On Wed, 2017-03-08 at 11:07 +0100, Wouter Verhelst wrote: > On Tue, Feb 28, 2017 at 11:57:11AM -0500, Josef Bacik wrote: > > > > The existing ioctl interface for configuring NBD devices is a bit > > cumbersome and hard to extend. The other problem is we leave a > > userspace app sitting in it's syscall until the device disconnects, > > which is less than ideal. > True. > > On the other hand, it has the advantage that you leave a userspace > app > sitting around until the device disconnects, which allows for some > form > of recovery in case you're doing root-on-NBD and the server has a > hiccup. Don't underestimate the advantage of that. > > (of course, that requires that the return value of NBD_DO_IT makes a > difference between "unexpected connection drop" and "we sent > NBD_CMD_DISC", but that's a different matter entirely) > Stay tuned for further developments ;). Yeah the problem is that even though we can return and allow the user to reconnect, we completely tear down the device and will return EIO to anything that comes in while we're reconnecting, which sucks for users. The patches that I'm testing now will multi-cast messages over netlink when a link goes down so a user space application can reconnect and provide a new connection seamlessly. The next step after that is to allow a complete failure of all connections and we will simply sit there and queue IO until userspace reconnects or the configured timeout elapses at which point we'll tear down the device. The end goal of all of this is seamless reconnects without throwing errors. Thanks, Josef
