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 mount *next_peer(struct mount *p)
  17 {
  18         return list_entry(p->mnt_share.next, struct mount, mnt_share);
  19 }
  20
  21 static inline struct mount *first_slave(struct mount *p)
  22 {
  23         return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
  24 }
  25
  26 static inline struct mount *next_slave(struct mount *p)
  27 {
  28         return list_entry(p->mnt_slave.next, struct mount, 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_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
  40                         return m;
  41
  42                 m = next_peer(m);
  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 = mnt->mnt_master; m != NULL; m = m->mnt_master) {
  59                 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
  60                 if (d)
  61                         return d->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 = mnt->mnt_master;
  70         struct mount *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 = next_peer(peer_mnt)) != mnt &&
  78                peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
  79
  80         if (peer_mnt == mnt) {
  81                 peer_mnt = next_peer(mnt);
  82                 if (peer_mnt == mnt)
  83                         peer_mnt = NULL;
  84         }
  85         if (IS_MNT_SHARED(mnt) && list_empty(&mnt->mnt_share))
  86                 mnt_release_group_id(mnt);
  87
  88         list_del_init(&mnt->mnt_share);
  89         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_slave_list, mnt_slave)
  96                         slave_mnt->mnt_master = master;
  97                 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
  98                 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
  99                 INIT_LIST_HEAD(&mnt->mnt_slave_list);
 100         } else {
 101                 struct list_head *p = &mnt->mnt_slave_list;
 102                 while (!list_empty(p)) {
 103                         slave_mnt = list_first_entry(p,
 104                                         struct mount, mnt_slave);
 105                         list_del_init(&slave_mnt->mnt_slave);
 106                         slave_mnt->mnt_master = NULL;
 107                 }
 108         }
 109         mnt->mnt_master = master;
 110         CLEAR_MNT_SHARED(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_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 /*
 166  * return the source mount to be used for cloning
 167  *
 168  * @dest        the current destination mount
 169  * @last_dest   the last seen destination mount
 170  * @last_src    the last seen source mount
 171  * @type        return CL_SLAVE if the new mount has to be
 172  *              cloned as a slave.
 173  */
 174 static struct mount *get_source(struct mount *dest,
 175                                 struct mount *last_dest,
 176                                 struct mount *last_src,
 177                                 int *type)
 178 {
 179         struct mount *p_last_src = NULL;
 180         struct mount *p_last_dest = NULL;
 181
 182         while (last_dest != dest->mnt_master) {
 183                 p_last_dest = last_dest;
 184                 p_last_src = last_src;
 185                 last_dest = last_dest->mnt_master;
 186                 last_src = last_src->mnt_master;
 187         }
 188
 189         if (p_last_dest) {
 190                 do {
 191                         p_last_dest = next_peer(p_last_dest);
 192                 } while (IS_MNT_NEW(p_last_dest));
 193                 /* is that a peer of the earlier? */
 194                 if (dest == p_last_dest) {
 195                         *type = CL_MAKE_SHARED;
 196                         return p_last_src;
 197                 }
 198         }
 199         /* slave of the earlier, then */
 200         *type = CL_SLAVE;
 201         /* beginning of peer group among the slaves? */
 202         if (IS_MNT_SHARED(dest))
 203                 *type |= CL_MAKE_SHARED;
 204         return last_src;
 205 }
 206
 207 /*
 208  * mount 'source_mnt' under the destination 'dest_mnt' at
 209  * dentry 'dest_dentry'. And propagate that mount to
 210  * all the peer and slave mounts of 'dest_mnt'.
 211  * Link all the new mounts into a propagation tree headed at
 212  * source_mnt. Also link all the new mounts using ->mnt_list
 213  * headed at source_mnt's ->mnt_list
 214  *
 215  * @dest_mnt: destination mount.
 216  * @dest_dentry: destination dentry.
 217  * @source_mnt: source mount.
 218  * @tree_list : list of heads of trees to be attached.
 219  */
 220 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 221                     struct mount *source_mnt, struct list_head *tree_list)
 222 {
 223         struct mount *m, *child;
 224         int ret = 0;
 225         struct mount *prev_dest_mnt = dest_mnt;
 226         struct mount *prev_src_mnt  = source_mnt;
 227         LIST_HEAD(tmp_list);
 228
 229         for (m = propagation_next(dest_mnt, dest_mnt); m;
 230                         m = propagation_next(m, dest_mnt)) {
 231                 int type;
 232                 struct mount *source;
 233
 234                 if (IS_MNT_NEW(m))
 235                         continue;
 236
 237                 source =  get_source(m, prev_dest_mnt, prev_src_mnt, &type);
 238
 239                 child = copy_tree(source, source->mnt.mnt_root, type);
 240                 if (IS_ERR(child)) {
 241                         ret = PTR_ERR(child);
 242                         list_splice(tree_list, tmp_list.prev);
 243                         goto out;
 244                 }
 245
 246                 if (is_subdir(dest_mp->m_dentry, m->mnt.mnt_root)) {
 247                         mnt_set_mountpoint(m, dest_mp, child);
 248                         list_add_tail(&child->mnt_hash, tree_list);
 249                 } else {
 250                         /*
 251                          * This can happen if the parent mount was bind mounted
 252                          * on some subdirectory of a shared/slave mount.
 253                          */
 254                         list_add_tail(&child->mnt_hash, &tmp_list);
 255                 }
 256                 prev_dest_mnt = m;
 257                 prev_src_mnt  = child;
 258         }
 259 out:
 260         br_write_lock(&vfsmount_lock);
 261         while (!list_empty(&tmp_list)) {
 262                 child = list_first_entry(&tmp_list, struct mount, mnt_hash);
 263                 umount_tree(child, 0);
 264         }
 265         br_write_unlock(&vfsmount_lock);
 266         return ret;
 267 }
 268
 269 /*
 270  * return true if the refcount is greater than count
 271  */
 272 static inline int do_refcount_check(struct mount *mnt, int count)
 273 {
 274         int mycount = mnt_get_count(mnt) - mnt->mnt_ghosts;
 275         return (mycount > count);
 276 }
 277
 278 /*
 279  * check if the mount 'mnt' can be unmounted successfully.
 280  * @mnt: the mount to be checked for unmount
 281  * NOTE: unmounting 'mnt' would naturally propagate to all
 282  * other mounts its parent propagates to.
 283  * Check if any of these mounts that **do not have submounts**
 284  * have more references than 'refcnt'. If so return busy.
 285  *
 286  * vfsmount lock must be held for write
 287  */
 288 int propagate_mount_busy(struct mount *mnt, int refcnt)
 289 {
 290         struct mount *m, *child;
 291         struct mount *parent = mnt->mnt_parent;
 292         int ret = 0;
 293
 294         if (mnt == parent)
 295                 return do_refcount_check(mnt, refcnt);
 296
 297         /*
 298          * quickly check if the current mount can be unmounted.
 299          * If not, we don't have to go checking for all other
 300          * mounts
 301          */
 302         if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 303                 return 1;
 304
 305         for (m = propagation_next(parent, parent); m;
 306                         m = propagation_next(m, parent)) {
 307                 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint, 0);
 308                 if (child && list_empty(&child->mnt_mounts) &&
 309                     (ret = do_refcount_check(child, 1)))
 310                         break;
 311         }
 312         return ret;
 313 }
 314
 315 /*
 316  * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
 317  * parent propagates to.
 318  */
 319 static void __propagate_umount(struct mount *mnt)
 320 {
 321         struct mount *parent = mnt->mnt_parent;
 322         struct mount *m;
 323
 324         BUG_ON(parent == mnt);
 325
 326         for (m = propagation_next(parent, parent); m;
 327                         m = propagation_next(m, parent)) {
 328
 329                 struct mount *child = __lookup_mnt(&m->mnt,
 330                                         mnt->mnt_mountpoint, 0);
 331                 /*
 332                  * umount the child only if the child has no
 333                  * other children
 334                  */
 335                 if (child && list_empty(&child->mnt_mounts))
 336                         list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
 337         }
 338 }
 339
 340 /*
 341  * collect all mounts that receive propagation from the mount in @list,
 342  * and return these additional mounts in the same list.
 343  * @list: the list of mounts to be unmounted.
 344  *
 345  * vfsmount lock must be held for write
 346  */
 347 int propagate_umount(struct list_head *list)
 348 {
 349         struct mount *mnt;
 350
 351         list_for_each_entry(mnt, list, mnt_hash)
 352                 __propagate_umount(mnt);
 353         return 0;
 354 }