[PATCH RFC 0/5] Reduce reply buffer requirements

[PATCH RFC 0/5] Reduce reply buffer requirements

[Chuck Lever]

Hi-

Last set I have for consideration for v5.1, please review and
comment.

My interest is in reducing the size of the RPC reply buffer, since
for RDMA, this is often registered memory. If I can get away with
one or two fewer chunks here and there, that helps improve transport
scalability.

These are a little incomplete and tasteless. I'm hoping someone can
help me give them a little more shape.

---

Chuck Lever (5):
      SUNRPC: Introduce rpc_prepare_reply_pages()
      NFS: Account for XDR pad of buf->pages
      SUNRPC: Make AUTH_SYS and AUTH_NULL set au_verfsize
      SUNRPC: Add rpc_auth::au_ralign field
      SUNRPC: Use au_rslack when computing reply buffer size


 fs/nfs/nfs2xdr.c               |   33 +++++---------------
 fs/nfs/nfs3xdr.c               |   39 +++++++-----------------
 fs/nfs/nfs4xdr.c               |   66 ++++++++++++++++++----------------------
 include/linux/sunrpc/auth.h    |   14 ++++----
 include/linux/sunrpc/clnt.h    |    3 ++
 include/trace/events/sunrpc.h  |   37 ++++++++++++++++++++++
 net/sunrpc/auth_gss/auth_gss.c |   19 ++++++++----
 net/sunrpc/auth_null.c         |    2 +
 net/sunrpc/auth_unix.c         |    6 +++-
 net/sunrpc/clnt.c              |   30 ++++++++++++++++--
 net/sunrpc/xdr.c               |   11 +++++++
 11 files changed, 155 insertions(+), 105 deletions(-)

--
Chuck Lever

[PATCH RFC 1/5] SUNRPC: Introduce rpc_prepare_reply_pages()

[Chuck Lever]

prepare_reply_buffer() and its NFSv4 equivalents expose the details
of the RPC header and the auth slack values to upper layer
consumers, creating a layering violation, and duplicating code.

Remedy these issues by adding a new RPC client API that hides those
details from upper layers in a common helper function.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/nfs/nfs2xdr.c              |   27 +++++-----------------
 fs/nfs/nfs3xdr.c              |   29 ++++++-----------------
 fs/nfs/nfs4xdr.c              |   51 +++++++++++++++++------------------------
 include/linux/sunrpc/clnt.h   |    3 ++
 include/trace/events/sunrpc.h |   37 ++++++++++++++++++++++++++++++
 net/sunrpc/clnt.c             |   19 +++++++++++++++
 net/sunrpc/xdr.c              |    9 +++++++
 7 files changed, 102 insertions(+), 73 deletions(-)

diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c
index bac3a4e..1dcd0fe 100644
--- a/fs/nfs/nfs2xdr.c
+++ b/fs/nfs/nfs2xdr.c
@@ -66,21 +66,6 @@
 static int nfs_stat_to_errno(enum nfs_stat);
 
 /*
- * While encoding arguments, set up the reply buffer in advance to
- * receive reply data directly into the page cache.
- */
-static void prepare_reply_buffer(struct rpc_rqst *req, struct page **pages,
-				 unsigned int base, unsigned int len,
-				 unsigned int bufsize)
-{
-	struct rpc_auth	*auth = req->rq_cred->cr_auth;
-	unsigned int replen;
-
-	replen = RPC_REPHDRSIZE + auth->au_rslack + bufsize;
-	xdr_inline_pages(&req->rq_rcv_buf, replen << 2, pages, base, len);
-}
-
-/*
  * Encode/decode NFSv2 basic data types
  *
  * Basic NFSv2 data types are defined in section 2.3 of RFC 1094:
@@ -593,8 +578,8 @@ static void nfs2_xdr_enc_readlinkargs(struct rpc_rqst *req,
 	const struct nfs_readlinkargs *args = data;
 
 	encode_fhandle(xdr, args->fh);
-	prepare_reply_buffer(req, args->pages, args->pgbase,
-					args->pglen, NFS_readlinkres_sz);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->pglen, NFS_readlinkres_sz);
 }
 
 /*
@@ -629,8 +614,8 @@ static void nfs2_xdr_enc_readargs(struct rpc_rqst *req,
 	const struct nfs_pgio_args *args = data;
 
 	encode_readargs(xdr, args);
-	prepare_reply_buffer(req, args->pages, args->pgbase,
-					args->count, NFS_readres_sz);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->count, NFS_readres_sz);
 	req->rq_rcv_buf.flags |= XDRBUF_READ;
 }
 
@@ -787,8 +772,8 @@ static void nfs2_xdr_enc_readdirargs(struct rpc_rqst *req,
 	const struct nfs_readdirargs *args = data;
 
 	encode_readdirargs(xdr, args);
-	prepare_reply_buffer(req, args->pages, 0,
-					args->count, NFS_readdirres_sz);
+	rpc_prepare_reply_pages(req, args->pages, 0,
+				args->count, NFS_readdirres_sz);
 }
 
 /*
diff --git a/fs/nfs/nfs3xdr.c b/fs/nfs/nfs3xdr.c
index 4aa3ffe..a54dcf4 100644
--- a/fs/nfs/nfs3xdr.c
+++ b/fs/nfs/nfs3xdr.c
@@ -105,21 +105,6 @@
 };
 
 /*
- * While encoding arguments, set up the reply buffer in advance to
- * receive reply data directly into the page cache.
- */
-static void prepare_reply_buffer(struct rpc_rqst *req, struct page **pages,
-				 unsigned int base, unsigned int len,
-				 unsigned int bufsize)
-{
-	struct rpc_auth	*auth = req->rq_cred->cr_auth;
-	unsigned int replen;
-
-	replen = RPC_REPHDRSIZE + auth->au_rslack + bufsize;
-	xdr_inline_pages(&req->rq_rcv_buf, replen << 2, pages, base, len);
-}
-
-/*
  * Encode/decode NFSv3 basic data types
  *
  * Basic NFSv3 data types are defined in section 2.5 of RFC 1813:
@@ -910,8 +895,8 @@ static void nfs3_xdr_enc_readlink3args(struct rpc_rqst *req,
 	const struct nfs3_readlinkargs *args = data;
 
 	encode_nfs_fh3(xdr, args->fh);
-	prepare_reply_buffer(req, args->pages, args->pgbase,
-					args->pglen, NFS3_readlinkres_sz);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->pglen, NFS3_readlinkres_sz);
 }
 
 /*
@@ -943,8 +928,8 @@ static void nfs3_xdr_enc_read3args(struct rpc_rqst *req,
 	unsigned int replen = args->replen ? args->replen : NFS3_readres_sz;
 
 	encode_read3args(xdr, args);
-	prepare_reply_buffer(req, args->pages, args->pgbase,
-					args->count, replen);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->count, replen);
 	req->rq_rcv_buf.flags |= XDRBUF_READ;
 }
 
@@ -1236,7 +1221,7 @@ static void nfs3_xdr_enc_readdir3args(struct rpc_rqst *req,
 	const struct nfs3_readdirargs *args = data;
 
 	encode_readdir3args(xdr, args);
-	prepare_reply_buffer(req, args->pages, 0,
+	rpc_prepare_reply_pages(req, args->pages, 0,
 				args->count, NFS3_readdirres_sz);
 }
 
@@ -1278,7 +1263,7 @@ static void nfs3_xdr_enc_readdirplus3args(struct rpc_rqst *req,
 	const struct nfs3_readdirargs *args = data;
 
 	encode_readdirplus3args(xdr, args);
-	prepare_reply_buffer(req, args->pages, 0,
+	rpc_prepare_reply_pages(req, args->pages, 0,
 				args->count, NFS3_readdirres_sz);
 }
 
@@ -1323,7 +1308,7 @@ static void nfs3_xdr_enc_getacl3args(struct rpc_rqst *req,
 	encode_nfs_fh3(xdr, args->fh);
 	encode_uint32(xdr, args->mask);
 	if (args->mask & (NFS_ACL | NFS_DFACL)) {
-		prepare_reply_buffer(req, args->pages, 0,
+		rpc_prepare_reply_pages(req, args->pages, 0,
 					NFSACL_MAXPAGES << PAGE_SHIFT,
 					ACL3_getaclres_sz);
 		req->rq_rcv_buf.flags |= XDRBUF_SPARSE_PAGES;
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 38a4cbc..d0fa18d 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -1016,12 +1016,11 @@ static void encode_compound_hdr(struct xdr_stream *xdr,
 				struct compound_hdr *hdr)
 {
 	__be32 *p;
-	struct rpc_auth *auth = req->rq_cred->cr_auth;
 
 	/* initialize running count of expected bytes in reply.
 	 * NOTE: the replied tag SHOULD be the same is the one sent,
 	 * but this is not required as a MUST for the server to do so. */
-	hdr->replen = RPC_REPHDRSIZE + auth->au_rslack + 3 + hdr->taglen;
+	hdr->replen = 3 + hdr->taglen;
 
 	WARN_ON_ONCE(hdr->taglen > NFS4_MAXTAGLEN);
 	encode_string(xdr, hdr->taglen, hdr->tag);
@@ -2341,9 +2340,9 @@ static void nfs4_xdr_enc_open(struct rpc_rqst *req, struct xdr_stream *xdr,
 	encode_getfattr_open(xdr, args->bitmask, args->open_bitmap, &hdr);
 	if (args->lg_args) {
 		encode_layoutget(xdr, args->lg_args, &hdr);
-		xdr_inline_pages(&req->rq_rcv_buf, hdr.replen << 2,
-				 args->lg_args->layout.pages,
-				 0, args->lg_args->layout.pglen);
+		rpc_prepare_reply_pages(req, args->lg_args->layout.pages, 0,
+					args->lg_args->layout.pglen,
+					hdr.replen);
 	}
 	encode_nops(&hdr);
 }
@@ -2387,9 +2386,9 @@ static void nfs4_xdr_enc_open_noattr(struct rpc_rqst *req,
 	encode_getfattr_open(xdr, args->bitmask, args->open_bitmap, &hdr);
 	if (args->lg_args) {
 		encode_layoutget(xdr, args->lg_args, &hdr);
-		xdr_inline_pages(&req->rq_rcv_buf, hdr.replen << 2,
-				 args->lg_args->layout.pages,
-				 0, args->lg_args->layout.pglen);
+		rpc_prepare_reply_pages(req, args->lg_args->layout.pages, 0,
+					args->lg_args->layout.pglen,
+					hdr.replen);
 	}
 	encode_nops(&hdr);
 }
@@ -2499,8 +2498,8 @@ static void nfs4_xdr_enc_readlink(struct rpc_rqst *req, struct xdr_stream *xdr,
 	encode_putfh(xdr, args->fh, &hdr);
 	encode_readlink(xdr, args, req, &hdr);
 
-	xdr_inline_pages(&req->rq_rcv_buf, hdr.replen << 2, args->pages,
-			args->pgbase, args->pglen);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->pglen, hdr.replen);
 	encode_nops(&hdr);
 }
 
@@ -2520,11 +2519,8 @@ static void nfs4_xdr_enc_readdir(struct rpc_rqst *req, struct xdr_stream *xdr,
 	encode_putfh(xdr, args->fh, &hdr);
 	encode_readdir(xdr, args, req, &hdr);
 
-	xdr_inline_pages(&req->rq_rcv_buf, hdr.replen << 2, args->pages,
-			 args->pgbase, args->count);
-	dprintk("%s: inlined page args = (%u, %p, %u, %u)\n",
-			__func__, hdr.replen << 2, args->pages,
-			args->pgbase, args->count);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->count, hdr.replen);
 	encode_nops(&hdr);
 }
 
@@ -2544,8 +2540,8 @@ static void nfs4_xdr_enc_read(struct rpc_rqst *req, struct xdr_stream *xdr,
 	encode_putfh(xdr, args->fh, &hdr);
 	encode_read(xdr, args, &hdr);
 
-	xdr_inline_pages(&req->rq_rcv_buf, hdr.replen << 2,
-			 args->pages, args->pgbase, args->count);
+	rpc_prepare_reply_pages(req, args->pages, args->pgbase,
+				args->count, hdr.replen);
 	req->rq_rcv_buf.flags |= XDRBUF_READ;
 	encode_nops(&hdr);
 }
@@ -2591,9 +2587,8 @@ static void nfs4_xdr_enc_getacl(struct rpc_rqst *req, struct xdr_stream *xdr,
 	encode_getattr(xdr, nfs4_acl_bitmap, NULL,
 			ARRAY_SIZE(nfs4_acl_bitmap), &hdr);
 
-	xdr_inline_pages(&req->rq_rcv_buf, replen << 2,
-		args->acl_pages, 0, args->acl_len);
-
+	rpc_prepare_reply_pages(req, args->acl_pages, 0,
+				args->acl_len, replen);
 	encode_nops(&hdr);
 }
 
@@ -2814,9 +2809,8 @@ static void nfs4_xdr_enc_fs_locations(struct rpc_rqst *req,
 		encode_fs_locations(xdr, args->bitmask, &hdr);
 	}
 
-	/* Set up reply kvec to capture returned fs_locations array. */
-	xdr_inline_pages(&req->rq_rcv_buf, replen << 2,
-			 (struct page **)&args->page, 0, PAGE_SIZE);
+	rpc_prepare_reply_pages(req, (struct page **)&args->page, 0,
+				PAGE_SIZE, replen);
 	encode_nops(&hdr);
 }
 
@@ -3018,10 +3012,8 @@ static void nfs4_xdr_enc_getdeviceinfo(struct rpc_rqst *req,
 
 	/* set up reply kvec. Subtract notification bitmap max size (2)
 	 * so that notification bitmap is put in xdr_buf tail */
-	xdr_inline_pages(&req->rq_rcv_buf, (hdr.replen - 2) << 2,
-			 args->pdev->pages, args->pdev->pgbase,
-			 args->pdev->pglen);
-
+	rpc_prepare_reply_pages(req, args->pdev->pages, args->pdev->pgbase,
+				args->pdev->pglen, hdr.replen - 2);
 	encode_nops(&hdr);
 }
 
@@ -3042,9 +3034,8 @@ static void nfs4_xdr_enc_layoutget(struct rpc_rqst *req,
 	encode_putfh(xdr, NFS_FH(args->inode), &hdr);
 	encode_layoutget(xdr, args, &hdr);
 
-	xdr_inline_pages(&req->rq_rcv_buf, hdr.replen << 2,
-	    args->layout.pages, 0, args->layout.pglen);
-
+	rpc_prepare_reply_pages(req, args->layout.pages, 0,
+				args->layout.pglen, hdr.replen);
 	encode_nops(&hdr);
 }
 
diff --git a/include/linux/sunrpc/clnt.h b/include/linux/sunrpc/clnt.h
index 1c44171..98bc988 100644
--- a/include/linux/sunrpc/clnt.h
+++ b/include/linux/sunrpc/clnt.h
@@ -169,6 +169,9 @@ int		rpcb_v4_register(struct net *net, const u32 program,
 				 const char *netid);
 void		rpcb_getport_async(struct rpc_task *);
 
+void rpc_prepare_reply_pages(struct rpc_rqst *req, struct page **pages,
+			     unsigned int base, unsigned int len,
+			     unsigned int hdrsize);
 void		rpc_call_start(struct rpc_task *);
 int		rpc_call_async(struct rpc_clnt *clnt,
 			       const struct rpc_message *msg, int flags,
diff --git a/include/trace/events/sunrpc.h b/include/trace/events/sunrpc.h
index e58dda8..8451f30 100644
--- a/include/trace/events/sunrpc.h
+++ b/include/trace/events/sunrpc.h
@@ -461,6 +461,43 @@
 	)
 );
 
+TRACE_EVENT(rpc_reply_pages,
+	TP_PROTO(
+		const struct rpc_rqst *req
+	),
+
+	TP_ARGS(req),
+
+	TP_STRUCT__entry(
+		__field(unsigned int, task_id)
+		__field(unsigned int, client_id)
+		__field(const void *, head_base)
+		__field(size_t, head_len)
+		__field(const void *, tail_base)
+		__field(size_t, tail_len)
+		__field(unsigned int, page_len)
+	),
+
+	TP_fast_assign(
+		__entry->task_id = req->rq_task->tk_pid;
+		__entry->client_id = req->rq_task->tk_client->cl_clid;
+
+		__entry->head_base = req->rq_rcv_buf.head[0].iov_base;
+		__entry->head_len = req->rq_rcv_buf.head[0].iov_len;
+		__entry->page_len = req->rq_rcv_buf.page_len;
+		__entry->tail_base = req->rq_rcv_buf.tail[0].iov_base;
+		__entry->tail_len = req->rq_rcv_buf.tail[0].iov_len;
+	),
+
+	TP_printk(
+		"task:%u@%u xdr=[%p,%zu]/%u/[%p,%zu]\n",
+		__entry->task_id, __entry->client_id,
+		__entry->head_base, __entry->head_len,
+		__entry->page_len,
+		__entry->tail_base, __entry->tail_len
+	)
+);
+
 /*
  * First define the enums in the below macros to be exported to userspace
  * via TRACE_DEFINE_ENUM().
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 803e931..f780605 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1164,6 +1164,25 @@ struct rpc_task *rpc_run_bc_task(struct rpc_rqst *req)
 }
 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 
+/**
+ * rpc_prepare_reply_pages - Prepare to receive a reply data payload into pages
+ * @req: RPC request to prepare
+ * @pages: vector of struct page pointers
+ * @base: offset in first page where receive should start, in bytes
+ * @len: expected size of the upper layer data payload, in bytes
+ * @hdrsize: expected size of upper layer reply header, in XDR words
+ *
+ */
+void rpc_prepare_reply_pages(struct rpc_rqst *req, struct page **pages,
+			     unsigned int base, unsigned int len,
+			     unsigned int hdrsize)
+{
+	hdrsize += RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_rslack;
+	xdr_inline_pages(&req->rq_rcv_buf, hdrsize << 2, pages, base, len);
+	trace_rpc_reply_pages(req);
+}
+EXPORT_SYMBOL_GPL(rpc_prepare_reply_pages);
+
 void
 rpc_call_start(struct rpc_task *task)
 {
diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
index 4bce619..7cca515 100644
--- a/net/sunrpc/xdr.c
+++ b/net/sunrpc/xdr.c
@@ -163,6 +163,15 @@ __be32 *xdr_encode_opaque(__be32 *p, const void *ptr, unsigned int nbytes)
 	buf->bvec = NULL;
 }
 
+/**
+ * xdr_inline_pages - Prepare receive buffer for a large reply
+ * @xdr: xdr_buf into which reply will be placed
+ * @offset: expected offset where data payload will start, in bytes
+ * @pages: vector of struct page pointers
+ * @base: offset in first page where receive should start, in bytes
+ * @len: expected size of the upper layer data payload, in bytes
+ *
+ */
 void
 xdr_inline_pages(struct xdr_buf *xdr, unsigned int offset,
 		 struct page **pages, unsigned int base, unsigned int len)

[PATCH RFC 2/5] NFS: Account for XDR pad of buf->pages

[Chuck Lever]

Certain NFS results (eg. READLINK) might expect a data payload that
is not an exact multiple of 4 bytes. In this case, XDR encoding
is required to pad that payload so its length on the wire is a
multiple of 4 bytes. The constants that define the maximum size of
each NFS result do not appear to account for this extra word.

In each case where the data payload is to be received into pages:

- 1 word is added to the size of the receive buffer allocated by
  call_allocate

- rpc_inline_rcv_pages subtracts 1 word from @hdrsize so that the
  extra buffer space falls into the rcv_buf's tail iovec

- If buf->pagelen is word-aligned, an XDR pad is not needed and
  is thus removed from the tail

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/nfs/nfs2xdr.c  |    6 +++---
 fs/nfs/nfs3xdr.c  |   10 +++++-----
 fs/nfs/nfs4xdr.c  |   15 ++++++++-------
 net/sunrpc/clnt.c |    6 +++++-
 net/sunrpc/xdr.c  |    2 ++
 5 files changed, 23 insertions(+), 16 deletions(-)

diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c
index 1dcd0fe..a7ed29d 100644
--- a/fs/nfs/nfs2xdr.c
+++ b/fs/nfs/nfs2xdr.c
@@ -56,11 +56,11 @@
 
 #define NFS_attrstat_sz		(1+NFS_fattr_sz)
 #define NFS_diropres_sz		(1+NFS_fhandle_sz+NFS_fattr_sz)
-#define NFS_readlinkres_sz	(2)
-#define NFS_readres_sz		(1+NFS_fattr_sz+1)
+#define NFS_readlinkres_sz	(2+1)
+#define NFS_readres_sz		(1+NFS_fattr_sz+1+1)
 #define NFS_writeres_sz         (NFS_attrstat_sz)
 #define NFS_stat_sz		(1)
-#define NFS_readdirres_sz	(1)
+#define NFS_readdirres_sz	(1+1)
 #define NFS_statfsres_sz	(1+NFS_info_sz)
 
 static int nfs_stat_to_errno(enum nfs_stat);
diff --git a/fs/nfs/nfs3xdr.c b/fs/nfs/nfs3xdr.c
index a54dcf4..110358f 100644
--- a/fs/nfs/nfs3xdr.c
+++ b/fs/nfs/nfs3xdr.c
@@ -69,13 +69,13 @@
 #define NFS3_removeres_sz	(NFS3_setattrres_sz)
 #define NFS3_lookupres_sz	(1+NFS3_fh_sz+(2 * NFS3_post_op_attr_sz))
 #define NFS3_accessres_sz	(1+NFS3_post_op_attr_sz+1)
-#define NFS3_readlinkres_sz	(1+NFS3_post_op_attr_sz+1)
-#define NFS3_readres_sz		(1+NFS3_post_op_attr_sz+3)
+#define NFS3_readlinkres_sz	(1+NFS3_post_op_attr_sz+1+1)
+#define NFS3_readres_sz		(1+NFS3_post_op_attr_sz+3+1)
 #define NFS3_writeres_sz	(1+NFS3_wcc_data_sz+4)
 #define NFS3_createres_sz	(1+NFS3_fh_sz+NFS3_post_op_attr_sz+NFS3_wcc_data_sz)
 #define NFS3_renameres_sz	(1+(2 * NFS3_wcc_data_sz))
 #define NFS3_linkres_sz		(1+NFS3_post_op_attr_sz+NFS3_wcc_data_sz)
-#define NFS3_readdirres_sz	(1+NFS3_post_op_attr_sz+2)
+#define NFS3_readdirres_sz	(1+NFS3_post_op_attr_sz+2+1)
 #define NFS3_fsstatres_sz	(1+NFS3_post_op_attr_sz+13)
 #define NFS3_fsinfores_sz	(1+NFS3_post_op_attr_sz+12)
 #define NFS3_pathconfres_sz	(1+NFS3_post_op_attr_sz+6)
@@ -85,7 +85,7 @@
 #define ACL3_setaclargs_sz	(NFS3_fh_sz+1+ \
 				XDR_QUADLEN(NFS_ACL_INLINE_BUFSIZE))
 #define ACL3_getaclres_sz	(1+NFS3_post_op_attr_sz+1+ \
-				XDR_QUADLEN(NFS_ACL_INLINE_BUFSIZE))
+				XDR_QUADLEN(NFS_ACL_INLINE_BUFSIZE)+1)
 #define ACL3_setaclres_sz	(1+NFS3_post_op_attr_sz)
 
 static int nfs3_stat_to_errno(enum nfs_stat);
@@ -1629,7 +1629,7 @@ static int nfs3_xdr_dec_read3res(struct rpc_rqst *req, struct xdr_stream *xdr,
 	result->op_status = status;
 	if (status != NFS3_OK)
 		goto out_status;
-	result->replen = 3 + ((xdr_stream_pos(xdr) - pos) >> 2);
+	result->replen = 4 + ((xdr_stream_pos(xdr) - pos) >> 2);
 	error = decode_read3resok(xdr, result);
 out:
 	return error;
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index d0fa18d..6d9d5e2 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -215,14 +215,14 @@ static int decode_layoutget(struct xdr_stream *xdr, struct rpc_rqst *req,
 				 nfs4_fattr_bitmap_maxsz)
 #define encode_read_maxsz	(op_encode_hdr_maxsz + \
 				 encode_stateid_maxsz + 3)
-#define decode_read_maxsz	(op_decode_hdr_maxsz + 2)
+#define decode_read_maxsz	(op_decode_hdr_maxsz + 2 + 1)
 #define encode_readdir_maxsz	(op_encode_hdr_maxsz + \
 				 2 + encode_verifier_maxsz + 5 + \
 				nfs4_label_maxsz)
 #define decode_readdir_maxsz	(op_decode_hdr_maxsz + \
-				 decode_verifier_maxsz)
+				 decode_verifier_maxsz + 1)
 #define encode_readlink_maxsz	(op_encode_hdr_maxsz)
-#define decode_readlink_maxsz	(op_decode_hdr_maxsz + 1)
+#define decode_readlink_maxsz	(op_decode_hdr_maxsz + 1 + 1)
 #define encode_write_maxsz	(op_encode_hdr_maxsz + \
 				 encode_stateid_maxsz + 4)
 #define decode_write_maxsz	(op_decode_hdr_maxsz + \
@@ -284,14 +284,14 @@ static int decode_layoutget(struct xdr_stream *xdr, struct rpc_rqst *req,
 #define decode_delegreturn_maxsz (op_decode_hdr_maxsz)
 #define encode_getacl_maxsz	(encode_getattr_maxsz)
 #define decode_getacl_maxsz	(op_decode_hdr_maxsz + \
-				 nfs4_fattr_bitmap_maxsz + 1)
+				 nfs4_fattr_bitmap_maxsz + 1 + 1)
 #define encode_setacl_maxsz	(op_encode_hdr_maxsz + \
 				 encode_stateid_maxsz + 3)
 #define decode_setacl_maxsz	(decode_setattr_maxsz)
 #define encode_fs_locations_maxsz \
 				(encode_getattr_maxsz)
 #define decode_fs_locations_maxsz \
-				(0)
+				(1)
 #define encode_secinfo_maxsz	(op_encode_hdr_maxsz + nfs4_name_maxsz)
 #define decode_secinfo_maxsz	(op_decode_hdr_maxsz + 1 + ((NFS_MAX_SECFLAVORS * (16 + GSS_OID_MAX_LEN)) / 4))
 
@@ -392,12 +392,13 @@ static int decode_layoutget(struct xdr_stream *xdr, struct rpc_rqst *req,
 				1 /* opaque devaddr4 length */ + \
 				  /* devaddr4 payload is read into page */ \
 				1 /* notification bitmap length */ + \
-				1 /* notification bitmap, word 0 */)
+				1 /* notification bitmap, word 0 */ + \
+				1 /* possible XDR padding */)
 #define encode_layoutget_maxsz	(op_encode_hdr_maxsz + 10 + \
 				encode_stateid_maxsz)
 #define decode_layoutget_maxsz	(op_decode_hdr_maxsz + 8 + \
 				decode_stateid_maxsz + \
-				XDR_QUADLEN(PNFS_LAYOUT_MAXSIZE))
+				XDR_QUADLEN(PNFS_LAYOUT_MAXSIZE) + 1)
 #define encode_layoutcommit_maxsz (op_encode_hdr_maxsz +          \
 				2 /* offset */ + \
 				2 /* length */ + \
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index f780605..4ea38b0 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1177,7 +1177,11 @@ void rpc_prepare_reply_pages(struct rpc_rqst *req, struct page **pages,
 			     unsigned int base, unsigned int len,
 			     unsigned int hdrsize)
 {
-	hdrsize += RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_rslack;
+	/* Subtract one to force an extra word of buffer space for the
+	 * payload's XDR pad to fall into the rcv_buf's tail iovec.
+	 */
+	hdrsize += RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_rslack - 1;
+
 	xdr_inline_pages(&req->rq_rcv_buf, hdrsize << 2, pages, base, len);
 	trace_rpc_reply_pages(req);
 }
diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
index 7cca515..aa8177d 100644
--- a/net/sunrpc/xdr.c
+++ b/net/sunrpc/xdr.c
@@ -189,6 +189,8 @@ __be32 *xdr_encode_opaque(__be32 *p, const void *ptr, unsigned int nbytes)
 
 	tail->iov_base = buf + offset;
 	tail->iov_len = buflen - offset;
+	if ((xdr->page_len & 3) == 0)
+		tail->iov_len -= sizeof(__be32);
 
 	xdr->buflen += len;
 }

[PATCH RFC 3/5] SUNRPC: Make AUTH_SYS and AUTH_NULL set au_verfsize

[Chuck Lever]

au_verfsize will be needed for a non-flavor-specific computation
in a subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 include/linux/sunrpc/auth.h    |    3 +--
 net/sunrpc/auth_gss/auth_gss.c |    1 +
 net/sunrpc/auth_null.c         |    1 +
 net/sunrpc/auth_unix.c         |    5 ++++-
 4 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/include/linux/sunrpc/auth.h b/include/linux/sunrpc/auth.h
index c51e189..359dfdd 100644
--- a/include/linux/sunrpc/auth.h
+++ b/include/linux/sunrpc/auth.h
@@ -77,8 +77,7 @@ struct rpc_auth {
 				/* guess at number of u32's auth adds before
 				 * reply data; normally the verifier size: */
 	unsigned int		au_rslack;
-				/* for gss, used to calculate au_rslack: */
-	unsigned int		au_verfsize;
+	unsigned int		au_verfsize;	/* size of reply verifier */
 
 	unsigned int		au_flags;	/* various flags */
 	const struct rpc_authops *au_ops;		/* operations */
diff --git a/net/sunrpc/auth_gss/auth_gss.c b/net/sunrpc/auth_gss/auth_gss.c
index fda454c..731e7a4 100644
--- a/net/sunrpc/auth_gss/auth_gss.c
+++ b/net/sunrpc/auth_gss/auth_gss.c
@@ -1016,6 +1016,7 @@ static void gss_pipe_free(struct gss_pipe *p)
 	auth = &gss_auth->rpc_auth;
 	auth->au_cslack = GSS_CRED_SLACK >> 2;
 	auth->au_rslack = GSS_VERF_SLACK >> 2;
+	auth->au_verfsize = GSS_VERF_SLACK >> 2;
 	auth->au_flags = 0;
 	auth->au_ops = &authgss_ops;
 	auth->au_flavor = flavor;
diff --git a/net/sunrpc/auth_null.c b/net/sunrpc/auth_null.c
index bf96975..9ae0824 100644
--- a/net/sunrpc/auth_null.c
+++ b/net/sunrpc/auth_null.c
@@ -114,6 +114,7 @@
 struct rpc_auth null_auth = {
 	.au_cslack	= NUL_CALLSLACK,
 	.au_rslack	= NUL_REPLYSLACK,
+	.au_verfsize	= NUL_REPLYSLACK,
 	.au_ops		= &authnull_ops,
 	.au_flavor	= RPC_AUTH_NULL,
 	.au_count	= REFCOUNT_INIT(1),
diff --git a/net/sunrpc/auth_unix.c b/net/sunrpc/auth_unix.c
index 5ea84a9..a93c564 100644
--- a/net/sunrpc/auth_unix.c
+++ b/net/sunrpc/auth_unix.c
@@ -163,6 +163,7 @@
 static int
 unx_validate(struct rpc_task *task, struct xdr_stream *xdr)
 {
+	struct rpc_auth *auth = task->tk_rqstp->rq_cred->cr_auth;
 	__be32 *p;
 	u32 size;
 
@@ -184,7 +185,8 @@
 	if (!p)
 		return -EIO;
 
-	task->tk_rqstp->rq_cred->cr_auth->au_rslack = (size >> 2) + 2;
+	auth->au_verfsize = XDR_QUADLEN(size) + 2;
+	auth->au_rslack = XDR_QUADLEN(size) + 2;
 	return 0;
 }
 
@@ -212,6 +214,7 @@ void rpc_destroy_authunix(void)
 struct rpc_auth		unix_auth = {
 	.au_cslack	= UNX_CALLSLACK,
 	.au_rslack	= NUL_REPLYSLACK,
+	.au_verfsize	= NUL_REPLYSLACK,
 	.au_ops		= &authunix_ops,
 	.au_flavor	= RPC_AUTH_UNIX,
 	.au_count	= REFCOUNT_INIT(1),

[PATCH RFC 4/5] SUNRPC: Add rpc_auth::au_ralign field

[Chuck Lever]

Currently rpc_inline_rcv_pages() uses au_rslack to estimate the
size of the upper layer reply header. This is fine for auth flavors
where au_verfsize == au_rslack.

However, some auth flavors have more going on. krb5i for example has
two more words after the verifier, and another blob following the
RPC message. The calculation involving au_rslack pushes the upper
layer reply header too far into the rcv_buf.

au_rslack is still valuable: it's the amount of buffer space needed
for the reply, and is used when allocating the reply buffer. We'll
keep that.

But, add a new field that can be used to properly estimate the
location of the upper layer header in each RPC reply, based on the
auth flavor in use.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 include/linux/sunrpc/auth.h    |    9 ++++-----
 net/sunrpc/auth_gss/auth_gss.c |   18 +++++++++++++-----
 net/sunrpc/auth_null.c         |    1 +
 net/sunrpc/auth_unix.c         |    1 +
 net/sunrpc/clnt.c              |    2 +-
 5 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/include/linux/sunrpc/auth.h b/include/linux/sunrpc/auth.h
index 359dfdd..5f9076f 100644
--- a/include/linux/sunrpc/auth.h
+++ b/include/linux/sunrpc/auth.h
@@ -74,13 +74,12 @@ struct rpc_cred {
 struct rpc_authops;
 struct rpc_auth {
 	unsigned int		au_cslack;	/* call cred size estimate */
-				/* guess at number of u32's auth adds before
-				 * reply data; normally the verifier size: */
-	unsigned int		au_rslack;
+	unsigned int		au_rslack;	/* reply cred size estimate */
 	unsigned int		au_verfsize;	/* size of reply verifier */
+	unsigned int		au_ralign;	/* words before UL header */
 
-	unsigned int		au_flags;	/* various flags */
-	const struct rpc_authops *au_ops;		/* operations */
+	unsigned int		au_flags;
+	const struct rpc_authops *au_ops;
 	rpc_authflavor_t	au_flavor;	/* pseudoflavor (note may
 						 * differ from the flavor in
 						 * au_ops->au_flavor in gss
diff --git a/net/sunrpc/auth_gss/auth_gss.c b/net/sunrpc/auth_gss/auth_gss.c
index 731e7a4..c67e2ad 100644
--- a/net/sunrpc/auth_gss/auth_gss.c
+++ b/net/sunrpc/auth_gss/auth_gss.c
@@ -1017,6 +1017,7 @@ static void gss_pipe_free(struct gss_pipe *p)
 	auth->au_cslack = GSS_CRED_SLACK >> 2;
 	auth->au_rslack = GSS_VERF_SLACK >> 2;
 	auth->au_verfsize = GSS_VERF_SLACK >> 2;
+	auth->au_ralign = GSS_VERF_SLACK >> 2;
 	auth->au_flags = 0;
 	auth->au_ops = &authgss_ops;
 	auth->au_flavor = flavor;
@@ -1891,7 +1892,10 @@ static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
 static int
 gss_unwrap_resp_auth(struct rpc_cred *cred)
 {
-	cred->cr_auth->au_rslack = cred->cr_auth->au_verfsize;
+	struct rpc_auth *auth = cred->cr_auth;
+
+	auth->au_rslack = auth->au_verfsize;
+	auth->au_ralign = auth->au_verfsize;
 	return 0;
 }
 
@@ -1902,6 +1906,7 @@ static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
 {
 	struct xdr_buf integ_buf, *rcv_buf = &rqstp->rq_rcv_buf;
 	u32 data_offset, mic_offset, integ_len, maj_stat;
+	struct rpc_auth *auth = cred->cr_auth;
 	struct xdr_netobj mic;
 	__be32 *p;
 
@@ -1928,8 +1933,8 @@ static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
 	if (maj_stat != GSS_S_COMPLETE)
 		goto bad_mic;
 
-	cred->cr_auth->au_rslack = cred->cr_auth->au_verfsize + 2 +
-				   1 + XDR_QUADLEN(mic.len);
+	auth->au_rslack = auth->au_verfsize + 2 + 1 + XDR_QUADLEN(mic.len);
+	auth->au_ralign = auth->au_verfsize + 2;
 	return 0;
 unwrap_failed:
 	trace_rpcgss_unwrap_failed(task);
@@ -1949,6 +1954,7 @@ static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
 {
 	struct xdr_buf *rcv_buf = &rqstp->rq_rcv_buf;
 	struct kvec *head = rqstp->rq_rcv_buf.head;
+	struct rpc_auth *auth = cred->cr_auth;
 	unsigned int savedlen = rcv_buf->len;
 	u32 offset, opaque_len, maj_stat;
 	__be32 *p;
@@ -1976,8 +1982,10 @@ static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
 	 */
 	xdr_init_decode(xdr, rcv_buf, p, rqstp);
 
-	cred->cr_auth->au_rslack = cred->cr_auth->au_verfsize + 2 +
-				   XDR_QUADLEN(savedlen - rcv_buf->len);
+	auth->au_rslack = auth->au_verfsize + 2 +
+			  XDR_QUADLEN(savedlen - rcv_buf->len);
+	auth->au_ralign = auth->au_verfsize + 2 +
+			  XDR_QUADLEN(savedlen - rcv_buf->len);
 	return 0;
 unwrap_failed:
 	trace_rpcgss_unwrap_failed(task);
diff --git a/net/sunrpc/auth_null.c b/net/sunrpc/auth_null.c
index 9ae0824..41a633a 100644
--- a/net/sunrpc/auth_null.c
+++ b/net/sunrpc/auth_null.c
@@ -115,6 +115,7 @@ struct rpc_auth null_auth = {
 	.au_cslack	= NUL_CALLSLACK,
 	.au_rslack	= NUL_REPLYSLACK,
 	.au_verfsize	= NUL_REPLYSLACK,
+	.au_ralign	= NUL_REPLYSLACK,
 	.au_ops		= &authnull_ops,
 	.au_flavor	= RPC_AUTH_NULL,
 	.au_count	= REFCOUNT_INIT(1),
diff --git a/net/sunrpc/auth_unix.c b/net/sunrpc/auth_unix.c
index a93c564..c048eb6 100644
--- a/net/sunrpc/auth_unix.c
+++ b/net/sunrpc/auth_unix.c
@@ -187,6 +187,7 @@
 
 	auth->au_verfsize = XDR_QUADLEN(size) + 2;
 	auth->au_rslack = XDR_QUADLEN(size) + 2;
+	auth->au_ralign = XDR_QUADLEN(size) + 2;
 	return 0;
 }
 
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 4ea38b0..99bfeb1 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1180,7 +1180,7 @@ void rpc_prepare_reply_pages(struct rpc_rqst *req, struct page **pages,
 	/* Subtract one to force an extra word of buffer space for the
 	 * payload's XDR pad to fall into the rcv_buf's tail iovec.
 	 */
-	hdrsize += RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_rslack - 1;
+	hdrsize += RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_ralign - 1;
 
 	xdr_inline_pages(&req->rq_rcv_buf, hdrsize << 2, pages, base, len);
 	trace_rpc_reply_pages(req);

[PATCH RFC 5/5] SUNRPC: Use au_rslack when computing reply buffer size

[Chuck Lever]

au_rslack is significantly smaller than (au_cslack << 2). Using
that value results in smaller receive buffers. In some cases this
eliminates an extra segment in Reply chunks (RPC/RDMA).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 net/sunrpc/clnt.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 99bfeb1..241e842 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1690,7 +1690,7 @@ void rpc_force_rebind(struct rpc_clnt *clnt)
 static void
 call_allocate(struct rpc_task *task)
 {
-	unsigned int slack = task->tk_rqstp->rq_cred->cr_auth->au_cslack;
+	const struct rpc_auth *auth = task->tk_rqstp->rq_cred->cr_auth;
 	struct rpc_rqst *req = task->tk_rqstp;
 	struct rpc_xprt *xprt = req->rq_xprt;
 	const struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
@@ -1715,9 +1715,10 @@ void rpc_force_rebind(struct rpc_clnt *clnt)
 	 * and reply headers, and convert both values
 	 * to byte sizes.
 	 */
-	req->rq_callsize = RPC_CALLHDRSIZE + (slack << 1) + proc->p_arglen;
+	req->rq_callsize = RPC_CALLHDRSIZE + (auth->au_cslack << 1) +
+			   proc->p_arglen;
 	req->rq_callsize <<= 2;
-	req->rq_rcvsize = RPC_REPHDRSIZE + slack + proc->p_replen;
+	req->rq_rcvsize = RPC_REPHDRSIZE + auth->au_rslack + proc->p_replen;
 	req->rq_rcvsize <<= 2;
 
 	status = xprt->ops->buf_alloc(task);