Re: [Libguestfs] [PATCH nbdkit v2 3/7] server: protocol: Send reply to NBD_INFO_BLOCK_SIZE if requested

> +{ > + GET_CONN; > + struct nbd_fixed_new_option_reply fixed_new_option_reply; > + struct nbd_fixed_new_option_reply_info_block_size block_size; > + > + fixed_new_option_reply.magic = htobe64 (NBD_REP_MAGIC); > + fixed_new_option_reply.option = htobe32 (option); > + fixed_new_option_reply.reply = htobe32 (reply); > + fixed_new_option_reply.replylen = htobe32 (14); Worth a sizeof computation instead of a magic number here, to explain why we got at 14?

> @@ -637,6 +669,35 @@ negotiate_handshake_newstyle_options (void) > return -1; > } > break; > + case NBD_INFO_BLOCK_SIZE: > + { > + uint32_t minimum, preferred, maximum; > + int r = backend_block_size (conn->top_context, > + &minimum, &preferred, &maximum); > + > + if (r == -1) { > + debug ("newstyle negotiation: %s: " > + "NBD_INFO_BLOCK_SIZE: error from plugin, " > + "ignoring client request", > + optname); > + break; Should this be 'return -1', so that the plugin isn't forced to continue serving later requests after reporting a failure at determining block sizes it wants to use? But I can also see your argument that if we trust the plugin to be robust, our communication with the client is still in sync so it isn't a fatal error...