Ram Pai | 07b2088 | 2005-11-07 17:19:07 -0500 | [diff] [blame] | 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 | */ |
Kirill Korotaev | 6b3286e | 2006-12-08 02:37:56 -0800 | [diff] [blame] | 9 | #include <linux/mnt_namespace.h> |
Ram Pai | 07b2088 | 2005-11-07 17:19:07 -0500 | [diff] [blame] | 10 | #include <linux/mount.h> |
| 11 | #include <linux/fs.h> |
| 12 | #include "pnode.h" |
| 13 | |
Ram Pai | 03e06e6 | 2005-11-07 17:19:33 -0500 | [diff] [blame] | 14 | /* return the next shared peer mount of @p */ |
| 15 | static inline struct vfsmount *next_peer(struct vfsmount *p) |
| 16 | { |
| 17 | return list_entry(p->mnt_share.next, struct vfsmount, mnt_share); |
| 18 | } |
| 19 | |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 20 | static inline struct vfsmount *first_slave(struct vfsmount *p) |
| 21 | { |
| 22 | return list_entry(p->mnt_slave_list.next, struct vfsmount, mnt_slave); |
| 23 | } |
| 24 | |
| 25 | static inline struct vfsmount *next_slave(struct vfsmount *p) |
| 26 | { |
| 27 | return list_entry(p->mnt_slave.next, struct vfsmount, mnt_slave); |
| 28 | } |
| 29 | |
Ram Pai | a58b0eb | 2005-11-07 17:20:48 -0500 | [diff] [blame] | 30 | static int do_make_slave(struct vfsmount *mnt) |
| 31 | { |
| 32 | struct vfsmount *peer_mnt = mnt, *master = mnt->mnt_master; |
| 33 | struct vfsmount *slave_mnt; |
| 34 | |
| 35 | /* |
| 36 | * slave 'mnt' to a peer mount that has the |
| 37 | * same root dentry. If none is available than |
| 38 | * slave it to anything that is available. |
| 39 | */ |
| 40 | while ((peer_mnt = next_peer(peer_mnt)) != mnt && |
| 41 | peer_mnt->mnt_root != mnt->mnt_root) ; |
| 42 | |
| 43 | if (peer_mnt == mnt) { |
| 44 | peer_mnt = next_peer(mnt); |
| 45 | if (peer_mnt == mnt) |
| 46 | peer_mnt = NULL; |
| 47 | } |
| 48 | list_del_init(&mnt->mnt_share); |
| 49 | |
| 50 | if (peer_mnt) |
| 51 | master = peer_mnt; |
| 52 | |
| 53 | if (master) { |
| 54 | list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave) |
| 55 | slave_mnt->mnt_master = master; |
Akinobu Mita | 1bfba4e | 2006-06-26 00:24:40 -0700 | [diff] [blame] | 56 | list_move(&mnt->mnt_slave, &master->mnt_slave_list); |
Ram Pai | a58b0eb | 2005-11-07 17:20:48 -0500 | [diff] [blame] | 57 | list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev); |
| 58 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
| 59 | } else { |
| 60 | struct list_head *p = &mnt->mnt_slave_list; |
| 61 | while (!list_empty(p)) { |
Pavel Emelianov | b5e6181 | 2007-05-08 00:30:19 -0700 | [diff] [blame^] | 62 | slave_mnt = list_first_entry(p, |
Ram Pai | a58b0eb | 2005-11-07 17:20:48 -0500 | [diff] [blame] | 63 | struct vfsmount, mnt_slave); |
| 64 | list_del_init(&slave_mnt->mnt_slave); |
| 65 | slave_mnt->mnt_master = NULL; |
| 66 | } |
| 67 | } |
| 68 | mnt->mnt_master = master; |
| 69 | CLEAR_MNT_SHARED(mnt); |
| 70 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
| 71 | return 0; |
| 72 | } |
| 73 | |
Ram Pai | 07b2088 | 2005-11-07 17:19:07 -0500 | [diff] [blame] | 74 | void change_mnt_propagation(struct vfsmount *mnt, int type) |
| 75 | { |
Ram Pai | 03e06e6 | 2005-11-07 17:19:33 -0500 | [diff] [blame] | 76 | if (type == MS_SHARED) { |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 77 | set_mnt_shared(mnt); |
Ram Pai | a58b0eb | 2005-11-07 17:20:48 -0500 | [diff] [blame] | 78 | return; |
| 79 | } |
| 80 | do_make_slave(mnt); |
| 81 | if (type != MS_SLAVE) { |
| 82 | list_del_init(&mnt->mnt_slave); |
| 83 | mnt->mnt_master = NULL; |
Ram Pai | 9676f0c | 2005-11-07 17:21:20 -0500 | [diff] [blame] | 84 | if (type == MS_UNBINDABLE) |
| 85 | mnt->mnt_flags |= MNT_UNBINDABLE; |
Ram Pai | 03e06e6 | 2005-11-07 17:19:33 -0500 | [diff] [blame] | 86 | } |
Ram Pai | 07b2088 | 2005-11-07 17:19:07 -0500 | [diff] [blame] | 87 | } |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 88 | |
| 89 | /* |
| 90 | * get the next mount in the propagation tree. |
| 91 | * @m: the mount seen last |
| 92 | * @origin: the original mount from where the tree walk initiated |
| 93 | */ |
| 94 | static struct vfsmount *propagation_next(struct vfsmount *m, |
| 95 | struct vfsmount *origin) |
| 96 | { |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 97 | /* are there any slaves of this mount? */ |
| 98 | if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list)) |
| 99 | return first_slave(m); |
| 100 | |
| 101 | while (1) { |
| 102 | struct vfsmount *next; |
| 103 | struct vfsmount *master = m->mnt_master; |
| 104 | |
Miklos Szeredi | bf066c7 | 2006-01-08 01:03:19 -0800 | [diff] [blame] | 105 | if (master == origin->mnt_master) { |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 106 | next = next_peer(m); |
| 107 | return ((next == origin) ? NULL : next); |
| 108 | } else if (m->mnt_slave.next != &master->mnt_slave_list) |
| 109 | return next_slave(m); |
| 110 | |
| 111 | /* back at master */ |
| 112 | m = master; |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * return the source mount to be used for cloning |
| 118 | * |
| 119 | * @dest the current destination mount |
| 120 | * @last_dest the last seen destination mount |
| 121 | * @last_src the last seen source mount |
| 122 | * @type return CL_SLAVE if the new mount has to be |
| 123 | * cloned as a slave. |
| 124 | */ |
| 125 | static struct vfsmount *get_source(struct vfsmount *dest, |
| 126 | struct vfsmount *last_dest, |
| 127 | struct vfsmount *last_src, |
| 128 | int *type) |
| 129 | { |
| 130 | struct vfsmount *p_last_src = NULL; |
| 131 | struct vfsmount *p_last_dest = NULL; |
Alexey Dobriyan | 53b3531 | 2006-03-24 03:16:13 -0800 | [diff] [blame] | 132 | *type = CL_PROPAGATION; |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 133 | |
| 134 | if (IS_MNT_SHARED(dest)) |
| 135 | *type |= CL_MAKE_SHARED; |
| 136 | |
| 137 | while (last_dest != dest->mnt_master) { |
| 138 | p_last_dest = last_dest; |
| 139 | p_last_src = last_src; |
| 140 | last_dest = last_dest->mnt_master; |
| 141 | last_src = last_src->mnt_master; |
| 142 | } |
| 143 | |
| 144 | if (p_last_dest) { |
| 145 | do { |
| 146 | p_last_dest = next_peer(p_last_dest); |
| 147 | } while (IS_MNT_NEW(p_last_dest)); |
| 148 | } |
| 149 | |
| 150 | if (dest != p_last_dest) { |
| 151 | *type |= CL_SLAVE; |
| 152 | return last_src; |
| 153 | } else |
| 154 | return p_last_src; |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 155 | } |
| 156 | |
| 157 | /* |
| 158 | * mount 'source_mnt' under the destination 'dest_mnt' at |
| 159 | * dentry 'dest_dentry'. And propagate that mount to |
| 160 | * all the peer and slave mounts of 'dest_mnt'. |
| 161 | * Link all the new mounts into a propagation tree headed at |
| 162 | * source_mnt. Also link all the new mounts using ->mnt_list |
| 163 | * headed at source_mnt's ->mnt_list |
| 164 | * |
| 165 | * @dest_mnt: destination mount. |
| 166 | * @dest_dentry: destination dentry. |
| 167 | * @source_mnt: source mount. |
| 168 | * @tree_list : list of heads of trees to be attached. |
| 169 | */ |
| 170 | int propagate_mnt(struct vfsmount *dest_mnt, struct dentry *dest_dentry, |
| 171 | struct vfsmount *source_mnt, struct list_head *tree_list) |
| 172 | { |
| 173 | struct vfsmount *m, *child; |
| 174 | int ret = 0; |
| 175 | struct vfsmount *prev_dest_mnt = dest_mnt; |
| 176 | struct vfsmount *prev_src_mnt = source_mnt; |
| 177 | LIST_HEAD(tmp_list); |
| 178 | LIST_HEAD(umount_list); |
| 179 | |
| 180 | for (m = propagation_next(dest_mnt, dest_mnt); m; |
| 181 | m = propagation_next(m, dest_mnt)) { |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 182 | int type; |
| 183 | struct vfsmount *source; |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 184 | |
| 185 | if (IS_MNT_NEW(m)) |
| 186 | continue; |
| 187 | |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 188 | source = get_source(m, prev_dest_mnt, prev_src_mnt, &type); |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 189 | |
Ram Pai | 5afe002 | 2005-11-07 17:21:01 -0500 | [diff] [blame] | 190 | if (!(child = copy_tree(source, source->mnt_root, type))) { |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 191 | ret = -ENOMEM; |
| 192 | list_splice(tree_list, tmp_list.prev); |
| 193 | goto out; |
| 194 | } |
| 195 | |
| 196 | if (is_subdir(dest_dentry, m->mnt_root)) { |
| 197 | mnt_set_mountpoint(m, dest_dentry, child); |
| 198 | list_add_tail(&child->mnt_hash, tree_list); |
| 199 | } else { |
| 200 | /* |
| 201 | * This can happen if the parent mount was bind mounted |
| 202 | * on some subdirectory of a shared/slave mount. |
| 203 | */ |
| 204 | list_add_tail(&child->mnt_hash, &tmp_list); |
| 205 | } |
| 206 | prev_dest_mnt = m; |
| 207 | prev_src_mnt = child; |
| 208 | } |
| 209 | out: |
| 210 | spin_lock(&vfsmount_lock); |
| 211 | while (!list_empty(&tmp_list)) { |
| 212 | child = list_entry(tmp_list.next, struct vfsmount, mnt_hash); |
| 213 | list_del_init(&child->mnt_hash); |
Ram Pai | a05964f | 2005-11-07 17:20:17 -0500 | [diff] [blame] | 214 | umount_tree(child, 0, &umount_list); |
Ram Pai | b90fa9a | 2005-11-07 17:19:50 -0500 | [diff] [blame] | 215 | } |
| 216 | spin_unlock(&vfsmount_lock); |
| 217 | release_mounts(&umount_list); |
| 218 | return ret; |
| 219 | } |
Ram Pai | a05964f | 2005-11-07 17:20:17 -0500 | [diff] [blame] | 220 | |
| 221 | /* |
| 222 | * return true if the refcount is greater than count |
| 223 | */ |
| 224 | static inline int do_refcount_check(struct vfsmount *mnt, int count) |
| 225 | { |
| 226 | int mycount = atomic_read(&mnt->mnt_count); |
| 227 | return (mycount > count); |
| 228 | } |
| 229 | |
| 230 | /* |
| 231 | * check if the mount 'mnt' can be unmounted successfully. |
| 232 | * @mnt: the mount to be checked for unmount |
| 233 | * NOTE: unmounting 'mnt' would naturally propagate to all |
| 234 | * other mounts its parent propagates to. |
| 235 | * Check if any of these mounts that **do not have submounts** |
| 236 | * have more references than 'refcnt'. If so return busy. |
| 237 | */ |
| 238 | int propagate_mount_busy(struct vfsmount *mnt, int refcnt) |
| 239 | { |
| 240 | struct vfsmount *m, *child; |
| 241 | struct vfsmount *parent = mnt->mnt_parent; |
| 242 | int ret = 0; |
| 243 | |
| 244 | if (mnt == parent) |
| 245 | return do_refcount_check(mnt, refcnt); |
| 246 | |
| 247 | /* |
| 248 | * quickly check if the current mount can be unmounted. |
| 249 | * If not, we don't have to go checking for all other |
| 250 | * mounts |
| 251 | */ |
| 252 | if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt)) |
| 253 | return 1; |
| 254 | |
| 255 | for (m = propagation_next(parent, parent); m; |
| 256 | m = propagation_next(m, parent)) { |
| 257 | child = __lookup_mnt(m, mnt->mnt_mountpoint, 0); |
| 258 | if (child && list_empty(&child->mnt_mounts) && |
| 259 | (ret = do_refcount_check(child, 1))) |
| 260 | break; |
| 261 | } |
| 262 | return ret; |
| 263 | } |
| 264 | |
| 265 | /* |
| 266 | * NOTE: unmounting 'mnt' naturally propagates to all other mounts its |
| 267 | * parent propagates to. |
| 268 | */ |
| 269 | static void __propagate_umount(struct vfsmount *mnt) |
| 270 | { |
| 271 | struct vfsmount *parent = mnt->mnt_parent; |
| 272 | struct vfsmount *m; |
| 273 | |
| 274 | BUG_ON(parent == mnt); |
| 275 | |
| 276 | for (m = propagation_next(parent, parent); m; |
| 277 | m = propagation_next(m, parent)) { |
| 278 | |
| 279 | struct vfsmount *child = __lookup_mnt(m, |
| 280 | mnt->mnt_mountpoint, 0); |
| 281 | /* |
| 282 | * umount the child only if the child has no |
| 283 | * other children |
| 284 | */ |
Akinobu Mita | 1bfba4e | 2006-06-26 00:24:40 -0700 | [diff] [blame] | 285 | if (child && list_empty(&child->mnt_mounts)) |
| 286 | list_move_tail(&child->mnt_hash, &mnt->mnt_hash); |
Ram Pai | a05964f | 2005-11-07 17:20:17 -0500 | [diff] [blame] | 287 | } |
| 288 | } |
| 289 | |
| 290 | /* |
| 291 | * collect all mounts that receive propagation from the mount in @list, |
| 292 | * and return these additional mounts in the same list. |
| 293 | * @list: the list of mounts to be unmounted. |
| 294 | */ |
| 295 | int propagate_umount(struct list_head *list) |
| 296 | { |
| 297 | struct vfsmount *mnt; |
| 298 | |
| 299 | list_for_each_entry(mnt, list, mnt_hash) |
| 300 | __propagate_umount(mnt); |
| 301 | return 0; |
| 302 | } |