[sheepdog] [PATCH 5/7 v2] sheep: split io_op_done

[Liu Yuan]

On 05/31/2012 09:26 PM, Christoph Hellwig wrote:

> Use a trivial completion handler for peer I/O requests and a slightly more
> complicated one for gateway requests instead of interwinding the two.
> 
> Signed-off-by: Christoph Hellwig <hch at lst.de>
> 
> Index: sheepdog/sheep/sdnet.c
> ===================================================================
> --- sheepdog.orig/sheep/sdnet.c	2012-05-31 15:12:43.639734147 +0200
> +++ sheepdog/sheep/sdnet.c	2012-05-31 15:14:21.031733213 +0200
> @@ -116,35 +116,48 @@ static void check_object_consistency(str
>  static void io_op_done(struct work *work)
>  {
>  	struct request *req = container_of(work, struct request, work);
> +
> +	list_del(&req->request_list);
> +
> +	if (req->rp.result == SD_RES_EIO) {
> +		req->rp.result = SD_RES_NETWORK_ERROR;
> +
> +		eprintf("leaving sheepdog cluster\n");
> +		leave_cluster();
> +	}
> +
> +	resume_pending_requests();
> +	resume_recovery_work();
> +
> +	req_done(req);
> +	return;
> +}
> +
> +static void gateway_op_done(struct work *work)
> +{
> +	struct request *req = container_of(work, struct request, work);
>  	struct sd_req *hdr = &req->rq;
>  
>  	list_del(&req->request_list);
>  
>  	switch (req->rp.result) {
>  	case SD_RES_OLD_NODE_VER:
> +		if (req->rp.epoch > sys->epoch) {
> +			list_add_tail(&req->request_list,
> +				      &sys->wait_rw_queue);
> +			break;
> +		}
> +		/*FALLTHRU*/
>  	case SD_RES_NEW_NODE_VER:
>  	case SD_RES_NETWORK_ERROR:
>  	case SD_RES_WAIT_FOR_JOIN:
>  	case SD_RES_WAIT_FOR_FORMAT:
> -		if (!(req->rq.flags & SD_FLAG_CMD_IO_LOCAL)) {
> -			if (req->rp.epoch > sys->epoch &&
> -			    req->rp.result == SD_RES_OLD_NODE_VER) {
> -				list_add_tail(&req->request_list,
> -						&sys->wait_rw_queue);
> -			} else
> -				goto retry;
> -		}
> -		break;
> +		goto retry;
>  	case SD_RES_EIO:
>  		if (is_access_local(req, hdr->obj.oid)) {
>  			eprintf("leaving sheepdog cluster\n");
>  			leave_cluster();
> -
> -			if (!(req->rq.flags & SD_FLAG_CMD_IO_LOCAL))
> -				goto retry;
> -
> -			/* hack to retry */
> -			req->rp.result = SD_RES_NETWORK_ERROR;
> +			goto retry;
>  		}
>  		break;
>  	case SD_RES_SUCCESS:
> @@ -364,7 +377,7 @@ static void queue_gateway_request(struct
>  		set_consistency_check(req);
>  
>  	req->work.fn = do_gateway_request;
> -	req->work.done = io_op_done;
> +	req->work.done = gateway_op_done;
>  	queue_work(sys->gateway_wqueue, &req->work);
>  }
>  


Both gateway_op_done() and io_op_done() calls
list_del(&req->request_list) and then requeue_request() calls
list_del(&req->request_list) again, looks to me segfault operation?

Thanks,
Yuan