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 "internal.h"
  13 #include "pnode.h"
  14
  15 /* return the next shared peer mount of @p */
  16 static inline struct vfsmount *next_peer(struct vfsmount *p)
  17 {
  18         return list_entry(p->mnt_share.next, struct vfsmount, mnt_share);
  19 }
  20
  21 static inline struct vfsmount *first_slave(struct vfsmount *p)
  22 {
  23         return list_entry(p->mnt_slave_list.next, struct vfsmount, mnt_slave);
  24 }
  25
  26 static inline struct vfsmount *next_slave(struct vfsmount *p)
  27 {
  28         return list_entry(p->mnt_slave.next, struct vfsmount, mnt_slave);
  29 }
  30
  31 static struct mount *get_peer_under_root(struct mount *mnt,
  32                                          struct mnt_namespace *ns,
  33                                          const struct path *root)
  34 {
  35         struct mount *m = mnt;
  36
  37         do {
  38                 /* Check the namespace first for optimization */
  39                 if (m->mnt.mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
  40                         return m;
  41
  42                 m = real_mount(next_peer(&m->mnt));
  43         } while (m != mnt);
  44
  45         return NULL;
  46 }
  47
  48 /*
  49  * Get ID of closest dominating peer group having a representative
  50  * under the given root.
  51  *
  52  * Caller must hold namespace_sem
  53  */
  54 int get_dominating_id(struct mount *mnt, const struct path *root)
  55 {
  56         struct mount *m;
  57
  58         for (m = real_mount(mnt->mnt.mnt_master); m != NULL; m = real_mount(m->mnt.mnt_master)) {
  59                 struct mount *d = get_peer_under_root(m, mnt->mnt.mnt_ns, root);
  60                 if (d)
  61                         return d->mnt.mnt_group_id;
  62         }
  63
  64         return 0;
  65 }
  66
  67 static int do_make_slave(struct mount *mnt)
  68 {
  69         struct mount *peer_mnt = mnt, *master = real_mount(mnt->mnt.mnt_master);
  70         struct vfsmount *slave_mnt;
  71
  72         /*
  73          * slave 'mnt' to a peer mount that has the
  74          * same root dentry. If none is available then
  75          * slave it to anything that is available.
  76          */
  77         while ((peer_mnt = real_mount(next_peer(&peer_mnt->mnt))) != mnt &&
  78                peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
  79
  80         if (peer_mnt == mnt) {
  81                 peer_mnt = real_mount(next_peer(&mnt->mnt));
  82                 if (peer_mnt == mnt)
  83                         peer_mnt = NULL;
  84         }
  85         if (IS_MNT_SHARED(&mnt->mnt) && list_empty(&mnt->mnt.mnt_share))
  86                 mnt_release_group_id(mnt);
  87
  88         list_del_init(&mnt->mnt.mnt_share);
  89         mnt->mnt.mnt_group_id = 0;
  90
  91         if (peer_mnt)
  92                 master = peer_mnt;
  93
  94         if (master) {
  95                 list_for_each_entry(slave_mnt, &mnt->mnt.mnt_slave_list, mnt_slave)
  96                         slave_mnt->mnt_master = &master->mnt;
  97                 list_move(&mnt->mnt.mnt_slave, &master->mnt.mnt_slave_list);
  98                 list_splice(&mnt->mnt.mnt_slave_list, master->mnt.mnt_slave_list.prev);
  99                 INIT_LIST_HEAD(&mnt->mnt.mnt_slave_list);
 100         } else {
 101                 struct list_head *p = &mnt->mnt.mnt_slave_list;
 102                 while (!list_empty(p)) {
 103                         slave_mnt = list_first_entry(p,
 104                                         struct vfsmount, mnt_slave);
 105                         list_del_init(&slave_mnt->mnt_slave);
 106                         slave_mnt->mnt_master = NULL;
 107                 }
 108         }
 109         mnt->mnt.mnt_master = &master->mnt;
 110         CLEAR_MNT_SHARED(&mnt->mnt);
 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.mnt_slave);
 126                 mnt->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 vfsmount *propagation_next(struct vfsmount *m,
 145                                          struct vfsmount *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 vfsmount *next;
 153                 struct vfsmount *master = m->mnt_master;
 154
 155                 if (master == origin->mnt_master) {
 156                         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 vfsmount *get_source(struct vfsmount *dest,
 176                                         struct vfsmount *last_dest,
 177                                         struct vfsmount *last_src,
 178                                         int *type)
 179 {
 180         struct vfsmount *p_last_src = NULL;
 181         struct vfsmount *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 vfsmount *dest_mnt, struct dentry *dest_dentry,
 222                     struct vfsmount *source_mnt, struct list_head *tree_list)
 223 {
 224         struct vfsmount *m;
 225         struct mount *child;
 226         int ret = 0;
 227         struct vfsmount *prev_dest_mnt = dest_mnt;
 228         struct vfsmount *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 vfsmount *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                 if (!(child = copy_tree(real_mount(source), source->mnt_root, type))) {
 243                         ret = -ENOMEM;
 244                         list_splice(tree_list, tmp_list.prev);
 245                         goto out;
 246                 }
 247
 248                 if (is_subdir(dest_dentry, m->mnt_root)) {
 249                         mnt_set_mountpoint(m, dest_dentry, child);
 250                         list_add_tail(&child->mnt_hash, tree_list);
 251                 } else {
 252                         /*
 253                          * This can happen if the parent mount was bind mounted
 254                          * on some subdirectory of a shared/slave mount.
 255                          */
 256                         list_add_tail(&child->mnt_hash, &tmp_list);
 257                 }
 258                 prev_dest_mnt = m;
 259                 prev_src_mnt  = &child->mnt;
 260         }
 261 out:
 262         br_write_lock(vfsmount_lock);
 263         while (!list_empty(&tmp_list)) {
 264                 child = list_first_entry(&tmp_list, struct mount, mnt_hash);
 265                 umount_tree(child, 0, &umount_list);
 266         }
 267         br_write_unlock(vfsmount_lock);
 268         release_mounts(&umount_list);
 269         return ret;
 270 }
 271
 272 /*
 273  * return true if the refcount is greater than count
 274  */
 275 static inline int do_refcount_check(struct mount *mnt, int count)
 276 {
 277         int mycount = mnt_get_count(mnt) - mnt->mnt.mnt_ghosts;
 278         return (mycount > count);
 279 }
 280
 281 /*
 282  * check if the mount 'mnt' can be unmounted successfully.
 283  * @mnt: the mount to be checked for unmount
 284  * NOTE: unmounting 'mnt' would naturally propagate to all
 285  * other mounts its parent propagates to.
 286  * Check if any of these mounts that **do not have submounts**
 287  * have more references than 'refcnt'. If so return busy.
 288  *
 289  * vfsmount lock must be held for write
 290  */
 291 int propagate_mount_busy(struct mount *mnt, int refcnt)
 292 {
 293         struct vfsmount *m;
 294         struct mount *child;
 295         struct mount *parent = mnt->mnt_parent;
 296         int ret = 0;
 297
 298         if (mnt == parent)
 299                 return do_refcount_check(mnt, refcnt);
 300
 301         /*
 302          * quickly check if the current mount can be unmounted.
 303          * If not, we don't have to go checking for all other
 304          * mounts
 305          */
 306         if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 307                 return 1;
 308
 309         for (m = propagation_next(&parent->mnt, &parent->mnt); m;
 310                         m = propagation_next(m, &parent->mnt)) {
 311                 child = __lookup_mnt(m, mnt->mnt_mountpoint, 0);
 312                 if (child && list_empty(&child->mnt_mounts) &&
 313                     (ret = do_refcount_check(child, 1)))
 314                         break;
 315         }
 316         return ret;
 317 }
 318
 319 /*
 320  * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
 321  * parent propagates to.
 322  */
 323 static void __propagate_umount(struct mount *mnt)
 324 {
 325         struct mount *parent = mnt->mnt_parent;
 326         struct vfsmount *m;
 327
 328         BUG_ON(parent == mnt);
 329
 330         for (m = propagation_next(&parent->mnt, &parent->mnt); m;
 331                         m = propagation_next(m, &parent->mnt)) {
 332
 333                 struct mount *child = __lookup_mnt(m,
 334                                         mnt->mnt_mountpoint, 0);
 335                 /*
 336                  * umount the child only if the child has no
 337                  * other children
 338                  */
 339                 if (child && list_empty(&child->mnt_mounts))
 340                         list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
 341         }
 342 }
 343
 344 /*
 345  * collect all mounts that receive propagation from the mount in @list,
 346  * and return these additional mounts in the same list.
 347  * @list: the list of mounts to be unmounted.
 348  *
 349  * vfsmount lock must be held for write
 350  */
 351 int propagate_umount(struct list_head *list)
 352 {
 353         struct mount *mnt;
 354
 355         list_for_each_entry(mnt, list, mnt_hash)
 356                 __propagate_umount(mnt);
 357         return 0;
 358 }