NFS: support large reads and writes on the wire

 Most NFS server implementations allow up to 64KB reads and writes on the
 wire.  The Solaris NFS server allows up to a megabyte, for instance.

 Now the Linux NFS client supports transfer sizes up to 1MB, too.  This will
 help reduce protocol and context switch overhead on read/write intensive NFS
 workloads, and support larger atomic read and write operations on servers
 that support them.

 Test-plan:
 Connectathon and iozone on mount point with wsize=rsize>32768 over TCP.
 Tests with NFS over UDP to verify the maximum RPC payload size cap.

 Signed-off-by: Chuck Lever <cel@netapp.com>
 Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

3 files changed, 51 insertions, 24 deletions

diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h
index 4dff705d2ff..d38010ba647 100644
--- a/include/linux/nfs_fs.h
+++ b/include/linux/nfs_fs.h
@@ -38,9 +38,6 @@
 # define NFS_DEBUG
 #endif
 
-#define NFS_MAX_FILE_IO_BUFFER_SIZE	32768
-#define NFS_DEF_FILE_IO_BUFFER_SIZE	4096
-
 /* Default timeout values */
 #define NFS_MAX_UDP_TIMEOUT	(60*HZ)
 #define NFS_MAX_TCP_TIMEOUT	(600*HZ)
@@ -462,18 +459,33 @@ static inline int nfs_wb_page(struct inode *inode, struct page* page)
  */
 extern mempool_t *nfs_wdata_mempool;
 
-static inline struct nfs_write_data *nfs_writedata_alloc(void)
+static inline struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
 {
 	struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS);
+
 	if (p) {
 		memset(p, 0, sizeof(*p));
 		INIT_LIST_HEAD(&p->pages);
+		if (pagecount < NFS_PAGEVEC_SIZE)
+			p->pagevec = &p->page_array[0];
+		else {
+			size_t size = ++pagecount * sizeof(struct page *);
+			p->pagevec = kmalloc(size, GFP_NOFS);
+			if (p->pagevec) {
+				memset(p->pagevec, 0, size);
+			} else {
+				mempool_free(p, nfs_wdata_mempool);
+				p = NULL;
+			}
+		}
 	}
 	return p;
 }
 
 static inline void nfs_writedata_free(struct nfs_write_data *p)
 {
+	if (p && (p->pagevec != &p->page_array[0]))
+		kfree(p->pagevec);
 	mempool_free(p, nfs_wdata_mempool);
 }
 
@@ -492,16 +504,33 @@ extern void  nfs_readdata_release(void *data);
  */
 extern mempool_t *nfs_rdata_mempool;
 
-static inline struct nfs_read_data *nfs_readdata_alloc(void)
+static inline struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
 {
 	struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, SLAB_NOFS);
-	if (p)
+
+	if (p) {
 		memset(p, 0, sizeof(*p));
+		INIT_LIST_HEAD(&p->pages);
+		if (pagecount < NFS_PAGEVEC_SIZE)
+			p->pagevec = &p->page_array[0];
+		else {
+			size_t size = ++pagecount * sizeof(struct page *);
+			p->pagevec = kmalloc(size, GFP_NOFS);
+			if (p->pagevec) {
+				memset(p->pagevec, 0, size);
+			} else {
+				mempool_free(p, nfs_rdata_mempool);
+				p = NULL;
+			}
+		}
+	}
 	return p;
 }
 
 static inline void nfs_readdata_free(struct nfs_read_data *p)
 {
+	if (p && (p->pagevec != &p->page_array[0]))
+		kfree(p->pagevec);
 	mempool_free(p, nfs_rdata_mempool);
 }
 
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index b8b0eed98ec..9f422fd8767 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -4,6 +4,16 @@
 #include <linux/sunrpc/xprt.h>
 #include <linux/nfsacl.h>
 
+/*
+ * To change the maximum rsize and wsize supported by the NFS client, adjust
+ * NFS_MAX_FILE_IO_SIZE.  64KB is a typical maximum, but some servers can
+ * support a megabyte or more.  The default is left at 4096 bytes, which is
+ * reasonable for NFS over UDP.
+ */
+#define NFS_MAX_FILE_IO_SIZE	(1048576U)
+#define NFS_DEF_FILE_IO_SIZE	(4096U)
+#define NFS_MIN_FILE_IO_SIZE	(1024U)
+
 struct nfs4_fsid {
 	__u64 major;
 	__u64 minor;
@@ -215,12 +225,6 @@ struct nfs4_delegreturnargs {
 /*
  * Arguments to the read call.
  */
-
-#define NFS_READ_MAXIOV		(9U)
-#if (NFS_READ_MAXIOV > (MAX_IOVEC -2))
-#error "NFS_READ_MAXIOV is too large"
-#endif
-
 struct nfs_readargs {
 	struct nfs_fh *		fh;
 	struct nfs_open_context *context;
@@ -239,11 +243,6 @@ struct nfs_readres {
 /*
  * Arguments to the write call.
  */
-#define NFS_WRITE_MAXIOV	(9U)
-#if (NFS_WRITE_MAXIOV > (MAX_IOVEC -2))
-#error "NFS_WRITE_MAXIOV is too large"
-#endif
-
 struct nfs_writeargs {
 	struct nfs_fh *		fh;
 	struct nfs_open_context *context;
@@ -674,6 +673,8 @@ struct nfs4_server_caps_res {
 
 struct nfs_page;
 
+#define NFS_PAGEVEC_SIZE	(8U)
+
 struct nfs_read_data {
 	int			flags;
 	struct rpc_task		task;
@@ -682,13 +683,14 @@ struct nfs_read_data {
 	struct nfs_fattr	fattr;	/* fattr storage */
 	struct list_head	pages;	/* Coalesced read requests */
 	struct nfs_page		*req;	/* multi ops per nfs_page */
-	struct page		*pagevec[NFS_READ_MAXIOV];
+	struct page		**pagevec;
 	struct nfs_readargs args;
 	struct nfs_readres  res;
 #ifdef CONFIG_NFS_V4
 	unsigned long		timestamp;	/* For lease renewal */
 #endif
 	void (*complete) (struct nfs_read_data *, int);
+	struct page		*page_array[NFS_PAGEVEC_SIZE + 1];
 };
 
 struct nfs_write_data {
@@ -700,13 +702,14 @@ struct nfs_write_data {
 	struct nfs_writeverf	verf;
 	struct list_head	pages;		/* Coalesced requests we wish to flush */
 	struct nfs_page		*req;		/* multi ops per nfs_page */
-	struct page		*pagevec[NFS_WRITE_MAXIOV];
+	struct page		**pagevec;
 	struct nfs_writeargs	args;		/* argument struct */
 	struct nfs_writeres	res;		/* result struct */
 #ifdef CONFIG_NFS_V4
 	unsigned long		timestamp;	/* For lease renewal */
 #endif
 	void (*complete) (struct nfs_write_data *, int);
+	struct page		*page_array[NFS_PAGEVEC_SIZE + 1];
 };
 
 struct nfs_access_entry;
diff --git a/include/linux/sunrpc/xdr.h b/include/linux/sunrpc/xdr.h
index 5da968729cf..5676794ee34 100644
--- a/include/linux/sunrpc/xdr.h
+++ b/include/linux/sunrpc/xdr.h
@@ -135,11 +135,6 @@ xdr_adjust_iovec(struct kvec *iov, u32 *p)
 }
 
 /*
- * Maximum number of iov's we use.
- */
-#define MAX_IOVEC	(12)
-
-/*
  * XDR buffer helper functions
  */
 extern void xdr_shift_buf(struct xdr_buf *, size_t);