近来碰到一个关于SCTP的问题:某个产品跑在Monta Vista上,其内核版本号为2.6.21;可每次该产品泡在我们的Red Hat Enterprise Linux 5.5上时,都会报错错误信息如下:
^?0 2011-08-16 09:11:57.316267 coolnjmcl018 AMMServ 26555[26566] ../com/AMMSSctpLib.cc,369 [EAMMS1315] AMMSSctpLib::ammsSetSockOpt() setsockopt - Invalid argument ^?1 2011-08-16 09:11:57.316326 coolnjmcl018 AMMServ 26555[26566] ../com/AMMSAccessServer.cc,779 [EAMMS1036] AMMSAccessServer::setSocketOpts(): failed to set SCTP_EVENTS for socket 6
ammsSetSockOpt()调用了Linux API setsockopt()。产品的同事告诉我们他们升级过libsctp.so,于是我们copy了MV中的这个shared lib到RHEL中,问题依旧。后来我们以为是RHEL5.5的内核和Monta Vista的内核不一致,于是我把RHEL的内核从2.6.18-194.el5升级到2.6.18-238.9.1.el5,问题还在。当我们还在考虑要不要把我们的server替换为Monta Vista时,一位同事受http://centos.org/modules/newbb/viewtopic.php?viewmode=threaded&order=ASC&topic_id=18994&forum=37&move=next&topic_time=1236769500
的启发,把setsockopt()最后一个参数由sizeof(events)改为8,竟然成功了! events定义如下。
struct sctp_event_subscribe events;
最终,发现RHEL5和RHEL6中struct sctp_event_subscribe的定义竟然完全不同!要想根本解决上面的问题,只能升级操作系统到RHEL6了。
RHEL5中的定义:http://rhkernel.org/RHEL5+2.6.18-194.el5/include/net/sctp/user.h#L367
/* 364 * Described in Section 7.3 365 * Ancillary Data and Notification Interest Options 366 */ 367struct sctp_event_subscribe { 368 __u8 sctp_data_io_event; 369 __u8 sctp_association_event; 370 __u8 sctp_address_event; 371 __u8 sctp_send_failure_event; 372 __u8 sctp_peer_error_event; 373 __u8 sctp_shutdown_event; 374 __u8 sctp_partial_delivery_event; 375 __u8 sctp_adaption_layer_event; 376};
RHEL 6中的定义:http://rhkernel.org/RHEL6+2.6.32-71.18.2.el6/include/net/sctp/user.h#L406
/* 403 * Described in Section 7.3 404 * Ancillary Data and Notification Interest Options 405 */ 406struct sctp_event_subscribe { 407 __u8 sctp_data_io_event; 408 __u8 sctp_association_event; 409 __u8 sctp_address_event; 410 __u8 sctp_send_failure_event; 411 __u8 sctp_peer_error_event; 412 __u8 sctp_shutdown_event; 413 __u8 sctp_partial_delivery_event; 414 __u8 sctp_adaptation_layer_event; 415 __u8 sctp_authentication_event; 416}; 417