Revert "ext4: use old interface for ext4_readdir()"
[linux-3.10.git] / fs / namespace.c
index 7641234..d0244c8 100644 (file)
@@ -9,30 +9,20 @@
  */
 
 #include <linux/syscalls.h>
-#include <linux/slab.h>
-#include <linux/sched.h>
-#include <linux/spinlock.h>
-#include <linux/percpu.h>
-#include <linux/init.h>
-#include <linux/kernel.h>
-#include <linux/acct.h>
+#include <linux/export.h>
 #include <linux/capability.h>
-#include <linux/cpumask.h>
-#include <linux/module.h>
-#include <linux/sysfs.h>
-#include <linux/seq_file.h>
 #include <linux/mnt_namespace.h>
+#include <linux/user_namespace.h>
 #include <linux/namei.h>
-#include <linux/nsproxy.h>
 #include <linux/security.h>
-#include <linux/mount.h>
-#include <linux/ramfs.h>
-#include <linux/log2.h>
 #include <linux/idr.h>
-#include <linux/fs_struct.h>
-#include <linux/fsnotify.h>
-#include <asm/uaccess.h>
-#include <asm/unistd.h>
+#include <linux/acct.h>                /* acct_auto_close_mnt */
+#include <linux/ramfs.h>       /* init_rootfs */
+#include <linux/fs_struct.h>   /* get_fs_root et.al. */
+#include <linux/fsnotify.h>    /* fsnotify_vfsmount_delete */
+#include <linux/uaccess.h>
+#include <linux/proc_ns.h>
+#include <linux/magic.h>
 #include "pnode.h"
 #include "internal.h"
 
@@ -47,6 +37,7 @@ static int mnt_id_start = 0;
 static int mnt_group_start = 1;
 
 static struct list_head *mount_hashtable __read_mostly;
+static struct list_head *mountpoint_hashtable __read_mostly;
 static struct kmem_cache *mnt_cache __read_mostly;
 static struct rw_semaphore namespace_sem;
 
@@ -85,9 +76,9 @@ static int mnt_alloc_id(struct mount *mnt)
 retry:
        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
        spin_lock(&mnt_id_lock);
-       res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt.mnt_id);
+       res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
        if (!res)
-               mnt_id_start = mnt->mnt.mnt_id + 1;
+               mnt_id_start = mnt->mnt_id + 1;
        spin_unlock(&mnt_id_lock);
        if (res == -EAGAIN)
                goto retry;
@@ -97,7 +88,7 @@ retry:
 
 static void mnt_free_id(struct mount *mnt)
 {
-       int id = mnt->mnt.mnt_id;
+       int id = mnt->mnt_id;
        spin_lock(&mnt_id_lock);
        ida_remove(&mnt_id_ida, id);
        if (mnt_id_start > id)
@@ -119,9 +110,9 @@ static int mnt_alloc_group_id(struct mount *mnt)
 
        res = ida_get_new_above(&mnt_group_ida,
                                mnt_group_start,
-                               &mnt->mnt.mnt_group_id);
+                               &mnt->mnt_group_id);
        if (!res)
-               mnt_group_start = mnt->mnt.mnt_group_id + 1;
+               mnt_group_start = mnt->mnt_group_id + 1;
 
        return res;
 }
@@ -131,17 +122,17 @@ static int mnt_alloc_group_id(struct mount *mnt)
  */
 void mnt_release_group_id(struct mount *mnt)
 {
-       int id = mnt->mnt.mnt_group_id;
+       int id = mnt->mnt_group_id;
        ida_remove(&mnt_group_ida, id);
        if (mnt_group_start > id)
                mnt_group_start = id;
-       mnt->mnt.mnt_group_id = 0;
+       mnt->mnt_group_id = 0;
 }
 
 /*
  * vfsmount lock must be held for read
  */
-static inline void mnt_add_count(struct vfsmount *mnt, int n)
+static inline void mnt_add_count(struct mount *mnt, int n)
 {
 #ifdef CONFIG_SMP
        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
@@ -155,7 +146,7 @@ static inline void mnt_add_count(struct vfsmount *mnt, int n)
 /*
  * vfsmount lock must be held for write
  */
-unsigned int mnt_get_count(struct vfsmount *mnt)
+unsigned int mnt_get_count(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
        unsigned int count = 0;
@@ -173,12 +164,11 @@ unsigned int mnt_get_count(struct vfsmount *mnt)
 
 static struct mount *alloc_vfsmnt(const char *name)
 {
-       struct mount *p = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
-       if (p) {
-               struct vfsmount *mnt = &p->mnt;
+       struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
+       if (mnt) {
                int err;
 
-               err = mnt_alloc_id(p);
+               err = mnt_alloc_id(mnt);
                if (err)
                        goto out_free_cache;
 
@@ -199,7 +189,7 @@ static struct mount *alloc_vfsmnt(const char *name)
                mnt->mnt_writers = 0;
 #endif
 
-               INIT_LIST_HEAD(&p->mnt.mnt_hash);
+               INIT_LIST_HEAD(&mnt->mnt_hash);
                INIT_LIST_HEAD(&mnt->mnt_child);
                INIT_LIST_HEAD(&mnt->mnt_mounts);
                INIT_LIST_HEAD(&mnt->mnt_list);
@@ -211,16 +201,16 @@ static struct mount *alloc_vfsmnt(const char *name)
                INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
 #endif
        }
-       return p;
+       return mnt;
 
 #ifdef CONFIG_SMP
 out_free_devname:
-       kfree(p->mnt.mnt_devname);
+       kfree(mnt->mnt_devname);
 #endif
 out_free_id:
-       mnt_free_id(p);
+       mnt_free_id(mnt);
 out_free_cache:
-       kmem_cache_free(mnt_cache, p);
+       kmem_cache_free(mnt_cache, mnt);
        return NULL;
 }
 
@@ -253,7 +243,7 @@ int __mnt_is_readonly(struct vfsmount *mnt)
 }
 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 
-static inline void mnt_inc_writers(struct vfsmount *mnt)
+static inline void mnt_inc_writers(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
@@ -262,7 +252,7 @@ static inline void mnt_inc_writers(struct vfsmount *mnt)
 #endif
 }
 
-static inline void mnt_dec_writers(struct vfsmount *mnt)
+static inline void mnt_dec_writers(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
@@ -271,7 +261,7 @@ static inline void mnt_dec_writers(struct vfsmount *mnt)
 #endif
 }
 
-static unsigned int mnt_get_writers(struct vfsmount *mnt)
+static unsigned int mnt_get_writers(struct mount *mnt)
 {
 #ifdef CONFIG_SMP
        unsigned int count = 0;
@@ -287,26 +277,34 @@ static unsigned int mnt_get_writers(struct vfsmount *mnt)
 #endif
 }
 
+static int mnt_is_readonly(struct vfsmount *mnt)
+{
+       if (mnt->mnt_sb->s_readonly_remount)
+               return 1;
+       /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
+       smp_rmb();
+       return __mnt_is_readonly(mnt);
+}
+
 /*
- * Most r/o checks on a fs are for operations that take
- * discrete amounts of time, like a write() or unlink().
- * We must keep track of when those operations start
- * (for permission checks) and when they end, so that
- * we can determine when writes are able to occur to
- * a filesystem.
+ * Most r/o & frozen checks on a fs are for operations that take discrete
+ * amounts of time, like a write() or unlink().  We must keep track of when
+ * those operations start (for permission checks) and when they end, so that we
+ * can determine when writes are able to occur to a filesystem.
  */
 /**
- * mnt_want_write - get write access to a mount
- * @mnt: the mount on which to take a write
+ * __mnt_want_write - get write access to a mount without freeze protection
+ * @m: the mount on which to take a write
  *
- * This tells the low-level filesystem that a write is
- * about to be performed to it, and makes sure that
- * writes are allowed before returning success.  When
- * the write operation is finished, mnt_drop_write()
- * must be called.  This is effectively a refcount.
+ * This tells the low-level filesystem that a write is about to be performed to
+ * it, and makes sure that writes are allowed (mnt it read-write) before
+ * returning success. This operation does not protect against filesystem being
+ * frozen. When the write operation is finished, __mnt_drop_write() must be
+ * called. This is effectively a refcount.
  */
-int mnt_want_write(struct vfsmount *mnt)
+int __mnt_want_write(struct vfsmount *m)
 {
+       struct mount *mnt = real_mount(m);
        int ret = 0;
 
        preempt_disable();
@@ -317,7 +315,7 @@ int mnt_want_write(struct vfsmount *mnt)
         * incremented count after it has set MNT_WRITE_HOLD.
         */
        smp_mb();
-       while (mnt->mnt_flags & MNT_WRITE_HOLD)
+       while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
                cpu_relax();
        /*
         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
@@ -325,13 +323,32 @@ int mnt_want_write(struct vfsmount *mnt)
         * MNT_WRITE_HOLD is cleared.
         */
        smp_rmb();
-       if (__mnt_is_readonly(mnt)) {
+       if (mnt_is_readonly(m)) {
                mnt_dec_writers(mnt);
                ret = -EROFS;
-               goto out;
        }
-out:
        preempt_enable();
+
+       return ret;
+}
+
+/**
+ * mnt_want_write - get write access to a mount
+ * @m: the mount on which to take a write
+ *
+ * This tells the low-level filesystem that a write is about to be performed to
+ * it, and makes sure that writes are allowed (mount is read-write, filesystem
+ * is not frozen) before returning success.  When the write operation is
+ * finished, mnt_drop_write() must be called.  This is effectively a refcount.
+ */
+int mnt_want_write(struct vfsmount *m)
+{
+       int ret;
+
+       sb_start_write(m->mnt_sb);
+       ret = __mnt_want_write(m);
+       if (ret)
+               sb_end_write(m->mnt_sb);
        return ret;
 }
 EXPORT_SYMBOL_GPL(mnt_want_write);
@@ -354,57 +371,95 @@ int mnt_clone_write(struct vfsmount *mnt)
        if (__mnt_is_readonly(mnt))
                return -EROFS;
        preempt_disable();
-       mnt_inc_writers(mnt);
+       mnt_inc_writers(real_mount(mnt));
        preempt_enable();
        return 0;
 }
 EXPORT_SYMBOL_GPL(mnt_clone_write);
 
 /**
- * mnt_want_write_file - get write access to a file's mount
+ * __mnt_want_write_file - get write access to a file's mount
  * @file: the file who's mount on which to take a write
  *
- * This is like mnt_want_write, but it takes a file and can
+ * This is like __mnt_want_write, but it takes a file and can
  * do some optimisations if the file is open for write already
  */
-int mnt_want_write_file(struct file *file)
+int __mnt_want_write_file(struct file *file)
 {
-       struct inode *inode = file->f_dentry->d_inode;
+       struct inode *inode = file_inode(file);
+
        if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
-               return mnt_want_write(file->f_path.mnt);
+               return __mnt_want_write(file->f_path.mnt);
        else
                return mnt_clone_write(file->f_path.mnt);
 }
+
+/**
+ * mnt_want_write_file - get write access to a file's mount
+ * @file: the file who's mount on which to take a write
+ *
+ * This is like mnt_want_write, but it takes a file and can
+ * do some optimisations if the file is open for write already
+ */
+int mnt_want_write_file(struct file *file)
+{
+       int ret;
+
+       sb_start_write(file->f_path.mnt->mnt_sb);
+       ret = __mnt_want_write_file(file);
+       if (ret)
+               sb_end_write(file->f_path.mnt->mnt_sb);
+       return ret;
+}
 EXPORT_SYMBOL_GPL(mnt_want_write_file);
 
 /**
- * mnt_drop_write - give up write access to a mount
+ * __mnt_drop_write - give up write access to a mount
  * @mnt: the mount on which to give up write access
  *
  * Tells the low-level filesystem that we are done
  * performing writes to it.  Must be matched with
- * mnt_want_write() call above.
+ * __mnt_want_write() call above.
  */
-void mnt_drop_write(struct vfsmount *mnt)
+void __mnt_drop_write(struct vfsmount *mnt)
 {
        preempt_disable();
-       mnt_dec_writers(mnt);
+       mnt_dec_writers(real_mount(mnt));
        preempt_enable();
 }
+
+/**
+ * mnt_drop_write - give up write access to a mount
+ * @mnt: the mount on which to give up write access
+ *
+ * Tells the low-level filesystem that we are done performing writes to it and
+ * also allows filesystem to be frozen again.  Must be matched with
+ * mnt_want_write() call above.
+ */
+void mnt_drop_write(struct vfsmount *mnt)
+{
+       __mnt_drop_write(mnt);
+       sb_end_write(mnt->mnt_sb);
+}
 EXPORT_SYMBOL_GPL(mnt_drop_write);
 
+void __mnt_drop_write_file(struct file *file)
+{
+       __mnt_drop_write(file->f_path.mnt);
+}
+
 void mnt_drop_write_file(struct file *file)
 {
        mnt_drop_write(file->f_path.mnt);
 }
 EXPORT_SYMBOL(mnt_drop_write_file);
 
-static int mnt_make_readonly(struct vfsmount *mnt)
+static int mnt_make_readonly(struct mount *mnt)
 {
        int ret = 0;
 
-       br_write_lock(vfsmount_lock);
-       mnt->mnt_flags |= MNT_WRITE_HOLD;
+       br_write_lock(&vfsmount_lock);
+       mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
        /*
         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
         * should be visible before we do.
@@ -430,30 +485,66 @@ static int mnt_make_readonly(struct vfsmount *mnt)
        if (mnt_get_writers(mnt) > 0)
                ret = -EBUSY;
        else
-               mnt->mnt_flags |= MNT_READONLY;
+               mnt->mnt.mnt_flags |= MNT_READONLY;
        /*
         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
         * that become unheld will see MNT_READONLY.
         */
        smp_wmb();
-       mnt->mnt_flags &= ~MNT_WRITE_HOLD;
-       br_write_unlock(vfsmount_lock);
+       mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
+       br_write_unlock(&vfsmount_lock);
        return ret;
 }
 
-static void __mnt_unmake_readonly(struct vfsmount *mnt)
+static void __mnt_unmake_readonly(struct mount *mnt)
 {
-       br_write_lock(vfsmount_lock);
-       mnt->mnt_flags &= ~MNT_READONLY;
-       br_write_unlock(vfsmount_lock);
+       br_write_lock(&vfsmount_lock);
+       mnt->mnt.mnt_flags &= ~MNT_READONLY;
+       br_write_unlock(&vfsmount_lock);
+}
+
+int sb_prepare_remount_readonly(struct super_block *sb)
+{
+       struct mount *mnt;
+       int err = 0;
+
+       /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
+       if (atomic_long_read(&sb->s_remove_count))
+               return -EBUSY;
+
+       br_write_lock(&vfsmount_lock);
+       list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
+               if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
+                       mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
+                       smp_mb();
+                       if (mnt_get_writers(mnt) > 0) {
+                               err = -EBUSY;
+                               break;
+                       }
+               }
+       }
+       if (!err && atomic_long_read(&sb->s_remove_count))
+               err = -EBUSY;
+
+       if (!err) {
+               sb->s_readonly_remount = 1;
+               smp_wmb();
+       }
+       list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
+               if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
+                       mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
+       }
+       br_write_unlock(&vfsmount_lock);
+
+       return err;
 }
 
 static void free_vfsmnt(struct mount *mnt)
 {
-       kfree(mnt->mnt.mnt_devname);
+       kfree(mnt->mnt_devname);
        mnt_free_id(mnt);
 #ifdef CONFIG_SMP
-       free_percpu(mnt->mnt.mnt_pcp);
+       free_percpu(mnt->mnt_pcp);
 #endif
        kmem_cache_free(mnt_cache, mnt);
 }
@@ -475,8 +566,8 @@ struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
                p = NULL;
                if (tmp == head)
                        break;
-               p = list_entry(tmp, struct mount, mnt.mnt_hash);
-               if (p->mnt.mnt_parent == mnt && p->mnt.mnt_mountpoint == dentry) {
+               p = list_entry(tmp, struct mount, mnt_hash);
+               if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
                        found = p;
                        break;
                }
@@ -485,26 +576,83 @@ struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
 }
 
 /*
- * lookup_mnt increments the ref count before returning
- * the vfsmount struct.
+ * lookup_mnt - Return the first child mount mounted at path
+ *
+ * "First" means first mounted chronologically.  If you create the
+ * following mounts:
+ *
+ * mount /dev/sda1 /mnt
+ * mount /dev/sda2 /mnt
+ * mount /dev/sda3 /mnt
+ *
+ * Then lookup_mnt() on the base /mnt dentry in the root mount will
+ * return successively the root dentry and vfsmount of /dev/sda1, then
+ * /dev/sda2, then /dev/sda3, then NULL.
+ *
+ * lookup_mnt takes a reference to the found vfsmount.
  */
 struct vfsmount *lookup_mnt(struct path *path)
 {
        struct mount *child_mnt;
 
-       br_read_lock(vfsmount_lock);
+       br_read_lock(&vfsmount_lock);
        child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
        if (child_mnt) {
                mnt_add_count(child_mnt, 1);
-               br_read_unlock(vfsmount_lock);
+               br_read_unlock(&vfsmount_lock);
                return &child_mnt->mnt;
        } else {
-               br_read_unlock(vfsmount_lock);
+               br_read_unlock(&vfsmount_lock);
                return NULL;
        }
 }
 
-static inline int check_mnt(struct vfsmount *mnt)
+static struct mountpoint *new_mountpoint(struct dentry *dentry)
+{
+       struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
+       struct mountpoint *mp;
+
+       list_for_each_entry(mp, chain, m_hash) {
+               if (mp->m_dentry == dentry) {
+                       /* might be worth a WARN_ON() */
+                       if (d_unlinked(dentry))
+                               return ERR_PTR(-ENOENT);
+                       mp->m_count++;
+                       return mp;
+               }
+       }
+
+       mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
+       if (!mp)
+               return ERR_PTR(-ENOMEM);
+
+       spin_lock(&dentry->d_lock);
+       if (d_unlinked(dentry)) {
+               spin_unlock(&dentry->d_lock);
+               kfree(mp);
+               return ERR_PTR(-ENOENT);
+       }
+       dentry->d_flags |= DCACHE_MOUNTED;
+       spin_unlock(&dentry->d_lock);
+       mp->m_dentry = dentry;
+       mp->m_count = 1;
+       list_add(&mp->m_hash, chain);
+       return mp;
+}
+
+static void put_mountpoint(struct mountpoint *mp)
+{
+       if (!--mp->m_count) {
+               struct dentry *dentry = mp->m_dentry;
+               spin_lock(&dentry->d_lock);
+               dentry->d_flags &= ~DCACHE_MOUNTED;
+               spin_unlock(&dentry->d_lock);
+               list_del(&mp->m_hash);
+               kfree(mp);
+       }
+}
+
+static inline int check_mnt(struct mount *mnt)
 {
        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 }
@@ -532,77 +680,45 @@ static void __touch_mnt_namespace(struct mnt_namespace *ns)
 }
 
 /*
- * Clear dentry's mounted state if it has no remaining mounts.
- * vfsmount_lock must be held for write.
- */
-static void dentry_reset_mounted(struct dentry *dentry)
-{
-       unsigned u;
-
-       for (u = 0; u < HASH_SIZE; u++) {
-               struct mount *p;
-
-               list_for_each_entry(p, &mount_hashtable[u], mnt.mnt_hash) {
-                       if (p->mnt.mnt_mountpoint == dentry)
-                               return;
-               }
-       }
-       spin_lock(&dentry->d_lock);
-       dentry->d_flags &= ~DCACHE_MOUNTED;
-       spin_unlock(&dentry->d_lock);
-}
-
-/*
  * vfsmount lock must be held for write
  */
 static void detach_mnt(struct mount *mnt, struct path *old_path)
 {
-       old_path->dentry = mnt->mnt.mnt_mountpoint;
-       old_path->mnt = mnt->mnt.mnt_parent;
-       mnt->mnt.mnt_parent = &mnt->mnt;
-       mnt->mnt.mnt_mountpoint = mnt->mnt.mnt_root;
-       list_del_init(&mnt->mnt.mnt_child);
-       list_del_init(&mnt->mnt.mnt_hash);
-       dentry_reset_mounted(old_path->dentry);
+       old_path->dentry = mnt->mnt_mountpoint;
+       old_path->mnt = &mnt->mnt_parent->mnt;
+       mnt->mnt_parent = mnt;
+       mnt->mnt_mountpoint = mnt->mnt.mnt_root;
+       list_del_init(&mnt->mnt_child);
+       list_del_init(&mnt->mnt_hash);
+       put_mountpoint(mnt->mnt_mp);
+       mnt->mnt_mp = NULL;
 }
 
 /*
  * vfsmount lock must be held for write
  */
-void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
-                       struct vfsmount *child_mnt)
+void mnt_set_mountpoint(struct mount *mnt,
+                       struct mountpoint *mp,
+                       struct mount *child_mnt)
 {
-       child_mnt->mnt_parent = mntget(mnt);
-       child_mnt->mnt_mountpoint = dget(dentry);
-       spin_lock(&dentry->d_lock);
-       dentry->d_flags |= DCACHE_MOUNTED;
-       spin_unlock(&dentry->d_lock);
+       mp->m_count++;
+       mnt_add_count(mnt, 1);  /* essentially, that's mntget */
+       child_mnt->mnt_mountpoint = dget(mp->m_dentry);
+       child_mnt->mnt_parent = mnt;
+       child_mnt->mnt_mp = mp;
 }
 
 /*
  * vfsmount lock must be held for write
  */
-static void attach_mnt(struct mount *mnt, struct path *path)
-{
-       mnt_set_mountpoint(path->mnt, path->dentry, &mnt->mnt);
-       list_add_tail(&mnt->mnt.mnt_hash, mount_hashtable +
-                       hash(path->mnt, path->dentry));
-       list_add_tail(&mnt->mnt.mnt_child, &path->mnt->mnt_mounts);
-}
-
-static inline void __mnt_make_longterm(struct vfsmount *mnt)
+static void attach_mnt(struct mount *mnt,
+                       struct mount *parent,
+                       struct mountpoint *mp)
 {
-#ifdef CONFIG_SMP
-       atomic_inc(&mnt->mnt_longterm);
-#endif
-}
-
-/* needs vfsmount lock for write */
-static inline void __mnt_make_shortterm(struct vfsmount *mnt)
-{
-#ifdef CONFIG_SMP
-       atomic_dec(&mnt->mnt_longterm);
-#endif
+       mnt_set_mountpoint(parent, mp, mnt);
+       list_add_tail(&mnt->mnt_hash, mount_hashtable +
+                       hash(&parent->mnt, mp->m_dentry));
+       list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 }
 
 /*
@@ -610,49 +726,47 @@ static inline void __mnt_make_shortterm(struct vfsmount *mnt)
  */
 static void commit_tree(struct mount *mnt)
 {
-       struct vfsmount *parent = mnt->mnt.mnt_parent;
-       struct vfsmount *m;
+       struct mount *parent = mnt->mnt_parent;
+       struct mount *m;
        LIST_HEAD(head);
        struct mnt_namespace *n = parent->mnt_ns;
 
-       BUG_ON(parent == &mnt->mnt);
+       BUG_ON(parent == mnt);
 
-       list_add_tail(&head, &mnt->mnt.mnt_list);
-       list_for_each_entry(m, &head, mnt_list) {
+       list_add_tail(&head, &mnt->mnt_list);
+       list_for_each_entry(m, &head, mnt_list)
                m->mnt_ns = n;
-               __mnt_make_longterm(m);
-       }
 
        list_splice(&head, n->list.prev);
 
-       list_add_tail(&mnt->mnt.mnt_hash, mount_hashtable +
-                               hash(parent, mnt->mnt.mnt_mountpoint));
-       list_add_tail(&mnt->mnt.mnt_child, &parent->mnt_mounts);
+       list_add_tail(&mnt->mnt_hash, mount_hashtable +
+                               hash(&parent->mnt, mnt->mnt_mountpoint));
+       list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
        touch_mnt_namespace(n);
 }
 
-static struct mount *next_mnt(struct mount *p, struct vfsmount *root)
+static struct mount *next_mnt(struct mount *p, struct mount *root)
 {
-       struct list_head *next = p->mnt.mnt_mounts.next;
-       if (next == &p->mnt.mnt_mounts) {
+       struct list_head *next = p->mnt_mounts.next;
+       if (next == &p->mnt_mounts) {
                while (1) {
-                       if (&p->mnt == root)
+                       if (p == root)
                                return NULL;
-                       next = p->mnt.mnt_child.next;
-                       if (next != &p->mnt.mnt_parent->mnt_mounts)
+                       next = p->mnt_child.next;
+                       if (next != &p->mnt_parent->mnt_mounts)
                                break;
-                       p = real_mount(p->mnt.mnt_parent);
+                       p = p->mnt_parent;
                }
        }
-       return list_entry(next, struct mount, mnt.mnt_child);
+       return list_entry(next, struct mount, mnt_child);
 }
 
 static struct mount *skip_mnt_tree(struct mount *p)
 {
-       struct list_head *prev = p->mnt.mnt_mounts.prev;
-       while (prev != &p->mnt.mnt_mounts) {
-               p = list_entry(prev, struct mount, mnt.mnt_child);
-               prev = p->mnt.mnt_mounts.prev;
+       struct list_head *prev = p->mnt_mounts.prev;
+       while (prev != &p->mnt_mounts) {
+               p = list_entry(prev, struct mount, mnt_child);
+               prev = p->mnt_mounts.prev;
        }
        return p;
 }
@@ -681,68 +795,97 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
 
        mnt->mnt.mnt_root = root;
        mnt->mnt.mnt_sb = root->d_sb;
-       mnt->mnt.mnt_mountpoint = mnt->mnt.mnt_root;
-       mnt->mnt.mnt_parent = &mnt->mnt;
+       mnt->mnt_mountpoint = mnt->mnt.mnt_root;
+       mnt->mnt_parent = mnt;
+       br_write_lock(&vfsmount_lock);
+       list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
+       br_write_unlock(&vfsmount_lock);
        return &mnt->mnt;
 }
 EXPORT_SYMBOL_GPL(vfs_kern_mount);
 
-static struct mount *clone_mnt(struct vfsmount *old, struct dentry *root,
+static struct mount *clone_mnt(struct mount *old, struct dentry *root,
                                        int flag)
 {
-       struct super_block *sb = old->mnt_sb;
-       struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
+       struct super_block *sb = old->mnt.mnt_sb;
+       struct mount *mnt;
+       int err;
 
-       if (mnt) {
-               if (flag & (CL_SLAVE | CL_PRIVATE))
-                       mnt->mnt.mnt_group_id = 0; /* not a peer of original */
-               else
-                       mnt->mnt.mnt_group_id = old->mnt_group_id;
-
-               if ((flag & CL_MAKE_SHARED) && !mnt->mnt.mnt_group_id) {
-                       int err = mnt_alloc_group_id(mnt);
-                       if (err)
-                               goto out_free;
-               }
+       mnt = alloc_vfsmnt(old->mnt_devname);
+       if (!mnt)
+               return ERR_PTR(-ENOMEM);
 
-               mnt->mnt.mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
-               atomic_inc(&sb->s_active);
-               mnt->mnt.mnt_sb = sb;
-               mnt->mnt.mnt_root = dget(root);
-               mnt->mnt.mnt_mountpoint = mnt->mnt.mnt_root;
-               mnt->mnt.mnt_parent = &mnt->mnt;
-
-               if (flag & CL_SLAVE) {
-                       list_add(&mnt->mnt.mnt_slave, &old->mnt_slave_list);
-                       mnt->mnt.mnt_master = old;
-                       CLEAR_MNT_SHARED(&mnt->mnt);
-               } else if (!(flag & CL_PRIVATE)) {
-                       if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
-                               list_add(&mnt->mnt.mnt_share, &old->mnt_share);
-                       if (IS_MNT_SLAVE(old))
-                               list_add(&mnt->mnt.mnt_slave, &old->mnt_slave);
-                       mnt->mnt.mnt_master = old->mnt_master;
-               }
-               if (flag & CL_MAKE_SHARED)
-                       set_mnt_shared(mnt);
-
-               /* stick the duplicate mount on the same expiry list
-                * as the original if that was on one */
-               if (flag & CL_EXPIRE) {
-                       if (!list_empty(&old->mnt_expire))
-                               list_add(&mnt->mnt.mnt_expire, &old->mnt_expire);
-               }
+       if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
+               mnt->mnt_group_id = 0; /* not a peer of original */
+       else
+               mnt->mnt_group_id = old->mnt_group_id;
+
+       if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
+               err = mnt_alloc_group_id(mnt);
+               if (err)
+                       goto out_free;
+       }
+
+       mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
+       /* Don't allow unprivileged users to change mount flags */
+       if (flag & CL_UNPRIVILEGED) {
+               mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
+
+               if (mnt->mnt.mnt_flags & MNT_READONLY)
+                       mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
+
+               if (mnt->mnt.mnt_flags & MNT_NODEV)
+                       mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
+
+               if (mnt->mnt.mnt_flags & MNT_NOSUID)
+                       mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
+
+               if (mnt->mnt.mnt_flags & MNT_NOEXEC)
+                       mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
+       }
+
+       atomic_inc(&sb->s_active);
+       mnt->mnt.mnt_sb = sb;
+       mnt->mnt.mnt_root = dget(root);
+       mnt->mnt_mountpoint = mnt->mnt.mnt_root;
+       mnt->mnt_parent = mnt;
+       br_write_lock(&vfsmount_lock);
+       list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
+       br_write_unlock(&vfsmount_lock);
+
+       if ((flag & CL_SLAVE) ||
+           ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
+               list_add(&mnt->mnt_slave, &old->mnt_slave_list);
+               mnt->mnt_master = old;
+               CLEAR_MNT_SHARED(mnt);
+       } else if (!(flag & CL_PRIVATE)) {
+               if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
+                       list_add(&mnt->mnt_share, &old->mnt_share);
+               if (IS_MNT_SLAVE(old))
+                       list_add(&mnt->mnt_slave, &old->mnt_slave);
+               mnt->mnt_master = old->mnt_master;
+       }
+       if (flag & CL_MAKE_SHARED)
+               set_mnt_shared(mnt);
+
+       /* stick the duplicate mount on the same expiry list
+        * as the original if that was on one */
+       if (flag & CL_EXPIRE) {
+               if (!list_empty(&old->mnt_expire))
+                       list_add(&mnt->mnt_expire, &old->mnt_expire);
        }
+
        return mnt;
 
  out_free:
        free_vfsmnt(mnt);
-       return NULL;
+       return ERR_PTR(err);
 }
 
-static inline void mntfree(struct vfsmount *mnt)
+static inline void mntfree(struct mount *mnt)
 {
-       struct super_block *sb = mnt->mnt_sb;
+       struct vfsmount *m = &mnt->mnt;
+       struct super_block *sb = m->mnt_sb;
 
        /*
         * This probably indicates that somebody messed
@@ -755,54 +898,58 @@ static inline void mntfree(struct vfsmount *mnt)
         * so mnt_get_writers() below is safe.
         */
        WARN_ON(mnt_get_writers(mnt));
-       fsnotify_vfsmount_delete(mnt);
-       dput(mnt->mnt_root);
-       free_vfsmnt(real_mount(mnt));
+       fsnotify_vfsmount_delete(m);
+       dput(m->mnt_root);
+       free_vfsmnt(mnt);
        deactivate_super(sb);
 }
 
-static void mntput_no_expire(struct vfsmount *mnt)
+static void mntput_no_expire(struct mount *mnt)
 {
 put_again:
 #ifdef CONFIG_SMP
-       br_read_lock(vfsmount_lock);
-       if (likely(atomic_read(&mnt->mnt_longterm))) {
+       br_read_lock(&vfsmount_lock);
+       if (likely(mnt->mnt_ns)) {
+               /* shouldn't be the last one */
                mnt_add_count(mnt, -1);
-               br_read_unlock(vfsmount_lock);
+               br_read_unlock(&vfsmount_lock);
                return;
        }
-       br_read_unlock(vfsmount_lock);
+       br_read_unlock(&vfsmount_lock);
 
-       br_write_lock(vfsmount_lock);
+       br_write_lock(&vfsmount_lock);
        mnt_add_count(mnt, -1);
        if (mnt_get_count(mnt)) {
-               br_write_unlock(vfsmount_lock);
+               br_write_unlock(&vfsmount_lock);
                return;
        }
 #else
        mnt_add_count(mnt, -1);
        if (likely(mnt_get_count(mnt)))
                return;
-       br_write_lock(vfsmount_lock);
+       br_write_lock(&vfsmount_lock);
 #endif
        if (unlikely(mnt->mnt_pinned)) {
                mnt_add_count(mnt, mnt->mnt_pinned + 1);
                mnt->mnt_pinned = 0;
-               br_write_unlock(vfsmount_lock);
-               acct_auto_close_mnt(mnt);
+               br_write_unlock(&vfsmount_lock);
+               acct_auto_close_mnt(&mnt->mnt);
                goto put_again;
        }
-       br_write_unlock(vfsmount_lock);
+
+       list_del(&mnt->mnt_instance);
+       br_write_unlock(&vfsmount_lock);
        mntfree(mnt);
 }
 
 void mntput(struct vfsmount *mnt)
 {
        if (mnt) {
+               struct mount *m = real_mount(mnt);
                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
-               if (unlikely(mnt->mnt_expiry_mark))
-                       mnt->mnt_expiry_mark = 0;
-               mntput_no_expire(mnt);
+               if (unlikely(m->mnt_expiry_mark))
+                       m->mnt_expiry_mark = 0;
+               mntput_no_expire(m);
        }
 }
 EXPORT_SYMBOL(mntput);
@@ -810,27 +957,28 @@ EXPORT_SYMBOL(mntput);
 struct vfsmount *mntget(struct vfsmount *mnt)
 {
        if (mnt)
-               mnt_add_count(mnt, 1);
+               mnt_add_count(real_mount(mnt), 1);
        return mnt;
 }
 EXPORT_SYMBOL(mntget);
 
 void mnt_pin(struct vfsmount *mnt)
 {
-       br_write_lock(vfsmount_lock);
-       mnt->mnt_pinned++;
-       br_write_unlock(vfsmount_lock);
+       br_write_lock(&vfsmount_lock);
+       real_mount(mnt)->mnt_pinned++;
+       br_write_unlock(&vfsmount_lock);
 }
 EXPORT_SYMBOL(mnt_pin);
 
-void mnt_unpin(struct vfsmount *mnt)
+void mnt_unpin(struct vfsmount *m)
 {
-       br_write_lock(vfsmount_lock);
+       struct mount *mnt = real_mount(m);
+       br_write_lock(&vfsmount_lock);
        if (mnt->mnt_pinned) {
                mnt_add_count(mnt, 1);
                mnt->mnt_pinned--;
        }
-       br_write_unlock(vfsmount_lock);
+       br_write_unlock(&vfsmount_lock);
 }
 EXPORT_SYMBOL(mnt_unpin);
 
@@ -845,12 +993,12 @@ static inline void mangle(struct seq_file *m, const char *s)
  *
  * See also save_mount_options().
  */
-int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
+int generic_show_options(struct seq_file *m, struct dentry *root)
 {
        const char *options;
 
        rcu_read_lock();
-       options = rcu_dereference(mnt->mnt_sb->s_options);
+       options = rcu_dereference(root->d_sb->s_options);
 
        if (options != NULL && options[0]) {
                seq_putc(m, ',');
@@ -894,10 +1042,10 @@ void replace_mount_options(struct super_block *sb, char *options)
 EXPORT_SYMBOL(replace_mount_options);
 
 #ifdef CONFIG_PROC_FS
-/* iterator */
+/* iterator; we want it to have access to namespace_sem, thus here... */
 static void *m_start(struct seq_file *m, loff_t *pos)
 {
-       struct proc_mounts *p = m->private;
+       struct proc_mounts *p = proc_mounts(m);
 
        down_read(&namespace_sem);
        return seq_list_start(&p->ns->list, *pos);
@@ -905,7 +1053,7 @@ static void *m_start(struct seq_file *m, loff_t *pos)
 
 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       struct proc_mounts *p = m->private;
+       struct proc_mounts *p = proc_mounts(m);
 
        return seq_list_next(v, &p->ns->list, pos);
 }
@@ -915,219 +1063,18 @@ static void m_stop(struct seq_file *m, void *v)
        up_read(&namespace_sem);
 }
 
-int mnt_had_events(struct proc_mounts *p)
+static int m_show(struct seq_file *m, void *v)
 {
-       struct mnt_namespace *ns = p->ns;
-       int res = 0;
-
-       br_read_lock(vfsmount_lock);
-       if (p->m.poll_event != ns->event) {
-               p->m.poll_event = ns->event;
-               res = 1;
-       }
-       br_read_unlock(vfsmount_lock);
-
-       return res;
-}
-
-struct proc_fs_info {
-       int flag;
-       const char *str;
-};
-
-static int show_sb_opts(struct seq_file *m, struct super_block *sb)
-{
-       static const struct proc_fs_info fs_info[] = {
-               { MS_SYNCHRONOUS, ",sync" },
-               { MS_DIRSYNC, ",dirsync" },
-               { MS_MANDLOCK, ",mand" },
-               { 0, NULL }
-       };
-       const struct proc_fs_info *fs_infop;
-
-       for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
-               if (sb->s_flags & fs_infop->flag)
-                       seq_puts(m, fs_infop->str);
-       }
-
-       return security_sb_show_options(m, sb);
-}
-
-static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
-{
-       static const struct proc_fs_info mnt_info[] = {
-               { MNT_NOSUID, ",nosuid" },
-               { MNT_NODEV, ",nodev" },
-               { MNT_NOEXEC, ",noexec" },
-               { MNT_NOATIME, ",noatime" },
-               { MNT_NODIRATIME, ",nodiratime" },
-               { MNT_RELATIME, ",relatime" },
-               { 0, NULL }
-       };
-       const struct proc_fs_info *fs_infop;
-
-       for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
-               if (mnt->mnt_flags & fs_infop->flag)
-                       seq_puts(m, fs_infop->str);
-       }
-}
-
-static void show_type(struct seq_file *m, struct super_block *sb)
-{
-       mangle(m, sb->s_type->name);
-       if (sb->s_subtype && sb->s_subtype[0]) {
-               seq_putc(m, '.');
-               mangle(m, sb->s_subtype);
-       }
-}
-
-static int show_vfsmnt(struct seq_file *m, void *v)
-{
-       struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
-       int err = 0;
-       struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
-
-       if (mnt->mnt_sb->s_op->show_devname) {
-               err = mnt->mnt_sb->s_op->show_devname(m, mnt);
-               if (err)
-                       goto out;
-       } else {
-               mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
-       }
-       seq_putc(m, ' ');
-       seq_path(m, &mnt_path, " \t\n\\");
-       seq_putc(m, ' ');
-       show_type(m, mnt->mnt_sb);
-       seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
-       err = show_sb_opts(m, mnt->mnt_sb);
-       if (err)
-               goto out;
-       show_mnt_opts(m, mnt);
-       if (mnt->mnt_sb->s_op->show_options)
-               err = mnt->mnt_sb->s_op->show_options(m, mnt);
-       seq_puts(m, " 0 0\n");
-out:
-       return err;
+       struct proc_mounts *p = proc_mounts(m);
+       struct mount *r = list_entry(v, struct mount, mnt_list);
+       return p->show(m, &r->mnt);
 }
 
 const struct seq_operations mounts_op = {
        .start  = m_start,
        .next   = m_next,
        .stop   = m_stop,
-       .show   = show_vfsmnt
-};
-
-static int show_mountinfo(struct seq_file *m, void *v)
-{
-       struct proc_mounts *p = m->private;
-       struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
-       struct super_block *sb = mnt->mnt_sb;
-       struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
-       struct path root = p->root;
-       int err = 0;
-
-       seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
-                  MAJOR(sb->s_dev), MINOR(sb->s_dev));
-       if (sb->s_op->show_path)
-               err = sb->s_op->show_path(m, mnt);
-       else
-               seq_dentry(m, mnt->mnt_root, " \t\n\\");
-       if (err)
-               goto out;
-       seq_putc(m, ' ');
-
-       /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
-       err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
-       if (err)
-               goto out;
-
-       seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
-       show_mnt_opts(m, mnt);
-
-       /* Tagged fields ("foo:X" or "bar") */
-       if (IS_MNT_SHARED(mnt))
-               seq_printf(m, " shared:%i", mnt->mnt_group_id);
-       if (IS_MNT_SLAVE(mnt)) {
-               int master = mnt->mnt_master->mnt_group_id;
-               int dom = get_dominating_id(mnt, &p->root);
-               seq_printf(m, " master:%i", master);
-               if (dom && dom != master)
-                       seq_printf(m, " propagate_from:%i", dom);
-       }
-       if (IS_MNT_UNBINDABLE(mnt))
-               seq_puts(m, " unbindable");
-
-       /* Filesystem specific data */
-       seq_puts(m, " - ");
-       show_type(m, sb);
-       seq_putc(m, ' ');
-       if (sb->s_op->show_devname)
-               err = sb->s_op->show_devname(m, mnt);
-       else
-               mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
-       if (err)
-               goto out;
-       seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
-       err = show_sb_opts(m, sb);
-       if (err)
-               goto out;
-       if (sb->s_op->show_options)
-               err = sb->s_op->show_options(m, mnt);
-       seq_putc(m, '\n');
-out:
-       return err;
-}
-
-const struct seq_operations mountinfo_op = {
-       .start  = m_start,
-       .next   = m_next,
-       .stop   = m_stop,
-       .show   = show_mountinfo,
-};
-
-static int show_vfsstat(struct seq_file *m, void *v)
-{
-       struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
-       struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
-       int err = 0;
-
-       /* device */
-       if (mnt->mnt_sb->s_op->show_devname) {
-               seq_puts(m, "device ");
-               err = mnt->mnt_sb->s_op->show_devname(m, mnt);
-       } else {
-               if (mnt->mnt_devname) {
-                       seq_puts(m, "device ");
-                       mangle(m, mnt->mnt_devname);
-               } else
-                       seq_puts(m, "no device");
-       }
-
-       /* mount point */
-       seq_puts(m, " mounted on ");
-       seq_path(m, &mnt_path, " \t\n\\");
-       seq_putc(m, ' ');
-
-       /* file system type */
-       seq_puts(m, "with fstype ");
-       show_type(m, mnt->mnt_sb);
-
-       /* optional statistics */
-       if (mnt->mnt_sb->s_op->show_stats) {
-               seq_putc(m, ' ');
-               if (!err)
-                       err = mnt->mnt_sb->s_op->show_stats(m, mnt);
-       }
-
-       seq_putc(m, '\n');
-       return err;
-}
-
-const struct seq_operations mountstats_op = {
-       .start  = m_start,
-       .next   = m_next,
-       .stop   = m_stop,
-       .show   = show_vfsstat,
+       .show   = m_show,
 };
 #endif  /* CONFIG_PROC_FS */
 
@@ -1139,20 +1086,21 @@ const struct seq_operations mountstats_op = {
  * open files, pwds, chroots or sub mounts that are
  * busy.
  */
-int may_umount_tree(struct vfsmount *mnt)
+int may_umount_tree(struct vfsmount *m)
 {
+       struct mount *mnt = real_mount(m);
        int actual_refs = 0;
        int minimum_refs = 0;
        struct mount *p;
-       BUG_ON(!mnt);
+       BUG_ON(!m);
 
        /* write lock needed for mnt_get_count */
-       br_write_lock(vfsmount_lock);
-       for (p = real_mount(mnt); p; p = next_mnt(p, mnt)) {
-               actual_refs += mnt_get_count(&p->mnt);
+       br_write_lock(&vfsmount_lock);
+       for (p = mnt; p; p = next_mnt(p, mnt)) {
+               actual_refs += mnt_get_count(p);
                minimum_refs += 2;
        }
-       br_write_unlock(vfsmount_lock);
+       br_write_unlock(&vfsmount_lock);
 
        if (actual_refs > minimum_refs)
                return 0;
@@ -1179,80 +1127,96 @@ int may_umount(struct vfsmount *mnt)
 {
        int ret = 1;
        down_read(&namespace_sem);
-       br_write_lock(vfsmount_lock);
-       if (propagate_mount_busy(mnt, 2))
+       br_write_lock(&vfsmount_lock);
+       if (propagate_mount_busy(real_mount(mnt), 2))
                ret = 0;
-       br_write_unlock(vfsmount_lock);
+       br_write_unlock(&vfsmount_lock);
        up_read(&namespace_sem);
        return ret;
 }
 
 EXPORT_SYMBOL(may_umount);
 
-void release_mounts(struct list_head *head)
+static LIST_HEAD(unmounted);   /* protected by namespace_sem */
+
+static void namespace_unlock(void)
 {
        struct mount *mnt;
-       while (!list_empty(head)) {
-               mnt = list_first_entry(head, struct mount, mnt.mnt_hash);
-               list_del_init(&mnt->mnt.mnt_hash);
-               if (mnt_has_parent(&mnt->mnt)) {
+       LIST_HEAD(head);
+
+       if (likely(list_empty(&unmounted))) {
+               up_write(&namespace_sem);
+               return;
+       }
+
+       list_splice_init(&unmounted, &head);
+       up_write(&namespace_sem);
+
+       while (!list_empty(&head)) {
+               mnt = list_first_entry(&head, struct mount, mnt_hash);
+               list_del_init(&mnt->mnt_hash);
+               if (mnt_has_parent(mnt)) {
                        struct dentry *dentry;
-                       struct vfsmount *m;
+                       struct mount *m;
 
-                       br_write_lock(vfsmount_lock);
-                       dentry = mnt->mnt.mnt_mountpoint;
-                       m = mnt->mnt.mnt_parent;
-                       mnt->mnt.mnt_mountpoint = mnt->mnt.mnt_root;
-                       mnt->mnt.mnt_parent = &mnt->mnt;
+                       br_write_lock(&vfsmount_lock);
+                       dentry = mnt->mnt_mountpoint;
+                       m = mnt->mnt_parent;
+                       mnt->mnt_mountpoint = mnt->mnt.mnt_root;
+                       mnt->mnt_parent = mnt;
                        m->mnt_ghosts--;
-                       br_write_unlock(vfsmount_lock);
+                       br_write_unlock(&vfsmount_lock);
                        dput(dentry);
-                       mntput(m);
+                       mntput(&m->mnt);
                }
                mntput(&mnt->mnt);
        }
 }
 
+static inline void namespace_lock(void)
+{
+       down_write(&namespace_sem);
+}
+
 /*
  * vfsmount lock must be held for write
  * namespace_sem must be held for write
  */
-void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
+void umount_tree(struct mount *mnt, int propagate)
 {
        LIST_HEAD(tmp_list);
        struct mount *p;
 
-       for (p = real_mount(mnt); p; p = next_mnt(p, mnt))
-               list_move(&p->mnt.mnt_hash, &tmp_list);
+       for (p = mnt; p; p = next_mnt(p, mnt))
+               list_move(&p->mnt_hash, &tmp_list);
 
        if (propagate)
                propagate_umount(&tmp_list);
 
-       list_for_each_entry(p, &tmp_list, mnt.mnt_hash) {
-               list_del_init(&p->mnt.mnt_expire);
-               list_del_init(&p->mnt.mnt_list);
-               __touch_mnt_namespace(p->mnt.mnt_ns);
-               p->mnt.mnt_ns = NULL;
-               __mnt_make_shortterm(&p->mnt);
-               list_del_init(&p->mnt.mnt_child);
-               if (mnt_has_parent(&p->mnt)) {
-                       p->mnt.mnt_parent->mnt_ghosts++;
-                       dentry_reset_mounted(p->mnt.mnt_mountpoint);
+       list_for_each_entry(p, &tmp_list, mnt_hash) {
+               list_del_init(&p->mnt_expire);
+               list_del_init(&p->mnt_list);
+               __touch_mnt_namespace(p->mnt_ns);
+               p->mnt_ns = NULL;
+               list_del_init(&p->mnt_child);
+               if (mnt_has_parent(p)) {
+                       p->mnt_parent->mnt_ghosts++;
+                       put_mountpoint(p->mnt_mp);
+                       p->mnt_mp = NULL;
                }
                change_mnt_propagation(p, MS_PRIVATE);
        }
-       list_splice(&tmp_list, kill);
+       list_splice(&tmp_list, &unmounted);
 }
 
-static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
+static void shrink_submounts(struct mount *mnt);
 
-static int do_umount(struct vfsmount *mnt, int flags)
+static int do_umount(struct mount *mnt, int flags)
 {
-       struct super_block *sb = mnt->mnt_sb;
+       struct super_block *sb = mnt->mnt.mnt_sb;
        int retval;
-       LIST_HEAD(umount_list);
 
-       retval = security_sb_umount(mnt, flags);
+       retval = security_sb_umount(&mnt->mnt, flags);
        if (retval)
                return retval;
 
@@ -1263,7 +1227,7 @@ static int do_umount(struct vfsmount *mnt, int flags)
         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
         */
        if (flags & MNT_EXPIRE) {
-               if (mnt == current->fs->root.mnt ||
+               if (&mnt->mnt == current->fs->root.mnt ||
                    flags & (MNT_FORCE | MNT_DETACH))
                        return -EINVAL;
 
@@ -1271,12 +1235,12 @@ static int do_umount(struct vfsmount *mnt, int flags)
                 * probably don't strictly need the lock here if we examined
                 * all race cases, but it's a slowpath.
                 */
-               br_write_lock(vfsmount_lock);
+               br_write_lock(&vfsmount_lock);
                if (mnt_get_count(mnt) != 2) {
-                       br_write_unlock(vfsmount_lock);
+                       br_write_unlock(&vfsmount_lock);
                        return -EBUSY;
                }
-               br_write_unlock(vfsmount_lock);
+               br_write_unlock(&vfsmount_lock);
 
                if (!xchg(&mnt->mnt_expiry_mark, 1))
                        return -EAGAIN;
@@ -1305,11 +1269,13 @@ static int do_umount(struct vfsmount *mnt, int flags)
         * /reboot - static binary that would close all descriptors and
         * call reboot(9). Then init(8) could umount root and exec /reboot.
         */
-       if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
+       if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
                /*
                 * Special case for "unmounting" root ...
                 * we just try to remount it readonly.
                 */
+               if (!capable(CAP_SYS_ADMIN))
+                       return -EPERM;
                down_write(&sb->s_umount);
                if (!(sb->s_flags & MS_RDONLY))
                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
@@ -1317,25 +1283,32 @@ static int do_umount(struct vfsmount *mnt, int flags)
                return retval;
        }
 
-       down_write(&namespace_sem);
-       br_write_lock(vfsmount_lock);
+       namespace_lock();
+       br_write_lock(&vfsmount_lock);
        event++;
 
        if (!(flags & MNT_DETACH))
-               shrink_submounts(mnt, &umount_list);
+               shrink_submounts(mnt);
 
        retval = -EBUSY;
        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
                if (!list_empty(&mnt->mnt_list))
-                       umount_tree(mnt, 1, &umount_list);
+                       umount_tree(mnt, 1);
                retval = 0;
        }
-       br_write_unlock(vfsmount_lock);
-       up_write(&namespace_sem);
-       release_mounts(&umount_list);
+       br_write_unlock(&vfsmount_lock);
+       namespace_unlock();
        return retval;
 }
 
+/* 
+ * Is the caller allowed to modify his namespace?
+ */
+static inline bool may_mount(void)
+{
+       return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
+}
+
 /*
  * Now umount can handle mount points as well as block devices.
  * This is important for filesystems which use unnamed block devices.
@@ -1347,33 +1320,37 @@ static int do_umount(struct vfsmount *mnt, int flags)
 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
 {
        struct path path;
+       struct mount *mnt;
        int retval;
        int lookup_flags = 0;
 
        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
                return -EINVAL;
 
+       if (!may_mount())
+               return -EPERM;
+
        if (!(flags & UMOUNT_NOFOLLOW))
                lookup_flags |= LOOKUP_FOLLOW;
 
        retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
        if (retval)
                goto out;
+       mnt = real_mount(path.mnt);
        retval = -EINVAL;
        if (path.dentry != path.mnt->mnt_root)
                goto dput_and_out;
-       if (!check_mnt(path.mnt))
+       if (!check_mnt(mnt))
                goto dput_and_out;
-
        retval = -EPERM;
-       if (!capable(CAP_SYS_ADMIN))
+       if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
                goto dput_and_out;
 
-       retval = do_umount(path.mnt, flags);
+       retval = do_umount(mnt, flags);
 dput_and_out:
        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
        dput(path.dentry);
-       mntput_no_expire(path.mnt);
+       mntput_no_expire(mnt);
 out:
        return retval;
 }
@@ -1390,38 +1367,39 @@ SYSCALL_DEFINE1(oldumount, char __user *, name)
 
 #endif
 
-static int mount_is_safe(struct path *path)
+static bool mnt_ns_loop(struct path *path)
 {
-       if (capable(CAP_SYS_ADMIN))
-               return 0;
-       return -EPERM;
-#ifdef notyet
-       if (S_ISLNK(path->dentry->d_inode->i_mode))
-               return -EPERM;
-       if (path->dentry->d_inode->i_mode & S_ISVTX) {
-               if (current_uid() != path->dentry->d_inode->i_uid)
-                       return -EPERM;
-       }
-       if (inode_permission(path->dentry->d_inode, MAY_WRITE))
-               return -EPERM;
-       return 0;
-#endif
+       /* Could bind mounting the mount namespace inode cause a
+        * mount namespace loop?
+        */
+       struct inode *inode = path->dentry->d_inode;
+       struct proc_ns *ei;
+       struct mnt_namespace *mnt_ns;
+
+       if (!proc_ns_inode(inode))
+               return false;
+
+       ei = get_proc_ns(inode);
+       if (ei->ns_ops != &mntns_operations)
+               return false;
+
+       mnt_ns = ei->ns;
+       return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
 }
 
-struct mount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
+struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
                                        int flag)
 {
-       struct mount *res, *q;
-       struct vfsmount *p, *r;
-       struct path path;
+       struct mount *res, *p, *q, *r, *parent;
 
        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
-               return NULL;
+               return ERR_PTR(-EINVAL);
 
        res = q = clone_mnt(mnt, dentry, flag);
-       if (!q)
-               goto Enomem;
-       q->mnt.mnt_mountpoint = mnt->mnt_mountpoint;
+       if (IS_ERR(q))
+               return q;
+
+       q->mnt_mountpoint = mnt->mnt_mountpoint;
 
        p = mnt;
        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
@@ -1429,68 +1407,68 @@ struct mount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
                if (!is_subdir(r->mnt_mountpoint, dentry))
                        continue;
 
-               for (s = real_mount(r); s; s = next_mnt(s, r)) {
-                       if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(&s->mnt)) {
+               for (s = r; s; s = next_mnt(s, r)) {
+                       if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
                                s = skip_mnt_tree(s);
                                continue;
                        }
-                       while (p != s->mnt.mnt_parent) {
+                       while (p != s->mnt_parent) {
                                p = p->mnt_parent;
-                               q = real_mount(q->mnt.mnt_parent);
+                               q = q->mnt_parent;
                        }
-                       p = &s->mnt;
-                       path.mnt = &q->mnt;
-                       path.dentry = p->mnt_mountpoint;
-                       q = clone_mnt(p, p->mnt_root, flag);
-                       if (!q)
-                               goto Enomem;
-                       br_write_lock(vfsmount_lock);
-                       list_add_tail(&q->mnt.mnt_list, &res->mnt.mnt_list);
-                       attach_mnt(q, &path);
-                       br_write_unlock(vfsmount_lock);
+                       p = s;
+                       parent = q;
+                       q = clone_mnt(p, p->mnt.mnt_root, flag);
+                       if (IS_ERR(q))
+                               goto out;
+                       br_write_lock(&vfsmount_lock);
+                       list_add_tail(&q->mnt_list, &res->mnt_list);
+                       attach_mnt(q, parent, p->mnt_mp);
+                       br_write_unlock(&vfsmount_lock);
                }
        }
        return res;
-Enomem:
+out:
        if (res) {
-               LIST_HEAD(umount_list);
-               br_write_lock(vfsmount_lock);
-               umount_tree(&res->mnt, 0, &umount_list);
-               br_write_unlock(vfsmount_lock);
-               release_mounts(&umount_list);
+               br_write_lock(&vfsmount_lock);
+               umount_tree(res, 0);
+               br_write_unlock(&vfsmount_lock);
        }
-       return NULL;
+       return q;
 }
 
+/* Caller should check returned pointer for errors */
+
 struct vfsmount *collect_mounts(struct path *path)
 {
        struct mount *tree;
-       down_write(&namespace_sem);
-       tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
-       up_write(&namespace_sem);
-       return tree ? &tree->mnt : NULL;
+       namespace_lock();
+       tree = copy_tree(real_mount(path->mnt), path->dentry,
+                        CL_COPY_ALL | CL_PRIVATE);
+       namespace_unlock();
+       if (IS_ERR(tree))
+               return ERR_CAST(tree);
+       return &tree->mnt;
 }
 
 void drop_collected_mounts(struct vfsmount *mnt)
 {
-       LIST_HEAD(umount_list);
-       down_write(&namespace_sem);
-       br_write_lock(vfsmount_lock);
-       umount_tree(mnt, 0, &umount_list);
-       br_write_unlock(vfsmount_lock);
-       up_write(&namespace_sem);
-       release_mounts(&umount_list);
+       namespace_lock();
+       br_write_lock(&vfsmount_lock);
+       umount_tree(real_mount(mnt), 0);
+       br_write_unlock(&vfsmount_lock);
+       namespace_unlock();
 }
 
 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
                   struct vfsmount *root)
 {
-       struct vfsmount *mnt;
+       struct mount *mnt;
        int res = f(root, arg);
        if (res)
                return res;
-       list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
-               res = f(mnt, arg);
+       list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
+               res = f(&mnt->mnt, arg);
                if (res)
                        return res;
        }
@@ -1501,8 +1479,8 @@ static void cleanup_group_ids(struct mount *mnt, struct mount *end)
 {
        struct mount *p;
 
-       for (p = mnt; p != end; p = next_mnt(p, &mnt->mnt)) {
-               if (p->mnt.mnt_group_id && !IS_MNT_SHARED(&p->mnt))
+       for (p = mnt; p != end; p = next_mnt(p, mnt)) {
+               if (p->mnt_group_id && !IS_MNT_SHARED(p))
                        mnt_release_group_id(p);
        }
 }
@@ -1511,8 +1489,8 @@ static int invent_group_ids(struct mount *mnt, bool recurse)
 {
        struct mount *p;
 
-       for (p = mnt; p; p = recurse ? next_mnt(p, &mnt->mnt) : NULL) {
-               if (!p->mnt.mnt_group_id && !IS_MNT_SHARED(&p->mnt)) {
+       for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
+               if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
                        int err = mnt_alloc_group_id(p);
                        if (err) {
                                cleanup_group_ids(mnt, p);
@@ -1588,11 +1566,11 @@ static int invent_group_ids(struct mount *mnt, bool recurse)
  * in allocations.
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
-                       struct path *path, struct path *parent_path)
+                       struct mount *dest_mnt,
+                       struct mountpoint *dest_mp,
+                       struct path *parent_path)
 {
        LIST_HEAD(tree_list);
-       struct vfsmount *dest_mnt = path->mnt;
-       struct dentry *dest_dentry = path->dentry;
        struct mount *child, *p;
        int err;
 
@@ -1601,30 +1579,30 @@ static int attach_recursive_mnt(struct mount *source_mnt,
                if (err)
                        goto out;
        }
-       err = propagate_mnt(dest_mnt, dest_dentry, &source_mnt->mnt, &tree_list);
+       err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
        if (err)
                goto out_cleanup_ids;
 
-       br_write_lock(vfsmount_lock);
+       br_write_lock(&vfsmount_lock);
 
        if (IS_MNT_SHARED(dest_mnt)) {
-               for (p = source_mnt; p; p = next_mnt(p, &source_mnt->mnt))
+               for (p = source_mnt; p; p = next_mnt(p, source_mnt))
                        set_mnt_shared(p);
        }
        if (parent_path) {
                detach_mnt(source_mnt, parent_path);
-               attach_mnt(source_mnt, path);
-               touch_mnt_namespace(parent_path->mnt->mnt_ns);
+               attach_mnt(source_mnt, dest_mnt, dest_mp);
+               touch_mnt_namespace(source_mnt->mnt_ns);
        } else {
-               mnt_set_mountpoint(dest_mnt, dest_dentry, &source_mnt->mnt);
+               mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
                commit_tree(source_mnt);
        }
 
-       list_for_each_entry_safe(child, p, &tree_list, mnt.mnt_hash) {
-               list_del_init(&child->mnt.mnt_hash);
+       list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
+               list_del_init(&child->mnt_hash);
                commit_tree(child);
        }
-       br_write_unlock(vfsmount_lock);
+       br_write_unlock(&vfsmount_lock);
 
        return 0;
 
@@ -1635,46 +1613,53 @@ static int attach_recursive_mnt(struct mount *source_mnt,
        return err;
 }
 
-static int lock_mount(struct path *path)
+static struct mountpoint *lock_mount(struct path *path)
 {
        struct vfsmount *mnt;
+       struct dentry *dentry = path->dentry;
 retry:
-       mutex_lock(&path->dentry->d_inode->i_mutex);
-       if (unlikely(cant_mount(path->dentry))) {
-               mutex_unlock(&path->dentry->d_inode->i_mutex);
-               return -ENOENT;
+       mutex_lock(&dentry->d_inode->i_mutex);
+       if (unlikely(cant_mount(dentry))) {
+               mutex_unlock(&dentry->d_inode->i_mutex);
+               return ERR_PTR(-ENOENT);
        }
-       down_write(&namespace_sem);
+       namespace_lock();
        mnt = lookup_mnt(path);
-       if (likely(!mnt))
-               return 0;
-       up_write(&namespace_sem);
+       if (likely(!mnt)) {
+               struct mountpoint *mp = new_mountpoint(dentry);
+               if (IS_ERR(mp)) {
+                       namespace_unlock();
+                       mutex_unlock(&dentry->d_inode->i_mutex);
+                       return mp;
+               }
+               return mp;
+       }
+       namespace_unlock();
        mutex_unlock(&path->dentry->d_inode->i_mutex);
        path_put(path);
        path->mnt = mnt;
-       path->dentry = dget(mnt->mnt_root);
+       dentry = path->dentry = dget(mnt->mnt_root);
        goto retry;
 }
 
-static void unlock_mount(struct path *path)
+static void unlock_mount(struct mountpoint *where)
 {
-       up_write(&namespace_sem);
-       mutex_unlock(&path->dentry->d_inode->i_mutex);
+       struct dentry *dentry = where->m_dentry;
+       put_mountpoint(where);
+       namespace_unlock();
+       mutex_unlock(&dentry->d_inode->i_mutex);
 }
 
-static int graft_tree(struct vfsmount *mnt, struct path *path)
+static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
 {
-       if (mnt->mnt_sb->s_flags & MS_NOUSER)
+       if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
                return -EINVAL;
 
-       if (S_ISDIR(path->dentry->d_inode->i_mode) !=
-             S_ISDIR(mnt->mnt_root->d_inode->i_mode))
+       if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
+             S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
                return -ENOTDIR;
 
-       if (d_unlinked(path->dentry))
-               return -ENOENT;
-
-       return attach_recursive_mnt(real_mount(mnt), path, NULL);
+       return attach_recursive_mnt(mnt, p, mp, NULL);
 }
 
 /*
@@ -1705,9 +1690,6 @@ static int do_change_type(struct path *path, int flag)
        int type;
        int err = 0;
 
-       if (!capable(CAP_SYS_ADMIN))
-               return -EPERM;
-
        if (path->dentry != path->mnt->mnt_root)
                return -EINVAL;
 
@@ -1715,70 +1697,76 @@ static int do_change_type(struct path *path, int flag)
        if (!type)
                return -EINVAL;
 
-       down_write(&namespace_sem);
+       namespace_lock();
        if (type == MS_SHARED) {
                err = invent_group_ids(mnt, recurse);
                if (err)
                        goto out_unlock;
        }
 
-       br_write_lock(vfsmount_lock);
-       for (m = mnt; m; m = (recurse ? next_mnt(m, &mnt->mnt) : NULL))
+       br_write_lock(&vfsmount_lock);
+       for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
                change_mnt_propagation(m, type);
-       br_write_unlock(vfsmount_lock);
+       br_write_unlock(&vfsmount_lock);
 
  out_unlock:
-       up_write(&namespace_sem);
+       namespace_unlock();
        return err;
 }
 
 /*
  * do loopback mount.
  */
-static int do_loopback(struct path *path, char *old_name,
+static int do_loopback(struct path *path, const char *old_name,
                                int recurse)
 {
-       LIST_HEAD(umount_list);
        struct path old_path;
-       struct mount *mnt = NULL;
-       int err = mount_is_safe(path);
-       if (err)
-               return err;
+       struct mount *mnt = NULL, *old, *parent;
+       struct mountpoint *mp;
+       int err;
        if (!old_name || !*old_name)
                return -EINVAL;
        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
        if (err)
                return err;
 
-       err = lock_mount(path);
-       if (err)
+       err = -EINVAL;
+       if (mnt_ns_loop(&old_path))
+               goto out; 
+
+       mp = lock_mount(path);
+       err = PTR_ERR(mp);
+       if (IS_ERR(mp))
                goto out;
 
+       old = real_mount(old_path.mnt);
+       parent = real_mount(path->mnt);
+
        err = -EINVAL;
-       if (IS_MNT_UNBINDABLE(old_path.mnt))
+       if (IS_MNT_UNBINDABLE(old))
                goto out2;
 
-       if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
+       if (!check_mnt(parent) || !check_mnt(old))
                goto out2;
 
-       err = -ENOMEM;
        if (recurse)
-               mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
+               mnt = copy_tree(old, old_path.dentry, 0);
        else
-               mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
+               mnt = clone_mnt(old, old_path.dentry, 0);
 
-       if (!mnt)
+       if (IS_ERR(mnt)) {
+               err = PTR_ERR(mnt);
                goto out2;
+       }
 
-       err = graft_tree(&mnt->mnt, path);
+       err = graft_tree(mnt, parent, mp);
        if (err) {
-               br_write_lock(vfsmount_lock);
-               umount_tree(&mnt->mnt, 0, &umount_list);
-               br_write_unlock(vfsmount_lock);
+               br_write_lock(&vfsmount_lock);
+               umount_tree(mnt, 0);
+               br_write_unlock(&vfsmount_lock);
        }
 out2:
-       unlock_mount(path);
-       release_mounts(&umount_list);
+       unlock_mount(mp);
 out:
        path_put(&old_path);
        return err;
@@ -1795,9 +1783,9 @@ static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
                return 0;
 
        if (readonly_request)
-               error = mnt_make_readonly(mnt);
+               error = mnt_make_readonly(real_mount(mnt));
        else
-               __mnt_unmake_readonly(mnt);
+               __mnt_unmake_readonly(real_mount(mnt));
        return error;
 }
 
@@ -1811,16 +1799,47 @@ static int do_remount(struct path *path, int flags, int mnt_flags,
 {
        int err;
        struct super_block *sb = path->mnt->mnt_sb;
+       struct mount *mnt = real_mount(path->mnt);
 
-       if (!capable(CAP_SYS_ADMIN))
-               return -EPERM;
-
-       if (!check_mnt(path->mnt))
+       if (!check_mnt(mnt))
                return -EINVAL;
 
        if (path->dentry != path->mnt->mnt_root)
                return -EINVAL;
 
+       /* Don't allow changing of locked mnt flags.
+        *
+        * No locks need to be held here while testing the various
+        * MNT_LOCK flags because those flags can never be cleared
+        * once they are set.
+        */
+       if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
+           !(mnt_flags & MNT_READONLY)) {
+               return -EPERM;
+       }
+       if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
+           !(mnt_flags & MNT_NODEV)) {
+               /* Was the nodev implicitly added in mount? */
+               if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
+                   !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
+                       mnt_flags |= MNT_NODEV;
+               } else {
+                       return -EPERM;
+               }
+       }
+       if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
+           !(mnt_flags & MNT_NOSUID)) {
+               return -EPERM;
+       }
+       if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
+           !(mnt_flags & MNT_NOEXEC)) {
+               return -EPERM;
+       }
+       if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
+           ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
+               return -EPERM;
+       }
+
        err = security_sb_remount(sb, data);
        if (err)
                return err;
@@ -1828,19 +1847,21 @@ static int do_remount(struct path *path, int flags, int mnt_flags,
        down_write(&sb->s_umount);
        if (flags & MS_BIND)
                err = change_mount_flags(path->mnt, flags);
+       else if (!capable(CAP_SYS_ADMIN))
+               err = -EPERM;
        else
                err = do_remount_sb(sb, flags, data, 0);
        if (!err) {
-               br_write_lock(vfsmount_lock);
-               mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
-               path->mnt->mnt_flags = mnt_flags;
-               br_write_unlock(vfsmount_lock);
+               br_write_lock(&vfsmount_lock);
+               mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
+               mnt->mnt.mnt_flags = mnt_flags;
+               br_write_unlock(&vfsmount_lock);
        }
        up_write(&sb->s_umount);
        if (!err) {
-               br_write_lock(vfsmount_lock);
-               touch_mnt_namespace(path->mnt->mnt_ns);
-               br_write_unlock(vfsmount_lock);
+               br_write_lock(&vfsmount_lock);
+               touch_mnt_namespace(mnt->mnt_ns);
+               br_write_unlock(&vfsmount_lock);
        }
        return err;
 }
@@ -1848,45 +1869,43 @@ static int do_remount(struct path *path, int flags, int mnt_flags,
 static inline int tree_contains_unbindable(struct mount *mnt)
 {
        struct mount *p;
-       for (p = mnt; p; p = next_mnt(p, &mnt->mnt)) {
-               if (IS_MNT_UNBINDABLE(&p->mnt))
+       for (p = mnt; p; p = next_mnt(p, mnt)) {
+               if (IS_MNT_UNBINDABLE(p))
                        return 1;
        }
        return 0;
 }
 
-static int do_move_mount(struct path *path, char *old_name)
+static int do_move_mount(struct path *path, const char *old_name)
 {
        struct path old_path, parent_path;
-       struct vfsmount *p;
+       struct mount *p;
        struct mount *old;
-       int err = 0;
-       if (!capable(CAP_SYS_ADMIN))
-               return -EPERM;
+       struct mountpoint *mp;
+       int err;
        if (!old_name || !*old_name)
                return -EINVAL;
        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
        if (err)
                return err;
 
-       err = lock_mount(path);
-       if (err < 0)
+       mp = lock_mount(path);
+       err = PTR_ERR(mp);
+       if (IS_ERR(mp))
                goto out;
 
-       err = -EINVAL;
-       if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
-               goto out1;
+       old = real_mount(old_path.mnt);
+       p = real_mount(path->mnt);
 
-       if (d_unlinked(path->dentry))
+       err = -EINVAL;
+       if (!check_mnt(p) || !check_mnt(old))
                goto out1;
 
        err = -EINVAL;
        if (old_path.dentry != old_path.mnt->mnt_root)
                goto out1;
 
-       old = real_mount(old_path.mnt);
-
-       if (!mnt_has_parent(old_path.mnt))
+       if (!mnt_has_parent(old))
                goto out1;
 
        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
@@ -1895,29 +1914,28 @@ static int do_move_mount(struct path *path, char *old_name)
        /*
         * Don't move a mount residing in a shared parent.
         */
-       if (IS_MNT_SHARED(old_path.mnt->mnt_parent))
+       if (IS_MNT_SHARED(old->mnt_parent))
                goto out1;
        /*
         * Don't move a mount tree containing unbindable mounts to a destination
         * mount which is shared.
         */
-       if (IS_MNT_SHARED(path->mnt) &&
-           tree_contains_unbindable(old))
+       if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
                goto out1;
        err = -ELOOP;
-       for (p = path->mnt; mnt_has_parent(p); p = p->mnt_parent)
-               if (p == old_path.mnt)
+       for (; mnt_has_parent(p); p = p->mnt_parent)
+               if (p == old)
                        goto out1;
 
-       err = attach_recursive_mnt(old, path, &parent_path);
+       err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
        if (err)
                goto out1;
 
        /* if the mount is moved, it should no longer be expire
         * automatically */
-       list_del_init(&old_path.mnt->mnt_expire);
+       list_del_init(&old->mnt_expire);
 out1:
-       unlock_mount(path);
+       unlock_mount(mp);
 out:
        if (!err)
                path_put(&parent_path);
@@ -1948,53 +1966,47 @@ static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
        return ERR_PTR(err);
 }
 
-static struct vfsmount *
-do_kern_mount(const char *fstype, int flags, const char *name, void *data)
-{
-       struct file_system_type *type = get_fs_type(fstype);
-       struct vfsmount *mnt;
-       if (!type)
-               return ERR_PTR(-ENODEV);
-       mnt = vfs_kern_mount(type, flags, name, data);
-       if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
-           !mnt->mnt_sb->s_subtype)
-               mnt = fs_set_subtype(mnt, fstype);
-       put_filesystem(type);
-       return mnt;
-}
-
 /*
  * add a mount into a namespace's mount tree
  */
-static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
+static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
 {
+       struct mountpoint *mp;
+       struct mount *parent;
        int err;
 
        mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
 
-       err = lock_mount(path);
-       if (err)
-               return err;
+       mp = lock_mount(path);
+       if (IS_ERR(mp))
+               return PTR_ERR(mp);
 
+       parent = real_mount(path->mnt);
        err = -EINVAL;
-       if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
-               goto unlock;
+       if (unlikely(!check_mnt(parent))) {
+               /* that's acceptable only for automounts done in private ns */
+               if (!(mnt_flags & MNT_SHRINKABLE))
+                       goto unlock;
+               /* ... and for those we'd better have mountpoint still alive */
+               if (!parent->mnt_ns)
+                       goto unlock;
+       }
 
        /* Refuse the same filesystem on the same mount point */
        err = -EBUSY;
-       if (path->mnt->mnt_sb == newmnt->mnt_sb &&
+       if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
            path->mnt->mnt_root == path->dentry)
                goto unlock;
 
        err = -EINVAL;
-       if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
+       if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
                goto unlock;
 
-       newmnt->mnt_flags = mnt_flags;
-       err = graft_tree(newmnt, path);
+       newmnt->mnt.mnt_flags = mnt_flags;
+       err = graft_tree(newmnt, parent, mp);
 
 unlock:
-       unlock_mount(path);
+       unlock_mount(mp);
        return err;
 }
 
@@ -2002,24 +2014,45 @@ unlock:
  * create a new mount for userspace and request it to be added into the
  * namespace's tree
  */
-static int do_new_mount(struct path *path, char *type, int flags,
-                       int mnt_flags, char *name, void *data)
+static int do_new_mount(struct path *path, const char *fstype, int flags,
+                       int mnt_flags, const char *name, void *data)
 {
+       struct file_system_type *type;
+       struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
        struct vfsmount *mnt;
        int err;
 
-       if (!type)
+       if (!fstype)
                return -EINVAL;
 
-       /* we need capabilities... */
-       if (!capable(CAP_SYS_ADMIN))
-               return -EPERM;
+       type = get_fs_type(fstype);
+       if (!type)
+               return -ENODEV;
 
-       mnt = do_kern_mount(type, flags, name, data);
+       if (user_ns != &init_user_ns) {
+               if (!(type->fs_flags & FS_USERNS_MOUNT)) {
+                       put_filesystem(type);
+                       return -EPERM;
+               }
+               /* Only in special cases allow devices from mounts
+                * created outside the initial user namespace.
+                */
+               if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
+                       flags |= MS_NODEV;
+                       mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
+               }
+       }
+
+       mnt = vfs_kern_mount(type, flags, name, data);
+       if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
+           !mnt->mnt_sb->s_subtype)
+               mnt = fs_set_subtype(mnt, fstype);
+
+       put_filesystem(type);
        if (IS_ERR(mnt))
                return PTR_ERR(mnt);
 
-       err = do_add_mount(mnt, path, mnt_flags);
+       err = do_add_mount(real_mount(mnt), path, mnt_flags);
        if (err)
                mntput(mnt);
        return err;
@@ -2027,11 +2060,12 @@ static int do_new_mount(struct path *path, char *type, int flags,
 
 int finish_automount(struct vfsmount *m, struct path *path)
 {
+       struct mount *mnt = real_mount(m);
        int err;
        /* The new mount record should have at least 2 refs to prevent it being
         * expired before we get a chance to add it
         */
-       BUG_ON(mnt_get_count(m) < 2);
+       BUG_ON(mnt_get_count(mnt) < 2);
 
        if (m->mnt_sb == path->mnt->mnt_sb &&
            m->mnt_root == path->dentry) {
@@ -2039,17 +2073,17 @@ int finish_automount(struct vfsmount *m, struct path *path)
                goto fail;
        }
 
-       err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
+       err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
        if (!err)
                return 0;
 fail:
        /* remove m from any expiration list it may be on */
-       if (!list_empty(&m->mnt_expire)) {
-               down_write(&namespace_sem);
-               br_write_lock(vfsmount_lock);
-               list_del_init(&m->mnt_expire);
-               br_write_unlock(vfsmount_lock);
-               up_write(&namespace_sem);
+       if (!list_empty(&mnt->mnt_expire)) {
+               namespace_lock();
+               br_write_lock(&vfsmount_lock);
+               list_del_init(&mnt->mnt_expire);
+               br_write_unlock(&vfsmount_lock);
+               namespace_unlock();
        }
        mntput(m);
        mntput(m);
@@ -2063,13 +2097,13 @@ fail:
  */
 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
 {
-       down_write(&namespace_sem);
-       br_write_lock(vfsmount_lock);
+       namespace_lock();
+       br_write_lock(&vfsmount_lock);
 
-       list_add_tail(&mnt->mnt_expire, expiry_list);
+       list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
 
-       br_write_unlock(vfsmount_lock);
-       up_write(&namespace_sem);
+       br_write_unlock(&vfsmount_lock);
+       namespace_unlock();
 }
 EXPORT_SYMBOL(mnt_set_expiry);
 
@@ -2080,15 +2114,14 @@ EXPORT_SYMBOL(mnt_set_expiry);
  */
 void mark_mounts_for_expiry(struct list_head *mounts)
 {
-       struct vfsmount *mnt, *next;
+       struct mount *mnt, *next;
        LIST_HEAD(graveyard);
-       LIST_HEAD(umounts);
 
        if (list_empty(mounts))
                return;
 
-       down_write(&namespace_sem);
-       br_write_lock(vfsmount_lock);
+       namespace_lock();
+       br_write_lock(&vfsmount_lock);
 
        /* extract from the expiration list every vfsmount that matches the
         * following criteria:
@@ -2103,14 +2136,12 @@ void mark_mounts_for_expiry(struct list_head *mounts)
                list_move(&mnt->mnt_expire, &graveyard);
        }
        while (!list_empty(&graveyard)) {
-               mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
+               mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
                touch_mnt_namespace(mnt->mnt_ns);
-               umount_tree(mnt, 1, &umounts);
+               umount_tree(mnt, 1);
        }
-       br_write_unlock(vfsmount_lock);
-       up_write(&namespace_sem);
-
-       release_mounts(&umounts);
+       br_write_unlock(&vfsmount_lock);
+       namespace_unlock();
 }
 
 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
@@ -2121,9 +2152,9 @@ EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
  * search the list of submounts for a given mountpoint, and move any
  * shrinkable submounts to the 'graveyard' list.
  */
-static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
+static int select_submounts(struct mount *parent, struct list_head *graveyard)
 {
-       struct vfsmount *this_parent = parent;
+       struct mount *this_parent = parent;
        struct list_head *next;
        int found = 0;
 
@@ -2132,10 +2163,10 @@ repeat:
 resume:
        while (next != &this_parent->mnt_mounts) {
                struct list_head *tmp = next;
-               struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
+               struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
 
                next = tmp->next;
-               if (!(mnt->mnt_flags & MNT_SHRINKABLE))
+               if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
                        continue;
                /*
                 * Descend a level if the d_mounts list is non-empty.
@@ -2167,18 +2198,18 @@ resume:
  *
  * vfsmount_lock must be held for write
  */
-static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
+static void shrink_submounts(struct mount *mnt)
 {
        LIST_HEAD(graveyard);
-       struct vfsmount *m;
+       struct mount *m;
 
        /* extract submounts of 'mountpoint' from the expiration list */
        while (select_submounts(mnt, &graveyard)) {
                while (!list_empty(&graveyard)) {
-                       m = list_first_entry(&graveyard, struct vfsmount,
+                       m = list_first_entry(&graveyard, struct mount,
                                                mnt_expire);
                        touch_mnt_namespace(m->mnt_ns);
-                       umount_tree(m, 1, umounts);
+                       umount_tree(m, 1);
                }
        }
 }
@@ -2275,8 +2306,8 @@ int copy_mount_string(const void __user *data, char **where)
  * Therefore, if this magic number is present, it carries no information
  * and must be discarded.
  */
-long do_mount(char *dev_name, char *dir_name, char *type_page,
-                 unsigned long flags, void *data_page)
+long do_mount(const char *dev_name, const char *dir_name,
+               const char *type_page, unsigned long flags, void *data_page)
 {
        struct path path;
        int retval = 0;
@@ -2301,6 +2332,8 @@ long do_mount(char *dev_name, char *dir_name, char *type_page,
 
        retval = security_sb_mount(dev_name, &path,
                                   type_page, flags, data_page);
+       if (!retval && !may_mount())
+               retval = -EPERM;
        if (retval)
                goto dput_out;
 
@@ -2324,6 +2357,14 @@ long do_mount(char *dev_name, char *dir_name, char *type_page,
        if (flags & MS_RDONLY)
                mnt_flags |= MNT_READONLY;
 
+       /* The default atime for remount is preservation */
+       if ((flags & MS_REMOUNT) &&
+           ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
+                      MS_STRICTATIME)) == 0)) {
+               mnt_flags &= ~MNT_ATIME_MASK;
+               mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
+       }
+
        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
                   MS_STRICTATIME);
@@ -2345,95 +2386,102 @@ dput_out:
        return retval;
 }
 
-static struct mnt_namespace *alloc_mnt_ns(void)
+static void free_mnt_ns(struct mnt_namespace *ns)
+{
+       proc_free_inum(ns->proc_inum);
+       put_user_ns(ns->user_ns);
+       kfree(ns);
+}
+
+/*
+ * Assign a sequence number so we can detect when we attempt to bind
+ * mount a reference to an older mount namespace into the current
+ * mount namespace, preventing reference counting loops.  A 64bit
+ * number incrementing at 10Ghz will take 12,427 years to wrap which
+ * is effectively never, so we can ignore the possibility.
+ */
+static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
+
+static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
 {
        struct mnt_namespace *new_ns;
+       int ret;
 
        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
        if (!new_ns)
                return ERR_PTR(-ENOMEM);
+       ret = proc_alloc_inum(&new_ns->proc_inum);
+       if (ret) {
+               kfree(new_ns);
+               return ERR_PTR(ret);
+       }
+       new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
        atomic_set(&new_ns->count, 1);
        new_ns->root = NULL;
        INIT_LIST_HEAD(&new_ns->list);
        init_waitqueue_head(&new_ns->poll);
        new_ns->event = 0;
+       new_ns->user_ns = get_user_ns(user_ns);
        return new_ns;
 }
 
-void mnt_make_longterm(struct vfsmount *mnt)
-{
-       __mnt_make_longterm(mnt);
-}
-
-void mnt_make_shortterm(struct vfsmount *mnt)
-{
-#ifdef CONFIG_SMP
-       if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
-               return;
-       br_write_lock(vfsmount_lock);
-       atomic_dec(&mnt->mnt_longterm);
-       br_write_unlock(vfsmount_lock);
-#endif
-}
-
 /*
  * Allocate a new namespace structure and populate it with contents
  * copied from the namespace of the passed in task structure.
  */
 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
-               struct fs_struct *fs)
+               struct user_namespace *user_ns, struct fs_struct *fs)
 {
        struct mnt_namespace *new_ns;
        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
        struct mount *p, *q;
+       struct mount *old = mnt_ns->root;
        struct mount *new;
+       int copy_flags;
 
-       new_ns = alloc_mnt_ns();
+       new_ns = alloc_mnt_ns(user_ns);
        if (IS_ERR(new_ns))
                return new_ns;
 
-       down_write(&namespace_sem);
+       namespace_lock();
        /* First pass: copy the tree topology */
-       new = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
-                                       CL_COPY_ALL | CL_EXPIRE);
-       if (!new) {
-               up_write(&namespace_sem);
-               kfree(new_ns);
-               return ERR_PTR(-ENOMEM);
-       }
-       new_ns->root = &new->mnt;
-       br_write_lock(vfsmount_lock);
-       list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
-       br_write_unlock(vfsmount_lock);
+       copy_flags = CL_COPY_ALL | CL_EXPIRE;
+       if (user_ns != mnt_ns->user_ns)
+               copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
+       new = copy_tree(old, old->mnt.mnt_root, copy_flags);
+       if (IS_ERR(new)) {
+               namespace_unlock();
+               free_mnt_ns(new_ns);
+               return ERR_CAST(new);
+       }
+       new_ns->root = new;
+       br_write_lock(&vfsmount_lock);
+       list_add_tail(&new_ns->list, &new->mnt_list);
+       br_write_unlock(&vfsmount_lock);
 
        /*
         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
         * as belonging to new namespace.  We have already acquired a private
         * fs_struct, so tsk->fs->lock is not needed.
         */
-       p = real_mount(mnt_ns->root);
+       p = old;
        q = new;
        while (p) {
-               q->mnt.mnt_ns = new_ns;
-               __mnt_make_longterm(&q->mnt);
+               q->mnt_ns = new_ns;
                if (fs) {
                        if (&p->mnt == fs->root.mnt) {
                                fs->root.mnt = mntget(&q->mnt);
-                               __mnt_make_longterm(&q->mnt);
-                               mnt_make_shortterm(&p->mnt);
                                rootmnt = &p->mnt;
                        }
                        if (&p->mnt == fs->pwd.mnt) {
                                fs->pwd.mnt = mntget(&q->mnt);
-                               __mnt_make_longterm(&q->mnt);
-                               mnt_make_shortterm(&p->mnt);
                                pwdmnt = &p->mnt;
                        }
                }
-               p = next_mnt(p, mnt_ns->root);
-               q = next_mnt(q, new_ns->root);
+               p = next_mnt(p, old);
+               q = next_mnt(q, new);
        }
-       up_write(&namespace_sem);
+       namespace_unlock();
 
        if (rootmnt)
                mntput(rootmnt);
@@ -2444,7 +2492,7 @@ static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
 }
 
 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
-               struct fs_struct *new_fs)
+               struct user_namespace *user_ns, struct fs_struct *new_fs)
 {
        struct mnt_namespace *new_ns;
 
@@ -2454,7 +2502,7 @@ struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
        if (!(flags & CLONE_NEWNS))
                return ns;
 
-       new_ns = dup_mnt_ns(ns, new_fs);
+       new_ns = dup_mnt_ns(ns, user_ns, new_fs);
 
        put_mnt_ns(ns);
        return new_ns;
@@ -2464,18 +2512,16 @@ struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
  * create_mnt_ns - creates a private namespace and adds a root filesystem
  * @mnt: pointer to the new root filesystem mountpoint
  */
-static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
+static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
 {
-       struct mnt_namespace *new_ns;
-
-       new_ns = alloc_mnt_ns();
+       struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
        if (!IS_ERR(new_ns)) {
+               struct mount *mnt = real_mount(m);
                mnt->mnt_ns = new_ns;
-               __mnt_make_longterm(mnt);
                new_ns->root = mnt;
-               list_add(&new_ns->list, &new_ns->root->mnt_list);
+               list_add(&mnt->mnt_list, &new_ns->list);
        } else {
-               mntput(mnt);
+               mntput(m);
        }
        return new_ns;
 }
@@ -2515,7 +2561,7 @@ SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
 {
        int ret;
        char *kernel_type;
-       char *kernel_dir;
+       struct filename *kernel_dir;
        char *kernel_dev;
        unsigned long data_page;
 
@@ -2537,7 +2583,7 @@ SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
        if (ret < 0)
                goto out_data;
 
-       ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
+       ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
                (void *) data_page);
 
        free_page(data_page);
@@ -2556,22 +2602,22 @@ out_type:
  *
  * namespace_sem or vfsmount_lock is held
  */
-bool is_path_reachable(struct vfsmount *mnt, struct dentry *dentry,
+bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
                         const struct path *root)
 {
-       while (mnt != root->mnt && mnt_has_parent(mnt)) {
+       while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
                dentry = mnt->mnt_mountpoint;
                mnt = mnt->mnt_parent;
        }
-       return mnt == root->mnt && is_subdir(dentry, root->dentry);
+       return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
 }
 
 int path_is_under(struct path *path1, struct path *path2)
 {
        int res;
-       br_read_lock(vfsmount_lock);
-       res = is_path_reachable(path1->mnt, path1->dentry, path2);
-       br_read_unlock(vfsmount_lock);
+       br_read_lock(&vfsmount_lock);
+       res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
+       br_read_unlock(&vfsmount_lock);
        return res;
 }
 EXPORT_SYMBOL(path_is_under);
@@ -2605,10 +2651,11 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
                const char __user *, put_old)
 {
        struct path new, old, parent_path, root_parent, root;
-       struct mount *new_mnt, *root_mnt;
+       struct mount *new_mnt, *root_mnt, *old_mnt;
+       struct mountpoint *old_mp, *root_mp;
        int error;
 
-       if (!capable(CAP_SYS_ADMIN))
+       if (!may_mount())
                return -EPERM;
 
        error = user_path_dir(new_root, &new);
@@ -2624,53 +2671,58 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
                goto out2;
 
        get_fs_root(current->fs, &root);
-       error = lock_mount(&old);
-       if (error)
+       old_mp = lock_mount(&old);
+       error = PTR_ERR(old_mp);
+       if (IS_ERR(old_mp))
                goto out3;
 
        error = -EINVAL;
        new_mnt = real_mount(new.mnt);
        root_mnt = real_mount(root.mnt);
-       if (IS_MNT_SHARED(old.mnt) ||
-               IS_MNT_SHARED(new.mnt->mnt_parent) ||
-               IS_MNT_SHARED(root.mnt->mnt_parent))
+       old_mnt = real_mount(old.mnt);
+       if (IS_MNT_SHARED(old_mnt) ||
+               IS_MNT_SHARED(new_mnt->mnt_parent) ||
+               IS_MNT_SHARED(root_mnt->mnt_parent))
                goto out4;
-       if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
+       if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
                goto out4;
        error = -ENOENT;
        if (d_unlinked(new.dentry))
                goto out4;
-       if (d_unlinked(old.dentry))
-               goto out4;
        error = -EBUSY;
-       if (new.mnt == root.mnt ||
-           old.mnt == root.mnt)
+       if (new_mnt == root_mnt || old_mnt == root_mnt)
                goto out4; /* loop, on the same file system  */
        error = -EINVAL;
        if (root.mnt->mnt_root != root.dentry)
                goto out4; /* not a mountpoint */
-       if (!mnt_has_parent(root.mnt))
+       if (!mnt_has_parent(root_mnt))
                goto out4; /* not attached */
+       root_mp = root_mnt->mnt_mp;
        if (new.mnt->mnt_root != new.dentry)
                goto out4; /* not a mountpoint */
-       if (!mnt_has_parent(new.mnt))
+       if (!mnt_has_parent(new_mnt))
                goto out4; /* not attached */
        /* make sure we can reach put_old from new_root */
-       if (!is_path_reachable(old.mnt, old.dentry, &new))
+       if (!is_path_reachable(old_mnt, old.dentry, &new))
+               goto out4;
+       /* make certain new is below the root */
+       if (!is_path_reachable(new_mnt, new.dentry, &root))
                goto out4;
-       br_write_lock(vfsmount_lock);
+       root_mp->m_count++; /* pin it so it won't go away */
+       br_write_lock(&vfsmount_lock);
        detach_mnt(new_mnt, &parent_path);
        detach_mnt(root_mnt, &root_parent);
        /* mount old root on put_old */
-       attach_mnt(root_mnt, &old);
+       attach_mnt(root_mnt, old_mnt, old_mp);
        /* mount new_root on / */
-       attach_mnt(new_mnt, &root_parent);
+       attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
        touch_mnt_namespace(current->nsproxy->mnt_ns);
-       br_write_unlock(vfsmount_lock);
+       br_write_unlock(&vfsmount_lock);
        chroot_fs_refs(&root, &new);
+       put_mountpoint(root_mp);
        error = 0;
 out4:
-       unlock_mount(&old);
+       unlock_mount(old_mp);
        if (!error) {
                path_put(&root_parent);
                path_put(&parent_path);
@@ -2690,8 +2742,13 @@ static void __init init_mount_tree(void)
        struct vfsmount *mnt;
        struct mnt_namespace *ns;
        struct path root;
+       struct file_system_type *type;
 
-       mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
+       type = get_fs_type("rootfs");
+       if (!type)
+               panic("Can't find rootfs type");
+       mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
+       put_filesystem(type);
        if (IS_ERR(mnt))
                panic("Can't create rootfs");
 
@@ -2702,8 +2759,8 @@ static void __init init_mount_tree(void)
        init_task.nsproxy->mnt_ns = ns;
        get_mnt_ns(ns);
 
-       root.mnt = ns->root;
-       root.dentry = ns->root->mnt_root;
+       root.mnt = mnt;
+       root.dentry = mnt->mnt_root;
 
        set_fs_pwd(current->fs, &root);
        set_fs_root(current->fs, &root);
@@ -2720,16 +2777,19 @@ void __init mnt_init(void)
                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 
        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
+       mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
 
-       if (!mount_hashtable)
+       if (!mount_hashtable || !mountpoint_hashtable)
                panic("Failed to allocate mount hash table\n");
 
        printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
 
        for (u = 0; u < HASH_SIZE; u++)
                INIT_LIST_HEAD(&mount_hashtable[u]);
+       for (u = 0; u < HASH_SIZE; u++)
+               INIT_LIST_HEAD(&mountpoint_hashtable[u]);
 
-       br_lock_init(vfsmount_lock);
+       br_lock_init(&vfsmount_lock);
 
        err = sysfs_init();
        if (err)
@@ -2744,17 +2804,14 @@ void __init mnt_init(void)
 
 void put_mnt_ns(struct mnt_namespace *ns)
 {
-       LIST_HEAD(umount_list);
-
        if (!atomic_dec_and_test(&ns->count))
                return;
-       down_write(&namespace_sem);
-       br_write_lock(vfsmount_lock);
-       umount_tree(ns->root, 0, &umount_list);
-       br_write_unlock(vfsmount_lock);
-       up_write(&namespace_sem);
-       release_mounts(&umount_list);
-       kfree(ns);
+       namespace_lock();
+       br_write_lock(&vfsmount_lock);
+       umount_tree(ns->root, 0);
+       br_write_unlock(&vfsmount_lock);
+       namespace_unlock();
+       free_mnt_ns(ns);
 }
 
 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
@@ -2766,7 +2823,7 @@ struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
                 * it is a longterm mount, don't release mnt until
                 * we unmount before file sys is unregistered
                */
-               mnt_make_longterm(mnt);
+               real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
        }
        return mnt;
 }
@@ -2776,7 +2833,9 @@ void kern_unmount(struct vfsmount *mnt)
 {
        /* release long term mount so mount point can be released */
        if (!IS_ERR_OR_NULL(mnt)) {
-               mnt_make_shortterm(mnt);
+               br_write_lock(&vfsmount_lock);
+               real_mount(mnt)->mnt_ns = NULL;
+               br_write_unlock(&vfsmount_lock);
                mntput(mnt);
        }
 }
@@ -2784,5 +2843,119 @@ EXPORT_SYMBOL(kern_unmount);
 
 bool our_mnt(struct vfsmount *mnt)
 {
-       return check_mnt(mnt);
+       return check_mnt(real_mount(mnt));
 }
+
+bool current_chrooted(void)
+{
+       /* Does the current process have a non-standard root */
+       struct path ns_root;
+       struct path fs_root;
+       bool chrooted;
+
+       /* Find the namespace root */
+       ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
+       ns_root.dentry = ns_root.mnt->mnt_root;
+       path_get(&ns_root);
+       while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
+               ;
+
+       get_fs_root(current->fs, &fs_root);
+
+       chrooted = !path_equal(&fs_root, &ns_root);
+
+       path_put(&fs_root);
+       path_put(&ns_root);
+
+       return chrooted;
+}
+
+void update_mnt_policy(struct user_namespace *userns)
+{
+       struct mnt_namespace *ns = current->nsproxy->mnt_ns;
+       struct mount *mnt;
+
+       down_read(&namespace_sem);
+       list_for_each_entry(mnt, &ns->list, mnt_list) {
+               switch (mnt->mnt.mnt_sb->s_magic) {
+               case SYSFS_MAGIC:
+                       userns->may_mount_sysfs = true;
+                       break;
+               case PROC_SUPER_MAGIC:
+                       userns->may_mount_proc = true;
+                       break;
+               }
+               if (userns->may_mount_sysfs && userns->may_mount_proc)
+                       break;
+       }
+       up_read(&namespace_sem);
+}
+
+static void *mntns_get(struct task_struct *task)
+{
+       struct mnt_namespace *ns = NULL;
+       struct nsproxy *nsproxy;
+
+       rcu_read_lock();
+       nsproxy = task_nsproxy(task);
+       if (nsproxy) {
+               ns = nsproxy->mnt_ns;
+               get_mnt_ns(ns);
+       }
+       rcu_read_unlock();
+
+       return ns;
+}
+
+static void mntns_put(void *ns)
+{
+       put_mnt_ns(ns);
+}
+
+static int mntns_install(struct nsproxy *nsproxy, void *ns)
+{
+       struct fs_struct *fs = current->fs;
+       struct mnt_namespace *mnt_ns = ns;
+       struct path root;
+
+       if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
+           !nsown_capable(CAP_SYS_CHROOT) ||
+           !nsown_capable(CAP_SYS_ADMIN))
+               return -EPERM;
+
+       if (fs->users != 1)
+               return -EINVAL;
+
+       get_mnt_ns(mnt_ns);
+       put_mnt_ns(nsproxy->mnt_ns);
+       nsproxy->mnt_ns = mnt_ns;
+
+       /* Find the root */
+       root.mnt    = &mnt_ns->root->mnt;
+       root.dentry = mnt_ns->root->mnt.mnt_root;
+       path_get(&root);
+       while(d_mountpoint(root.dentry) && follow_down_one(&root))
+               ;
+
+       /* Update the pwd and root */
+       set_fs_pwd(fs, &root);
+       set_fs_root(fs, &root);
+
+       path_put(&root);
+       return 0;
+}
+
+static unsigned int mntns_inum(void *ns)
+{
+       struct mnt_namespace *mnt_ns = ns;
+       return mnt_ns->proc_inum;
+}
+
+const struct proc_ns_operations mntns_operations = {
+       .name           = "mnt",
+       .type           = CLONE_NEWNS,
+       .get            = mntns_get,
+       .put            = mntns_put,
+       .install        = mntns_install,
+       .inum           = mntns_inum,
+};