On Tue 16-04-19 00:54:28, Al Viro wrote: > On Mon, Apr 15, 2019 at 04:20:17PM -0700, Khazhismel Kumykov wrote: > > I was able to reproduce this by setting security.capability xattr on a > > blockdev file, then writing to it - when writing to the blockdev we > > never lock the inode, so when we clear the capability we hit this > > lockdep warning. > > > > Is the issue here that we can set this xattr in the first place so we > > have to clear it at all? Or should we really be locking the inode for > > blockdevs after all? I'm not too familiar, but my gut says former > > More interesting question is, WTF do we even touch that thing for > bdev? The thing is, mknod will cheerfully create any number of > different filesystem objects, all giving access to the same block > device. Which of them should have that xattr removed? It makes > no sense whatsoever; moreover, who *cares* about caps for block > device in the first place? > > And if we did, what of another way to modify the block device? > You know, mount it read-write... Yes, Alexander Lochman has sent a patch to silence this warning back in February [1] by just bailing out from file_remove_privs() for non-regular files. But so far you've ignored that patch... Will you pick it up please? Honza [1] https://lore.kernel.org/lkml/cbdc8071-de76-bb0a-6890-15ef21023a70@xxxxxxxxxxxxxx -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
