fs/pnode.c

   1 /*
   2  *  linux/fs/pnode.c
   3  *
   4  * (C) Copyright IBM Corporation 2005.
   5  *      Released under GPL v2.
   6  *      Author : Ram Pai (linuxram@us.ibm.com)
   7  *
   8  */
   9 #include <linux/mnt_namespace.h>
  10 #include <linux/mount.h>
  11 #include <linux/fs.h>
  12 #include <linux/nsproxy.h>
  13 #include <uapi/linux/mount.h>
  14 #include "internal.h"
  15 #include "pnode.h"
  16
  17 /* return the next shared peer mount of @p */
  18 static inline struct mount *next_peer(struct mount *p)
  19 {
  20         return list_entry(p->mnt_share.next, struct mount, mnt_share);
  21 }
  22
  23 static inline struct mount *first_slave(struct mount *p)
  24 {
  25         return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
  26 }
  27
  28 static inline struct mount *last_slave(struct mount *p)
  29 {
  30         return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
  31 }
  32
  33 static inline struct mount *next_slave(struct mount *p)
  34 {
  35         return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
  36 }
  37
  38 static struct mount *get_peer_under_root(struct mount *mnt,
  39                                          struct mnt_namespace *ns,
  40                                          const struct path *root)
  41 {
  42         struct mount *m = mnt;
  43
  44         do {
  45                 /* Check the namespace first for optimization */
  46                 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
  47                         return m;
  48
  49                 m = next_peer(m);
  50         } while (m != mnt);
  51
  52         return NULL;
  53 }
  54
  55 /*
  56  * Get ID of closest dominating peer group having a representative
  57  * under the given root.
  58  *
  59  * Caller must hold namespace_sem
  60  */
  61 int get_dominating_id(struct mount *mnt, const struct path *root)
  62 {
  63         struct mount *m;
  64
  65         for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
  66                 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
  67                 if (d)
  68                         return d->mnt_group_id;
  69         }
  70
  71         return 0;
  72 }
  73
  74 static int do_make_slave(struct mount *mnt)
  75 {
  76         struct mount *master, *slave_mnt;
  77
  78         if (list_empty(&mnt->mnt_share)) {
  79                 if (IS_MNT_SHARED(mnt)) {
  80                         mnt_release_group_id(mnt);
  81                         CLEAR_MNT_SHARED(mnt);
  82                 }
  83                 master = mnt->mnt_master;
  84                 if (!master) {
  85                         struct list_head *p = &mnt->mnt_slave_list;
  86                         while (!list_empty(p)) {
  87                                 slave_mnt = list_first_entry(p,
  88                                                 struct mount, mnt_slave);
  89                                 list_del_init(&slave_mnt->mnt_slave);
  90                                 slave_mnt->mnt_master = NULL;
  91                         }
  92                         return 0;
  93                 }
  94         } else {
  95                 struct mount *m;
  96                 /*
  97                  * slave 'mnt' to a peer mount that has the
  98                  * same root dentry. If none is available then
  99                  * slave it to anything that is available.
 100                  */
 101                 for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
 102                         if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
 103                                 master = m;
 104                                 break;
 105                         }
 106                 }
 107                 list_del_init(&mnt->mnt_share);
 108                 mnt->mnt_group_id = 0;
 109                 CLEAR_MNT_SHARED(mnt);
 110         }
 111         list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
 112                 slave_mnt->mnt_master = master;
 113         list_move(&mnt->mnt_slave, &master->mnt_slave_list);
 114         list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
 115         INIT_LIST_HEAD(&mnt->mnt_slave_list);
 116         mnt->mnt_master = master;
 117         return 0;
 118 }
 119
 120 /*
 121  * vfsmount lock must be held for write
 122  */
 123 void change_mnt_propagation(struct mount *mnt, int type)
 124 {
 125         if (type == MS_SHARED) {
 126                 set_mnt_shared(mnt);
 127                 return;
 128         }
 129         do_make_slave(mnt);
 130         if (type != MS_SLAVE) {
 131                 list_del_init(&mnt->mnt_slave);
 132                 mnt->mnt_master = NULL;
 133                 if (type == MS_UNBINDABLE)
 134                         mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
 135                 else
 136                         mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
 137         }
 138 }
 139
 140 /*
 141  * get the next mount in the propagation tree.
 142  * @m: the mount seen last
 143  * @origin: the original mount from where the tree walk initiated
 144  *
 145  * Note that peer groups form contiguous segments of slave lists.
 146  * We rely on that in get_source() to be able to find out if
 147  * vfsmount found while iterating with propagation_next() is
 148  * a peer of one we'd found earlier.
 149  */
 150 static struct mount *propagation_next(struct mount *m,
 151                                          struct mount *origin)
 152 {
 153         /* are there any slaves of this mount? */
 154         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 155                 return first_slave(m);
 156
 157         while (1) {
 158                 struct mount *master = m->mnt_master;
 159
 160                 if (master == origin->mnt_master) {
 161                         struct mount *next = next_peer(m);
 162                         return (next == origin) ? NULL : next;
 163                 } else if (m->mnt_slave.next != &master->mnt_slave_list)
 164                         return next_slave(m);
 165
 166                 /* back at master */
 167                 m = master;
 168         }
 169 }
 170
 171 static struct mount *skip_propagation_subtree(struct mount *m,
 172                                                 struct mount *origin)
 173 {
 174         /*
 175          * Advance m such that propagation_next will not return
 176          * the slaves of m.
 177          */
 178         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 179                 m = last_slave(m);
 180
 181         return m;
 182 }
 183
 184 static struct mount *next_group(struct mount *m, struct mount *origin)
 185 {
 186         while (1) {
 187                 while (1) {
 188                         struct mount *next;
 189                         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 190                                 return first_slave(m);
 191                         next = next_peer(m);
 192                         if (m->mnt_group_id == origin->mnt_group_id) {
 193                                 if (next == origin)
 194                                         return NULL;
 195                         } else if (m->mnt_slave.next != &next->mnt_slave)
 196                                 break;
 197                         m = next;
 198                 }
 199                 /* m is the last peer */
 200                 while (1) {
 201                         struct mount *master = m->mnt_master;
 202                         if (m->mnt_slave.next != &master->mnt_slave_list)
 203                                 return next_slave(m);
 204                         m = next_peer(master);
 205                         if (master->mnt_group_id == origin->mnt_group_id)
 206                                 break;
 207                         if (master->mnt_slave.next == &m->mnt_slave)
 208                                 break;
 209                         m = master;
 210                 }
 211                 if (m == origin)
 212                         return NULL;
 213         }
 214 }
 215
 216 /* all accesses are serialized by namespace_sem */
 217 static struct user_namespace *user_ns;
 218 static struct mount *last_dest, *first_source, *last_source, *dest_master;
 219 static struct mountpoint *mp;
 220 static struct hlist_head *list;
 221
 222 static inline bool peers(struct mount *m1, struct mount *m2)
 223 {
 224         return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
 225 }
 226
 227 static int propagate_one(struct mount *m)
 228 {
 229         struct mount *child;
 230         int type;
 231         /* skip ones added by this propagate_mnt() */
 232         if (IS_MNT_NEW(m))
 233                 return 0;
 234         /* skip if mountpoint isn't covered by it */
 235         if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
 236                 return 0;
 237         if (peers(m, last_dest)) {
 238                 type = CL_MAKE_SHARED;
 239         } else {
 240                 struct mount *n, *p;
 241                 bool done;
 242                 for (n = m; ; n = p) {
 243                         p = n->mnt_master;
 244                         if (p == dest_master || IS_MNT_MARKED(p))
 245                                 break;
 246                 }
 247                 do {
 248                         struct mount *parent = last_source->mnt_parent;
 249                         if (last_source == first_source)
 250                                 break;
 251                         done = parent->mnt_master == p;
 252                         if (done && peers(n, parent))
 253                                 break;
 254                         last_source = last_source->mnt_master;
 255                 } while (!done);
 256
 257                 type = CL_SLAVE;
 258                 /* beginning of peer group among the slaves? */
 259                 if (IS_MNT_SHARED(m))
 260                         type |= CL_MAKE_SHARED;
 261         }
 262
 263         /* Notice when we are propagating across user namespaces */
 264         if (m->mnt_ns->user_ns != user_ns)
 265                 type |= CL_UNPRIVILEGED;
 266         child = copy_tree(last_source, last_source->mnt.mnt_root, type);
 267         if (IS_ERR(child))
 268                 return PTR_ERR(child);
 269         child->mnt.mnt_flags &= ~MNT_LOCKED;
 270         mnt_set_mountpoint(m, mp, child);
 271         last_dest = m;
 272         last_source = child;
 273         if (m->mnt_master != dest_master) {
 274                 read_seqlock_excl(&mount_lock);
 275                 SET_MNT_MARK(m->mnt_master);
 276                 read_sequnlock_excl(&mount_lock);
 277         }
 278         hlist_add_head(&child->mnt_hash, list);
 279         return count_mounts(m->mnt_ns, child);
 280 }
 281
 282 /*
 283  * mount 'source_mnt' under the destination 'dest_mnt' at
 284  * dentry 'dest_dentry'. And propagate that mount to
 285  * all the peer and slave mounts of 'dest_mnt'.
 286  * Link all the new mounts into a propagation tree headed at
 287  * source_mnt. Also link all the new mounts using ->mnt_list
 288  * headed at source_mnt's ->mnt_list
 289  *
 290  * @dest_mnt: destination mount.
 291  * @dest_dentry: destination dentry.
 292  * @source_mnt: source mount.
 293  * @tree_list : list of heads of trees to be attached.
 294  */
 295 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 296                     struct mount *source_mnt, struct hlist_head *tree_list)
 297 {
 298         struct mount *m, *n;
 299         int ret = 0;
 300
 301         /*
 302          * we don't want to bother passing tons of arguments to
 303          * propagate_one(); everything is serialized by namespace_sem,
 304          * so globals will do just fine.
 305          */
 306         user_ns = current->nsproxy->mnt_ns->user_ns;
 307         last_dest = dest_mnt;
 308         first_source = source_mnt;
 309         last_source = source_mnt;
 310         mp = dest_mp;
 311         list = tree_list;
 312         dest_master = dest_mnt->mnt_master;
 313
 314         /* all peers of dest_mnt, except dest_mnt itself */
 315         for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
 316                 ret = propagate_one(n);
 317                 if (ret)
 318                         goto out;
 319         }
 320
 321         /* all slave groups */
 322         for (m = next_group(dest_mnt, dest_mnt); m;
 323                         m = next_group(m, dest_mnt)) {
 324                 /* everything in that slave group */
 325                 n = m;
 326                 do {
 327                         ret = propagate_one(n);
 328                         if (ret)
 329                                 goto out;
 330                         n = next_peer(n);
 331                 } while (n != m);
 332         }
 333 out:
 334         read_seqlock_excl(&mount_lock);
 335         hlist_for_each_entry(n, tree_list, mnt_hash) {
 336                 m = n->mnt_parent;
 337                 if (m->mnt_master != dest_mnt->mnt_master)
 338                         CLEAR_MNT_MARK(m->mnt_master);
 339         }
 340         read_sequnlock_excl(&mount_lock);
 341         return ret;
 342 }
 343
 344 static struct mount *find_topper(struct mount *mnt)
 345 {
 346         /* If there is exactly one mount covering mnt completely return it. */
 347         struct mount *child;
 348
 349         if (!list_is_singular(&mnt->mnt_mounts))
 350                 return NULL;
 351
 352         child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
 353         if (child->mnt_mountpoint != mnt->mnt.mnt_root)
 354                 return NULL;
 355
 356         return child;
 357 }
 358
 359 /*
 360  * return true if the refcount is greater than count
 361  */
 362 static inline int do_refcount_check(struct mount *mnt, int count)
 363 {
 364         return mnt_get_count(mnt) > count;
 365 }
 366
 367 /*
 368  * check if the mount 'mnt' can be unmounted successfully.
 369  * @mnt: the mount to be checked for unmount
 370  * NOTE: unmounting 'mnt' would naturally propagate to all
 371  * other mounts its parent propagates to.
 372  * Check if any of these mounts that **do not have submounts**
 373  * have more references than 'refcnt'. If so return busy.
 374  *
 375  * vfsmount lock must be held for write
 376  */
 377 int propagate_mount_busy(struct mount *mnt, int refcnt)
 378 {
 379         struct mount *m, *child, *topper;
 380         struct mount *parent = mnt->mnt_parent;
 381
 382         if (mnt == parent)
 383                 return do_refcount_check(mnt, refcnt);
 384
 385         /*
 386          * quickly check if the current mount can be unmounted.
 387          * If not, we don't have to go checking for all other
 388          * mounts
 389          */
 390         if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 391                 return 1;
 392
 393         for (m = propagation_next(parent, parent); m;
 394                         m = propagation_next(m, parent)) {
 395                 int count = 1;
 396                 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 397                 if (!child)
 398                         continue;
 399
 400                 /* Is there exactly one mount on the child that covers
 401                  * it completely whose reference should be ignored?
 402                  */
 403                 topper = find_topper(child);
 404                 if (topper)
 405                         count += 1;
 406                 else if (!list_empty(&child->mnt_mounts))
 407                         continue;
 408
 409                 if (do_refcount_check(child, count))
 410                         return 1;
 411         }
 412         return 0;
 413 }
 414
 415 /*
 416  * Clear MNT_LOCKED when it can be shown to be safe.
 417  *
 418  * mount_lock lock must be held for write
 419  */
 420 void propagate_mount_unlock(struct mount *mnt)
 421 {
 422         struct mount *parent = mnt->mnt_parent;
 423         struct mount *m, *child;
 424
 425         BUG_ON(parent == mnt);
 426
 427         for (m = propagation_next(parent, parent); m;
 428                         m = propagation_next(m, parent)) {
 429                 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 430                 if (child)
 431                         child->mnt.mnt_flags &= ~MNT_LOCKED;
 432         }
 433 }
 434
 435 static void umount_one(struct mount *mnt, struct list_head *to_umount)
 436 {
 437         CLEAR_MNT_MARK(mnt);
 438         mnt->mnt.mnt_flags |= MNT_UMOUNT;
 439         list_del_init(&mnt->mnt_child);
 440         list_del_init(&mnt->mnt_umounting);
 441         list_move_tail(&mnt->mnt_list, to_umount);
 442 }
 443
 444 /*
 445  * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
 446  * parent propagates to.
 447  */
 448 static bool __propagate_umount(struct mount *mnt,
 449                                struct list_head *to_umount,
 450                                struct list_head *to_restore)
 451 {
 452         bool progress = false;
 453         struct mount *child;
 454
 455         /*
 456          * The state of the parent won't change if this mount is
 457          * already unmounted or marked as without children.
 458          */
 459         if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
 460                 goto out;
 461
 462         /* Verify topper is the only grandchild that has not been
 463          * speculatively unmounted.
 464          */
 465         list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
 466                 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
 467                         continue;
 468                 if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
 469                         continue;
 470                 /* Found a mounted child */
 471                 goto children;
 472         }
 473
 474         /* Mark mounts that can be unmounted if not locked */
 475         SET_MNT_MARK(mnt);
 476         progress = true;
 477
 478         /* If a mount is without children and not locked umount it. */
 479         if (!IS_MNT_LOCKED(mnt)) {
 480                 umount_one(mnt, to_umount);
 481         } else {
 482 children:
 483                 list_move_tail(&mnt->mnt_umounting, to_restore);
 484         }
 485 out:
 486         return progress;
 487 }
 488
 489 static void umount_list(struct list_head *to_umount,
 490                         struct list_head *to_restore)
 491 {
 492         struct mount *mnt, *child, *tmp;
 493         list_for_each_entry(mnt, to_umount, mnt_list) {
 494                 list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
 495                         /* topper? */
 496                         if (child->mnt_mountpoint == mnt->mnt.mnt_root)
 497                                 list_move_tail(&child->mnt_umounting, to_restore);
 498                         else
 499                                 umount_one(child, to_umount);
 500                 }
 501         }
 502 }
 503
 504 static void restore_mounts(struct list_head *to_restore)
 505 {
 506         /* Restore mounts to a clean working state */
 507         while (!list_empty(to_restore)) {
 508                 struct mount *mnt, *parent;
 509                 struct mountpoint *mp;
 510
 511                 mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
 512                 CLEAR_MNT_MARK(mnt);
 513                 list_del_init(&mnt->mnt_umounting);
 514
 515                 /* Should this mount be reparented? */
 516                 mp = mnt->mnt_mp;
 517                 parent = mnt->mnt_parent;
 518                 while (parent->mnt.mnt_flags & MNT_UMOUNT) {
 519                         mp = parent->mnt_mp;
 520                         parent = parent->mnt_parent;
 521                 }
 522                 if (parent != mnt->mnt_parent)
 523                         mnt_change_mountpoint(parent, mp, mnt);
 524         }
 525 }
 526
 527 static void cleanup_umount_visitations(struct list_head *visited)
 528 {
 529         while (!list_empty(visited)) {
 530                 struct mount *mnt =
 531                         list_first_entry(visited, struct mount, mnt_umounting);
 532                 list_del_init(&mnt->mnt_umounting);
 533         }
 534 }
 535
 536 /*
 537  * collect all mounts that receive propagation from the mount in @list,
 538  * and return these additional mounts in the same list.
 539  * @list: the list of mounts to be unmounted.
 540  *
 541  * vfsmount lock must be held for write
 542  */
 543 int propagate_umount(struct list_head *list)
 544 {
 545         struct mount *mnt;
 546         LIST_HEAD(to_restore);
 547         LIST_HEAD(to_umount);
 548         LIST_HEAD(visited);
 549
 550         /* Find candidates for unmounting */
 551         list_for_each_entry_reverse(mnt, list, mnt_list) {
 552                 struct mount *parent = mnt->mnt_parent;
 553                 struct mount *m;
 554
 555                 /*
 556                  * If this mount has already been visited it is known that it's
 557                  * entire peer group and all of their slaves in the propagation
 558                  * tree for the mountpoint has already been visited and there is
 559                  * no need to visit them again.
 560                  */
 561                 if (!list_empty(&mnt->mnt_umounting))
 562                         continue;
 563
 564                 list_add_tail(&mnt->mnt_umounting, &visited);
 565                 for (m = propagation_next(parent, parent); m;
 566                      m = propagation_next(m, parent)) {
 567                         struct mount *child = __lookup_mnt(&m->mnt,
 568                                                            mnt->mnt_mountpoint);
 569                         if (!child)
 570                                 continue;
 571
 572                         if (!list_empty(&child->mnt_umounting)) {
 573                                 /*
 574                                  * If the child has already been visited it is
 575                                  * know that it's entire peer group and all of
 576                                  * their slaves in the propgation tree for the
 577                                  * mountpoint has already been visited and there
 578                                  * is no need to visit this subtree again.
 579                                  */
 580                                 m = skip_propagation_subtree(m, parent);
 581                                 continue;
 582                         } else if (child->mnt.mnt_flags & MNT_UMOUNT) {
 583                                 /*
 584                                  * We have come accross an partially unmounted
 585                                  * mount in list that has not been visited yet.
 586                                  * Remember it has been visited and continue
 587                                  * about our merry way.
 588                                  */
 589                                 list_add_tail(&child->mnt_umounting, &visited);
 590                                 continue;
 591                         }
 592
 593                         /* Check the child and parents while progress is made */
 594                         while (__propagate_umount(child,
 595                                                   &to_umount, &to_restore)) {
 596                                 /* Is the parent a umount candidate? */
 597                                 child = child->mnt_parent;
 598                                 if (list_empty(&child->mnt_umounting))
 599                                         break;
 600                         }
 601                 }
 602         }
 603
 604         umount_list(&to_umount, &to_restore);
 605         restore_mounts(&to_restore);
 606         cleanup_umount_visitations(&visited);
 607         list_splice_tail(&to_umount, list);
 608
 609         return 0;
 610 }