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 (IS_MNT_SHARED(mnt) && list_empty(&mnt->mnt_share))
  87                 mnt_release_group_id(mnt);
  88
  89         list_del_init(&mnt->mnt_share);
  90         mnt->mnt_group_id = 0;
  91
  92         if (peer_mnt)
  93                 master = peer_mnt;
  94
  95         if (master) {
  96                 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
  97                         slave_mnt->mnt_master = master;
  98                 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
  99                 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
 100                 INIT_LIST_HEAD(&mnt->mnt_slave_list);
 101         } else {
 102                 struct list_head *p = &mnt->mnt_slave_list;
 103                 while (!list_empty(p)) {
 104                         slave_mnt = list_first_entry(p,
 105                                         struct mount, mnt_slave);
 106                         list_del_init(&slave_mnt->mnt_slave);
 107                         slave_mnt->mnt_master = NULL;
 108                 }
 109         }
 110         mnt->mnt_master = master;
 111         CLEAR_MNT_SHARED(mnt);
 112         return 0;
 113 }
 114
 115 /*
 116  * vfsmount lock must be held for write
 117  */
 118 void change_mnt_propagation(struct mount *mnt, int type)
 119 {
 120         if (type == MS_SHARED) {
 121                 set_mnt_shared(mnt);
 122                 return;
 123         }
 124         do_make_slave(mnt);
 125         if (type != MS_SLAVE) {
 126                 list_del_init(&mnt->mnt_slave);
 127                 mnt->mnt_master = NULL;
 128                 if (type == MS_UNBINDABLE)
 129                         mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
 130                 else
 131                         mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
 132         }
 133 }
 134
 135 /*
 136  * get the next mount in the propagation tree.
 137  * @m: the mount seen last
 138  * @origin: the original mount from where the tree walk initiated
 139  *
 140  * Note that peer groups form contiguous segments of slave lists.
 141  * We rely on that in get_source() to be able to find out if
 142  * vfsmount found while iterating with propagation_next() is
 143  * a peer of one we'd found earlier.
 144  */
 145 static struct mount *propagation_next(struct mount *m,
 146                                          struct mount *origin)
 147 {
 148         /* are there any slaves of this mount? */
 149         if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 150                 return first_slave(m);
 151
 152         while (1) {
 153                 struct mount *master = m->mnt_master;
 154
 155                 if (master == origin->mnt_master) {
 156                         struct mount *next = next_peer(m);
 157                         return (next == origin) ? NULL : next;
 158                 } else if (m->mnt_slave.next != &master->mnt_slave_list)
 159                         return next_slave(m);
 160
 161                 /* back at master */
 162                 m = master;
 163         }
 164 }
 165
 166 /*
 167  * return the source mount to be used for cloning
 168  *
 169  * @dest        the current destination mount
 170  * @last_dest   the last seen destination mount
 171  * @last_src    the last seen source mount
 172  * @type        return CL_SLAVE if the new mount has to be
 173  *              cloned as a slave.
 174  */
 175 static struct mount *get_source(struct mount *dest,
 176                                 struct mount *last_dest,
 177                                 struct mount *last_src,
 178                                 int *type)
 179 {
 180         struct mount *p_last_src = NULL;
 181         struct mount *p_last_dest = NULL;
 182
 183         while (last_dest != dest->mnt_master) {
 184                 p_last_dest = last_dest;
 185                 p_last_src = last_src;
 186                 last_dest = last_dest->mnt_master;
 187                 last_src = last_src->mnt_master;
 188         }
 189
 190         if (p_last_dest) {
 191                 do {
 192                         p_last_dest = next_peer(p_last_dest);
 193                 } while (IS_MNT_NEW(p_last_dest));
 194                 /* is that a peer of the earlier? */
 195                 if (dest == p_last_dest) {
 196                         *type = CL_MAKE_SHARED;
 197                         return p_last_src;
 198                 }
 199         }
 200         /* slave of the earlier, then */
 201         *type = CL_SLAVE;
 202         /* beginning of peer group among the slaves? */
 203         if (IS_MNT_SHARED(dest))
 204                 *type |= CL_MAKE_SHARED;
 205         return last_src;
 206 }
 207
 208 /*
 209  * mount 'source_mnt' under the destination 'dest_mnt' at
 210  * dentry 'dest_dentry'. And propagate that mount to
 211  * all the peer and slave mounts of 'dest_mnt'.
 212  * Link all the new mounts into a propagation tree headed at
 213  * source_mnt. Also link all the new mounts using ->mnt_list
 214  * headed at source_mnt's ->mnt_list
 215  *
 216  * @dest_mnt: destination mount.
 217  * @dest_dentry: destination dentry.
 218  * @source_mnt: source mount.
 219  * @tree_list : list of heads of trees to be attached.
 220  */
 221 int propagate_mnt(struct mount *dest_mnt, struct dentry *dest_dentry,
 222                     struct mount *source_mnt, struct list_head *tree_list)
 223 {
 224         struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 225         struct mount *m, *child;
 226         int ret = 0;
 227         struct mount *prev_dest_mnt = dest_mnt;
 228         struct mount *prev_src_mnt  = source_mnt;
 229         LIST_HEAD(tmp_list);
 230         LIST_HEAD(umount_list);
 231
 232         for (m = propagation_next(dest_mnt, dest_mnt); m;
 233                         m = propagation_next(m, dest_mnt)) {
 234                 int type;
 235                 struct mount *source;
 236
 237                 if (IS_MNT_NEW(m))
 238                         continue;
 239
 240                 source =  get_source(m, prev_dest_mnt, prev_src_mnt, &type);
 241
 242                 /* Notice when we are propagating across user namespaces */
 243                 if (m->mnt_ns->user_ns != user_ns)
 244                         type |= CL_UNPRIVILEGED;
 245
 246                 child = copy_tree(source, source->mnt.mnt_root, type);
 247                 if (IS_ERR(child)) {
 248                         ret = PTR_ERR(child);
 249                         list_splice(tree_list, tmp_list.prev);
 250                         goto out;
 251                 }
 252
 253                 if (is_subdir(dest_dentry, m->mnt.mnt_root)) {
 254                         mnt_set_mountpoint(m, dest_dentry, child);
 255                         list_add_tail(&child->mnt_hash, tree_list);
 256                 } else {
 257                         /*
 258                          * This can happen if the parent mount was bind mounted
 259                          * on some subdirectory of a shared/slave mount.
 260                          */
 261                         list_add_tail(&child->mnt_hash, &tmp_list);
 262                 }
 263                 prev_dest_mnt = m;
 264                 prev_src_mnt  = child;
 265         }
 266 out:
 267         br_write_lock(&vfsmount_lock);
 268         while (!list_empty(&tmp_list)) {
 269                 child = list_first_entry(&tmp_list, struct mount, mnt_hash);
 270                 umount_tree(child, 0, &umount_list);
 271         }
 272         br_write_unlock(&vfsmount_lock);
 273         release_mounts(&umount_list);
 274         return ret;
 275 }
 276
 277 /*
 278  * return true if the refcount is greater than count
 279  */
 280 static inline int do_refcount_check(struct mount *mnt, int count)
 281 {
 282         int mycount = mnt_get_count(mnt) - mnt->mnt_ghosts;
 283         return (mycount > count);
 284 }
 285
 286 /*
 287  * check if the mount 'mnt' can be unmounted successfully.
 288  * @mnt: the mount to be checked for unmount
 289  * NOTE: unmounting 'mnt' would naturally propagate to all
 290  * other mounts its parent propagates to.
 291  * Check if any of these mounts that **do not have submounts**
 292  * have more references than 'refcnt'. If so return busy.
 293  *
 294  * vfsmount lock must be held for write
 295  */
 296 int propagate_mount_busy(struct mount *mnt, int refcnt)
 297 {
 298         struct mount *m, *child;
 299         struct mount *parent = mnt->mnt_parent;
 300         int ret = 0;
 301
 302         if (mnt == parent)
 303                 return do_refcount_check(mnt, refcnt);
 304
 305         /*
 306          * quickly check if the current mount can be unmounted.
 307          * If not, we don't have to go checking for all other
 308          * mounts
 309          */
 310         if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 311                 return 1;
 312
 313         for (m = propagation_next(parent, parent); m;
 314                         m = propagation_next(m, parent)) {
 315                 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint, 0);
 316                 if (child && list_empty(&child->mnt_mounts) &&
 317                     (ret = do_refcount_check(child, 1)))
 318                         break;
 319         }
 320         return ret;
 321 }
 322
 323 /*
 324  * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
 325  * parent propagates to.
 326  */
 327 static void __propagate_umount(struct mount *mnt)
 328 {
 329         struct mount *parent = mnt->mnt_parent;
 330         struct mount *m;
 331
 332         BUG_ON(parent == mnt);
 333
 334         for (m = propagation_next(parent, parent); m;
 335                         m = propagation_next(m, parent)) {
 336
 337                 struct mount *child = __lookup_mnt(&m->mnt,
 338                                         mnt->mnt_mountpoint, 0);
 339                 /*
 340                  * umount the child only if the child has no
 341                  * other children
 342                  */
 343                 if (child && list_empty(&child->mnt_mounts))
 344                         list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
 345         }
 346 }
 347
 348 /*
 349  * collect all mounts that receive propagation from the mount in @list,
 350  * and return these additional mounts in the same list.
 351  * @list: the list of mounts to be unmounted.
 352  *
 353  * vfsmount lock must be held for write
 354  */
 355 int propagate_umount(struct list_head *list)
 356 {
 357         struct mount *mnt;
 358
 359         list_for_each_entry(mnt, list, mnt_hash)
 360                 __propagate_umount(mnt);
 361         return 0;
 362 }