Upgrade to 4.4.50-rt62
[kvmfornfv.git] / kernel / fs / namespace.c
index 2893702..c09f16e 100644 (file)
@@ -594,24 +594,35 @@ static void delayed_free_vfsmnt(struct rcu_head *head)
 }
 
 /* call under rcu_read_lock */
-bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
+int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
 {
        struct mount *mnt;
        if (read_seqretry(&mount_lock, seq))
-               return false;
+               return 1;
        if (bastard == NULL)
-               return true;
+               return 0;
        mnt = real_mount(bastard);
        mnt_add_count(mnt, 1);
        if (likely(!read_seqretry(&mount_lock, seq)))
-               return true;
+               return 0;
        if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
                mnt_add_count(mnt, -1);
-               return false;
+               return 1;
+       }
+       return -1;
+}
+
+/* call under rcu_read_lock */
+bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
+{
+       int res = __legitimize_mnt(bastard, seq);
+       if (likely(!res))
+               return true;
+       if (unlikely(res < 0)) {
+               rcu_read_unlock();
+               mntput(bastard);
+               rcu_read_lock();
        }
-       rcu_read_unlock();
-       mntput(bastard);
-       rcu_read_lock();
        return false;
 }
 
@@ -736,26 +747,50 @@ static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
        return NULL;
 }
 
-static struct mountpoint *new_mountpoint(struct dentry *dentry)
+static struct mountpoint *get_mountpoint(struct dentry *dentry)
 {
-       struct hlist_head *chain = mp_hash(dentry);
-       struct mountpoint *mp;
+       struct mountpoint *mp, *new = NULL;
        int ret;
 
-       mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
-       if (!mp)
+       if (d_mountpoint(dentry)) {
+mountpoint:
+               read_seqlock_excl(&mount_lock);
+               mp = lookup_mountpoint(dentry);
+               read_sequnlock_excl(&mount_lock);
+               if (mp)
+                       goto done;
+       }
+
+       if (!new)
+               new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
+       if (!new)
                return ERR_PTR(-ENOMEM);
 
+
+       /* Exactly one processes may set d_mounted */
        ret = d_set_mounted(dentry);
-       if (ret) {
-               kfree(mp);
-               return ERR_PTR(ret);
-       }
 
-       mp->m_dentry = dentry;
-       mp->m_count = 1;
-       hlist_add_head(&mp->m_hash, chain);
-       INIT_HLIST_HEAD(&mp->m_list);
+       /* Someone else set d_mounted? */
+       if (ret == -EBUSY)
+               goto mountpoint;
+
+       /* The dentry is not available as a mountpoint? */
+       mp = ERR_PTR(ret);
+       if (ret)
+               goto done;
+
+       /* Add the new mountpoint to the hash table */
+       read_seqlock_excl(&mount_lock);
+       new->m_dentry = dentry;
+       new->m_count = 1;
+       hlist_add_head(&new->m_hash, mp_hash(dentry));
+       INIT_HLIST_HEAD(&new->m_list);
+       read_sequnlock_excl(&mount_lock);
+
+       mp = new;
+       new = NULL;
+done:
+       kfree(new);
        return mp;
 }
 
@@ -1219,7 +1254,7 @@ EXPORT_SYMBOL(replace_mount_options);
 /* 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 = proc_mounts(m);
+       struct proc_mounts *p = m->private;
 
        down_read(&namespace_sem);
        if (p->cached_event == p->ns->event) {
@@ -1240,7 +1275,7 @@ 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 = proc_mounts(m);
+       struct proc_mounts *p = m->private;
 
        p->cached_mount = seq_list_next(v, &p->ns->list, pos);
        p->cached_index = *pos;
@@ -1254,7 +1289,7 @@ static void m_stop(struct seq_file *m, void *v)
 
 static int m_show(struct seq_file *m, void *v)
 {
-       struct proc_mounts *p = proc_mounts(m);
+       struct proc_mounts *p = m->private;
        struct mount *r = list_entry(v, struct mount, mnt_list);
        return p->show(m, &r->mnt);
 }
@@ -1550,11 +1585,12 @@ void __detach_mounts(struct dentry *dentry)
        struct mount *mnt;
 
        namespace_lock();
+       lock_mount_hash();
        mp = lookup_mountpoint(dentry);
        if (IS_ERR_OR_NULL(mp))
                goto out_unlock;
 
-       lock_mount_hash();
+       event++;
        while (!hlist_empty(&mp->m_list)) {
                mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
                if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
@@ -1563,9 +1599,9 @@ void __detach_mounts(struct dentry *dentry)
                }
                else umount_tree(mnt, UMOUNT_CONNECTED);
        }
-       unlock_mount_hash();
        put_mountpoint(mp);
 out_unlock:
+       unlock_mount_hash();
        namespace_unlock();
 }
 
@@ -1954,9 +1990,7 @@ retry:
        namespace_lock();
        mnt = lookup_mnt(path);
        if (likely(!mnt)) {
-               struct mountpoint *mp = lookup_mountpoint(dentry);
-               if (!mp)
-                       mp = new_mountpoint(dentry);
+               struct mountpoint *mp = get_mountpoint(dentry);
                if (IS_ERR(mp)) {
                        namespace_unlock();
                        mutex_unlock(&dentry->d_inode->i_mutex);
@@ -1975,7 +2009,11 @@ retry:
 static void unlock_mount(struct mountpoint *where)
 {
        struct dentry *dentry = where->m_dentry;
+
+       read_seqlock_excl(&mount_lock);
        put_mountpoint(where);
+       read_sequnlock_excl(&mount_lock);
+
        namespace_unlock();
        mutex_unlock(&dentry->d_inode->i_mutex);
 }
@@ -2394,8 +2432,10 @@ static int do_new_mount(struct path *path, const char *fstype, int flags,
                        mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
                }
                if (type->fs_flags & FS_USERNS_VISIBLE) {
-                       if (!fs_fully_visible(type, &mnt_flags))
+                       if (!fs_fully_visible(type, &mnt_flags)) {
+                               put_filesystem(type);
                                return -EPERM;
+                       }
                }
        }
 
@@ -3045,9 +3085,9 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
        touch_mnt_namespace(current->nsproxy->mnt_ns);
        /* A moved mount should not expire automatically */
        list_del_init(&new_mnt->mnt_expire);
+       put_mountpoint(root_mp);
        unlock_mount_hash();
        chroot_fs_refs(&root, &new);
-       put_mountpoint(root_mp);
        error = 0;
 out4:
        unlock_mount(old_mp);
@@ -3211,6 +3251,8 @@ static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
        down_read(&namespace_sem);
        list_for_each_entry(mnt, &ns->list, mnt_list) {
                struct mount *child;
+               int mnt_flags;
+
                if (mnt->mnt.mnt_sb->s_type != type)
                        continue;
 
@@ -3220,17 +3262,34 @@ static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
                if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
                        continue;
 
+               /* Read the mount flags and filter out flags that
+                * may safely be ignored.
+                */
+               mnt_flags = mnt->mnt.mnt_flags;
+               if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
+                       mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
+
+               /* Don't miss readonly hidden in the superblock flags */
+               if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
+                       mnt_flags |= MNT_LOCK_READONLY;
+
                /* Verify the mount flags are equal to or more permissive
                 * than the proposed new mount.
                 */
-               if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
+               if ((mnt_flags & MNT_LOCK_READONLY) &&
                    !(new_flags & MNT_READONLY))
                        continue;
-               if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
+               if ((mnt_flags & MNT_LOCK_NODEV) &&
                    !(new_flags & MNT_NODEV))
                        continue;
-               if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
-                   ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
+               if ((mnt_flags & MNT_LOCK_NOSUID) &&
+                   !(new_flags & MNT_NOSUID))
+                       continue;
+               if ((mnt_flags & MNT_LOCK_NOEXEC) &&
+                   !(new_flags & MNT_NOEXEC))
+                       continue;
+               if ((mnt_flags & MNT_LOCK_ATIME) &&
+                   ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
                        continue;
 
                /* This mount is not fully visible if there are any
@@ -3240,16 +3299,18 @@ static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
                        struct inode *inode = child->mnt_mountpoint->d_inode;
                        /* Only worry about locked mounts */
-                       if (!(mnt->mnt.mnt_flags & MNT_LOCKED))
+                       if (!(child->mnt.mnt_flags & MNT_LOCKED))
                                continue;
                        /* Is the directory permanetly empty? */
                        if (!is_empty_dir_inode(inode))
                                goto next;
                }
                /* Preserve the locked attributes */
-               *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \
-                                                       MNT_LOCK_NODEV    | \
-                                                       MNT_LOCK_ATIME);
+               *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
+                                              MNT_LOCK_NODEV    | \
+                                              MNT_LOCK_NOSUID   | \
+                                              MNT_LOCK_NOEXEC   | \
+                                              MNT_LOCK_ATIME);
                visible = true;
                goto found;
        next:   ;