[PATCH v3 0/7] NFS/RDMA server patches for 4.2

[PATCH v3 0/7] NFS/RDMA server patches for 4.2

[Chuck Lever]

Hi Bruce-

This is a simple update of last week's v2, just to keep up to date
with review comments. You can find these in my git repo in the
"nfsd-rdma-for-4.2" topic branch. See:

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


Changes since v2:
- Rebased on v4.1-rc6
- Dropped 2/10, 8/10, and 9/10, as discussed


Changes since v1:
- Rebased on v4.1-rc5
- Add patch to merge xprtrdma.ko and svcrdma.ko into one module

---

Chuck Lever (7):
      rpcrdma: Merge svcrdma and xprtrdma modules into one
      svcrdma: Add a separate "max data segs macro for svcrdma
      svcrdma: Replace GFP_KERNEL in a loop with GFP_NOFAIL
      svcrdma: Keep rpcrdma_msg fields in network byte-order
      svcrdma: Remove svc_rdma_xdr_decode_deferred_req()
      SUNRPC: Move EXPORT_SYMBOL for svc_process
      svcrdma: Fix byte-swapping in svc_rdma_sendto.c


 include/linux/sunrpc/svc_rdma.h          |   11 ++
 net/sunrpc/Kconfig                       |   28 ++----
 net/sunrpc/Makefile                      |    3 -
 net/sunrpc/svc.c                         |    2 
 net/sunrpc/xprtrdma/Makefile             |   14 +--
 net/sunrpc/xprtrdma/module.c             |   46 ++++++++++
 net/sunrpc/xprtrdma/svc_rdma.c           |    8 --
 net/sunrpc/xprtrdma/svc_rdma_marshal.c   |  140 ++++++++----------------------
 net/sunrpc/xprtrdma/svc_rdma_recvfrom.c  |    2 
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |   16 ++-
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   36 ++------
 net/sunrpc/xprtrdma/transport.c          |   13 ---
 net/sunrpc/xprtrdma/xprt_rdma.h          |   11 +-
 13 files changed, 135 insertions(+), 195 deletions(-)
 create mode 100644 net/sunrpc/xprtrdma/module.c

-- 
Chuck Lever

[PATCH v3 1/7] svcrdma: Fix byte-swapping in svc_rdma_sendto.c

[Chuck Lever]

In send_write_chunks(), we have:

	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
	     xfer_len && chunk_no < arg_ary->wc_nchunks;
	     chunk_no++) {
		 . . .
	}

Note that arg_ary->wc_nchunk is in network byte-order. For the
comparison to work correctly, both have to be in native byte-order.

In send_reply_chunks, we have:

	write_len = min(xfer_len, htonl(ch->rs_length));

xfer_len is in native byte-order, and ch->rs_length is in
network byte-order. be32_to_cpu() is the correct byte swap
for ch->rs_length.

As an additional clean up, replace ntohl() with be32_to_cpu() in
a few other places.

This appears to address a problem with large rsize hangs while
using PHYSICAL memory registration. I suspect that is the only
registration mode that uses more than one chunk element.

BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=248
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 net/sunrpc/xprtrdma/svc_rdma_sendto.c |   14 ++++++++------
 1 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 7de33d1..109e967 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -240,6 +240,7 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
 	u32 xdr_off;
 	int chunk_off;
 	int chunk_no;
+	int nchunks;
 	struct rpcrdma_write_array *arg_ary;
 	struct rpcrdma_write_array *res_ary;
 	int ret;
@@ -251,14 +252,15 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
 		&rdma_resp->rm_body.rm_chunks[1];
 
 	/* Write chunks start at the pagelist */
+	nchunks = be32_to_cpu(arg_ary->wc_nchunks);
 	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
-	     xfer_len && chunk_no < arg_ary->wc_nchunks;
+	     xfer_len && chunk_no < nchunks;
 	     chunk_no++) {
 		struct rpcrdma_segment *arg_ch;
 		u64 rs_offset;
 
 		arg_ch = &arg_ary->wc_array[chunk_no].wc_target;
-		write_len = min(xfer_len, ntohl(arg_ch->rs_length));
+		write_len = min(xfer_len, be32_to_cpu(arg_ch->rs_length));
 
 		/* Prepare the response chunk given the length actually
 		 * written */
@@ -270,7 +272,7 @@ static int send_write_chunks(struct svcxprt_rdma *xprt,
 		chunk_off = 0;
 		while (write_len) {
 			ret = send_write(xprt, rqstp,
-					 ntohl(arg_ch->rs_handle),
+					 be32_to_cpu(arg_ch->rs_handle),
 					 rs_offset + chunk_off,
 					 xdr_off,
 					 write_len,
@@ -318,13 +320,13 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
 		&rdma_resp->rm_body.rm_chunks[2];
 
 	/* xdr offset starts at RPC message */
-	nchunks = ntohl(arg_ary->wc_nchunks);
+	nchunks = be32_to_cpu(arg_ary->wc_nchunks);
 	for (xdr_off = 0, chunk_no = 0;
 	     xfer_len && chunk_no < nchunks;
 	     chunk_no++) {
 		u64 rs_offset;
 		ch = &arg_ary->wc_array[chunk_no].wc_target;
-		write_len = min(xfer_len, htonl(ch->rs_length));
+		write_len = min(xfer_len, be32_to_cpu(ch->rs_length));
 
 		/* Prepare the reply chunk given the length actually
 		 * written */
@@ -335,7 +337,7 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt,
 		chunk_off = 0;
 		while (write_len) {
 			ret = send_write(xprt, rqstp,
-					 ntohl(ch->rs_handle),
+					 be32_to_cpu(ch->rs_handle),
 					 rs_offset + chunk_off,
 					 xdr_off,
 					 write_len,

[PATCH v3 2/7] SUNRPC: Move EXPORT_SYMBOL for svc_process

[Chuck Lever]

Clean up.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 net/sunrpc/svc.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
index 78974e4..852ae60 100644
--- a/net/sunrpc/svc.c
+++ b/net/sunrpc/svc.c
@@ -1290,7 +1290,6 @@ err_bad:
 	svc_putnl(resv, ntohl(rpc_stat));
 	goto sendit;
 }
-EXPORT_SYMBOL_GPL(svc_process);
 
 /*
  * Process the RPC request.
@@ -1338,6 +1337,7 @@ out_drop:
 	svc_drop(rqstp);
 	return 0;
 }
+EXPORT_SYMBOL_GPL(svc_process);
 
 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 /*

[PATCH v3 3/7] svcrdma: Remove svc_rdma_xdr_decode_deferred_req()

[Chuck Lever]

svc_rdma_xdr_decode_deferred_req() indexes an array with an
un-byte-swapped value off the wire. Fortunately this function
isn't used anywhere, so simply remove it.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 include/linux/sunrpc/svc_rdma.h        |    1 -
 net/sunrpc/xprtrdma/svc_rdma_marshal.c |   56 --------------------------------
 2 files changed, 0 insertions(+), 57 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index df8edf8..8ad9b6d 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -182,7 +182,6 @@ struct svcxprt_rdma {
 
 /* svc_rdma_marshal.c */
 extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *);
-extern int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *);
 extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
 				     struct rpcrdma_msg *,
 				     enum rpcrdma_errcode, u32 *);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_marshal.c b/net/sunrpc/xprtrdma/svc_rdma_marshal.c
index b681855..45e22c9 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_marshal.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_marshal.c
@@ -211,62 +211,6 @@ int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
 	return hdr_len;
 }
 
-int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *rqstp)
-{
-	struct rpcrdma_msg *rmsgp = NULL;
-	struct rpcrdma_read_chunk *ch;
-	struct rpcrdma_write_array *ary;
-	u32 *va;
-	u32 hdrlen;
-
-	dprintk("svcrdma: processing deferred RDMA header on rqstp=%p\n",
-		rqstp);
-	rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
-
-	/* Pull in the extra for the padded case and bump our pointer */
-	if (rmsgp->rm_type == RDMA_MSGP) {
-		va = &rmsgp->rm_body.rm_padded.rm_pempty[4];
-		rqstp->rq_arg.head[0].iov_base = va;
-		hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp);
-		rqstp->rq_arg.head[0].iov_len -= hdrlen;
-		return hdrlen;
-	}
-
-	/*
-	 * Skip all chunks to find RPC msg. These were previously processed
-	 */
-	va = &rmsgp->rm_body.rm_chunks[0];
-
-	/* Skip read-list */
-	for (ch = (struct rpcrdma_read_chunk *)va;
-	     ch->rc_discrim != xdr_zero; ch++);
-	va = (u32 *)&ch->rc_position;
-
-	/* Skip write-list */
-	ary = (struct rpcrdma_write_array *)va;
-	if (ary->wc_discrim == xdr_zero)
-		va = (u32 *)&ary->wc_nchunks;
-	else
-		/*
-		 * rs_length is the 2nd 4B field in wc_target and taking its
-		 * address skips the list terminator
-		 */
-		va = (u32 *)&ary->wc_array[ary->wc_nchunks].wc_target.rs_length;
-
-	/* Skip reply-array */
-	ary = (struct rpcrdma_write_array *)va;
-	if (ary->wc_discrim == xdr_zero)
-		va = (u32 *)&ary->wc_nchunks;
-	else
-		va = (u32 *)&ary->wc_array[ary->wc_nchunks];
-
-	rqstp->rq_arg.head[0].iov_base = va;
-	hdrlen = (unsigned long)va - (unsigned long)rmsgp;
-	rqstp->rq_arg.head[0].iov_len -= hdrlen;
-
-	return hdrlen;
-}
-
 int svc_rdma_xdr_encode_error(struct svcxprt_rdma *xprt,
 			      struct rpcrdma_msg *rmsgp,
 			      enum rpcrdma_errcode err, u32 *va)

[PATCH v3 4/7] svcrdma: Keep rpcrdma_msg fields in network byte-order

[Chuck Lever]

Fields in struct rpcrdma_msg are __be32. Don't byte-swap these
fields when decoding RPC calls and then swap them back for the
reply. For the most part, they can be left alone.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 include/linux/sunrpc/svc_rdma.h          |    2 -
 net/sunrpc/xprtrdma/svc_rdma_marshal.c   |   84 ++++++++++++++----------------
 net/sunrpc/xprtrdma/svc_rdma_recvfrom.c  |    2 -
 net/sunrpc/xprtrdma/svc_rdma_transport.c |    2 -
 4 files changed, 42 insertions(+), 48 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index 8ad9b6d..c03ca0a 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -184,7 +184,7 @@ struct svcxprt_rdma {
 extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *);
 extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
 				     struct rpcrdma_msg *,
-				     enum rpcrdma_errcode, u32 *);
+				     enum rpcrdma_errcode, __be32 *);
 extern void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *, int);
 extern void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *, int);
 extern void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *, int,
diff --git a/net/sunrpc/xprtrdma/svc_rdma_marshal.c b/net/sunrpc/xprtrdma/svc_rdma_marshal.c
index 45e22c9..e2fca76 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_marshal.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_marshal.c
@@ -50,12 +50,12 @@
 /*
  * Decodes a read chunk list. The expected format is as follows:
  *    descrim  : xdr_one
- *    position : u32 offset into XDR stream
- *    handle   : u32 RKEY
+ *    position : __be32 offset into XDR stream
+ *    handle   : __be32 RKEY
  *    . . .
  *  end-of-list: xdr_zero
  */
-static u32 *decode_read_list(u32 *va, u32 *vaend)
+static __be32 *decode_read_list(__be32 *va, __be32 *vaend)
 {
 	struct rpcrdma_read_chunk *ch = (struct rpcrdma_read_chunk *)va;
 
@@ -67,20 +67,20 @@ static u32 *decode_read_list(u32 *va, u32 *vaend)
 		}
 		ch++;
 	}
-	return (u32 *)&ch->rc_position;
+	return &ch->rc_position;
 }
 
 /*
  * Decodes a write chunk list. The expected format is as follows:
  *    descrim  : xdr_one
  *    nchunks  : <count>
- *       handle   : u32 RKEY              ---+
- *       length   : u32 <len of segment>     |
+ *       handle   : __be32 RKEY           ---+
+ *       length   : __be32 <len of segment>  |
  *       offset   : remove va                + <count>
  *       . . .                               |
  *                                        ---+
  */
-static u32 *decode_write_list(u32 *va, u32 *vaend)
+static __be32 *decode_write_list(__be32 *va, __be32 *vaend)
 {
 	unsigned long start, end;
 	int nchunks;
@@ -90,14 +90,14 @@ static u32 *decode_write_list(u32 *va, u32 *vaend)
 
 	/* Check for not write-array */
 	if (ary->wc_discrim == xdr_zero)
-		return (u32 *)&ary->wc_nchunks;
+		return &ary->wc_nchunks;
 
 	if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) >
 	    (unsigned long)vaend) {
 		dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend);
 		return NULL;
 	}
-	nchunks = ntohl(ary->wc_nchunks);
+	nchunks = be32_to_cpu(ary->wc_nchunks);
 
 	start = (unsigned long)&ary->wc_array[0];
 	end = (unsigned long)vaend;
@@ -112,10 +112,10 @@ static u32 *decode_write_list(u32 *va, u32 *vaend)
 	 * rs_length is the 2nd 4B field in wc_target and taking its
 	 * address skips the list terminator
 	 */
-	return (u32 *)&ary->wc_array[nchunks].wc_target.rs_length;
+	return &ary->wc_array[nchunks].wc_target.rs_length;
 }
 
-static u32 *decode_reply_array(u32 *va, u32 *vaend)
+static __be32 *decode_reply_array(__be32 *va, __be32 *vaend)
 {
 	unsigned long start, end;
 	int nchunks;
@@ -124,14 +124,14 @@ static u32 *decode_reply_array(u32 *va, u32 *vaend)
 
 	/* Check for no reply-array */
 	if (ary->wc_discrim == xdr_zero)
-		return (u32 *)&ary->wc_nchunks;
+		return &ary->wc_nchunks;
 
 	if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) >
 	    (unsigned long)vaend) {
 		dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend);
 		return NULL;
 	}
-	nchunks = ntohl(ary->wc_nchunks);
+	nchunks = be32_to_cpu(ary->wc_nchunks);
 
 	start = (unsigned long)&ary->wc_array[0];
 	end = (unsigned long)vaend;
@@ -142,15 +142,14 @@ static u32 *decode_reply_array(u32 *va, u32 *vaend)
 			ary, nchunks, vaend);
 		return NULL;
 	}
-	return (u32 *)&ary->wc_array[nchunks];
+	return (__be32 *)&ary->wc_array[nchunks];
 }
 
 int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
 			    struct svc_rqst *rqstp)
 {
 	struct rpcrdma_msg *rmsgp = NULL;
-	u32 *va;
-	u32 *vaend;
+	__be32 *va, *vaend;
 	u32 hdr_len;
 
 	rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
@@ -162,22 +161,17 @@ int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
 		return -EINVAL;
 	}
 
-	/* Decode the header */
-	rmsgp->rm_xid = ntohl(rmsgp->rm_xid);
-	rmsgp->rm_vers = ntohl(rmsgp->rm_vers);
-	rmsgp->rm_credit = ntohl(rmsgp->rm_credit);
-	rmsgp->rm_type = ntohl(rmsgp->rm_type);
-
-	if (rmsgp->rm_vers != RPCRDMA_VERSION)
+	if (rmsgp->rm_vers != rpcrdma_version)
 		return -ENOSYS;
 
 	/* Pull in the extra for the padded case and bump our pointer */
-	if (rmsgp->rm_type == RDMA_MSGP) {
+	if (rmsgp->rm_type == rdma_msgp) {
 		int hdrlen;
+
 		rmsgp->rm_body.rm_padded.rm_align =
-			ntohl(rmsgp->rm_body.rm_padded.rm_align);
+			be32_to_cpu(rmsgp->rm_body.rm_padded.rm_align);
 		rmsgp->rm_body.rm_padded.rm_thresh =
-			ntohl(rmsgp->rm_body.rm_padded.rm_thresh);
+			be32_to_cpu(rmsgp->rm_body.rm_padded.rm_thresh);
 
 		va = &rmsgp->rm_body.rm_padded.rm_pempty[4];
 		rqstp->rq_arg.head[0].iov_base = va;
@@ -192,7 +186,7 @@ int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
 	 * chunk list and a reply chunk list.
 	 */
 	va = &rmsgp->rm_body.rm_chunks[0];
-	vaend = (u32 *)((unsigned long)rmsgp + rqstp->rq_arg.len);
+	vaend = (__be32 *)((unsigned long)rmsgp + rqstp->rq_arg.len);
 	va = decode_read_list(va, vaend);
 	if (!va)
 		return -EINVAL;
@@ -213,18 +207,18 @@ int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
 
 int svc_rdma_xdr_encode_error(struct svcxprt_rdma *xprt,
 			      struct rpcrdma_msg *rmsgp,
-			      enum rpcrdma_errcode err, u32 *va)
+			      enum rpcrdma_errcode err, __be32 *va)
 {
-	u32 *startp = va;
+	__be32 *startp = va;
 
-	*va++ = htonl(rmsgp->rm_xid);
-	*va++ = htonl(rmsgp->rm_vers);
-	*va++ = htonl(xprt->sc_max_requests);
-	*va++ = htonl(RDMA_ERROR);
-	*va++ = htonl(err);
+	*va++ = rmsgp->rm_xid;
+	*va++ = rmsgp->rm_vers;
+	*va++ = cpu_to_be32(xprt->sc_max_requests);
+	*va++ = rdma_error;
+	*va++ = cpu_to_be32(err);
 	if (err == ERR_VERS) {
-		*va++ = htonl(RPCRDMA_VERSION);
-		*va++ = htonl(RPCRDMA_VERSION);
+		*va++ = rpcrdma_version;
+		*va++ = rpcrdma_version;
 	}
 
 	return (int)((unsigned long)va - (unsigned long)startp);
@@ -241,7 +235,7 @@ int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *rmsgp)
 		&rmsgp->rm_body.rm_chunks[1];
 	if (wr_ary->wc_discrim)
 		wr_ary = (struct rpcrdma_write_array *)
-			&wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)].
+			&wr_ary->wc_array[be32_to_cpu(wr_ary->wc_nchunks)].
 			wc_target.rs_length;
 	else
 		wr_ary = (struct rpcrdma_write_array *)
@@ -250,7 +244,7 @@ int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *rmsgp)
 	/* skip reply array */
 	if (wr_ary->wc_discrim)
 		wr_ary = (struct rpcrdma_write_array *)
-			&wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)];
+			&wr_ary->wc_array[be32_to_cpu(wr_ary->wc_nchunks)];
 	else
 		wr_ary = (struct rpcrdma_write_array *)
 			&wr_ary->wc_nchunks;
@@ -269,7 +263,7 @@ void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *rmsgp, int chunks)
 	ary = (struct rpcrdma_write_array *)
 		&rmsgp->rm_body.rm_chunks[1];
 	ary->wc_discrim = xdr_one;
-	ary->wc_nchunks = htonl(chunks);
+	ary->wc_nchunks = cpu_to_be32(chunks);
 
 	/* write-list terminator */
 	ary->wc_array[chunks].wc_target.rs_handle = xdr_zero;
@@ -282,7 +276,7 @@ void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *ary,
 				 int chunks)
 {
 	ary->wc_discrim = xdr_one;
-	ary->wc_nchunks = htonl(chunks);
+	ary->wc_nchunks = cpu_to_be32(chunks);
 }
 
 void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *ary,
@@ -294,7 +288,7 @@ void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *ary,
 	struct rpcrdma_segment *seg = &ary->wc_array[chunk_no].wc_target;
 	seg->rs_handle = rs_handle;
 	seg->rs_offset = rs_offset;
-	seg->rs_length = htonl(write_len);
+	seg->rs_length = cpu_to_be32(write_len);
 }
 
 void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *xprt,
@@ -302,10 +296,10 @@ void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *xprt,
 				  struct rpcrdma_msg *rdma_resp,
 				  enum rpcrdma_proc rdma_type)
 {
-	rdma_resp->rm_xid = htonl(rdma_argp->rm_xid);
-	rdma_resp->rm_vers = htonl(rdma_argp->rm_vers);
-	rdma_resp->rm_credit = htonl(xprt->sc_max_requests);
-	rdma_resp->rm_type = htonl(rdma_type);
+	rdma_resp->rm_xid = rdma_argp->rm_xid;
+	rdma_resp->rm_vers = rdma_argp->rm_vers;
+	rdma_resp->rm_credit = cpu_to_be32(xprt->sc_max_requests);
+	rdma_resp->rm_type = cpu_to_be32(rdma_type);
 
 	/* Encode <nul> chunks lists */
 	rdma_resp->rm_body.rm_chunks[0] = xdr_zero;
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index f9f13a3..ac93ce0 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -85,7 +85,7 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
 
 	/* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
 	rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
-	if (be32_to_cpu(rmsgp->rm_type) == RDMA_NOMSG)
+	if (rmsgp->rm_type == rdma_nomsg)
 		rqstp->rq_arg.pages = &rqstp->rq_pages[0];
 	else
 		rqstp->rq_arg.pages = &rqstp->rq_pages[1];
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index f609c1c..be08493 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -1319,7 +1319,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
 	struct ib_send_wr err_wr;
 	struct page *p;
 	struct svc_rdma_op_ctxt *ctxt;
-	u32 *va;
+	__be32 *va;
 	int length;
 	int ret;
 

[PATCH v3 5/7] svcrdma: Replace GFP_KERNEL in a loop with GFP_NOFAIL

[Chuck Lever]

At the 2015 LSF/MM, it was requested that memory allocation
call sites that request GFP_KERNEL allocations in a loop should be
annotated with __GFP_NOFAIL.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 include/linux/sunrpc/svc_rdma.h          |    1 -
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |    2 +-
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   32 ++++++------------------------
 3 files changed, 7 insertions(+), 28 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index c03ca0a..d26384b 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -211,7 +211,6 @@ extern int svc_rdma_sendto(struct svc_rqst *);
 extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *);
 extern void svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *,
 				enum rpcrdma_errcode);
-struct page *svc_rdma_get_page(void);
 extern int svc_rdma_post_recv(struct svcxprt_rdma *);
 extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
 extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 109e967..d25cd43 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -517,7 +517,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	inline_bytes = rqstp->rq_res.len;
 
 	/* Create the RDMA response header */
-	res_page = svc_rdma_get_page();
+	res_page = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
 	rdma_resp = page_address(res_page);
 	reply_ary = svc_rdma_get_reply_array(rdma_argp);
 	if (reply_ary)
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index be08493..1ed4740 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -99,12 +99,8 @@ struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
 {
 	struct svc_rdma_op_ctxt *ctxt;
 
-	while (1) {
-		ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep, GFP_KERNEL);
-		if (ctxt)
-			break;
-		schedule_timeout_uninterruptible(msecs_to_jiffies(500));
-	}
+	ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep,
+				GFP_KERNEL | __GFP_NOFAIL);
 	ctxt->xprt = xprt;
 	INIT_LIST_HEAD(&ctxt->dto_q);
 	ctxt->count = 0;
@@ -156,12 +152,8 @@ void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
 struct svc_rdma_req_map *svc_rdma_get_req_map(void)
 {
 	struct svc_rdma_req_map *map;
-	while (1) {
-		map = kmem_cache_alloc(svc_rdma_map_cachep, GFP_KERNEL);
-		if (map)
-			break;
-		schedule_timeout_uninterruptible(msecs_to_jiffies(500));
-	}
+	map = kmem_cache_alloc(svc_rdma_map_cachep,
+			       GFP_KERNEL | __GFP_NOFAIL);
 	map->count = 0;
 	return map;
 }
@@ -490,18 +482,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
 	return cma_xprt;
 }
 
-struct page *svc_rdma_get_page(void)
-{
-	struct page *page;
-
-	while ((page = alloc_page(GFP_KERNEL)) == NULL) {
-		/* If we can't get memory, wait a bit and try again */
-		printk(KERN_INFO "svcrdma: out of memory...retrying in 1s\n");
-		schedule_timeout_uninterruptible(msecs_to_jiffies(1000));
-	}
-	return page;
-}
-
 int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
 {
 	struct ib_recv_wr recv_wr, *bad_recv_wr;
@@ -520,7 +500,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
 			pr_err("svcrdma: Too many sges (%d)\n", sge_no);
 			goto err_put_ctxt;
 		}
-		page = svc_rdma_get_page();
+		page = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
 		ctxt->pages[sge_no] = page;
 		pa = ib_dma_map_page(xprt->sc_cm_id->device,
 				     page, 0, PAGE_SIZE,
@@ -1323,7 +1303,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
 	int length;
 	int ret;
 
-	p = svc_rdma_get_page();
+	p = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
 	va = page_address(p);
 
 	/* XDR encode error */

[PATCH v3 6/7] svcrdma: Add a separate "max data segs macro for svcrdma

[Chuck Lever]

The server and client maximum are architecturally independent.
Allow changing one without affecting the other.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 include/linux/sunrpc/svc_rdma.h          |    7 +++++++
 net/sunrpc/xprtrdma/svc_rdma_transport.c |    2 +-
 net/sunrpc/xprtrdma/xprt_rdma.h          |    6 ------
 3 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index d26384b..cb94ee4 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -172,6 +172,13 @@ struct svcxprt_rdma {
 #define RDMAXPRT_SQ_PENDING	2
 #define RDMAXPRT_CONN_PENDING	3
 
+#define RPCRDMA_MAX_SVC_SEGS	(64)	/* server max scatter/gather */
+#if RPCSVC_MAXPAYLOAD < (RPCRDMA_MAX_SVC_SEGS << PAGE_SHIFT)
+#define RPCRDMA_MAXPAYLOAD	RPCSVC_MAXPAYLOAD
+#else
+#define RPCRDMA_MAXPAYLOAD	(RPCRDMA_MAX_SVC_SEGS << PAGE_SHIFT)
+#endif
+
 #define RPCRDMA_LISTEN_BACKLOG  10
 /* The default ORD value is based on two outstanding full-size writes with a
  * page size of 4k, or 32k * 2 ops / 4k = 16 outstanding RDMA_READ.  */
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 1ed4740..3b4c2ff 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -91,7 +91,7 @@ struct svc_xprt_class svc_rdma_class = {
 	.xcl_name = "rdma",
 	.xcl_owner = THIS_MODULE,
 	.xcl_ops = &svc_rdma_ops,
-	.xcl_max_payload = RPCSVC_MAXPAYLOAD_RDMA,
+	.xcl_max_payload = RPCRDMA_MAXPAYLOAD,
 	.xcl_ident = XPRT_TRANSPORT_RDMA,
 };
 
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index 78e0b8b..e60907b 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -487,10 +487,4 @@ extern struct kmem_cache *svc_rdma_ctxt_cachep;
 /* Workqueue created in svc_rdma.c */
 extern struct workqueue_struct *svc_rdma_wq;
 
-#if RPCSVC_MAXPAYLOAD < (RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT)
-#define RPCSVC_MAXPAYLOAD_RDMA RPCSVC_MAXPAYLOAD
-#else
-#define RPCSVC_MAXPAYLOAD_RDMA (RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT)
-#endif
-
 #endif				/* _LINUX_SUNRPC_XPRT_RDMA_H */

[PATCH v3 7/7] rpcrdma: Merge svcrdma and xprtrdma modules into one

[Chuck Lever]

Bi-directional RPC support means code in svcrdma.ko invokes a bit of
code in xprtrdma.ko, and vice versa. To avoid loader/linker loops,
merge the server and client side modules together into a single
module.

When backchannel capabilities are added, the combined module will
register all needed transport capabilities so that Upper Layer
consumers automatically have everything needed to create a
bi-directional transport connection.

Module aliases are added for backwards compatibility with user
space, which still may expect svcrdma.ko or xprtrdma.ko to be
present.

This commit reverts commit 2e8c12e1b765 ("xprtrdma: add separate
Kconfig options for NFSoRDMA client and server support") and
provides a single CONFIG option for enabling the new module.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---

 net/sunrpc/Kconfig              |   28 +++++++-----------------
 net/sunrpc/Makefile             |    3 +--
 net/sunrpc/xprtrdma/Makefile    |   14 +++++-------
 net/sunrpc/xprtrdma/module.c    |   46 +++++++++++++++++++++++++++++++++++++++
 net/sunrpc/xprtrdma/svc_rdma.c  |    8 +------
 net/sunrpc/xprtrdma/transport.c |   13 ++---------
 net/sunrpc/xprtrdma/xprt_rdma.h |    5 ++++
 7 files changed, 69 insertions(+), 48 deletions(-)
 create mode 100644 net/sunrpc/xprtrdma/module.c

diff --git a/net/sunrpc/Kconfig b/net/sunrpc/Kconfig
index 9068e72..04ce2c0 100644
--- a/net/sunrpc/Kconfig
+++ b/net/sunrpc/Kconfig
@@ -48,28 +48,16 @@ config SUNRPC_DEBUG
 
 	  If unsure, say Y.
 
-config SUNRPC_XPRT_RDMA_CLIENT
-	tristate "RPC over RDMA Client Support"
+config SUNRPC_XPRT_RDMA
+	tristate "RPC-over-RDMA transport"
 	depends on SUNRPC && INFINIBAND && INFINIBAND_ADDR_TRANS
 	default SUNRPC && INFINIBAND
 	help
-	  This option allows the NFS client to support an RDMA-enabled
-	  transport.
+	  This option allows the NFS client and server to use RDMA
+	  transports (InfiniBand, iWARP, or RoCE).
 
-	  To compile RPC client RDMA transport support as a module,
-	  choose M here: the module will be called xprtrdma.
+	  To compile this support as a module, choose M. The module
+	  will be called rpcrdma.ko.
 
-	  If unsure, say N.
-
-config SUNRPC_XPRT_RDMA_SERVER
-	tristate "RPC over RDMA Server Support"
-	depends on SUNRPC && INFINIBAND && INFINIBAND_ADDR_TRANS
-	default SUNRPC && INFINIBAND
-	help
-	  This option allows the NFS server to support an RDMA-enabled
-	  transport.
-
-	  To compile RPC server RDMA transport support as a module,
-	  choose M here: the module will be called svcrdma.
-
-	  If unsure, say N.
+	  If unsure, or you know there is no RDMA capability on your
+	  hardware platform, say N.
diff --git a/net/sunrpc/Makefile b/net/sunrpc/Makefile
index 15e6f6c..936ad0a 100644
--- a/net/sunrpc/Makefile
+++ b/net/sunrpc/Makefile
@@ -5,8 +5,7 @@
 
 obj-$(CONFIG_SUNRPC) += sunrpc.o
 obj-$(CONFIG_SUNRPC_GSS) += auth_gss/
-
-obj-y += xprtrdma/
+obj-$(CONFIG_SUNRPC_XPRT_RDMA) += xprtrdma/
 
 sunrpc-y := clnt.o xprt.o socklib.o xprtsock.o sched.o \
 	    auth.o auth_null.o auth_unix.o auth_generic.o \
diff --git a/net/sunrpc/xprtrdma/Makefile b/net/sunrpc/xprtrdma/Makefile
index 579f72b..48913de 100644
--- a/net/sunrpc/xprtrdma/Makefile
+++ b/net/sunrpc/xprtrdma/Makefile
@@ -1,9 +1,7 @@
-obj-$(CONFIG_SUNRPC_XPRT_RDMA_CLIENT) += xprtrdma.o
+obj-$(CONFIG_SUNRPC_XPRT_RDMA) += rpcrdma.o
 
-xprtrdma-y := transport.o rpc_rdma.o verbs.o \
-	fmr_ops.o frwr_ops.o physical_ops.o
-
-obj-$(CONFIG_SUNRPC_XPRT_RDMA_SERVER) += svcrdma.o
-
-svcrdma-y := svc_rdma.o svc_rdma_transport.o \
-	svc_rdma_marshal.o svc_rdma_sendto.o svc_rdma_recvfrom.o
+rpcrdma-y := transport.o rpc_rdma.o verbs.o \
+	fmr_ops.o frwr_ops.o physical_ops.o \
+	svc_rdma.o svc_rdma_transport.o \
+	svc_rdma_marshal.o svc_rdma_sendto.o svc_rdma_recvfrom.o \
+	module.o
diff --git a/net/sunrpc/xprtrdma/module.c b/net/sunrpc/xprtrdma/module.c
new file mode 100644
index 0000000..560712b
--- /dev/null
+++ b/net/sunrpc/xprtrdma/module.c
@@ -0,0 +1,46 @@
+/*
+ * Copyright (c) 2015 Oracle.  All rights reserved.
+ */
+
+/* rpcrdma.ko module initialization
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/sunrpc/svc_rdma.h>
+#include "xprt_rdma.h"
+
+#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
+# define RPCDBG_FACILITY	RPCDBG_TRANS
+#endif
+
+MODULE_AUTHOR("Open Grid Computing and Network Appliance, Inc.");
+MODULE_DESCRIPTION("RPC/RDMA Transport");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_ALIAS("svcrdma");
+MODULE_ALIAS("xprtrdma");
+
+static void __exit rpc_rdma_cleanup(void)
+{
+	xprt_rdma_cleanup();
+	svc_rdma_cleanup();
+}
+
+static int __init rpc_rdma_init(void)
+{
+	int rc;
+
+	rc = svc_rdma_init();
+	if (rc)
+		goto out;
+
+	rc = xprt_rdma_init();
+	if (rc)
+		svc_rdma_cleanup();
+
+out:
+	return rc;
+}
+
+module_init(rpc_rdma_init);
+module_exit(rpc_rdma_cleanup);
diff --git a/net/sunrpc/xprtrdma/svc_rdma.c b/net/sunrpc/xprtrdma/svc_rdma.c
index c1b6270..2cd252f 100644
--- a/net/sunrpc/xprtrdma/svc_rdma.c
+++ b/net/sunrpc/xprtrdma/svc_rdma.c
@@ -38,8 +38,7 @@
  *
  * Author: Tom Tucker <tom@opengridcomputing.com>
  */
-#include <linux/module.h>
-#include <linux/init.h>
+
 #include <linux/slab.h>
 #include <linux/fs.h>
 #include <linux/sysctl.h>
@@ -295,8 +294,3 @@ int svc_rdma_init(void)
 	destroy_workqueue(svc_rdma_wq);
 	return -ENOMEM;
 }
-MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
-MODULE_DESCRIPTION("SVC RDMA Transport");
-MODULE_LICENSE("Dual BSD/GPL");
-module_init(svc_rdma_init);
-module_exit(svc_rdma_cleanup);
diff --git a/net/sunrpc/xprtrdma/transport.c b/net/sunrpc/xprtrdma/transport.c
index 54f23b1..436da2c 100644
--- a/net/sunrpc/xprtrdma/transport.c
+++ b/net/sunrpc/xprtrdma/transport.c
@@ -48,7 +48,6 @@
  */
 
 #include <linux/module.h>
-#include <linux/init.h>
 #include <linux/slab.h>
 #include <linux/seq_file.h>
 #include <linux/sunrpc/addr.h>
@@ -59,11 +58,6 @@
 # define RPCDBG_FACILITY	RPCDBG_TRANS
 #endif
 
-MODULE_LICENSE("Dual BSD/GPL");
-
-MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS");
-MODULE_AUTHOR("Network Appliance, Inc.");
-
 /*
  * tunables
  */
@@ -711,7 +705,7 @@ static struct xprt_class xprt_rdma = {
 	.setup			= xprt_setup_rdma,
 };
 
-static void __exit xprt_rdma_cleanup(void)
+void xprt_rdma_cleanup(void)
 {
 	int rc;
 
@@ -728,7 +722,7 @@ static void __exit xprt_rdma_cleanup(void)
 			__func__, rc);
 }
 
-static int __init xprt_rdma_init(void)
+int xprt_rdma_init(void)
 {
 	int rc;
 
@@ -753,6 +747,3 @@ static int __init xprt_rdma_init(void)
 #endif
 	return 0;
 }
-
-module_init(xprt_rdma_init);
-module_exit(xprt_rdma_cleanup);
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index e60907b..58163b8 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -480,6 +480,11 @@ void rpcrdma_reply_handler(struct rpcrdma_rep *);
  */
 int rpcrdma_marshal_req(struct rpc_rqst *);
 
+/* RPC/RDMA module init - xprtrdma/transport.c
+ */
+int xprt_rdma_init(void);
+void xprt_rdma_cleanup(void);
+
 /* Temporary NFS request map cache. Created in svc_rdma.c  */
 extern struct kmem_cache *svc_rdma_map_cachep;
 /* WR context cache. Created in svc_rdma.c  */

Re: [PATCH v3 0/7] NFS/RDMA server patches for 4.2

[J. Bruce Fields]

On Thu, Jun 04, 2015 at 11:20:25AM -0400, Chuck Lever wrote:
> Hi Bruce-
> 
> This is a simple update of last week's v2, just to keep up to date
> with review comments. You can find these in my git repo in the
> "nfsd-rdma-for-4.2" topic branch. See:

Thanks, applying for 4.2 pending some testing.

--b.

> 
>  git://git.linux-nfs.org/projects/cel/cel-2.6.git
> 
> 
> Changes since v2:
> - Rebased on v4.1-rc6
> - Dropped 2/10, 8/10, and 9/10, as discussed
> 
> 
> Changes since v1:
> - Rebased on v4.1-rc5
> - Add patch to merge xprtrdma.ko and svcrdma.ko into one module
> 
> ---
> 
> Chuck Lever (7):
>       rpcrdma: Merge svcrdma and xprtrdma modules into one
>       svcrdma: Add a separate "max data segs macro for svcrdma
>       svcrdma: Replace GFP_KERNEL in a loop with GFP_NOFAIL
>       svcrdma: Keep rpcrdma_msg fields in network byte-order
>       svcrdma: Remove svc_rdma_xdr_decode_deferred_req()
>       SUNRPC: Move EXPORT_SYMBOL for svc_process
>       svcrdma: Fix byte-swapping in svc_rdma_sendto.c
> 
> 
>  include/linux/sunrpc/svc_rdma.h          |   11 ++
>  net/sunrpc/Kconfig                       |   28 ++----
>  net/sunrpc/Makefile                      |    3 -
>  net/sunrpc/svc.c                         |    2 
>  net/sunrpc/xprtrdma/Makefile             |   14 +--
>  net/sunrpc/xprtrdma/module.c             |   46 ++++++++++
>  net/sunrpc/xprtrdma/svc_rdma.c           |    8 --
>  net/sunrpc/xprtrdma/svc_rdma_marshal.c   |  140 ++++++++----------------------
>  net/sunrpc/xprtrdma/svc_rdma_recvfrom.c  |    2 
>  net/sunrpc/xprtrdma/svc_rdma_sendto.c    |   16 ++-
>  net/sunrpc/xprtrdma/svc_rdma_transport.c |   36 ++------
>  net/sunrpc/xprtrdma/transport.c          |   13 ---
>  net/sunrpc/xprtrdma/xprt_rdma.h          |   11 +-
>  13 files changed, 135 insertions(+), 195 deletions(-)
>  create mode 100644 net/sunrpc/xprtrdma/module.c
> 
> -- 
> Chuck Lever