[PATCH v1 0/3] server-side NFS/RDMA patches proposed for v4.9

[PATCH v1 0/3] server-side NFS/RDMA patches proposed for v4.9

[Chuck Lever]

Posted for review, the following patch series makes these changes:

- Introduce simple RDMA-CM private message exchange
- Support Remote Invalidation

Available in the "nfsd-rdma-for-4.9" topic branch of this git repo:

git://git.linux-nfs.org/projects/cel/cel-2.6.git


Or for browsing:

http://git.linux-nfs.org/?p=cel/cel-2.6.git;a=log;h=refs/heads/nfsd-rdma-for-4.9


---

Chuck Lever (3):
      rpcrdma: RDMA/CM private message data structure
      svcrdma: Server-side support for rpcrdma_connect_private
      svcrdma: support Remote Invalidation for prototyping


 include/linux/sunrpc/rpc_rdma.h          |   35 ++++++++++++++++++
 include/linux/sunrpc/svc_rdma.h          |    1 +
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |   58 ++++++++++++++++++++++++++++--
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   42 +++++++++++++++++++---
 4 files changed, 128 insertions(+), 8 deletions(-)

--
Chuck Lever

[PATCH v1 1/3] rpcrdma: RDMA/CM private message data structure

[Chuck Lever]

Introduce data structure used by both client and server to exchange
implementation details during RDMA/CM connection establishment.

This is an experimental out-of-band exchange between Linux
RPC-over-RDMA Version One implementations, replacing the deprecated
CCP (see RFC 5666bis). The purpose of this extension is to enable
prototyping of features that might be introduced in a subsequent
version of RPC-over-RDMA.

Suggested by Christoph Hellwig and Devesh Sharma.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 include/linux/sunrpc/rpc_rdma.h |   35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/include/linux/sunrpc/rpc_rdma.h b/include/linux/sunrpc/rpc_rdma.h
index 3b1ff38..a7da6bf 100644
--- a/include/linux/sunrpc/rpc_rdma.h
+++ b/include/linux/sunrpc/rpc_rdma.h
@@ -41,6 +41,7 @@
 #define _LINUX_SUNRPC_RPC_RDMA_H
 
 #include <linux/types.h>
+#include <linux/bitops.h>
 
 #define RPCRDMA_VERSION		1
 #define rpcrdma_version		cpu_to_be32(RPCRDMA_VERSION)
@@ -129,4 +130,38 @@ enum rpcrdma_proc {
 #define rdma_done	cpu_to_be32(RDMA_DONE)
 #define rdma_error	cpu_to_be32(RDMA_ERROR)
 
+/*
+ * Private extension to RPC-over-RDMA Version One.
+ * Message passed during RDMA-CM connection set-up.
+ *
+ * Add new fields at the end, and don't permute existing
+ * fields.
+ */
+struct rpcrdma_connect_private {
+	__be32			cp_magic;
+	u8			cp_version;
+	u8			cp_flags;
+	u8			cp_send_size;
+	u8			cp_recv_size;
+} __packed;
+
+#define rpcrdma_cmp_magic	__cpu_to_be32(0xf6ab0e18)
+
+enum {
+	RPCRDMA_CMP_VERSION		= 1,
+	RPCRDMA_CMP_F_SND_W_INV_OK	= BIT(0),
+};
+
+static inline u8
+rpcrdma_encode_buffer_size(unsigned int size)
+{
+	return (size >> 10) - 1;
+}
+
+static inline unsigned int
+rpcrdma_decode_buffer_size(u8 val)
+{
+	return ((unsigned int)val + 1) << 10;
+}
+
 #endif				/* _LINUX_SUNRPC_RPC_RDMA_H */

[PATCH v1 2/3] svcrdma: Server-side support for rpcrdma_connect_private

[Chuck Lever]

Prepare to receive an RDMA-CM private message when handling a new
connection attempt, and send a similar message as part of connection
acceptance.

Both sides can communicate their various implementation limits.
Implementations that don't support this sideband protocol ignore it.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   34 ++++++++++++++++++++++++++----
 1 file changed, 30 insertions(+), 4 deletions(-)

diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index dd94401..4843824 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -642,6 +642,21 @@ int svc_rdma_repost_recv(struct svcxprt_rdma *xprt, gfp_t flags)
 	return ret;
 }
 
+static void
+svc_rdma_parse_connect_private(struct svcxprt_rdma *newxprt,
+			       struct rdma_conn_param *param)
+{
+	const struct rpcrdma_connect_private *pmsg = param->private_data;
+
+	if (pmsg &&
+	    pmsg->cp_magic == rpcrdma_cmp_magic &&
+	    pmsg->cp_version == RPCRDMA_CMP_VERSION) {
+		dprintk("svcrdma: client send_size %u, recv_size %u\n",
+			rpcrdma_decode_buffer_size(pmsg->cp_send_size),
+			rpcrdma_decode_buffer_size(pmsg->cp_recv_size));
+	}
+}
+
 /*
  * This function handles the CONNECT_REQUEST event on a listening
  * endpoint. It is passed the cma_id for the _new_ connection. The context in
@@ -653,7 +668,8 @@ int svc_rdma_repost_recv(struct svcxprt_rdma *xprt, gfp_t flags)
  * will call the recvfrom method on the listen xprt which will accept the new
  * connection.
  */
-static void handle_connect_req(struct rdma_cm_id *new_cma_id, size_t client_ird)
+static void handle_connect_req(struct rdma_cm_id *new_cma_id,
+			       struct rdma_conn_param *param)
 {
 	struct svcxprt_rdma *listen_xprt = new_cma_id->context;
 	struct svcxprt_rdma *newxprt;
@@ -669,9 +685,10 @@ static void handle_connect_req(struct rdma_cm_id *new_cma_id, size_t client_ird)
 	new_cma_id->context = newxprt;
 	dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
 		newxprt, newxprt->sc_cm_id, listen_xprt);
+	svc_rdma_parse_connect_private(newxprt, param);
 
 	/* Save client advertised inbound read limit for use later in accept. */
-	newxprt->sc_ord = client_ird;
+	newxprt->sc_ord = param->initiator_depth;
 
 	/* Set the local and remote addresses in the transport */
 	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
@@ -706,8 +723,7 @@ static int rdma_listen_handler(struct rdma_cm_id *cma_id,
 		dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
 			"event = %s (%d)\n", cma_id, cma_id->context,
 			rdma_event_msg(event->event), event->event);
-		handle_connect_req(cma_id,
-				   event->param.conn.initiator_depth);
+		handle_connect_req(cma_id, &event->param.conn);
 		break;
 
 	case RDMA_CM_EVENT_ESTABLISHED:
@@ -941,6 +957,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	struct svcxprt_rdma *listen_rdma;
 	struct svcxprt_rdma *newxprt = NULL;
 	struct rdma_conn_param conn_param;
+	struct rpcrdma_connect_private pmsg;
 	struct ib_qp_init_attr qp_attr;
 	struct ib_device *dev;
 	unsigned int i;
@@ -1094,11 +1111,20 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	/* Swap out the handler */
 	newxprt->sc_cm_id->event_handler = rdma_cma_handler;
 
+	/* Construct RDMA-CM private message */
+	pmsg.cp_magic = rpcrdma_cmp_magic;
+	pmsg.cp_version = RPCRDMA_CMP_VERSION;
+	pmsg.cp_flags = 0;
+	pmsg.cp_send_size = pmsg.cp_recv_size =
+		rpcrdma_encode_buffer_size(newxprt->sc_max_req_size);
+
 	/* Accept Connection */
 	set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
 	memset(&conn_param, 0, sizeof conn_param);
 	conn_param.responder_resources = 0;
 	conn_param.initiator_depth = newxprt->sc_ord;
+	conn_param.private_data = &pmsg;
+	conn_param.private_data_len = sizeof(pmsg);
 	ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
 	if (ret) {
 		dprintk("svcrdma: failed to accept new connection, ret=%d\n",

[PATCH v1 3/3] svcrdma: support Remote Invalidation for prototyping

[Chuck Lever]

To allow testing, add a sysctl that enables the use of Send With
Invalidate in place of Send when transmitting RPC replies. The
invalidate_rkey is arbitrarily chosen from among rkeys present
in the RPC-over-RDMA header's chunk lists.

Send With Invalidate can be enabled when all client and server HCAs
support it, and the client does not send persistently registered
rkeys (like a local DMA rkey).

Send With Invalidate improves performance only when clients can
recognize, while processing an RPC reply, that an rkey has already
been invalidated. That is a separate change.

In the future, the RPC-over-RDMA protocol might support Remote
Invalidation properly. The protocol needs to enable signaling
between peers to indicate when Remote Invalidation can be used.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 include/linux/sunrpc/svc_rdma.h          |    1 +
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |   58 ++++++++++++++++++++++++++++--
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   12 +++++-
 3 files changed, 65 insertions(+), 6 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index d6917b8..8a43650 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -136,6 +136,7 @@ struct svcxprt_rdma {
 	int		     sc_ord;		/* RDMA read limit */
 	int                  sc_max_sge;
 	int                  sc_max_sge_rd;	/* max sge for read target */
+	bool		     sc_snd_w_inv;	/* OK to use Send With Invalidate */
 
 	atomic_t             sc_sq_count;	/* Number of SQ WR on queue */
 	unsigned int	     sc_sq_depth;	/* Depth of SQ */
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 54d53330..e9adbe5 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -225,6 +225,48 @@ svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp,
 	return rp_ary;
 }
 
+/* RPC-over-RDMA Version One private extension: Remote Invalidation.
+ * Responder's choice: requester signals it can handle Send With
+ * Invalidate, and responder chooses one rkey to invalidate.
+ *
+ * Find a candidate rkey to invalidate when sending a reply.  Picks the
+ * first rkey it finds in the chunks lists.
+ *
+ * Returns zero if RPC's chunk lists are empty.
+ */
+static u32 svc_rdma_get_inv_rkey(struct rpcrdma_msg *rdma_argp,
+				 struct rpcrdma_write_array *wr_ary,
+				 struct rpcrdma_write_array *rp_ary)
+{
+	struct rpcrdma_read_chunk *rd_ary;
+	struct rpcrdma_segment *arg_ch;
+	u32 inv_rkey;
+
+	inv_rkey = 0;
+
+	rd_ary = svc_rdma_get_read_chunk(rdma_argp);
+	if (rd_ary) {
+		inv_rkey = be32_to_cpu(rd_ary->rc_target.rs_handle);
+		goto out;
+	}
+
+	if (wr_ary && be32_to_cpu(wr_ary->wc_nchunks)) {
+		arg_ch = &wr_ary->wc_array[0].wc_target;
+		inv_rkey = be32_to_cpu(arg_ch->rs_handle);
+		goto out;
+	}
+
+	if (rp_ary && be32_to_cpu(rp_ary->wc_nchunks)) {
+		arg_ch = &rp_ary->wc_array[0].wc_target;
+		inv_rkey = be32_to_cpu(arg_ch->rs_handle);
+		goto out;
+	}
+
+out:
+	dprintk("svcrdma: Send With Invalidate rkey=%08x\n", inv_rkey);
+	return inv_rkey;
+}
+
 /* Assumptions:
  * - The specified write_len can be represented in sc_max_sge * PAGE_SIZE
  */
@@ -464,7 +506,8 @@ static int send_reply(struct svcxprt_rdma *rdma,
 		      struct page *page,
 		      struct rpcrdma_msg *rdma_resp,
 		      struct svc_rdma_req_map *vec,
-		      int byte_count)
+		      int byte_count,
+		      u32 inv_rkey)
 {
 	struct svc_rdma_op_ctxt *ctxt;
 	struct ib_send_wr send_wr;
@@ -549,7 +592,11 @@ static int send_reply(struct svcxprt_rdma *rdma,
 	send_wr.wr_cqe = &ctxt->cqe;
 	send_wr.sg_list = ctxt->sge;
 	send_wr.num_sge = sge_no;
-	send_wr.opcode = IB_WR_SEND;
+	if (inv_rkey) {
+		send_wr.opcode = IB_WR_SEND_WITH_INV;
+		send_wr.ex.invalidate_rkey = inv_rkey;
+	} else
+		send_wr.opcode = IB_WR_SEND;
 	send_wr.send_flags =  IB_SEND_SIGNALED;
 
 	ret = svc_rdma_send(rdma, &send_wr);
@@ -581,6 +628,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	int inline_bytes;
 	struct page *res_page;
 	struct svc_rdma_req_map *vec;
+	u32 inv_rkey;
 
 	dprintk("svcrdma: sending response for rqstp=%p\n", rqstp);
 
@@ -591,6 +639,10 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	wr_ary = svc_rdma_get_write_array(rdma_argp);
 	rp_ary = svc_rdma_get_reply_array(rdma_argp, wr_ary);
 
+	inv_rkey = 0;
+	if (rdma->sc_snd_w_inv)
+		inv_rkey = svc_rdma_get_inv_rkey(rdma_argp, wr_ary, rp_ary);
+
 	/* Build an req vec for the XDR */
 	vec = svc_rdma_get_req_map(rdma);
 	ret = svc_rdma_map_xdr(rdma, &rqstp->rq_res, vec, wr_ary != NULL);
@@ -633,7 +685,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 		goto err1;
 
 	ret = send_reply(rdma, rqstp, res_page, rdma_resp, vec,
-			 inline_bytes);
+			 inline_bytes, inv_rkey);
 	if (ret < 0)
 		goto err1;
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 4843824..1fe34f6 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -651,9 +651,14 @@ svc_rdma_parse_connect_private(struct svcxprt_rdma *newxprt,
 	if (pmsg &&
 	    pmsg->cp_magic == rpcrdma_cmp_magic &&
 	    pmsg->cp_version == RPCRDMA_CMP_VERSION) {
-		dprintk("svcrdma: client send_size %u, recv_size %u\n",
+		newxprt->sc_snd_w_inv = pmsg->cp_flags &
+					RPCRDMA_CMP_F_SND_W_INV_OK;
+
+		dprintk("svcrdma: client send_size %u, recv_size %u "
+			"remote inv %ssupported\n",
 			rpcrdma_decode_buffer_size(pmsg->cp_send_size),
-			rpcrdma_decode_buffer_size(pmsg->cp_recv_size));
+			rpcrdma_decode_buffer_size(pmsg->cp_recv_size),
+			newxprt->sc_snd_w_inv ? "" : "un");
 	}
 }
 
@@ -1087,7 +1092,8 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 			dev->attrs.max_fast_reg_page_list_len;
 		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_FAST_REG;
 		newxprt->sc_reader = rdma_read_chunk_frmr;
-	}
+	} else
+		newxprt->sc_snd_w_inv = false;
 
 	/*
 	 * Determine if a DMA MR is required and if so, what privs are required

Re: [PATCH v1 0/3] server-side NFS/RDMA patches proposed for v4.9

[Sagi Grimberg]

The series looks good to me Chuck,

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

Re: [PATCH v1 0/3] server-side NFS/RDMA patches proposed for v4.9

[Chuck Lever]

> On Aug 16, 2016, at 6:29 AM, Sagi Grimberg <sagi@grimberg.me> wrote:
> 
> The series looks good to me Chuck,
> 
> Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

Thanks, Sagi.


--
Chuck Lever