On Thu, 27 Nov 2014 10:40:06 +0100 Jan Kara <jack@xxxxxxx> wrote: > > so we no longer require that the address_space be stabilized after > > lock_page(). Of course something needs to protect the bdi and I'm not > > sure what that is, but we're talking about umount and that quiesces and > > evicts lots of things before proceeding, so surely there's something in > > there which will save us ;) > In do_wp_page() the process doing the fault and ending in > balance_dirty_pages() has to have the page mapped, thus it has to have the > file open => no umount. Actually, umount isn't enough to kill the backing_dev_info. It's an attribute of the device itself (for blockdevs it's a field in request_queue) so I assume it will be stable until device hot-unplug, losetup -d, rmmod, etc. If the backing_dev can go away in the middle of a pagefault against that device then we have bigger problems ;) -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
