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         int n = 0, r;
  33         bool again;
  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                         k = cunescape(path, UNESCAPE_RELAX, &p);
  77                         if (k < 0)
  78                                 return k;
  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 < 0 ? 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 you do not have direct access to /proc/self/mountinfo but the caller can open it
 111  * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
 112 int bind_remount_recursive_with_mountinfo(
 113                 const char *prefix,
 114                 unsigned long new_flags,
 115                 unsigned long flags_mask,
 116                 char **blacklist,
 117                 FILE *proc_self_mountinfo) {
 118
 119         _cleanup_set_free_free_ Set *done = NULL;
 120         _cleanup_free_ char *cleaned = NULL;
 121         int r;
 122
 123         assert(proc_self_mountinfo);
 124
 125         /* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
 126          * mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
 127          * operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
 128          * all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
 129          * individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
 130          * do not have any effect on future submounts that might get propagated, they migt be writable. This includes
 131          * future submounts that have been triggered via autofs.
 132          *
 133          * If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
 134          * remount operation. Note that we'll ignore the blacklist for the top-level path. */
 135
 136         cleaned = strdup(prefix);
 137         if (!cleaned)
 138                 return -ENOMEM;
 139
 140         path_simplify(cleaned, false);
 141
 142         done = set_new(&path_hash_ops);
 143         if (!done)
 144                 return -ENOMEM;
 145
 146         for (;;) {
 147                 _cleanup_set_free_free_ Set *todo = NULL;
 148                 bool top_autofs = false;
 149                 char *x;
 150                 unsigned long orig_flags;
 151
 152                 todo = set_new(&path_hash_ops);
 153                 if (!todo)
 154                         return -ENOMEM;
 155
 156                 rewind(proc_self_mountinfo);
 157
 158                 for (;;) {
 159                         _cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
 160                         int k;
 161
 162                         k = fscanf(proc_self_mountinfo,
 163                                    "%*s "       /* (1) mount id */
 164                                    "%*s "       /* (2) parent id */
 165                                    "%*s "       /* (3) major:minor */
 166                                    "%*s "       /* (4) root */
 167                                    "%ms "       /* (5) mount point */
 168                                    "%*s"        /* (6) mount options (superblock) */
 169                                    "%*[^-]"     /* (7) optional fields */
 170                                    "- "         /* (8) separator */
 171                                    "%ms "       /* (9) file system type */
 172                                    "%*s"        /* (10) mount source */
 173                                    "%*s"        /* (11) mount options (bind mount) */
 174                                    "%*[^\n]",   /* some rubbish at the end */
 175                                    &path,
 176                                    &type);
 177                         if (k != 2) {
 178                                 if (k == EOF)
 179                                         break;
 180
 181                                 continue;
 182                         }
 183
 184                         r = cunescape(path, UNESCAPE_RELAX, &p);
 185                         if (r < 0)
 186                                 return r;
 187
 188                         if (!path_startswith(p, cleaned))
 189                                 continue;
 190
 191                         /* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
 192                          * operate on. */
 193                         if (!path_equal(cleaned, p)) {
 194                                 bool blacklisted = false;
 195                                 char **i;
 196
 197                                 STRV_FOREACH(i, blacklist) {
 198
 199                                         if (path_equal(*i, cleaned))
 200                                                 continue;
 201
 202                                         if (!path_startswith(*i, cleaned))
 203                                                 continue;
 204
 205                                         if (path_startswith(p, *i)) {
 206                                                 blacklisted = true;
 207                                                 log_debug("Not remounting %s blacklisted by %s, called for %s", p, *i, cleaned);
 208                                                 break;
 209                                         }
 210                                 }
 211                                 if (blacklisted)
 212                                         continue;
 213                         }
 214
 215                         /* Let's ignore autofs mounts.  If they aren't
 216                          * triggered yet, we want to avoid triggering
 217                          * them, as we don't make any guarantees for
 218                          * future submounts anyway.  If they are
 219                          * already triggered, then we will find
 220                          * another entry for this. */
 221                         if (streq(type, "autofs")) {
 222                                 top_autofs = top_autofs || path_equal(cleaned, p);
 223                                 continue;
 224                         }
 225
 226                         if (!set_contains(done, p)) {
 227                                 r = set_consume(todo, p);
 228                                 p = NULL;
 229                                 if (r == -EEXIST)
 230                                         continue;
 231                                 if (r < 0)
 232                                         return r;
 233                         }
 234                 }
 235
 236                 /* If we have no submounts to process anymore and if
 237                  * the root is either already done, or an autofs, we
 238                  * are done */
 239                 if (set_isempty(todo) &&
 240                     (top_autofs || set_contains(done, cleaned)))
 241                         return 0;
 242
 243                 if (!set_contains(done, cleaned) &&
 244                     !set_contains(todo, cleaned)) {
 245                         /* The prefix directory itself is not yet a mount, make it one. */
 246                         if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
 247                                 return -errno;
 248
 249                         orig_flags = 0;
 250                         (void) get_mount_flags(cleaned, &orig_flags);
 251                         orig_flags &= ~MS_RDONLY;
 252
 253                         if (mount(NULL, cleaned, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 254                                 return -errno;
 255
 256                         log_debug("Made top-level directory %s a mount point.", prefix);
 257
 258                         r = set_put_strdup(done, cleaned);
 259                         if (r < 0)
 260                                 return r;
 261                 }
 262
 263                 while ((x = set_steal_first(todo))) {
 264
 265                         r = set_consume(done, x);
 266                         if (IN_SET(r, 0, -EEXIST))
 267                                 continue;
 268                         if (r < 0)
 269                                 return r;
 270
 271                         /* Deal with mount points that are obstructed by a later mount */
 272                         r = path_is_mount_point(x, NULL, 0);
 273                         if (IN_SET(r, 0, -ENOENT))
 274                                 continue;
 275                         if (IN_SET(r, -EACCES, -EPERM)) {
 276                                 /* Even if root user invoke this, submounts under private FUSE or NFS mount points
 277                                  * may not be acceessed. E.g.,
 278                                  *
 279                                  * $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
 280                                  * $ bindfs --no-allow-other ~/mnt ~/mnt
 281                                  *
 282                                  * Then, root user cannot access the mount point ~/mnt/mnt.
 283                                  * In such cases, the submounts are ignored, as we have no way to manage them. */
 284                                 log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
 285                                 continue;
 286                         }
 287                         if (r < 0)
 288                                 return r;
 289
 290                         /* Try to reuse the original flag set */
 291                         orig_flags = 0;
 292                         (void) get_mount_flags(x, &orig_flags);
 293                         orig_flags &= ~MS_RDONLY;
 294
 295                         if (mount(NULL, x, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
 296                                 return -errno;
 297
 298                         log_debug("Remounted %s read-only.", x);
 299                 }
 300         }
 301 }
 302
 303 int bind_remount_recursive(const char *prefix, unsigned long new_flags, unsigned long flags_mask, char **blacklist) {
 304         _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
 305
 306         proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
 307         if (!proc_self_mountinfo)
 308                 return -errno;
 309
 310         (void) __fsetlocking(proc_self_mountinfo, FSETLOCKING_BYCALLER);
 311
 312         return bind_remount_recursive_with_mountinfo(prefix, new_flags, flags_mask, blacklist, proc_self_mountinfo);
 313 }
 314
 315 int mount_move_root(const char *path) {
 316         assert(path);
 317
 318         if (chdir(path) < 0)
 319                 return -errno;
 320
 321         if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
 322                 return -errno;
 323
 324         if (chroot(".") < 0)
 325                 return -errno;
 326
 327         if (chdir("/") < 0)
 328                 return -errno;
 329
 330         return 0;
 331 }
 332
 333 int repeat_unmount(const char *path, int flags) {
 334         bool done = false;
 335
 336         assert(path);
 337
 338         /* If there are multiple mounts on a mount point, this
 339          * removes them all */
 340
 341         for (;;) {
 342                 if (umount2(path, flags) < 0) {
 343
 344                         if (errno == EINVAL)
 345                                 return done;
 346
 347                         return -errno;
 348                 }
 349
 350                 done = true;
 351         }
 352 }
 353
 354 const char* mode_to_inaccessible_node(mode_t mode) {
 355         /* This function maps a node type to a corresponding inaccessible file node. These nodes are created during
 356          * early boot by PID 1. In some cases we lacked the privs to create the character and block devices (maybe
 357          * because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a devices policy that excludes
 358          * device nodes with major and minor of 0), but that's fine, in that case we use an AF_UNIX file node instead,
 359          * which is not the same, but close enough for most uses. And most importantly, the kernel allows bind mounts
 360          * from socket nodes to any non-directory file nodes, and that's the most important thing that matters. */
 361
 362         switch(mode & S_IFMT) {
 363                 case S_IFREG:
 364                         return "/run/systemd/inaccessible/reg";
 365
 366                 case S_IFDIR:
 367                         return "/run/systemd/inaccessible/dir";
 368
 369                 case S_IFCHR:
 370                         if (access("/run/systemd/inaccessible/chr", F_OK) == 0)
 371                                 return "/run/systemd/inaccessible/chr";
 372                         return "/run/systemd/inaccessible/sock";
 373
 374                 case S_IFBLK:
 375                         if (access("/run/systemd/inaccessible/blk", F_OK) == 0)
 376                                 return "/run/systemd/inaccessible/blk";
 377                         return "/run/systemd/inaccessible/sock";
 378
 379                 case S_IFIFO:
 380                         return "/run/systemd/inaccessible/fifo";
 381
 382                 case S_IFSOCK:
 383                         return "/run/systemd/inaccessible/sock";
 384         }
 385         return NULL;
 386 }
 387
 388 #define FLAG(name) (flags & name ? STRINGIFY(name) "|" : "")
 389 static char* mount_flags_to_string(long unsigned flags) {
 390         char *x;
 391         _cleanup_free_ char *y = NULL;
 392         long unsigned overflow;
 393
 394         overflow = flags & ~(MS_RDONLY |
 395                              MS_NOSUID |
 396                              MS_NODEV |
 397                              MS_NOEXEC |
 398                              MS_SYNCHRONOUS |
 399                              MS_REMOUNT |
 400                              MS_MANDLOCK |
 401                              MS_DIRSYNC |
 402                              MS_NOATIME |
 403                              MS_NODIRATIME |
 404                              MS_BIND |
 405                              MS_MOVE |
 406                              MS_REC |
 407                              MS_SILENT |
 408                              MS_POSIXACL |
 409                              MS_UNBINDABLE |
 410                              MS_PRIVATE |
 411                              MS_SLAVE |
 412                              MS_SHARED |
 413                              MS_RELATIME |
 414                              MS_KERNMOUNT |
 415                              MS_I_VERSION |
 416                              MS_STRICTATIME |
 417                              MS_LAZYTIME);
 418
 419         if (flags == 0 || overflow != 0)
 420                 if (asprintf(&y, "%lx", overflow) < 0)
 421                         return NULL;
 422
 423         x = strjoin(FLAG(MS_RDONLY),
 424                     FLAG(MS_NOSUID),
 425                     FLAG(MS_NODEV),
 426                     FLAG(MS_NOEXEC),
 427                     FLAG(MS_SYNCHRONOUS),
 428                     FLAG(MS_REMOUNT),
 429                     FLAG(MS_MANDLOCK),
 430                     FLAG(MS_DIRSYNC),
 431                     FLAG(MS_NOATIME),
 432                     FLAG(MS_NODIRATIME),
 433                     FLAG(MS_BIND),
 434                     FLAG(MS_MOVE),
 435                     FLAG(MS_REC),
 436                     FLAG(MS_SILENT),
 437                     FLAG(MS_POSIXACL),
 438                     FLAG(MS_UNBINDABLE),
 439                     FLAG(MS_PRIVATE),
 440                     FLAG(MS_SLAVE),
 441                     FLAG(MS_SHARED),
 442                     FLAG(MS_RELATIME),
 443                     FLAG(MS_KERNMOUNT),
 444                     FLAG(MS_I_VERSION),
 445                     FLAG(MS_STRICTATIME),
 446                     FLAG(MS_LAZYTIME),
 447                     y);
 448         if (!x)
 449                 return NULL;
 450         if (!y)
 451                 x[strlen(x) - 1] = '\0'; /* truncate the last | */
 452         return x;
 453 }
 454
 455 int mount_verbose(
 456                 int error_log_level,
 457                 const char *what,
 458                 const char *where,
 459                 const char *type,
 460                 unsigned long flags,
 461                 const char *options) {
 462
 463         _cleanup_free_ char *fl = NULL, *o = NULL;
 464         unsigned long f;
 465         int r;
 466
 467         r = mount_option_mangle(options, flags, &f, &o);
 468         if (r < 0)
 469                 return log_full_errno(error_log_level, r,
 470                                       "Failed to mangle mount options %s: %m",
 471                                       strempty(options));
 472
 473         fl = mount_flags_to_string(f);
 474
 475         if ((f & MS_REMOUNT) && !what && !type)
 476                 log_debug("Remounting %s (%s \"%s\")...",
 477                           where, strnull(fl), strempty(o));
 478         else if (!what && !type)
 479                 log_debug("Mounting %s (%s \"%s\")...",
 480                           where, strnull(fl), strempty(o));
 481         else if ((f & MS_BIND) && !type)
 482                 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
 483                           what, where, strnull(fl), strempty(o));
 484         else if (f & MS_MOVE)
 485                 log_debug("Moving mount %s → %s (%s \"%s\")...",
 486                           what, where, strnull(fl), strempty(o));
 487         else
 488                 log_debug("Mounting %s on %s (%s \"%s\")...",
 489                           strna(type), where, strnull(fl), strempty(o));
 490         if (mount(what, where, type, f, o) < 0)
 491                 return log_full_errno(error_log_level, errno,
 492                                       "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
 493                                       strna(what), strna(type), where, strnull(fl), strempty(o));
 494         return 0;
 495 }
 496
 497 int umount_verbose(const char *what) {
 498         log_debug("Umounting %s...", what);
 499         if (umount(what) < 0)
 500                 return log_error_errno(errno, "Failed to unmount %s: %m", what);
 501         return 0;
 502 }
 503
 504 int mount_option_mangle(
 505                 const char *options,
 506                 unsigned long mount_flags,
 507                 unsigned long *ret_mount_flags,
 508                 char **ret_remaining_options) {
 509
 510         const struct libmnt_optmap *map;
 511         _cleanup_free_ char *ret = NULL;
 512         const char *p;
 513         int r;
 514
 515         /* This extracts mount flags from the mount options, and store
 516          * non-mount-flag options to '*ret_remaining_options'.
 517          * E.g.,
 518          * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000"
 519          * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
 520          * "size=1630748k,mode=700,uid=1000,gid=1000".
 521          * See more examples in test-mount-utils.c.
 522          *
 523          * Note that if 'options' does not contain any non-mount-flag options,
 524          * then '*ret_remaining_options' is set to NULL instread of empty string.
 525          * Note that this does not check validity of options stored in
 526          * '*ret_remaining_options'.
 527          * Note that if 'options' is NULL, then this just copies 'mount_flags'
 528          * to '*ret_mount_flags'. */
 529
 530         assert(ret_mount_flags);
 531         assert(ret_remaining_options);
 532
 533         map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
 534         if (!map)
 535                 return -EINVAL;
 536
 537         p = options;
 538         for (;;) {
 539                 _cleanup_free_ char *word = NULL;
 540                 const struct libmnt_optmap *ent;
 541
 542                 r = extract_first_word(&p, &word, ",", EXTRACT_QUOTES);
 543                 if (r < 0)
 544                         return r;
 545                 if (r == 0)
 546                         break;
 547
 548                 for (ent = map; ent->name; ent++) {
 549                         /* All entries in MNT_LINUX_MAP do not take any argument.
 550                          * Thus, ent->name does not contain "=" or "[=]". */
 551                         if (!streq(word, ent->name))
 552                                 continue;
 553
 554                         if (!(ent->mask & MNT_INVERT))
 555                                 mount_flags |= ent->id;
 556                         else if (mount_flags & ent->id)
 557                                 mount_flags ^= ent->id;
 558
 559                         break;
 560                 }
 561
 562                 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
 563                 if (!ent->name && !strextend_with_separator(&ret, ",", word, NULL))
 564                         return -ENOMEM;
 565         }
 566
 567         *ret_mount_flags = mount_flags;
 568         *ret_remaining_options = TAKE_PTR(ret);
 569
 570         return 0;
 571 }