On Wed, Jun 14, 2017 at 05:01:11PM +0800, Anand Jain wrote: > > I have a prototype for the > > preallocated flush bio but was waiting until these cleanups are in > > before snding it. > > oh. sorry for the delay if any. -ENOMEM was only purpose I had/have > in mind for peralloc, but looks like there is another purpose which > I am not aware of. The bio preallocation is a slight optimization, we don't have to allocate and free the bio all the time. > > I've noticed that threre are two more "if (!device->bdev)" checks under > > the device lock in write_all_supers, that might be worth removing. > > > However, a NULL bdev and device->missing are related and I think there > > are some dark corners in dev replace where the invariant can be > > temporarily broken. > > Ok. Thanks for the hints. Trying to dig. > > > Given that we probably don't have great testing > > coverage of devices with the flush capabilities, > > Any hints on what type of test coverage ? Devices used for normal testing, any sort of test basically, that do not have the flush capabilities will never exercise the in write_dev_flush. So any reference counts or failures will not get covered. -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
