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 "internal.h"
  14 #include "pnode.h"
  15
  16 /* return the next shared peer mount of @p */
  17 static inline struct mount *next_peer(struct mount *p)
  18 {
  19         return list_entry(p->mnt_share.next, struct mount, mnt_share);
  20 }
  21
  22 static inline struct mount *first_slave(struct mount *p)
  23 {
  24         return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
  25 }
  26
  27 static inline struct mount *next_slave(struct mount *p)
  28 {
  29         return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
  30 }
  31
  32 static struct mount *get_peer_under_root(struct mount *mnt,
  33                                          struct mnt_namespace *ns,
  34                                          const struct path *root)
  35 {
  36         struct mount *m = mnt;
  37
  38         do {
  39                 /* Check the namespace first for optimization */
  40                 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
  41                         return m;
  42
  43                 m = next_peer(m);
  44         } while (m != mnt);
  45
  46         return NULL;
  47 }
  48
  49 /*
  50  * Get ID of closest dominating peer group having a representative
  51  * under the given root.
  52  *
  53  * Caller must hold namespace_sem
  54  */
  55 int get_dominating_id(struct mount *mnt, const struct path *root)
  56 {
  57         struct mount *m;
  58
  59         for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
  60                 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
  61                 if (d)
  62                         return d->mnt_group_id;
  63         }
  64
  65         return 0;
  66 }
  67
  68 static int do_make_slave(struct mount *mnt)
  69 {
  70         struct mount *master, *slave_mnt;
  71
  72         if (list_empty(&mnt->mnt_share)) {
  73                 if (IS_MNT_SHARED(mnt)) {
  74                         mnt_release_group_id(mnt);
  75                         CLEAR_MNT_SHARED(mnt);
  76                 }
  77                 master = mnt->mnt_master;
  78                 if (!master) {
  79                         struct list_head *p = &mnt->mnt_slave_list;
  80                         while (!list_empty(p)) {
  81                                 slave_mnt = list_first_entry(p,
  82                                                 struct mount, mnt_slave);
  83                                 list_del_init(&slave_mnt->mnt_slave);
  84                                 slave_mnt->mnt_master = NULL;
  85                         }
  86                         return 0;
  87                 }
  88         } else {
  89                 struct mount *m;
  90                 /*
  91                  * slave 'mnt' to a peer mount that has the
  92                  * same root dentry. If none is available then
  93                  * slave it to anything that is available.
  94                  */
  95                 for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
  96                         if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
  97                                 master = m;
  98                                 break;
  99                         }
 100                 }
 101                 list_del_init(&mnt->mnt_share);
 102                 mnt->mnt_group_id = 0;
 103                 CLEAR_MNT_SHARED(mnt);
 104         }
 105         list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
 106                 slave_mnt->mnt_master = master;
 107         list_move(&mnt->mnt_slave, &master->mnt_slave_list);
 108         list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
 109         INIT_LIST_HEAD(&mnt->mnt_slave_list);
 110         mnt->mnt_master = master;
 111         return 0;
 112 }
 113
 114 /*
 115  * vfsmount lock must be held for write
 116  */
 117 void change_mnt_propagation(struct mount *mnt, int type)
 118 {
 119         if (type == MS_SHARED) {
 120                 set_mnt_shared(mnt);
 121                 return;
 122         }
 123         do_make_slave(mnt);
 124         if (type != MS_SLAVE) {
 125                 list_del_init(&mnt->mnt_slave);
 126                 mnt->mnt_master = NULL;
 127                 if (type == MS_UNBINDABLE)
 128                         mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
 129                 else
 130                         mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
 131         }
 132 }
 133
 134 /*
 135  * get the next mount in the propagation tree.
 136  * @m: the mount seen last
 137  * @origin: the original mount from where the tree walk initiated
 138  *
 139  * Note that peer groups form contiguous segments of slave lists.
 140  * We rely on that in get_source() to be able to find out if
 141  * vfsmount found while iterating with propagation_next() is
 142  * a peer of one we'd found earlier.
 143  */
 144 static struct mount *propagation_next(struct mount *m,
 145                                          struct mount *origin)
 146 {
 147         /* are there any slaves of this mount? */
 148         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 149                 return first_slave(m);
 150
 151         while (1) {
 152                 struct mount *master = m->mnt_master;
 153
 154                 if (master == origin->mnt_master) {
 155                         struct mount *next = next_peer(m);
 156                         return (next == origin) ? NULL : next;
 157                 } else if (m->mnt_slave.next != &master->mnt_slave_list)
 158                         return next_slave(m);
 159
 160                 /* back at master */
 161                 m = master;
 162         }
 163 }
 164
 165 static struct mount *next_group(struct mount *m, struct mount *origin)
 166 {
 167         while (1) {
 168                 while (1) {
 169                         struct mount *next;
 170                         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 171                                 return first_slave(m);
 172                         next = next_peer(m);
 173                         if (m->mnt_group_id == origin->mnt_group_id) {
 174                                 if (next == origin)
 175                                         return NULL;
 176                         } else if (m->mnt_slave.next != &next->mnt_slave)
 177                                 break;
 178                         m = next;
 179                 }
 180                 /* m is the last peer */
 181                 while (1) {
 182                         struct mount *master = m->mnt_master;
 183                         if (m->mnt_slave.next != &master->mnt_slave_list)
 184                                 return next_slave(m);
 185                         m = next_peer(master);
 186                         if (master->mnt_group_id == origin->mnt_group_id)
 187                                 break;
 188                         if (master->mnt_slave.next == &m->mnt_slave)
 189                                 break;
 190                         m = master;
 191                 }
 192                 if (m == origin)
 193                         return NULL;
 194         }
 195 }
 196
 197 /* all accesses are serialized by namespace_sem */
 198 static struct user_namespace *user_ns;
 199 static struct mount *last_dest, *first_source, *last_source, *dest_master;
 200 static struct mountpoint *mp;
 201 static struct hlist_head *list;
 202
 203 static inline bool peers(struct mount *m1, struct mount *m2)
 204 {
 205         return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
 206 }
 207
 208 static int propagate_one(struct mount *m)
 209 {
 210         struct mount *child;
 211         int type;
 212         /* skip ones added by this propagate_mnt() */
 213         if (IS_MNT_NEW(m))
 214                 return 0;
 215         /* skip if mountpoint isn't covered by it */
 216         if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
 217                 return 0;
 218         if (peers(m, last_dest)) {
 219                 type = CL_MAKE_SHARED;
 220         } else {
 221                 struct mount *n, *p;
 222                 bool done;
 223                 for (n = m; ; n = p) {
 224                         p = n->mnt_master;
 225                         if (p == dest_master || IS_MNT_MARKED(p))
 226                                 break;
 227                 }
 228                 do {
 229                         struct mount *parent = last_source->mnt_parent;
 230                         if (last_source == first_source)
 231                                 break;
 232                         done = parent->mnt_master == p;
 233                         if (done && peers(n, parent))
 234                                 break;
 235                         last_source = last_source->mnt_master;
 236                 } while (!done);
 237
 238                 type = CL_SLAVE;
 239                 /* beginning of peer group among the slaves? */
 240                 if (IS_MNT_SHARED(m))
 241                         type |= CL_MAKE_SHARED;
 242         }
 243
 244         /* Notice when we are propagating across user namespaces */
 245         if (m->mnt_ns->user_ns != user_ns)
 246                 type |= CL_UNPRIVILEGED;
 247         child = copy_tree(last_source, last_source->mnt.mnt_root, type);
 248         if (IS_ERR(child))
 249                 return PTR_ERR(child);
 250         child->mnt.mnt_flags &= ~MNT_LOCKED;
 251         mnt_set_mountpoint(m, mp, child);
 252         last_dest = m;
 253         last_source = child;
 254         if (m->mnt_master != dest_master) {
 255                 read_seqlock_excl(&mount_lock);
 256                 SET_MNT_MARK(m->mnt_master);
 257                 read_sequnlock_excl(&mount_lock);
 258         }
 259         hlist_add_head(&child->mnt_hash, list);
 260         return count_mounts(m->mnt_ns, child);
 261 }
 262
 263 /*
 264  * mount 'source_mnt' under the destination 'dest_mnt' at
 265  * dentry 'dest_dentry'. And propagate that mount to
 266  * all the peer and slave mounts of 'dest_mnt'.
 267  * Link all the new mounts into a propagation tree headed at
 268  * source_mnt. Also link all the new mounts using ->mnt_list
 269  * headed at source_mnt's ->mnt_list
 270  *
 271  * @dest_mnt: destination mount.
 272  * @dest_dentry: destination dentry.
 273  * @source_mnt: source mount.
 274  * @tree_list : list of heads of trees to be attached.
 275  */
 276 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 277                     struct mount *source_mnt, struct hlist_head *tree_list)
 278 {
 279         struct mount *m, *n;
 280         int ret = 0;
 281
 282         /*
 283          * we don't want to bother passing tons of arguments to
 284          * propagate_one(); everything is serialized by namespace_sem,
 285          * so globals will do just fine.
 286          */
 287         user_ns = current->nsproxy->mnt_ns->user_ns;
 288         last_dest = dest_mnt;
 289         first_source = source_mnt;
 290         last_source = source_mnt;
 291         mp = dest_mp;
 292         list = tree_list;
 293         dest_master = dest_mnt->mnt_master;
 294
 295         /* all peers of dest_mnt, except dest_mnt itself */
 296         for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
 297                 ret = propagate_one(n);
 298                 if (ret)
 299                         goto out;
 300         }
 301
 302         /* all slave groups */
 303         for (m = next_group(dest_mnt, dest_mnt); m;
 304                         m = next_group(m, dest_mnt)) {
 305                 /* everything in that slave group */
 306                 n = m;
 307                 do {
 308                         ret = propagate_one(n);
 309                         if (ret)
 310                                 goto out;
 311                         n = next_peer(n);
 312                 } while (n != m);
 313         }
 314 out:
 315         read_seqlock_excl(&mount_lock);
 316         hlist_for_each_entry(n, tree_list, mnt_hash) {
 317                 m = n->mnt_parent;
 318                 if (m->mnt_master != dest_mnt->mnt_master)
 319                         CLEAR_MNT_MARK(m->mnt_master);
 320         }
 321         read_sequnlock_excl(&mount_lock);
 322         return ret;
 323 }
 324
 325 static struct mount *find_topper(struct mount *mnt)
 326 {
 327         /* If there is exactly one mount covering mnt completely return it. */
 328         struct mount *child;
 329
 330         if (!list_is_singular(&mnt->mnt_mounts))
 331                 return NULL;
 332
 333         child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
 334         if (child->mnt_mountpoint != mnt->mnt.mnt_root)
 335                 return NULL;
 336
 337         return child;
 338 }
 339
 340 /*
 341  * return true if the refcount is greater than count
 342  */
 343 static inline int do_refcount_check(struct mount *mnt, int count)
 344 {
 345         return mnt_get_count(mnt) > count;
 346 }
 347
 348 /*
 349  * check if the mount 'mnt' can be unmounted successfully.
 350  * @mnt: the mount to be checked for unmount
 351  * NOTE: unmounting 'mnt' would naturally propagate to all
 352  * other mounts its parent propagates to.
 353  * Check if any of these mounts that **do not have submounts**
 354  * have more references than 'refcnt'. If so return busy.
 355  *
 356  * vfsmount lock must be held for write
 357  */
 358 int propagate_mount_busy(struct mount *mnt, int refcnt)
 359 {
 360         struct mount *m, *child, *topper;
 361         struct mount *parent = mnt->mnt_parent;
 362
 363         if (mnt == parent)
 364                 return do_refcount_check(mnt, refcnt);
 365
 366         /*
 367          * quickly check if the current mount can be unmounted.
 368          * If not, we don't have to go checking for all other
 369          * mounts
 370          */
 371         if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 372                 return 1;
 373
 374         for (m = propagation_next(parent, parent); m;
 375                         m = propagation_next(m, parent)) {
 376                 int count = 1;
 377                 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 378                 if (!child)
 379                         continue;
 380
 381                 /* Is there exactly one mount on the child that covers
 382                  * it completely whose reference should be ignored?
 383                  */
 384                 topper = find_topper(child);
 385                 if (topper)
 386                         count += 1;
 387                 else if (!list_empty(&child->mnt_mounts))
 388                         continue;
 389
 390                 if (do_refcount_check(child, count))
 391                         return 1;
 392         }
 393         return 0;
 394 }
 395
 396 /*
 397  * Clear MNT_LOCKED when it can be shown to be safe.
 398  *
 399  * mount_lock lock must be held for write
 400  */
 401 void propagate_mount_unlock(struct mount *mnt)
 402 {
 403         struct mount *parent = mnt->mnt_parent;
 404         struct mount *m, *child;
 405
 406         BUG_ON(parent == mnt);
 407
 408         for (m = propagation_next(parent, parent); m;
 409                         m = propagation_next(m, parent)) {
 410                 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 411                 if (child)
 412                         child->mnt.mnt_flags &= ~MNT_LOCKED;
 413         }
 414 }
 415
 416 /*
 417  * Mark all mounts that the MNT_LOCKED logic will allow to be unmounted.
 418  */
 419 static void mark_umount_candidates(struct mount *mnt)
 420 {
 421         struct mount *parent = mnt->mnt_parent;
 422         struct mount *m;
 423
 424         BUG_ON(parent == mnt);
 425
 426         for (m = propagation_next(parent, parent); m;
 427                         m = propagation_next(m, parent)) {
 428                 struct mount *child = __lookup_mnt(&m->mnt,
 429                                                 mnt->mnt_mountpoint);
 430                 if (!child || (child->mnt.mnt_flags & MNT_UMOUNT))
 431                         continue;
 432                 if (!IS_MNT_LOCKED(child) || IS_MNT_MARKED(m)) {
 433                         SET_MNT_MARK(child);
 434                 }
 435         }
 436 }
 437
 438 /*
 439  * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
 440  * parent propagates to.
 441  */
 442 static void __propagate_umount(struct mount *mnt)
 443 {
 444         struct mount *parent = mnt->mnt_parent;
 445         struct mount *m;
 446
 447         BUG_ON(parent == mnt);
 448
 449         for (m = propagation_next(parent, parent); m;
 450                         m = propagation_next(m, parent)) {
 451                 struct mount *topper;
 452                 struct mount *child = __lookup_mnt(&m->mnt,
 453                                                 mnt->mnt_mountpoint);
 454                 /*
 455                  * umount the child only if the child has no children
 456                  * and the child is marked safe to unmount.
 457                  */
 458                 if (!child || !IS_MNT_MARKED(child))
 459                         continue;
 460                 CLEAR_MNT_MARK(child);
 461
 462                 /* If there is exactly one mount covering all of child
 463                  * replace child with that mount.
 464                  */
 465                 topper = find_topper(child);
 466                 if (topper)
 467                         mnt_change_mountpoint(child->mnt_parent, child->mnt_mp,
 468                                               topper);
 469
 470                 if (list_empty(&child->mnt_mounts)) {
 471                         list_del_init(&child->mnt_child);
 472                         child->mnt.mnt_flags |= MNT_UMOUNT;
 473                         list_move_tail(&child->mnt_list, &mnt->mnt_list);
 474                 }
 475         }
 476 }
 477
 478 /*
 479  * collect all mounts that receive propagation from the mount in @list,
 480  * and return these additional mounts in the same list.
 481  * @list: the list of mounts to be unmounted.
 482  *
 483  * vfsmount lock must be held for write
 484  */
 485 int propagate_umount(struct list_head *list)
 486 {
 487         struct mount *mnt;
 488
 489         list_for_each_entry_reverse(mnt, list, mnt_list)
 490                 mark_umount_candidates(mnt);
 491
 492         list_for_each_entry(mnt, list, mnt_list)
 493                 __propagate_umount(mnt);
 494         return 0;
 495 }