On Mon 28-11-16 10:57:18, Tejun Heo wrote: > Hello, Jan. > > On Mon, Nov 28, 2016 at 11:07:18AM +0100, Jan Kara wrote: > > As I'm looking into the code, we need a serialization between bdev writeback > > and blkdev_put(). That should be doable if we use writeback_single_inode() > > for writing bdev inode instead of simple filemap_fdatawrite() and then use > > inode_wait_for_writeback() in blkdev_put() but it needs some careful > > thought. > > It's kinda weird that sync() is ends up accessing bdev's without any > synchronization. Can't we just make iterate_bdevs() grab bd_mutex and > verify bd_disk isn't NULL before calling into the callback? This reminded me I've already seen something like this and indeed I've already had a very similar discussion in March - https://patchwork.kernel.org/patch/8556941/ Holding bd_mutex in iterate_devs() works but still nothing protects from flusher thread just walking across the block device inode and trying to write it which would result in the very same oops. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>