On Sat 08-12-18 00:49:44, Al Viro wrote: > On Fri, Dec 07, 2018 at 08:49:16PM +0100, Alexander Lochmann wrote: > > > > _What_ SUID bit? We are talking about a write to block device, for fsck sake... > > > > > That's the way I understood Jan's explanation: > > " > > Thinking more about this I'm not sure if this is actually the right > > solution. Because for example the write(2) can set S_NOSEC flag wrongly > > when it would race with chmod adding SUID bit. So probably we rather need > > to acquire i_rwsem in blkdev_write_iter() if file does not have S_NOSEC set > > (we don't want to acquire it unconditionally as that would heavily impact > > scalability of block device writes). > > IDGI. We are talking about a block device here. What business could > file_remove_privs() have doing _anything_ to it? should_remove_suid() returns > to return 0 for those; what case do you have in mind? Somebody setting > security.capabilities on a block device inode? > > IMO the bug here is file_remove_privs() not buggering off immediately > after having observed that we are dealing with a block device. It really > has nothing useful to do. I didn't notice that S_ISREG() check in should_remove_suid(). My bad. And I wasn't quite sure whether some security module does not rely on inode_need_killpriv security hook. But now when I grep I see that cap_inode_need_killpriv() is really the only user and security.capabilities probably don't make sense for it since block devices cannot be executed anyway. So yes, the easiest fix is to just bail from file_remove_privs(). Probably for anything that is not a regular file, right? Directories cannot be written anyway and for pipes and character devices same logic applies as for block devices. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
