xprtrdma: Add data structure to manage RDMA Send arguments

Problem statement:

Recently Sagi Grimberg <[email protected]> observed that kernel RDMA-
enabled storage initiators don't handle delayed Send completion
correctly. If Send completion is delayed beyond the end of a ULP
transaction, the ULP may release resources that are still being used
by the HCA to complete a long-running Send operation.

This is a common design trait amongst our initiators. Most Send
operations are faster than the ULP transaction they are part of.
Waiting for a completion for these is typically unnecessary.

Infrequently, a network partition or some other problem crops up
where an ordering problem can occur. In NFS parlance, the RPC Reply
arrives and completes the RPC, but the HCA is still retrying the
Send WR that conveyed the RPC Call. In this case, the HCA can try
to use memory that has been invalidated or DMA unmapped, and the
connection is lost. If that memory has been re-used for something
else (possibly not related to NFS), and the Send retransmission
exposes that data on the wire.

Thus we cannot assume that it is safe to release Send-related
resources just because a ULP reply has arrived.

After some analysis, we have determined that the completion
housekeeping will not be difficult for xprtrdma:

 - Inline Send buffers are registered via the local DMA key, and
   are already left DMA mapped for the lifetime of a transport
   connection, thus no additional handling is necessary for those
 - Gathered Sends involving page cache pages _will_ need to
   DMA unmap those pages after the Send completes. But like
   inline send buffers, they are registered via the local DMA key,
   and thus will not need to be invalidated

In addition, RPC completion will need to wait for Send completion
in the latter case. However, nearly always, the Send that conveys
the RPC Call will have completed long before the RPC Reply
arrives, and thus no additional latency will be accrued.

Design notes:

In this patch, the rpcrdma_sendctx object is introduced, and a
lock-free circular queue is added to manage a set of them per
transport.

The RPC client's send path already prevents sending more than one
RPC Call at the same time. This allows us to treat the consumer
side of the queue (rpcrdma_sendctx_get_locked) as if there is a
single consumer thread.

The producer side of the queue (rpcrdma_sendctx_put_locked) is
invoked only from the Send completion handler, which is a single
thread of execution (soft IRQ).

The only care that needs to be taken is with the tail index, which
is shared between the producer and consumer. Only the producer
updates the tail index. The consumer compares the head with the
tail to ensure that the a sendctx that is in use is never handed
out again (or, expressed more conventionally, the queue is empty).

When the sendctx queue empties completely, there are enough Sends
outstanding that posting more Send operations can result in a Send
Queue overflow. In this case, the ULP is told to wait and try again.
This introduces strong Send Queue accounting to xprtrdma.

As a final touch, Jason Gunthorpe <[email protected]>
suggested a mechanism that does not require signaling every Send.
We signal once every N Sends, and perform SGE unmapping of N Send
operations during that one completion.

Reported-by: Sagi Grimberg <[email protected]>
Suggested-by: Jason Gunthorpe <[email protected]>
Signed-off-by: Chuck Lever <[email protected]>
Signed-off-by: Anna Schumaker <[email protected]>

Author: chucklever

net/sunrpc/xprtrdma/rpc_rdma.c

@@ -512,23 +512,26 @@ rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req,
 }
 
 /**
- * rpcrdma_unmap_sges - DMA-unmap Send buffers
- * @ia: interface adapter (device)
- * @req: req with possibly some SGEs to be DMA unmapped
+ * rpcrdma_unmap_sendctx - DMA-unmap Send buffers
+ * @sc: sendctx containing SGEs to unmap
  *
  */
 void
-rpcrdma_unmap_sges(struct rpcrdma_ia *ia, struct rpcrdma_req *req)
+rpcrdma_unmap_sendctx(struct rpcrdma_sendctx *sc)
 {
+	struct rpcrdma_ia *ia = &sc->sc_xprt->rx_ia;
 	struct ib_sge *sge;
 	unsigned int count;
 
+	dprintk("RPC:       %s: unmapping %u sges for sc=%p\n",
+		__func__, sc->sc_unmap_count, sc);
+
 	/* The first two SGEs contain the transport header and
 	 * the inline buffer. These are always left mapped so
 	 * they can be cheaply re-used.
 	 */
-	sge = &req->rl_send_sge[2];
-	for (count = req->rl_mapped_sges; count--; sge++)
+	sge = &sc->sc_sges[2];
+	for (count = sc->sc_unmap_count; count; ++sge, --count)
 		ib_dma_unmap_page(ia->ri_device,
 				  sge->addr, sge->length, DMA_TO_DEVICE);
 }
@@ -539,8 +542,9 @@ static bool
 rpcrdma_prepare_hdr_sge(struct rpcrdma_ia *ia, struct rpcrdma_req *req,
 			u32 len)
 {
+	struct rpcrdma_sendctx *sc = req->rl_sendctx;
 	struct rpcrdma_regbuf *rb = req->rl_rdmabuf;
-	struct ib_sge *sge = &req->rl_send_sge[0];
+	struct ib_sge *sge = sc->sc_sges;
 
 	if (!rpcrdma_dma_map_regbuf(ia, rb))
 		goto out_regbuf;
@@ -550,7 +554,7 @@ rpcrdma_prepare_hdr_sge(struct rpcrdma_ia *ia, struct rpcrdma_req *req,
 
 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr,
 				      sge->length, DMA_TO_DEVICE);
-	req->rl_send_wr.num_sge++;
+	sc->sc_wr.num_sge++;
 	return true;
 
 out_regbuf:
@@ -565,10 +569,11 @@ static bool
 rpcrdma_prepare_msg_sges(struct rpcrdma_ia *ia, struct rpcrdma_req *req,
 			 struct xdr_buf *xdr, enum rpcrdma_chunktype rtype)
 {
+	struct rpcrdma_sendctx *sc = req->rl_sendctx;
 	unsigned int sge_no, page_base, len, remaining;
 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
 	struct ib_device *device = ia->ri_device;
-	struct ib_sge *sge = req->rl_send_sge;
+	struct ib_sge *sge = sc->sc_sges;
 	u32 lkey = ia->ri_pd->local_dma_lkey;
 	struct page *page, **ppages;
 
@@ -631,7 +636,7 @@ rpcrdma_prepare_msg_sges(struct rpcrdma_ia *ia, struct rpcrdma_req *req,
 			sge[sge_no].length = len;
 			sge[sge_no].lkey = lkey;
 
-			req->rl_mapped_sges++;
+			sc->sc_unmap_count++;
 			ppages++;
 			remaining -= len;
 			page_base = 0;
@@ -657,24 +662,24 @@ rpcrdma_prepare_msg_sges(struct rpcrdma_ia *ia, struct rpcrdma_req *req,
 			goto out_mapping_err;
 		sge[sge_no].length = len;
 		sge[sge_no].lkey = lkey;
-		req->rl_mapped_sges++;
+		sc->sc_unmap_count++;
 	}
 
 out:
-	req->rl_send_wr.num_sge += sge_no;
+	sc->sc_wr.num_sge += sge_no;
 	return true;
 
 out_regbuf:
 	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
 	return false;
 
 out_mapping_overflow:
-	rpcrdma_unmap_sges(ia, req);
+	rpcrdma_unmap_sendctx(sc);
 	pr_err("rpcrdma: too many Send SGEs (%u)\n", sge_no);
 	return false;
 
 out_mapping_err:
-	rpcrdma_unmap_sges(ia, req);
+	rpcrdma_unmap_sendctx(sc);
 	pr_err("rpcrdma: Send mapping error\n");
 	return false;
 }
@@ -694,8 +699,11 @@ rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
 			  struct rpcrdma_req *req, u32 hdrlen,
 			  struct xdr_buf *xdr, enum rpcrdma_chunktype rtype)
 {
-	req->rl_send_wr.num_sge = 0;
-	req->rl_mapped_sges = 0;
+	req->rl_sendctx = rpcrdma_sendctx_get_locked(&r_xprt->rx_buf);
+	if (!req->rl_sendctx)
+		return -ENOBUFS;
+	req->rl_sendctx->sc_wr.num_sge = 0;
+	req->rl_sendctx->sc_unmap_count = 0;
 
 	if (!rpcrdma_prepare_hdr_sge(&r_xprt->rx_ia, req, hdrlen))
 		return -EIO;

net/sunrpc/xprtrdma/transport.c

@@ -687,7 +687,6 @@ xprt_rdma_free(struct rpc_task *task)
 
 	if (!list_empty(&req->rl_registered))
 		ia->ri_ops->ro_unmap_sync(r_xprt, &req->rl_registered);
-	rpcrdma_unmap_sges(ia, req);
 	rpcrdma_buffer_put(req);
 }
 
@@ -790,11 +789,12 @@ void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
 		   r_xprt->rx_stats.failed_marshal_count,
 		   r_xprt->rx_stats.bad_reply_count,
 		   r_xprt->rx_stats.nomsg_call_count);
-	seq_printf(seq, "%lu %lu %lu %lu\n",
+	seq_printf(seq, "%lu %lu %lu %lu %lu\n",
 		   r_xprt->rx_stats.mrs_recovered,
 		   r_xprt->rx_stats.mrs_orphaned,
 		   r_xprt->rx_stats.mrs_allocated,
-		   r_xprt->rx_stats.local_inv_needed);
+		   r_xprt->rx_stats.local_inv_needed,
+		   r_xprt->rx_stats.empty_sendctx_q);
 }
 
 static int