the submounts are ignored, as we have no way to manage them. */
 283                                 log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
 284                                 continue;
 285                         }
 286                         if (r < 0)
 287                                 return r;
 288
 289                         /* Try to reuse the original flag set */
 290                         orig_flags = 0;
 291                         (void) get_mount_flags(x, &orig_flags);
 292                         orig_flags &= ~MS_RDONLY;
 293
 294                         if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
 295                                 return -errno;
 296
 297                         log_debug("Remounted %s read-only.", x);
 298                 }
 299         }
 300 }
 301
 302 int bind_remount_recursive(const char *prefix, bool ro, char **blacklist) {
 303         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 304
 305         proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
 306         if (!proc_self_mountinfo)
 307                 return -errno;
 308
 309         (void) __fsetlocking(proc_self_mountinfo, FSETLOCKING_BYCALLER);
 310
 311         return bind_remount_recursive_with_mountinfo(prefix, ro, blacklist, proc_self_mountinfo);
 312 }
 313
 314 int mount_move_root(const char *path) {
 315         assert(path);
 316
 317         if (chdir(path) < 0)
 318                 return -errno;
 319
 320         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 321                 return -errno;
 322
 323         if (chroot(".") < 0)
 324                 return -errno;
 325
 326         if (chdir("/") < 0)
 327                 return -errno;
 328
 329         return 0;
 330 }
 331
 332 int repeat_unmount(const char *path, int flags) {
 333         bool done = false;
 334
 335         assert(path);
 336
 337         /* If there are multiple mounts on a mount point, this
 338          * removes them all */
 339
 340         for (;;) {
 341                 if (umount2(path, flags) < 0) {
 342
 343                         if (errno == EINVAL)
 344                                 return done;
 345
 346                         return -errno;
 347                 }
 348
 349                 done = true;
 350         }
 351 }
 352
 353 const char* mode_to_inaccessible_node(mode_t mode) {
 354         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created during
 355          * early boot by PID 1. In some cases we lacked the privs to create the character and block devices (maybe
 356          * because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a devices policy that excludes
 357          * device nodes with major and minor of 0), but that's fine, in that case we use an AF_UNIX file node instead,
 358          * which is not the same, but close enough for most uses. And most importantly, the kernel allows bind mounts
 359          * from socket nodes to any non-directory file nodes, and that's the most important thing that matters. */
 360
 361         switch(mode & S_IFMT) {
 362                 case S_IFREG:
 363                         return "/run/systemd/inaccessible/reg";
 364
 365                 case S_IFDIR:
 366                         return "/run/systemd/inaccessible/dir";
 367
 368                 case S_IFCHR:
 369                         if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
 370                                 return "/run/systemd/inaccessible/chr";
 371                         return "/run/systemd/inaccessible/sock";
 372
 373                 case S_IFBLK:
 374                         if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
 375                                 return "/run/systemd/inaccessible/blk";
 376                         return "/run/systemd/inaccessible/sock";
 377
 378                 case S_IFIFO:
 379                         return "/run/systemd/inaccessible/fifo";
 380
 381                 case S_IFSOCK:
 382                         return "/run/systemd/inaccessible/sock";
 383         }
 384         return NULL;
 385 }
 386
 387 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 388 static char* mount_flags_to_string(long unsigned flags) {
 389         char *x;
 390         _cleanup_free_ char *y = NULL;
 391         long unsigned overflow;
 392
 393         overflow = flags & ~(MS_RDONLY |
 394                              MS_NOSUID |
 395                              MS_NODEV |
 396                              MS_NOEXEC |
 397                              MS_SYNCHRONOUS |
 398                              MS_REMOUNT |
 399                              MS_MANDLOCK |
 400                              MS_DIRSYNC |
 401                              MS_NOATIME |
 402                              MS_NODIRATIME |
 403                              MS_BIND |
 404                              MS_MOVE |
 405                              MS_REC |
 406                              MS_SILENT |
 407                              MS_POSIXACL |
 408                              MS_UNBINDABLE |
 409                              MS_PRIVATE |
 410                              MS_SLAVE |
 411                              MS_SHARED |
 412                              MS_RELATIME |
 413                              MS_KERNMOUNT |
 414                              MS_I_VERSION |
 415                              MS_STRICTATIME |
 416                              MS_LAZYTIME);
 417
 418         if (flags == 0 || overflow != 0)
 419                 if (asprintf(&y, "%lx", overflow) < 0)
 420                         return NULL;
 421
 422         x = strjoin(FLAG(MS_RDONLY),
 423                     FLAG(MS_NOSUID),
 424                     FLAG(MS_NODEV),
 425                     FLAG(MS_NOEXEC),
 426                     FLAG(MS_SYNCHRONOUS),
 427                     FLAG(MS_REMOUNT),
 428                     FLAG(MS_MANDLOCK),
 429                     FLAG(MS_DIRSYNC),
 430                     FLAG(MS_NOATIME),
 431                     FLAG(MS_NODIRATIME),
 432                     FLAG(MS_BIND),
 433                     FLAG(MS_MOVE),
 434                     FLAG(MS_REC),
 435                     FLAG(MS_SILENT),
 436                     FLAG(MS_POSIXACL),
 437                     FLAG(MS_UNBINDABLE),
 438                     FLAG(MS_PRIVATE),
 439                     FLAG(MS_SLAVE),
 440                     FLAG(MS_SHARED),
 441                     FLAG(MS_RELATIME),
 442                     FLAG(MS_KERNMOUNT),
 443                     FLAG(MS_I_VERSION),
 444                     FLAG(MS_STRICTATIME),
 445                     FLAG(MS_LAZYTIME),
 446                     y);
 447         if (!x)
 448                 return NULL;
 449         if (!y)
 450                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 451         return x;
 452 }
 453
 454 int mount_verbose(
 455                 int error_log_level,
 456                 const char *what,
 457                 const char *where,
 458                 const char *type,
 459                 unsigned long flags,
 460                 const char *options) {
 461
 462         _cleanup_free_ char *fl = NULL, *o = NULL;
 463         unsigned long f;
 464         int r;
 465
 466         r = mount_option_mangle(options, flags, &f, &o);
 467         if (r < 0)
 468                 return log_full_errno(error_log_level, r,
 469                                       "Failed to mangle mount options %s: %m",
 470                                       strempty(options));
 471
 472         fl = mount_flags_to_string(f);
 473
 474         if ((f & MS_REMOUNT) && !what && !type)
 475                 log_debug("Remounting %s (%s \"%s\")...",
 476                           where, strnull(fl), strempty(o));
 477         else if (!what && !type)
 478                 log_debug("Mounting %s (%s \"%s\")...",
 479                           where, strnull(fl), strempty(o));
 480         else if ((f & MS_BIND) && !type)
 481                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 482                           what, where, strnull(fl), strempty(o));
 483         else if (f & MS_MOVE)
 484                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 485                           what, where, strnull(fl), strempty(o));
 486         else
 487                 log_debug("Mounting %s on %s (%s \"%s\")...",
 488                           strna(type), where, strnull(fl), strempty(o));
 489         if (mount(what, where, type, f, o) < 0)
 490                 return log_full_errno(error_log_level, errno,
 491                                       "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
 492                                       strna(what), strna(type), where, strnull(fl), strempty(o));
 493         return 0;
 494 }
 495
 496 int umount_verbose(const char *what) {
 497         log_debug("Umounting %s...", what);
 498         if (umount(what) < 0)
 499                 return log_error_errno(errno, "Failed to unmount %s: %m", what);
 500         return 0;
 501 }
 502
 503 int mount_option_mangle(
 504                 const char *options,
 505                 unsigned long mount_flags,
 506                 unsigned long *ret_mount_flags,
 507                 char **ret_remaining_options) {
 508
 509         const struct libmnt_optmap *map;
 510         _cleanup_free_ char *ret = NULL;
 511         const char *p;
 512         int r;
 513
 514         /* This extracts mount flags from the mount options, and store
 515          * non-mount-flag options to '*ret_remaining_options'.
 516          * E.g.,
 517          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 518          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 519          * "size=1630748k,mode=700,uid=1000,gid=1000".
 520          * See more examples in test-mount-utils.c.
 521          *
 522          * Note that if 'options' does not contain any non-mount-flag options,
 523          * then '*ret_remaining_options' is set to NULL instread of empty string.
 524          * Note that this does not check validity of options stored in
 525          * '*ret_remaining_options'.
 526          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 527          * to '*ret_mount_flags'. */
 528
 529         assert(ret_mount_flags);
 530         assert(ret_remaining_options);
 531
 532         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 533         if (!map)
 534                 return -EINVAL;
 535
 536         p = options;
 537         for (;;) {
 538                 _cleanup_free_ char *word = NULL;
 539                 const struct libmnt_optmap *ent;
 540
 541                 r = extract_first_word(&p, &word, ",", EXTRACT_QUOTES);
 542                 if (r < 0)
 543                         return r;
 544                 if (r == 0)
 545                         break;
 546
 547                 for (ent = map; ent->name; ent++) {
 548                         /* All entries in MNT_LINUX_MAP do not take any argument.
 549                          * Thus, ent->name does not contain "=" or "[=]". */
 550                         if (!streq(word, ent->name))
 551                                 continue;
 552
 553                         if (!(ent->mask & MNT_INVERT))
 554                                 mount_flags |= ent->id;
 555                         else if (mount_flags & ent->id)
 556                                 mount_flags ^= ent->id;
 557
 558                         break;
 559                 }
 560
 561                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 562                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 563                         return -ENOMEM;
 564         }
 565
 566         *ret_mount_flags = mount_flags;
 567         *ret_remaining_options = TAKE_PTR(ret);
 568
 569         return 0;
 570 }