[PATCH] namespace: update event counter when umounting a deleted dentry
From: Andrey Ulanov
Date: Thu Apr 14 2016 - 19:49:05 EST
This change fixes a problem originally reported at:
https://github.com/coreos/bugs/issues/1153
You can find code that reproduces it at:
https://github.com/coreos/bugs/issues/1153#issuecomment-210201681
- m_start() in fs/namespace.c expects that ns->event is incremented each
time a mount added or removed from ns->list.
- umount_tree() removes items from the list but does not increment event
counter, expecting that it's done before the function is called.
- There are some codepaths that call umount_tree() without updating
"event" counter. e.g. from __detach_mounts().
- When this happens m_start may reuse a cached mount structure that no
longer belongs to ns->list (i.e. use after free which usually leads
to infinite loop).
This change fixes the above problem by incrementing global event counter
before invoking umount_tree().
Signed-off-by: Andrey Ulanov <andreyu@xxxxxxxxxx>
---
fs/namespace.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/fs/namespace.c b/fs/namespace.c
index 4fb1691..b06bb9a 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1562,6 +1562,7 @@ void __detach_mounts(struct dentry *dentry)
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) {
@@ -1576,7 +1577,7 @@ out_unlock:
namespace_unlock();
}
-/*
+/*
* Is the caller allowed to modify his namespace?
*/
static inline bool may_mount(void)
@@ -1765,6 +1766,7 @@ void drop_collected_mounts(struct vfsmount *mnt)
{
namespace_lock();
lock_mount_hash();
+ event++;
umount_tree(real_mount(mnt), UMOUNT_SYNC);
unlock_mount_hash();
namespace_unlock();
--
2.8.0.rc3.226.g39d4020