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 later */
  12 #include <libmount.h>
  13
  14 #include "alloc-util.h"
  15 #include "escape.h"
  16 #include "extract-word.h"
  17 #include "fd-util.h"
  18 #include "fileio.h"
  19 #include "fs-util.h"
  20 #include "hashmap.h"
  21 #include "mount-util.h"
  22 #include "mountpoint-util.h"
  23 #include "parse-util.h"
  24 #include "path-util.h"
  25 #include "set.h"
  26 #include "stdio-util.h"
  27 #include "string-util.h"
  28 #include "strv.h"
  29
  30 int umount_recursive(const char *prefix, int flags) {
  31         int n = 0, r;
  32         bool again;
  33
  34         /* Try to umount everything recursively below a
  35          * directory. Also, take care of stacked mounts, and keep
  36          * unmounting them until they are gone. */
  37
  38         do {
  39                 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
  40
  41                 again = false;
  42
  43                 r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo);
  44                 if (r < 0)
  45                         return r;
  46
  47                 for (;;) {
  48                         _cleanup_free_ char *path = NULL, *p = NULL;
  49                         int k;
  50
  51                         k = fscanf(proc_self_mountinfo,
  52                                    "%*s "       /* (1) mount id */
  53                                    "%*s "       /* (2) parent id */
  54                                    "%*s "       /* (3) major:minor */
  55                                    "%*s "       /* (4) root */
  56                                    "%ms "       /* (5) mount point */
  57                                    "%*s"        /* (6) mount options */
  58                                    "%*[^-]"     /* (7) optional fields */
  59                                    "- "         /* (8) separator */
  60                                    "%*s "       /* (9) file system type */
  61                                    "%*s"        /* (10) mount source */
  62                                    "%*s"        /* (11) mount options 2 */
  63                                    "%*[^\n]",   /* some rubbish at the end */
  64                                    &path);
  65                         if (k != 1) {
  66                                 if (k == EOF)
  67                                         break;
  68
  69                                 continue;
  70                         }
  71
  72                         k = cunescape(path, UNESCAPE_RELAX, &p);
  73                         if (k < 0)
  74                                 return k;
  75
  76                         if (!path_startswith(p, prefix))
  77                                 continue;
  78
  79                         if (umount2(p, flags) < 0) {
  80                                 r = log_debug_errno(errno, "Failed to umount %s: %m", p);
  81                                 continue;
  82                         }
  83
  84                         log_debug("Successfully unmounted %s", p);
  85
  86                         again = true;
  87                         n++;
  88
  89                         break;
  90                 }
  91
  92         } while (again);
  93
  94         return r < 0 ? r : n;
  95 }
  96
  97 static int get_mount_flags(const char *path, unsigned long *flags) {
  98         struct statvfs buf;
  99
 100         if (statvfs(path, &buf) < 0)
 101                 return -errno;
 102         *flags = buf.f_flag;
 103         return 0;
 104 }
 105
 106 /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
 107  * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
 108 int bind_remount_recursive_with_mountinfo(
 109                 const char *prefix,
 110                 unsigned long new_flags,
 111                 unsigned long flags_mask,
 112                 char **blacklist,
 113                 FILE *proc_self_mountinfo) {
 114
 115         _cleanup_set_free_free_ Set *done = NULL;
 116         _cleanup_free_ char *cleaned = NULL;
 117         int r;
 118
 119         assert(proc_self_mountinfo);
 120
 121         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 122          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 123          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 124          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 125          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 126          * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
 127          * future submounts that have been triggered via autofs.
 128          *
 129          * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
 130          * remount operation. Note that we'll ignore the blacklist for the top-level path. */
 131
 132         cleaned = strdup(prefix);
 133         if (!cleaned)
 134                 return -ENOMEM;
 135
 136         path_simplify(cleaned, false);
 137
 138         done = set_new(&path_hash_ops);
 139         if (!done)
 140                 return -ENOMEM;
 141
 142         for (;;) {
 143                 _cleanup_set_free_free_ Set *todo = NULL;
 144                 bool top_autofs = false;
 145                 char *x;
 146                 unsigned long orig_flags;
 147
 148                 todo = set_new(&path_hash_ops);
 149                 if (!todo)
 150                         return -ENOMEM;
 151
 152                 rewind(proc_self_mountinfo);
 153
 154                 for (;;) {
 155                         _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
 156                         int k;
 157
 158                         k = fscanf(proc_self_mountinfo,
 159                                    "%*s "       /* (1) mount id */
 160                                    "%*s "       /* (2) parent id */
 161                                    "%*s "       /* (3) major:minor */
 162                                    "%*s "       /* (4) root */
 163                                    "%ms "       /* (5) mount point */
 164                                    "%*s"        /* (6) mount options (superblock) */
 165                                    "%*[^-]"     /* (7) optional fields */
 166                                    "- "         /* (8) separator */
 167                                    "%ms "       /* (9) file system type */
 168                                    "%*s"        /* (10) mount source */
 169                                    "%*s"        /* (11) mount options (bind mount) */
 170                                    "%*[^\n]",   /* some rubbish at the end */
 171                                    &path,
 172                                    &type);
 173                         if (k != 2) {
 174                                 if (k == EOF)
 175                                         break;
 176
 177                                 continue;
 178                         }
 179
 180                         r = cunescape(path, UNESCAPE_RELAX, &p);
 181                         if (r < 0)
 182                                 return r;
 183
 184                         if (!path_startswith(p, cleaned))
 185                                 continue;
 186
 187                         /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
 188                          * operate on. */
 189                         if (!path_equal(cleaned, p)) {
 190                                 bool blacklisted = false;
 191                                 char **i;
 192
 193                                 STRV_FOREACH(i, blacklist) {
 194
 195                                         if (path_equal(*i, cleaned))
 196                                                 continue;
 197
 198                                         if (!path_startswith(*i, cleaned))
 199                                                 continue;
 200
 201                                         if (path_startswith(p, *i)) {
 202                                                 blacklisted = true;
 203                                                 log_debug("Not remounting %s blacklisted by %s, called for %s", p, *i, cleaned);
 204                                                 break;
 205                                         }
 206                                 }
 207                                 if (blacklisted)
 208                                         continue;
 209                         }
 210
 211                         /* Let's ignore autofs mounts.  If they aren't
 212                          * triggered yet, we want to avoid triggering
 213                          * them, as we don't make any guarantees for
 214                          * future submounts anyway.  If they are
 215                          * already triggered, then we will find
 216                          * another entry for this. */
 217                         if (streq(type, "autofs")) {
 218                                 top_autofs = top_autofs || path_equal(cleaned, p);
 219                                 continue;
 220                         }
 221
 222                         if (!set_contains(done, p)) {
 223                                 r = set_consume(todo, p);
 224                                 p = NULL;
 225                                 if (r == -EEXIST)
 226                                         continue;
 227                                 if (r < 0)
 228                                         return r;
 229                         }
 230                 }
 231
 232                 /* If we have no submounts to process anymore and if
 233                  * the root is either already done, or an autofs, we
 234                  * are done */
 235                 if (set_isempty(todo) &&
 236                     (top_autofs || set_contains(done, cleaned)))
 237                         return 0;
 238
 239                 if (!set_contains(done, cleaned) &&
 240                     !set_contains(todo, cleaned)) {
 241                         /* The prefix directory itself is not yet a mount, make it one. */
 242                         if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
 243                                 return -errno;
 244
 245                         orig_flags = 0;
 246                         (void) get_mount_flags(cleaned, &orig_flags);
 247                         orig_flags &= ~MS_RDONLY;
 248
 249                         if (mount(NULL, cleaned, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 250                                 return -errno;
 251
 252                         log_debug("Made top-level directory %s a mount point.", prefix);
 253
 254                         r = set_put_strdup(done, cleaned);
 255                         if (r < 0)
 256                                 return r;
 257                 }
 258
 259                 while ((x = set_steal_first(todo))) {
 260
 261                         r = set_consume(done, x);
 262                         if (IN_SET(r, 0, -EEXIST))
 263                                 continue;
 264                         if (r < 0)
 265                                 return r;
 266
 267                         /* Deal with mount points that are obstructed by a later mount */
 268                         r = path_is_mount_point(x, NULL, 0);
 269                         if (IN_SET(r, 0, -ENOENT))
 270                                 continue;
 271                         if (IN_SET(r, -EACCES, -EPERM)) {
 272                                 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
 273                                  * may not be acceessed. E.g.,
 274                                  *
 275                                  * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
 276                                  * $ bindfs --no-allow-other ~/mnt ~/mnt
 277                                  *
 278                                  * Then, root user cannot access the mount point ~/mnt/mnt.
 279                                  * In such cases, the submounts are ignored, as we have no way to manage them. */
 280                                 log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
 281                                 continue;
 282                         }
 283                         if (r < 0)
 284                                 return r;
 285
 286                         /* Try to reuse the original flag set */
 287                         orig_flags = 0;
 288                         (void) get_mount_flags(x, &orig_flags);
 289                         orig_flags &= ~MS_RDONLY;
 290
 291                         if (mount(NULL, x, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 292                                 return -errno;
 293
 294                         log_debug("Remounted %s read-only.", x);
 295                 }
 296         }
 297 }
 298
 299 int bind_remount_recursive(const char *prefix, unsigned long new_flags, unsigned long flags_mask, char **blacklist) {
 300         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 301         int r;
 302
 303         r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo);
 304         if (r < 0)
 305                 return r;
 306
 307         return bind_remount_recursive_with_mountinfo(prefix, new_flags, flags_mask, blacklist, proc_self_mountinfo);
 308 }
 309
 310 int mount_move_root(const char *path) {
 311         assert(path);
 312
 313         if (chdir(path) < 0)
 314                 return -errno;
 315
 316         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 317                 return -errno;
 318
 319         if (chroot(".") < 0)
 320                 return -errno;
 321
 322         if (chdir("/") < 0)
 323                 return -errno;
 324
 325         return 0;
 326 }
 327
 328 int repeat_unmount(const char *path, int flags) {
 329         bool done = false;
 330
 331         assert(path);
 332
 333         /* If there are multiple mounts on a mount point, this
 334          * removes them all */
 335
 336         for (;;) {
 337                 if (umount2(path, flags) < 0) {
 338
 339                         if (errno == EINVAL)
 340                                 return done;
 341
 342                         return -errno;
 343                 }
 344
 345                 done = true;
 346         }
 347 }
 348
 349 const char* mode_to_inaccessible_node(mode_t mode) {
 350         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created during
 351          * early boot by PID 1. In some cases we lacked the privs to create the character and block devices (maybe
 352          * because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a devices policy that excludes
 353          * device nodes with major and minor of 0), but that's fine, in that case we use an AF_UNIX file node instead,
 354          * which is not the same, but close enough for most uses. And most importantly, the kernel allows bind mounts
 355          * from socket nodes to any non-directory file nodes, and that's the most important thing that matters. */
 356
 357         switch(mode & S_IFMT) {
 358                 case S_IFREG:
 359                         return "/run/systemd/inaccessible/reg";
 360
 361                 case S_IFDIR:
 362                         return "/run/systemd/inaccessible/dir";
 363
 364                 case S_IFCHR:
 365                         if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
 366                                 return "/run/systemd/inaccessible/chr";
 367                         return "/run/systemd/inaccessible/sock";
 368
 369                 case S_IFBLK:
 370                         if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
 371                                 return "/run/systemd/inaccessible/blk";
 372                         return "/run/systemd/inaccessible/sock";
 373
 374                 case S_IFIFO:
 375                         return "/run/systemd/inaccessible/fifo";
 376
 377                 case S_IFSOCK:
 378                         return "/run/systemd/inaccessible/sock";
 379         }
 380         return NULL;
 381 }
 382
 383 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 384 static char* mount_flags_to_string(long unsigned flags) {
 385         char *x;
 386         _cleanup_free_ char *y = NULL;
 387         long unsigned overflow;
 388
 389         overflow = flags & ~(MS_RDONLY |
 390                              MS_NOSUID |
 391                              MS_NODEV |
 392                              MS_NOEXEC |
 393                              MS_SYNCHRONOUS |
 394                              MS_REMOUNT |
 395                              MS_MANDLOCK |
 396                              MS_DIRSYNC |
 397                              MS_NOATIME |
 398                              MS_NODIRATIME |
 399                              MS_BIND |
 400                              MS_MOVE |
 401                              MS_REC |
 402                              MS_SILENT |
 403                              MS_POSIXACL |
 404                              MS_UNBINDABLE |
 405                              MS_PRIVATE |
 406                              MS_SLAVE |
 407                              MS_SHARED |
 408                              MS_RELATIME |
 409                              MS_KERNMOUNT |
 410                              MS_I_VERSION |
 411                              MS_STRICTATIME |
 412                              MS_LAZYTIME);
 413
 414         if (flags == 0 || overflow != 0)
 415                 if (asprintf(&y, "%lx", overflow) < 0)
 416                         return NULL;
 417
 418         x = strjoin(FLAG(MS_RDONLY),
 419                     FLAG(MS_NOSUID),
 420                     FLAG(MS_NODEV),
 421                     FLAG(MS_NOEXEC),
 422                     FLAG(MS_SYNCHRONOUS),
 423                     FLAG(MS_REMOUNT),
 424                     FLAG(MS_MANDLOCK),
 425                     FLAG(MS_DIRSYNC),
 426                     FLAG(MS_NOATIME),
 427                     FLAG(MS_NODIRATIME),
 428                     FLAG(MS_BIND),
 429                     FLAG(MS_MOVE),
 430                     FLAG(MS_REC),
 431                     FLAG(MS_SILENT),
 432                     FLAG(MS_POSIXACL),
 433                     FLAG(MS_UNBINDABLE),
 434                     FLAG(MS_PRIVATE),
 435                     FLAG(MS_SLAVE),
 436                     FLAG(MS_SHARED),
 437                     FLAG(MS_RELATIME),
 438                     FLAG(MS_KERNMOUNT),
 439                     FLAG(MS_I_VERSION),
 440                     FLAG(MS_STRICTATIME),
 441                     FLAG(MS_LAZYTIME),
 442                     y);
 443         if (!x)
 444                 return NULL;
 445         if (!y)
 446                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 447         return x;
 448 }
 449
 450 int mount_verbose(
 451                 int error_log_level,
 452                 const char *what,
 453                 const char *where,
 454                 const char *type,
 455                 unsigned long flags,
 456                 const char *options) {
 457
 458         _cleanup_free_ char *fl = NULL, *o = NULL;
 459         unsigned long f;
 460         int r;
 461
 462         r = mount_option_mangle(options, flags, &f, &o);
 463         if (r < 0)
 464                 return log_full_errno(error_log_level, r,
 465                                       "Failed to mangle mount options %s: %m",
 466                                       strempty(options));
 467
 468         fl = mount_flags_to_string(f);
 469
 470         if ((f & MS_REMOUNT) && !what && !type)
 471                 log_debug("Remounting %s (%s \"%s\")...",
 472                           where, strnull(fl), strempty(o));
 473         else if (!what && !type)
 474                 log_debug("Mounting %s (%s \"%s\")...",
 475                           where, strnull(fl), strempty(o));
 476         else if ((f & MS_BIND) && !type)
 477                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 478                           what, where, strnull(fl), strempty(o));
 479         else if (f & MS_MOVE)
 480                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 481                           what, where, strnull(fl), strempty(o));
 482         else
 483                 log_debug("Mounting %s on %s (%s \"%s\")...",
 484                           strna(type), where, strnull(fl), strempty(o));
 485         if (mount(what, where, type, f, o) < 0)
 486                 return log_full_errno(error_log_level, errno,
 487                                       "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
 488                                       strna(what), strna(type), where, strnull(fl), strempty(o));
 489         return 0;
 490 }
 491
 492 int umount_verbose(const char *what) {
 493         log_debug("Umounting %s...", what);
 494         if (umount(what) < 0)
 495                 return log_error_errno(errno, "Failed to unmount %s: %m", what);
 496         return 0;
 497 }
 498
 499 int mount_option_mangle(
 500                 const char *options,
 501                 unsigned long mount_flags,
 502                 unsigned long *ret_mount_flags,
 503                 char **ret_remaining_options) {
 504
 505         const struct libmnt_optmap *map;
 506         _cleanup_free_ char *ret = NULL;
 507         const char *p;
 508         int r;
 509
 510         /* This extracts mount flags from the mount options, and store
 511          * non-mount-flag options to '*ret_remaining_options'.
 512          * E.g.,
 513          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 514          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 515          * "size=1630748k,mode=700,uid=1000,gid=1000".
 516          * See more examples in test-mount-utils.c.
 517          *
 518          * Note that if 'options' does not contain any non-mount-flag options,
 519          * then '*ret_remaining_options' is set to NULL instread of empty string.
 520          * Note that this does not check validity of options stored in
 521          * '*ret_remaining_options'.
 522          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 523          * to '*ret_mount_flags'. */
 524
 525         assert(ret_mount_flags);
 526         assert(ret_remaining_options);
 527
 528         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 529         if (!map)
 530                 return -EINVAL;
 531
 532         p = options;
 533         for (;;) {
 534                 _cleanup_free_ char *word = NULL;
 535                 const struct libmnt_optmap *ent;
 536
 537                 r = extract_first_word(&p, &word, ",", EXTRACT_QUOTES);
 538                 if (r < 0)
 539                         return r;
 540                 if (r == 0)
 541                         break;
 542
 543                 for (ent = map; ent->name; ent++) {
 544                         /* All entries in MNT_LINUX_MAP do not take any argument.
 545                          * Thus, ent->name does not contain "=" or "[=]". */
 546                         if (!streq(word, ent->name))
 547                                 continue;
 548
 549                         if (!(ent->mask & MNT_INVERT))
 550                                 mount_flags |= ent->id;
 551                         else if (mount_flags & ent->id)
 552                                 mount_flags ^= ent->id;
 553
 554                         break;
 555                 }
 556
 557                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 558                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 559                         return -ENOMEM;
 560         }
 561
 562         *ret_mount_flags = mount_flags;
 563         *ret_remaining_options = TAKE_PTR(ret);
 564
 565         return 0;
 566 }