On Mon, 05/18 20:35, Max Reitz wrote: > On 15.05.2015 08:04, Fam Zheng wrote: > >So that NBD export cannot submit IO during bdrv_drain_all(). > > > >Signed-off-by: Fam Zheng <f...@redhat.com> > >--- > > nbd.c | 18 ++++++++++++++++++ > > 1 file changed, 18 insertions(+) > > But can_read is not cleared until nbd_update_can_read() is called, which is > only in nbd_request_get(), nbd_request_put(), and nbd_co_receive_request(), > which are all called from nbd_trip() - so the one request which clears > can_read will most probably be itself processed. > > Maybe we should call nbd_update_can_read() in nbd_op_blocker_changed(), for > all clients of the export? Or just once somewhere at the beginning of > processing a request, and if io_blocked is set, just yield...
I think you're right. I'll fix that. Fam
