Fix nasty /proc vulnerability

We have a bad interaction with both the kernel and user space being able
to change some of the /proc file status.  This fixes the most obvious
part of it, but I expect we'll also make it harder for users to modify
even their "own" files in /proc.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 243a94af0427b2630fb85f489a5419410dac3bfc..0cb8f20d000c25118947fcafa81606300ced35f8 100644 (file)
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -1338,6 +1338,7 @@ static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
                } else {
                        inode->i_uid = 0;
                        inode->i_gid = 0;
+                       inode->i_mode = 0;
                }
                security_task_to_inode(task, inode);
                put_task_struct(task);