[PATCH] svcrdma: send_write() must not overflow the device's max sge

[PATCH] svcrdma: send_write() must not overflow the device's max sge

[Steve Wise]

Function send_write() must stop creating sges when it reaches the device
max and return the amount sent in the RDMA Write to the caller.

Signed-off-by: Steve Wise <swise-7bPotxP6k4+P2YhJcF5u+vpXobYPEAuW@public.gmane.org>
---

 net/sunrpc/xprtrdma/svc_rdma_sendto.c |   39 +++++++++++++--------------------
 1 files changed, 15 insertions(+), 24 deletions(-)

diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 49fd21a..9f1b506 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -192,6 +192,8 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
 		xdr_sge_no++;
 		BUG_ON(xdr_sge_no > vec->count);
 		bc -= sge_bytes;
+		if (sge_no == xprt->sc_max_sge)
+			break;
 	}
 
 	/* Prepare WRITE WR */
@@ -209,7 +211,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
 	atomic_inc(&rdma_stat_write);
 	if (svc_rdma_send(xprt, &write_wr))
 		goto err;
-	return 0;
+	return write_len - bc;
  err:
 	svc_rdma_unmap_dma(ctxt);
 	svc_rdma_put_context(ctxt, 0);
@@ -225,7 +227,6 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
 {
 	u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len;
 	int write_len;
-	int max_write;
 	u32 xdr_off;
 	int chunk_off;
 	int chunk_no;
@@ -239,8 +240,6 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
 	res_ary = (struct rpcrdma_write_array *)
 		&rdma_resp->rm_body.rm_chunks[1];
 
-	max_write = xprt->sc_max_sge * PAGE_SIZE;
-
 	/* Write chunks start at the pagelist */
 	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
 	     xfer_len && chunk_no < arg_ary->wc_nchunks;
@@ -260,23 +259,21 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
 						write_len);
 		chunk_off = 0;
 		while (write_len) {
-			int this_write;
-			this_write = min(write_len, max_write);
 			ret = send_write(xprt, rqstp,
 					 ntohl(arg_ch->rs_handle),
 					 rs_offset + chunk_off,
 					 xdr_off,
-					 this_write,
+					 write_len,
 					 vec);
-			if (ret) {
+			if (ret <= 0) {
 				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
 					ret);
 				return -EIO;
 			}
-			chunk_off += this_write;
-			xdr_off += this_write;
-			xfer_len -= this_write;
-			write_len -= this_write;
+			chunk_off += ret;
+			xdr_off += ret;
+			xfer_len -= ret;
+			write_len -= ret;
 		}
 	}
 	/* Update the req with the number of chunks actually used */
@@ -293,7 +290,6 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
 {
 	u32 xfer_len = rqstp->rq_res.len;
 	int write_len;
-	int max_write;
 	u32 xdr_off;
 	int chunk_no;
 	int chunk_off;
@@ -311,8 +307,6 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
 	res_ary = (struct rpcrdma_write_array *)
 		&rdma_resp->rm_body.rm_chunks[2];
 
-	max_write = xprt->sc_max_sge * PAGE_SIZE;
-
 	/* xdr offset starts at RPC message */
 	nchunks = ntohl(arg_ary->wc_nchunks);
 	for (xdr_off = 0, chunk_no = 0;
@@ -330,24 +324,21 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
 						write_len);
 		chunk_off = 0;
 		while (write_len) {
-			int this_write;
-
-			this_write = min(write_len, max_write);
 			ret = send_write(xprt, rqstp,
 					 ntohl(ch->rs_handle),
 					 rs_offset + chunk_off,
 					 xdr_off,
-					 this_write,
+					 write_len,
 					 vec);
-			if (ret) {
+			if (ret <= 0) {
 				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
 					ret);
 				return -EIO;
 			}
-			chunk_off += this_write;
-			xdr_off += this_write;
-			xfer_len -= this_write;
-			write_len -= this_write;
+			chunk_off += ret;
+			xdr_off += ret;
+			xfer_len -= ret;
+			write_len -= ret;
 		}
 	}
 	/* Update the req with the number of chunks actually used */

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH] svcrdma: send_write() must not overflow the device's max sge

[Steve Wise]

I tested this with nfsv3 and v4, over cxgb4 and mlx4: cthon04, iozone, fio, and xfs.  No regressions seen...

Steve.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] svcrdma: send_write() must not overflow the device's max sge

[J. Bruce Fields]

On Wed, Jul 09, 2014 at 01:49:15PM -0500, Steve Wise wrote:
> Function send_write() must stop creating sges when it reaches the device
> max and return the amount sent in the RDMA Write to the caller.

Queueing up for 3.17, thanks.--b.

> 
> Signed-off-by: Steve Wise <swise-7bPotxP6k4+P2YhJcF5u+vpXobYPEAuW@public.gmane.org>
> ---
> 
>  net/sunrpc/xprtrdma/svc_rdma_sendto.c |   39 +++++++++++++--------------------
>  1 files changed, 15 insertions(+), 24 deletions(-)
> 
> diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> index 49fd21a..9f1b506 100644
> --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> @@ -192,6 +192,8 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
>  		xdr_sge_no++;
>  		BUG_ON(xdr_sge_no > vec->count);
>  		bc -= sge_bytes;
> +		if (sge_no == xprt->sc_max_sge)
> +			break;
>  	}
>  
>  	/* Prepare WRITE WR */
> @@ -209,7 +211,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
>  	atomic_inc(&rdma_stat_write);
>  	if (svc_rdma_send(xprt, &write_wr))
>  		goto err;
> -	return 0;
> +	return write_len - bc;
>   err:
>  	svc_rdma_unmap_dma(ctxt);
>  	svc_rdma_put_context(ctxt, 0);
> @@ -225,7 +227,6 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
>  {
>  	u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len;
>  	int write_len;
> -	int max_write;
>  	u32 xdr_off;
>  	int chunk_off;
>  	int chunk_no;
> @@ -239,8 +240,6 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
>  	res_ary = (struct rpcrdma_write_array *)
>  		&rdma_resp->rm_body.rm_chunks[1];
>  
> -	max_write = xprt->sc_max_sge * PAGE_SIZE;
> -
>  	/* Write chunks start at the pagelist */
>  	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
>  	     xfer_len && chunk_no < arg_ary->wc_nchunks;
> @@ -260,23 +259,21 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
>  						write_len);
>  		chunk_off = 0;
>  		while (write_len) {
> -			int this_write;
> -			this_write = min(write_len, max_write);
>  			ret = send_write(xprt, rqstp,
>  					 ntohl(arg_ch->rs_handle),
>  					 rs_offset + chunk_off,
>  					 xdr_off,
> -					 this_write,
> +					 write_len,
>  					 vec);
> -			if (ret) {
> +			if (ret <= 0) {
>  				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
>  					ret);
>  				return -EIO;
>  			}
> -			chunk_off += this_write;
> -			xdr_off += this_write;
> -			xfer_len -= this_write;
> -			write_len -= this_write;
> +			chunk_off += ret;
> +			xdr_off += ret;
> +			xfer_len -= ret;
> +			write_len -= ret;
>  		}
>  	}
>  	/* Update the req with the number of chunks actually used */
> @@ -293,7 +290,6 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
>  {
>  	u32 xfer_len = rqstp->rq_res.len;
>  	int write_len;
> -	int max_write;
>  	u32 xdr_off;
>  	int chunk_no;
>  	int chunk_off;
> @@ -311,8 +307,6 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
>  	res_ary = (struct rpcrdma_write_array *)
>  		&rdma_resp->rm_body.rm_chunks[2];
>  
> -	max_write = xprt->sc_max_sge * PAGE_SIZE;
> -
>  	/* xdr offset starts at RPC message */
>  	nchunks = ntohl(arg_ary->wc_nchunks);
>  	for (xdr_off = 0, chunk_no = 0;
> @@ -330,24 +324,21 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
>  						write_len);
>  		chunk_off = 0;
>  		while (write_len) {
> -			int this_write;
> -
> -			this_write = min(write_len, max_write);
>  			ret = send_write(xprt, rqstp,
>  					 ntohl(ch->rs_handle),
>  					 rs_offset + chunk_off,
>  					 xdr_off,
> -					 this_write,
> +					 write_len,
>  					 vec);
> -			if (ret) {
> +			if (ret <= 0) {
>  				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
>  					ret);
>  				return -EIO;
>  			}
> -			chunk_off += this_write;
> -			xdr_off += this_write;
> -			xfer_len -= this_write;
> -			write_len -= this_write;
> +			chunk_off += ret;
> +			xdr_off += ret;
> +			xfer_len -= ret;
> +			write_len -= ret;
>  		}
>  	}
>  	/* Update the req with the number of chunks actually used */
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH] svcrdma: send_write() must not overflow the device's max sge

[Devesh Sharma]

Reported-By: Devesh Sharma <devesh.sharma at Emulex.com>
Acked-By: Devesh Sharma <Devesh.sharma at Emulex.com>

> -----Original Message-----
> From: linux-rdma-owner@vger.kernel.org [mailto:linux-rdma-
> owner@vger.kernel.org] On Behalf Of Steve Wise
> Sent: Thursday, July 10, 2014 12:19 AM
> To: bfields@fieldses.org
> Cc: linux-rdma@vger.kernel.org; linux-nfs@vger.kernel.org; tom@ogc.us
> Subject: [PATCH] svcrdma: send_write() must not overflow the device's max
> sge
> 
> Function send_write() must stop creating sges when it reaches the device
> max and return the amount sent in the RDMA Write to the caller.
> 
> Signed-off-by: Steve Wise <swise@opengridcomputing.com>
> ---
> 
>  net/sunrpc/xprtrdma/svc_rdma_sendto.c |   39 +++++++++++++--------------
> ------
>  1 files changed, 15 insertions(+), 24 deletions(-)
> 
> diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> index 49fd21a..9f1b506 100644
> --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> @@ -192,6 +192,8 @@ static int send_write(struct svcxprt_rdma *xprt, struct
> svc_rqst *rqstp,
>  		xdr_sge_no++;
>  		BUG_ON(xdr_sge_no > vec->count);
>  		bc -= sge_bytes;
> +		if (sge_no == xprt->sc_max_sge)
> +			break;
>  	}
> 
>  	/* Prepare WRITE WR */
> @@ -209,7 +211,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct
> svc_rqst *rqstp,
>  	atomic_inc(&rdma_stat_write);
>  	if (svc_rdma_send(xprt, &write_wr))
>  		goto err;
> -	return 0;
> +	return write_len - bc;
>   err:
>  	svc_rdma_unmap_dma(ctxt);
>  	svc_rdma_put_context(ctxt, 0);
> @@ -225,7 +227,6 @@ static int send_write_chunks(struct svcxprt_rdma
> *xprt,  {
>  	u32 xfer_len = rqstp->rq_res.page_len + rqstp-
> >rq_res.tail[0].iov_len;
>  	int write_len;
> -	int max_write;
>  	u32 xdr_off;
>  	int chunk_off;
>  	int chunk_no;
> @@ -239,8 +240,6 @@ static int send_write_chunks(struct svcxprt_rdma
> *xprt,
>  	res_ary = (struct rpcrdma_write_array *)
>  		&rdma_resp->rm_body.rm_chunks[1];
> 
> -	max_write = xprt->sc_max_sge * PAGE_SIZE;
> -
>  	/* Write chunks start at the pagelist */
>  	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
>  	     xfer_len && chunk_no < arg_ary->wc_nchunks; @@ -260,23
> +259,21 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
>  						write_len);
>  		chunk_off = 0;
>  		while (write_len) {
> -			int this_write;
> -			this_write = min(write_len, max_write);
>  			ret = send_write(xprt, rqstp,
>  					 ntohl(arg_ch->rs_handle),
>  					 rs_offset + chunk_off,
>  					 xdr_off,
> -					 this_write,
> +					 write_len,
>  					 vec);
> -			if (ret) {
> +			if (ret <= 0) {
>  				dprintk("svcrdma: RDMA_WRITE failed,
> ret=%d\n",
>  					ret);
>  				return -EIO;
>  			}
> -			chunk_off += this_write;
> -			xdr_off += this_write;
> -			xfer_len -= this_write;
> -			write_len -= this_write;
> +			chunk_off += ret;
> +			xdr_off += ret;
> +			xfer_len -= ret;
> +			write_len -= ret;
>  		}
>  	}
>  	/* Update the req with the number of chunks actually used */ @@ -
> 293,7 +290,6 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,  {
>  	u32 xfer_len = rqstp->rq_res.len;
>  	int write_len;
> -	int max_write;
>  	u32 xdr_off;
>  	int chunk_no;
>  	int chunk_off;
> @@ -311,8 +307,6 @@ static int send_reply_chunks(struct svcxprt_rdma
> *xprt,
>  	res_ary = (struct rpcrdma_write_array *)
>  		&rdma_resp->rm_body.rm_chunks[2];
> 
> -	max_write = xprt->sc_max_sge * PAGE_SIZE;
> -
>  	/* xdr offset starts at RPC message */
>  	nchunks = ntohl(arg_ary->wc_nchunks);
>  	for (xdr_off = 0, chunk_no = 0;
> @@ -330,24 +324,21 @@ static int send_reply_chunks(struct svcxprt_rdma
> *xprt,
>  						write_len);
>  		chunk_off = 0;
>  		while (write_len) {
> -			int this_write;
> -
> -			this_write = min(write_len, max_write);
>  			ret = send_write(xprt, rqstp,
>  					 ntohl(ch->rs_handle),
>  					 rs_offset + chunk_off,
>  					 xdr_off,
> -					 this_write,
> +					 write_len,
>  					 vec);
> -			if (ret) {
> +			if (ret <= 0) {
>  				dprintk("svcrdma: RDMA_WRITE failed,
> ret=%d\n",
>  					ret);
>  				return -EIO;
>  			}
> -			chunk_off += this_write;
> -			xdr_off += this_write;
> -			xfer_len -= this_write;
> -			write_len -= this_write;
> +			chunk_off += ret;
> +			xdr_off += ret;
> +			xfer_len -= ret;
> +			write_len -= ret;
>  		}
>  	}
>  	/* Update the req with the number of chunks actually used */
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the
> body of a message to majordomo@vger.kernel.org More majordomo info at
> http://vger.kernel.org/majordomo-info.html