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 *peer_mnt = mnt, *master = mnt->mnt_master;
  71         struct mount *slave_mnt;
  72
  73         /*
  74          * slave 'mnt' to a peer mount that has the
  75          * same root dentry. If none is available then
  76          * slave it to anything that is available.
  77          */
  78         while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
  79                peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
  80
  81         if (peer_mnt == mnt) {
  82                 peer_mnt = next_peer(mnt);
  83                 if (peer_mnt == mnt)
  84                         peer_mnt = NULL;
  85         }
  86         if (mnt->mnt_group_id && IS_MNT_SHARED(mnt) &&
  87             list_empty(&mnt->mnt_share))
  88                 mnt_release_group_id(mnt);
  89
  90         list_del_init(&mnt->mnt_share);
  91         mnt->mnt_group_id = 0;
  92
  93         if (peer_mnt)
  94                 master = peer_mnt;
  95
  96         if (master) {
  97                 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
  98                         slave_mnt->mnt_master = master;
  99                 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
 100                 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
 101                 INIT_LIST_HEAD(&mnt->mnt_slave_list);
 102         } else {
 103                 struct list_head *p = &mnt->mnt_slave_list;
 104                 while (!list_empty(p)) {
 105                         slave_mnt = list_first_entry(p,
 106                                         struct mount, mnt_slave);
 107                         list_del_init(&slave_mnt->mnt_slave);
 108                         slave_mnt->mnt_master = NULL;
 109                 }
 110         }
 111         mnt->mnt_master = master;
 112         CLEAR_MNT_SHARED(mnt);
 113         return 0;
 114 }
 115
 116 /*
 117  * vfsmount lock must be held for write
 118  */
 119 void change_mnt_propagation(struct mount *mnt, int type)
 120 {
 121         if (type == MS_SHARED) {
 122                 set_mnt_shared(mnt);
 123                 return;
 124         }
 125         do_make_slave(mnt);
 126         if (type != MS_SLAVE) {
 127                 list_del_init(&mnt->mnt_slave);
 128                 mnt->mnt_master = NULL;
 129                 if (type == MS_UNBINDABLE)
 130                         mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
 131                 else
 132                         mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
 133         }
 134 }
 135
 136 /*
 137  * get the next mount in the propagation tree.
 138  * @m: the mount seen last
 139  * @origin: the original mount from where the tree walk initiated
 140  *
 141  * Note that peer groups form contiguous segments of slave lists.
 142  * We rely on that in get_source() to be able to find out if
 143  * vfsmount found while iterating with propagation_next() is
 144  * a peer of one we'd found earlier.
 145  */
 146 static struct mount *propagation_next(struct mount *m,
 147                                          struct mount *origin)
 148 {
 149         /* are there any slaves of this mount? */
 150         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 151                 return first_slave(m);
 152
 153         while (1) {
 154                 struct mount *master = m->mnt_master;
 155
 156                 if (master == origin->mnt_master) {
 157                         struct mount *next = next_peer(m);
 158                         return (next == origin) ? NULL : next;
 159                 } else if (m->mnt_slave.next != &master->mnt_slave_list)
 160                         return next_slave(m);
 161
 162                 /* back at master */
 163                 m = master;
 164         }
 165 }
 166
 167 static struct mount *next_group(struct mount *m, struct mount *origin)
 168 {
 169         while (1) {
 170                 while (1) {
 171                         struct mount *next;
 172                         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 173                                 return first_slave(m);
 174                         next = next_peer(m);
 175                         if (m->mnt_group_id == origin->mnt_group_id) {
 176                                 if (next == origin)
 177                                         return NULL;
 178                         } else if (m->mnt_slave.next != &next->mnt_slave)
 179                                 break;
 180                         m = next;
 181                 }
 182                 /* m is the last peer */
 183                 while (1) {
 184                         struct mount *master = m->mnt_master;
 185                         if (m->mnt_slave.next != &master->mnt_slave_list)
 186                                 return next_slave(m);
 187                         m = next_peer(master);
 188                         if (master->mnt_group_id == origin->mnt_group_id)
 189                                 break;
 190                         if (master->mnt_slave.next == &m->mnt_slave)
 191                                 break;
 192                         m = master;
 193                 }
 194                 if (m == origin)
 195                         return NULL;
 196         }
 197 }
 198
 199 /* all accesses are serialized by namespace_sem */
 200 static struct user_namespace *user_ns;
 201 static struct mount *last_dest, *last_source, *dest_master;
 202 static struct mountpoint *mp;
 203 static struct hlist_head *list;
 204
 205 static inline bool peers(struct mount *m1, struct mount *m2)
 206 {
 207         return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
 208 }
 209
 210 static int propagate_one(struct mount *m)
 211 {
 212         struct mount *child;
 213         int type;
 214         /* skip ones added by this propagate_mnt() */
 215         if (IS_MNT_NEW(m))
 216                 return 0;
 217         /* skip if mountpoint isn't covered by it */
 218         if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
 219                 return 0;
 220         if (peers(m, last_dest)) {
 221                 type = CL_MAKE_SHARED;
 222         } else {
 223                 struct mount *n, *p;
 224                 for (n = m; ; n = p) {
 225                         p = n->mnt_master;
 226                         if (p == dest_master || IS_MNT_MARKED(p)) {
 227                                 while (last_dest->mnt_master != p) {
 228                                         last_source = last_source->mnt_master;
 229                                         last_dest = last_source->mnt_parent;
 230                                 }
 231                                 if (!peers(n, last_dest)) {
 232                                         last_source = last_source->mnt_master;
 233                                         last_dest = last_source->mnt_parent;
 234                                 }
 235                                 break;
 236                         }
 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 0;
 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         last_source = source_mnt;
 290         mp = dest_mp;
 291         list = tree_list;
 292         dest_master = dest_mnt->mnt_master;
 293
 294         /* all peers of dest_mnt, except dest_mnt itself */
 295         for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
 296                 ret = propagate_one(n);
 297                 if (ret)
 298                         goto out;
 299         }
 300
 301         /* all slave groups */
 302         for (m = next_group(dest_mnt, dest_mnt); m;
 303                         m = next_group(m, dest_mnt)) {
 304                 /* everything in that slave group */
 305                 n = m;
 306                 do {
 307                         ret = propagate_one(n);
 308                         if (ret)
 309                                 goto out;
 310                         n = next_peer(n);
 311                 } while (n != m);
 312         }
 313 out:
 314         read_seqlock_excl(&mount_lock);
 315         hlist_for_each_entry(n, tree_list, mnt_hash) {
 316                 m = n->mnt_parent;
 317                 if (m->mnt_master != dest_mnt->mnt_master)
 318                         CLEAR_MNT_MARK(m->mnt_master);
 319         }
 320         read_sequnlock_excl(&mount_lock);
 321         return ret;
 322 }
 323
 324 /*
 325  * return true if the refcount is greater than count
 326  */
 327 static inline int do_refcount_check(struct mount *mnt, int count)
 328 {
 329         return mnt_get_count(mnt) > count;
 330 }
 331
 332 /*
 333  * check if the mount 'mnt' can be unmounted successfully.
 334  * @mnt: the mount to be checked for unmount
 335  * NOTE: unmounting 'mnt' would naturally propagate to all
 336  * other mounts its parent propagates to.
 337  * Check if any of these mounts that **do not have submounts**
 338  * have more references than 'refcnt'. If so return busy.
 339  *
 340  * vfsmount lock must be held for write
 341  */
 342 int propagate_mount_busy(struct mount *mnt, int refcnt)
 343 {
 344         struct mount *m, *child;
 345         struct mount *parent = mnt->mnt_parent;
 346         int ret = 0;
 347
 348         if (mnt == parent)
 349                 return do_refcount_check(mnt, refcnt);
 350
 351         /*
 352          * quickly check if the current mount can be unmounted.
 353          * If not, we don't have to go checking for all other
 354          * mounts
 355          */
 356         if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 357                 return 1;
 358
 359         for (m = propagation_next(parent, parent); m;
 360                         m = propagation_next(m, parent)) {
 361                 child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
 362                 if (child && list_empty(&child->mnt_mounts) &&
 363                     (ret = do_refcount_check(child, 1)))
 364                         break;
 365         }
 366         return ret;
 367 }
 368
 369 /*
 370  * Clear MNT_LOCKED when it can be shown to be safe.
 371  *
 372  * mount_lock lock must be held for write
 373  */
 374 void propagate_mount_unlock(struct mount *mnt)
 375 {
 376         struct mount *parent = mnt->mnt_parent;
 377         struct mount *m, *child;
 378
 379         BUG_ON(parent == mnt);
 380
 381         for (m = propagation_next(parent, parent); m;
 382                         m = propagation_next(m, parent)) {
 383                 child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
 384                 if (child)
 385                         child->mnt.mnt_flags &= ~MNT_LOCKED;
 386         }
 387 }
 388
 389 /*
 390  * Mark all mounts that the MNT_LOCKED logic will allow to be unmounted.
 391  */
 392 static void mark_umount_candidates(struct mount *mnt)
 393 {
 394         struct mount *parent = mnt->mnt_parent;
 395         struct mount *m;
 396
 397         BUG_ON(parent == mnt);
 398
 399         for (m = propagation_next(parent, parent); m;
 400                         m = propagation_next(m, parent)) {
 401                 struct mount *child = __lookup_mnt_last(&m->mnt,
 402                                                 mnt->mnt_mountpoint);
 403                 if (child && (!IS_MNT_LOCKED(child) || IS_MNT_MARKED(m))) {
 404                         SET_MNT_MARK(child);
 405                 }
 406         }
 407 }
 408
 409 /*
 410  * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
 411  * parent propagates to.
 412  */
 413 static void __propagate_umount(struct mount *mnt)
 414 {
 415         struct mount *parent = mnt->mnt_parent;
 416         struct mount *m;
 417
 418         BUG_ON(parent == mnt);
 419
 420         for (m = propagation_next(parent, parent); m;
 421                         m = propagation_next(m, parent)) {
 422
 423                 struct mount *child = __lookup_mnt_last(&m->mnt,
 424                                                 mnt->mnt_mountpoint);
 425                 /*
 426                  * umount the child only if the child has no children
 427                  * and the child is marked safe to unmount.
 428                  */
 429                 if (!child || !IS_MNT_MARKED(child))
 430                         continue;
 431                 CLEAR_MNT_MARK(child);
 432                 if (list_empty(&child->mnt_mounts)) {
 433                         list_del_init(&child->mnt_child);
 434                         child->mnt.mnt_flags |= MNT_UMOUNT;
 435                         list_move_tail(&child->mnt_list, &mnt->mnt_list);
 436                 }
 437         }
 438 }
 439
 440 /*
 441  * collect all mounts that receive propagation from the mount in @list,
 442  * and return these additional mounts in the same list.
 443  * @list: the list of mounts to be unmounted.
 444  *
 445  * vfsmount lock must be held for write
 446  */
 447 int propagate_umount(struct list_head *list)
 448 {
 449         struct mount *mnt;
 450
 451         list_for_each_entry_reverse(mnt, list, mnt_list)
 452                 mark_umount_candidates(mnt);
 453
 454         list_for_each_entry(mnt, list, mnt_list)
 455                 __propagate_umount(mnt);
 456         return 0;
 457 }