[PATCH RFC 08/11] svcrdma: Add svc_rdma_skip_payloads()

[Chuck Lever <chuck.lever@oracle.com>]

We'll need a generic mechanism for processing only the parts of an
egress RPC message that are _not_ a READ payload. This will be used
in subsequent patches.

This is a separate patch to reduce the complexity of subsequent
patches, so that the logic of this new mechanism can be separately
reviewed.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 include/linux/sunrpc/svc_rdma.h       |    4 ++
 net/sunrpc/xprtrdma/svc_rdma_sendto.c |   72 +++++++++++++++++++++++++++++++++
 2 files changed, 76 insertions(+)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index 37e4c597dc71..93642a889535 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -195,6 +195,10 @@ extern struct svc_rdma_send_ctxt *
 		svc_rdma_send_ctxt_get(struct svcxprt_rdma *rdma);
 extern void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
 				   struct svc_rdma_send_ctxt *ctxt);
+extern int svc_rdma_skip_payloads(const struct xdr_buf *xdr,
+				  const struct svc_rdma_recv_ctxt *rctxt,
+				  int (*actor)(const struct xdr_buf *, void *),
+				  void *data);
 extern int svc_rdma_send(struct svcxprt_rdma *rdma, struct ib_send_wr *wr);
 extern int svc_rdma_map_reply_msg(struct svcxprt_rdma *rdma,
 				  struct svc_rdma_send_ctxt *sctxt,
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 9fe7b0d1e335..85c91d0debb4 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -515,6 +515,78 @@ svc_rdma_encode_reply_chunk(const struct svc_rdma_recv_ctxt *rctxt,
 		return svc_rdma_encode_write_chunk(sctxt, &payload);
 }
 
+static inline int
+xdr_buf_process_region(const struct xdr_buf *xdr,
+		       unsigned int offset, unsigned int length,
+		       int (*actor)(const struct xdr_buf *, void *),
+		       void *data)
+{
+	struct xdr_buf subbuf;
+
+	if (!length)
+		return 0;
+	if (xdr_buf_subsegment(xdr, &subbuf, offset, length))
+		return -EMSGSIZE;
+	return actor(&subbuf, data);
+}
+
+/**
+ * svc_rdma_skip_payloads - Call an actor for non-payload regions of @xdr
+ * @xdr: xdr_buf to process
+ * @rctxt: Write and Reply chunks provided by client
+ * @actor: function to invoke on that region
+ * @data: pointer to arguments for @actor
+ *
+ * This mechanism must ignore not only READ payloads that were already
+ * sent via RDMA Write, but also XDR padding for those payloads that
+ * the upper layer has added.
+ *
+ * Assumptions:
+ *  The xdr->len and rp_ fields are aligned to 4-byte multiples.
+ *
+ * Returns:
+ *   On success, zero,
+ *   %-EMSGSIZE on XDR buffer overflow, or
+ *   The return value of @actor
+ */
+int svc_rdma_skip_payloads(const struct xdr_buf *xdr,
+			   const struct svc_rdma_recv_ctxt *rctxt,
+			   int (*actor)(const struct xdr_buf *, void *),
+			   void *data)
+{
+	const unsigned int num_payloads = rctxt ? rctxt->rc_cur_payload : 0;
+	unsigned int offset, length;
+	int i, ret;
+
+	if (likely(!num_payloads))
+		return actor(xdr, data);
+
+	/* Before the first READ payload */
+	offset = 0;
+	length = rctxt->rc_read_payloads[0].rp_offset;
+	ret = xdr_buf_process_region(xdr, offset, length, actor, data);
+	if (ret < 0)
+		return ret;
+
+	/* Any middle READ payloads */
+	for (i = 0; i + 1 < num_payloads; i++) {
+		offset = xdr_align_size(length + rctxt->rc_read_payloads[i].rp_length);
+		length = rctxt->rc_read_payloads[i + 1].rp_offset - offset;
+		ret = xdr_buf_process_region(xdr, offset, length, actor, data);
+		if (ret < 0)
+			return ret;
+	}
+
+	/* After the last READ payload */
+	offset = xdr_align_size(length + rctxt->rc_read_payloads[i].rp_length);
+	length = xdr->len - offset;
+	ret = xdr_buf_process_region(xdr, offset, length, actor, data);
+	if (ret < 0)
+		return ret;
+
+	return 0;
+}
+
 static int svc_rdma_dma_map_page(struct svcxprt_rdma *rdma,
 				 struct svc_rdma_send_ctxt *ctxt,
 				 struct page *page,