| 1 | |
| 2 | * linux/fs/pnode.c
|
| 3 | *
|
| 4 | * (C) Copyright IBM Corporation 2005.
|
| 5 | * Released under GPL v2.
|
| 6 | * Author : Ram Pai ([email protected])
|
| 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 |
|
| 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 |
|
| 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 |
|
| 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 |
|
| 163 | m = master;
|
| 164 | }
|
| 165 | }
|
| 166 |
|
| 167 | |
| 168 | * return the source mount to be used for cloning
|
| 169 | *
|
| 170 | * @dest the current destination mount
|
| 171 | * @last_dest the last seen destination mount
|
| 172 | * @last_src the last seen source mount
|
| 173 | * @type return CL_SLAVE if the new mount has to be
|
| 174 | * cloned as a slave.
|
| 175 |
|
| 176 | static struct mount *get_source(struct mount *dest,
|
| 177 | struct mount *last_dest,
|
| 178 | struct mount *last_src,
|
| 179 | int *type)
|
| 180 | {
|
| 181 | struct mount *p_last_src = NULL;
|
| 182 | struct mount *p_last_dest = NULL;
|
| 183 |
|
| 184 | while (last_dest != dest->mnt_master) {
|
| 185 | p_last_dest = last_dest;
|
| 186 | p_last_src = last_src;
|
| 187 | last_dest = last_dest->mnt_master;
|
| 188 | last_src = last_src->mnt_master;
|
| 189 | }
|
| 190 |
|
| 191 | if (p_last_dest) {
|
| 192 | do {
|
| 193 | p_last_dest = next_peer(p_last_dest);
|
| 194 | } while (IS_MNT_NEW(p_last_dest));
|
| 195 |
|
| 196 | if (dest == p_last_dest) {
|
| 197 | *type = CL_MAKE_SHARED;
|
| 198 | return p_last_src;
|
| 199 | }
|
| 200 | }
|
| 201 |
|
| 202 | *type = CL_SLAVE;
|
| 203 |
|
| 204 | if (IS_MNT_SHARED(dest))
|
| 205 | *type |= CL_MAKE_SHARED;
|
| 206 | return last_src;
|
| 207 | }
|
| 208 |
|
| 209 | |
| 210 | * mount 'source_mnt' under the destination 'dest_mnt' at
|
| 211 | * dentry 'dest_dentry'. And propagate that mount to
|
| 212 | * all the peer and slave mounts of 'dest_mnt'.
|
| 213 | * Link all the new mounts into a propagation tree headed at
|
| 214 | * source_mnt. Also link all the new mounts using ->mnt_list
|
| 215 | * headed at source_mnt's ->mnt_list
|
| 216 | *
|
| 217 | * @dest_mnt: destination mount.
|
| 218 | * @dest_dentry: destination dentry.
|
| 219 | * @source_mnt: source mount.
|
| 220 | * @tree_list : list of heads of trees to be attached.
|
| 221 |
|
| 222 | int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
|
| 223 | struct mount *source_mnt, struct list_head *tree_list)
|
| 224 | {
|
| 225 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
|
| 226 | struct mount *m, *child;
|
| 227 | int ret = 0;
|
| 228 | struct mount *prev_dest_mnt = dest_mnt;
|
| 229 | struct mount *prev_src_mnt = source_mnt;
|
| 230 | LIST_HEAD(tmp_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 |
|
| 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_mp->m_dentry, m->mnt.mnt_root)) {
|
| 254 | mnt_set_mountpoint(m, dest_mp, 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 | lock_mount_hash();
|
| 268 | while (!list_empty(&tmp_list)) {
|
| 269 | child = list_first_entry(&tmp_list, struct mount, mnt_hash);
|
| 270 | umount_tree(child, 0);
|
| 271 | }
|
| 272 | unlock_mount_hash();
|
| 273 | return ret;
|
| 274 | }
|
| 275 |
|
| 276 | |
| 277 | * return true if the refcount is greater than count
|
| 278 |
|
| 279 | static inline int do_refcount_check(struct mount *mnt, int count)
|
| 280 | {
|
| 281 | return mnt_get_count(mnt) > count;
|
| 282 | }
|
| 283 |
|
| 284 | |
| 285 | * check if the mount 'mnt' can be unmounted successfully.
|
| 286 | * @mnt: the mount to be checked for unmount
|
| 287 | * NOTE: unmounting 'mnt' would naturally propagate to all
|
| 288 | * other mounts its parent propagates to.
|
| 289 | * Check if any of these mounts that **do not have submounts**
|
| 290 | * have more references than 'refcnt'. If so return busy.
|
| 291 | *
|
| 292 | * vfsmount lock must be held for write
|
| 293 |
|
| 294 | int propagate_mount_busy(struct mount *mnt, int refcnt)
|
| 295 | {
|
| 296 | struct mount *m, *child;
|
| 297 | struct mount *parent = mnt->mnt_parent;
|
| 298 | int ret = 0;
|
| 299 |
|
| 300 | if (mnt == parent)
|
| 301 | return do_refcount_check(mnt, refcnt);
|
| 302 |
|
| 303 | |
| 304 | * quickly check if the current mount can be unmounted.
|
| 305 | * If not, we don't have to go checking for all other
|
| 306 | * mounts
|
| 307 |
|
| 308 | if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
|
| 309 | return 1;
|
| 310 |
|
| 311 | for (m = propagation_next(parent, parent); m;
|
| 312 | m = propagation_next(m, parent)) {
|
| 313 | child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
|
| 314 | if (child && list_empty(&child->mnt_mounts) &&
|
| 315 | (ret = do_refcount_check(child, 1)))
|
| 316 | break;
|
| 317 | }
|
| 318 | return ret;
|
| 319 | }
|
| 320 |
|
| 321 | |
| 322 | * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
|
| 323 | * parent propagates to.
|
| 324 |
|
| 325 | static void __propagate_umount(struct mount *mnt)
|
| 326 | {
|
| 327 | struct mount *parent = mnt->mnt_parent;
|
| 328 | struct mount *m;
|
| 329 |
|
| 330 | BUG_ON(parent == mnt);
|
| 331 |
|
| 332 | for (m = propagation_next(parent, parent); m;
|
| 333 | m = propagation_next(m, parent)) {
|
| 334 |
|
| 335 | struct mount *child = __lookup_mnt_last(&m->mnt,
|
| 336 | mnt->mnt_mountpoint);
|
| 337 | |
| 338 | * umount the child only if the child has no
|
| 339 | * other children
|
| 340 |
|
| 341 | if (child && list_empty(&child->mnt_mounts))
|
| 342 | list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
|
| 343 | }
|
| 344 | }
|
| 345 |
|
| 346 | |
| 347 | * collect all mounts that receive propagation from the mount in @list,
|
| 348 | * and return these additional mounts in the same list.
|
| 349 | * @list: the list of mounts to be unmounted.
|
| 350 | *
|
| 351 | * vfsmount lock must be held for write
|
| 352 |
|
| 353 | int propagate_umount(struct list_head *list)
|
| 354 | {
|
| 355 | struct mount *mnt;
|
| 356 |
|
| 357 | list_for_each_entry(mnt, list, mnt_hash)
|
| 358 | __propagate_umount(mnt);
|
| 359 | return 0;
|
| 360 | }
|