Hello, NBD server could be upgraded if we have multiple connections. But if we have only one connection, after we take down NBD server, all inflight IO could finally timeout and return error. These patches fix this using current reconfiguration framework. I noticed that Mike has following patchset nbd: local daemon restart support https://lore.kernel.org/linux-block/5DD41C49.3080209@xxxxxxxxxx/ It add another netlink interface (NBD_ATTR_SWAP_SOCKETS) and requeue request immediately after recongirure/swap socket. It do not need to wait for timeout to fire and requeue in timeout handler, which seems more like an improvement. Let fix this in current framework first. Changes compared to v2: Fix comments in nbd_read_stat() to be aligned with the code change suggested by Mike Christie. Hou Pu (2): nbd: enable replace socket if only one connection is configured nbd: requeue command if the soecket is changed drivers/block/nbd.c | 27 +++++++++++++++++++-------- 1 file changed, 19 insertions(+), 8 deletions(-) -- 2.11.0
