[sheepdog] [PATCH 7/7] sheep: fix SD_RES_OLD_NODE_VER handling in check_request_epoch

[Liu Yuan]

On 06/01/2012 03:15 PM, Liu Yuan wrote:

> On 05/31/2012 05:48 PM, Christoph Hellwig wrote:
> 
>>  		/* ask gateway to retry. */
>>  		req->rp.result = SD_RES_OLD_NODE_VER;
>>  		req->rp.epoch = sys->epoch;
>> -		req->work.done(&req->work);
>> +		req_done(req);
>>  		return -1;
>>  	} else if (after(req->rq.epoch, sys->epoch)) {
>>  		eprintf("new node version %u, %u, %x\n",
> 
> 
> So for the gateway happens to be this local node, we response with
> SD_RES_OLD_NODE_VER to upper layer? I don't think upper layer should
> handle this error code, we need to call work.done() to queue the request
> again and fix double deletion problem.
> 


By a second thought, it seems that this case won't happen at all. Now
check req epoch is always for IO requests, which are from peers not itself.

Thanks,
Yuan