From: Trond Myklebust <trond.myklebust@fys.uio.no>

NFSv2/v3/v4: Patch by Chuck Lever that add support for asynchronous
writes even if wsize<PAGE_CACHE_SIZE.


---

 25-akpm/fs/nfs/direct.c         |    1 
 25-akpm/fs/nfs/nfs3proc.c       |   34 ---
 25-akpm/fs/nfs/nfs4proc.c       |   69 +------
 25-akpm/fs/nfs/nfs4xdr.c        |    7 
 25-akpm/fs/nfs/proc.c           |   21 --
 25-akpm/fs/nfs/write.c          |  364 +++++++++++++++++++++++++++++-----------
 25-akpm/include/linux/nfs_fs.h  |    2 
 25-akpm/include/linux/nfs_xdr.h |    8 
 8 files changed, 294 insertions(+), 212 deletions(-)

diff -puN fs/nfs/direct.c~nfs-03-small_wsize fs/nfs/direct.c
--- 25/fs/nfs/direct.c~nfs-03-small_wsize	2004-04-03 15:02:43.857923120 -0800
+++ 25-akpm/fs/nfs/direct.c	2004-04-03 15:02:43.877920080 -0800
@@ -259,6 +259,7 @@ nfs_direct_write_seg(struct inode *inode
 		.inode		= inode,
 		.args		= {
 			.fh		= NFS_FH(inode),
+			.lockowner	= current->files,
 		},
 		.res		= {
 			.fattr		= &wdata.fattr,
diff -puN fs/nfs/nfs3proc.c~nfs-03-small_wsize fs/nfs/nfs3proc.c
--- 25/fs/nfs/nfs3proc.c~nfs-03-small_wsize	2004-04-03 15:02:43.858922968 -0800
+++ 25-akpm/fs/nfs/nfs3proc.c	2004-04-03 15:02:43.878919928 -0800
@@ -763,11 +763,10 @@ nfs3_write_done(struct rpc_task *task)
 }
 
 static void
-nfs3_proc_write_setup(struct nfs_write_data *data, unsigned int count, int how)
+nfs3_proc_write_setup(struct nfs_write_data *data, int how)
 {
 	struct rpc_task		*task = &data->task;
 	struct inode		*inode = data->inode;
-	struct nfs_page		*req;
 	int			stable;
 	int			flags;
 	struct rpc_message	msg = {
@@ -784,28 +783,14 @@ nfs3_proc_write_setup(struct nfs_write_d
 			stable = NFS_DATA_SYNC;
 	} else
 		stable = NFS_UNSTABLE;
-	
-	req = nfs_list_entry(data->pages.next);
-	data->args.fh     = NFS_FH(inode);
-	data->args.offset = req_offset(req);
-	data->args.pgbase = req->wb_pgbase;
-	data->args.count  = count;
 	data->args.stable = stable;
-	data->args.pages  = data->pagevec;
-	data->res.fattr   = &data->fattr;
-	data->res.count   = count;
-	data->res.verf    = &data->verf;
 
 	/* Set the initial flags for the task.  */
 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
 
 	/* Finalize the task. */
 	rpc_init_task(task, NFS_CLIENT(inode), nfs3_write_done, flags);
-	task->tk_calldata = data;
-	/* Release requests */
-	task->tk_release = nfs_writedata_release;
-
-	rpc_call_setup(&data->task, &msg, 0);
+	rpc_call_setup(task, &msg, 0);
 }
 
 static void
@@ -822,7 +807,7 @@ nfs3_commit_done(struct rpc_task *task)
 }
 
 static void
-nfs3_proc_commit_setup(struct nfs_write_data *data, u64 start, u32 len, int how)
+nfs3_proc_commit_setup(struct nfs_write_data *data, int how)
 {
 	struct rpc_task		*task = &data->task;
 	struct inode		*inode = data->inode;
@@ -834,23 +819,12 @@ nfs3_proc_commit_setup(struct nfs_write_
 		.rpc_cred	= data->cred,
 	};
 
-	data->args.fh     = NFS_FH(data->inode);
-	data->args.offset = start;
-	data->args.count  = len;
-	data->res.count   = len;
-	data->res.fattr   = &data->fattr;
-	data->res.verf    = &data->verf;
-	
 	/* Set the initial flags for the task.  */
 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
 
 	/* Finalize the task. */
 	rpc_init_task(task, NFS_CLIENT(inode), nfs3_commit_done, flags);
-	task->tk_calldata = data;
-	/* Release requests */
-	task->tk_release = nfs_commit_release;
-	
-	rpc_call_setup(&data->task, &msg, 0);
+	rpc_call_setup(task, &msg, 0);
 }
 
 /*
diff -puN fs/nfs/nfs4proc.c~nfs-03-small_wsize fs/nfs/nfs4proc.c
--- 25/fs/nfs/nfs4proc.c~nfs-03-small_wsize	2004-04-03 15:02:43.860922664 -0800
+++ 25-akpm/fs/nfs/nfs4proc.c	2004-04-03 15:02:43.880919624 -0800
@@ -1134,10 +1134,10 @@ nfs4_proc_write(struct nfs_write_data *w
 	if (filp) {
 		struct nfs4_state *state;
 		state = (struct nfs4_state *)filp->private_data;
-		nfs4_copy_stateid(&wdata->args.stateid, state, wdata->lockowner);
+		wdata->args.state = state;
 		msg.rpc_cred = state->owner->so_cred;
 	} else {
-		memcpy(&wdata->args.stateid, &zero_stateid, sizeof(wdata->args.stateid));
+		wdata->args.state = NULL;
 		msg.rpc_cred = NFS_I(inode)->mm_cred;
 	}
 
@@ -1166,15 +1166,10 @@ nfs4_proc_commit(struct nfs_write_data *
 	/*
 	 * Try first to use O_WRONLY, then O_RDWR stateid.
 	 */
-	if (filp) {
-		struct nfs4_state *state;
-		state = (struct nfs4_state *)filp->private_data;
-		nfs4_copy_stateid(&cdata->args.stateid, state, cdata->lockowner);
-		msg.rpc_cred = state->owner->so_cred;
-	} else {
-		memcpy(&cdata->args.stateid, &zero_stateid, sizeof(cdata->args.stateid));
+	if (filp)
+		msg.rpc_cred = ((struct nfs4_state *)filp->private_data)->owner->so_cred;
+	else
 		msg.rpc_cred = NFS_I(inode)->mm_cred;
-	}
 
 	fattr->valid = 0;
 	status = rpc_call_sync(server->client, &msg, 0);
@@ -1549,27 +1544,13 @@ nfs4_proc_read_setup(struct nfs_read_dat
 }
 
 static void
-nfs4_restart_write(struct rpc_task *task)
-{
-	struct nfs_write_data *data = (struct nfs_write_data *)task->tk_calldata;
-	struct nfs_page *req;
-
-	rpc_restart_call(task);
-	req = nfs_list_entry(data->pages.next);
-	if (req->wb_state)
-		nfs4_copy_stateid(&data->args.stateid, req->wb_state, req->wb_lockowner);
-	else
-		memcpy(&data->args.stateid, &zero_stateid, sizeof(data->args.stateid));
-}
-
-static void
 nfs4_write_done(struct rpc_task *task)
 {
 	struct nfs_write_data *data = (struct nfs_write_data *) task->tk_calldata;
 	struct inode *inode = data->inode;
 	
 	if (nfs4_async_handle_error(task, NFS_SERVER(inode)) == -EAGAIN) {
-		task->tk_action = nfs4_restart_write;
+		rpc_restart_call(task);
 		return;
 	}
 	if (task->tk_status >= 0)
@@ -1579,7 +1560,7 @@ nfs4_write_done(struct rpc_task *task)
 }
 
 static void
-nfs4_proc_write_setup(struct nfs_write_data *data, unsigned int count, int how)
+nfs4_proc_write_setup(struct nfs_write_data *data, int how)
 {
 	struct rpc_task	*task = &data->task;
 	struct rpc_message msg = {
@@ -1589,7 +1570,6 @@ nfs4_proc_write_setup(struct nfs_write_d
 		.rpc_cred = data->cred,
 	};
 	struct inode *inode = data->inode;
-	struct nfs_page *req = nfs_list_entry(data->pages.next);
 	int stable;
 	int flags;
 	
@@ -1600,33 +1580,15 @@ nfs4_proc_write_setup(struct nfs_write_d
 			stable = NFS_DATA_SYNC;
 	} else
 		stable = NFS_UNSTABLE;
-
-	data->args.fh     = NFS_FH(inode);
-	data->args.offset = req_offset(req);
-	data->args.pgbase = req->wb_pgbase;
-	data->args.count  = count;
 	data->args.stable = stable;
-	data->args.pages  = data->pagevec;
-	data->res.fattr   = &data->fattr;
-	data->res.count   = count;
-	data->res.verf    = &data->verf;
-	data->timestamp   = jiffies;
 
-	data->lockowner = req->wb_lockowner;
-	if (req->wb_state)
-		nfs4_copy_stateid(&data->args.stateid, req->wb_state, req->wb_lockowner);
-	else
-		memcpy(&data->args.stateid, &zero_stateid, sizeof(data->args.stateid));
+	data->timestamp   = jiffies;
 
 	/* Set the initial flags for the task.  */
 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
 
 	/* Finalize the task. */
 	rpc_init_task(task, NFS_CLIENT(inode), nfs4_write_done, flags);
-	task->tk_calldata = data;
-	/* Release requests */
-	task->tk_release = nfs_writedata_release;
-
 	rpc_call_setup(task, &msg, 0);
 }
 
@@ -1637,7 +1599,7 @@ nfs4_commit_done(struct rpc_task *task)
 	struct inode *inode = data->inode;
 	
 	if (nfs4_async_handle_error(task, NFS_SERVER(inode)) == -EAGAIN) {
-		task->tk_action = nfs4_restart_write;
+		rpc_restart_call(task);
 		return;
 	}
 	/* Call back common NFS writeback processing */
@@ -1645,7 +1607,7 @@ nfs4_commit_done(struct rpc_task *task)
 }
 
 static void
-nfs4_proc_commit_setup(struct nfs_write_data *data, u64 start, u32 len, int how)
+nfs4_proc_commit_setup(struct nfs_write_data *data, int how)
 {
 	struct rpc_task	*task = &data->task;
 	struct rpc_message msg = {
@@ -1657,22 +1619,11 @@ nfs4_proc_commit_setup(struct nfs_write_
 	struct inode *inode = data->inode;
 	int flags;
 	
-	data->args.fh     = NFS_FH(data->inode);
-	data->args.offset = start;
-	data->args.count  = len;
-	data->res.count   = len;
-	data->res.fattr   = &data->fattr;
-	data->res.verf    = &data->verf;
-	
 	/* Set the initial flags for the task.  */
 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
 
 	/* Finalize the task. */
 	rpc_init_task(task, NFS_CLIENT(inode), nfs4_commit_done, flags);
-	task->tk_calldata = data;
-	/* Release requests */
-	task->tk_release = nfs_commit_release;
-	
 	rpc_call_setup(task, &msg, 0);	
 }
 
diff -puN fs/nfs/nfs4xdr.c~nfs-03-small_wsize fs/nfs/nfs4xdr.c
--- 25/fs/nfs/nfs4xdr.c~nfs-03-small_wsize	2004-04-03 15:02:43.862922360 -0800
+++ 25-akpm/fs/nfs/nfs4xdr.c	2004-04-03 15:02:43.882919320 -0800
@@ -1075,9 +1075,12 @@ encode_write(struct xdr_stream *xdr, str
 {
 	uint32_t *p;
 
-	RESERVE_SPACE(36);
+	RESERVE_SPACE(4);
 	WRITE32(OP_WRITE);
-	WRITEMEM(args->stateid.data, sizeof(args->stateid.data));
+
+	encode_stateid(xdr, args->state, args->lockowner);
+
+	RESERVE_SPACE(16);
 	WRITE64(args->offset);
 	WRITE32(args->stable);
 	WRITE32(args->count);
diff -puN fs/nfs/proc.c~nfs-03-small_wsize fs/nfs/proc.c
--- 25/fs/nfs/proc.c~nfs-03-small_wsize	2004-04-03 15:02:43.864922056 -0800
+++ 25-akpm/fs/nfs/proc.c	2004-04-03 15:02:43.883919168 -0800
@@ -590,11 +590,10 @@ nfs_write_done(struct rpc_task *task)
 }
 
 static void
-nfs_proc_write_setup(struct nfs_write_data *data, unsigned int count, int how)
+nfs_proc_write_setup(struct nfs_write_data *data, int how)
 {
 	struct rpc_task		*task = &data->task;
 	struct inode		*inode = data->inode;
-	struct nfs_page		*req;
 	int			flags;
 	struct rpc_message	msg = {
 		.rpc_proc	= &nfs_procedures[NFSPROC_WRITE],
@@ -604,32 +603,18 @@ nfs_proc_write_setup(struct nfs_write_da
 	};
 
 	/* Note: NFSv2 ignores @stable and always uses NFS_FILE_SYNC */
-	
-	req = nfs_list_entry(data->pages.next);
-	data->args.fh     = NFS_FH(inode);
-	data->args.offset = req_offset(req);
-	data->args.pgbase = req->wb_pgbase;
-	data->args.count  = count;
 	data->args.stable = NFS_FILE_SYNC;
-	data->args.pages  = data->pagevec;
-	data->res.fattr   = &data->fattr;
-	data->res.count   = count;
-	data->res.verf    = &data->verf;
 
 	/* Set the initial flags for the task.  */
 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
 
 	/* Finalize the task. */
 	rpc_init_task(task, NFS_CLIENT(inode), nfs_write_done, flags);
-	task->tk_calldata = data;
-	/* Release requests */
-	task->tk_release = nfs_writedata_release;
-
-	rpc_call_setup(&data->task, &msg, 0);
+	rpc_call_setup(task, &msg, 0);
 }
 
 static void
-nfs_proc_commit_setup(struct nfs_write_data *data, u64 start, u32 len, int how)
+nfs_proc_commit_setup(struct nfs_write_data *data, int how)
 {
 	BUG();
 }
diff -puN fs/nfs/write.c~nfs-03-small_wsize fs/nfs/write.c
--- 25/fs/nfs/write.c~nfs-03-small_wsize	2004-04-03 15:02:43.865921904 -0800
+++ 25-akpm/fs/nfs/write.c	2004-04-03 15:02:43.887918560 -0800
@@ -74,6 +74,8 @@
 static struct nfs_page * nfs_update_request(struct file*, struct inode *,
 					    struct page *,
 					    unsigned int, unsigned int);
+static void nfs_writeback_done_partial(struct nfs_write_data *, int);
+static void nfs_writeback_done_full(struct nfs_write_data *, int);
 
 static kmem_cache_t *nfs_wdata_cachep;
 static mempool_t *nfs_wdata_mempool;
@@ -95,7 +97,7 @@ static __inline__ void nfs_writedata_fre
 	mempool_free(p, nfs_wdata_mempool);
 }
 
-void nfs_writedata_release(struct rpc_task *task)
+static void nfs_writedata_release(struct rpc_task *task)
 {
 	struct nfs_write_data	*wdata = (struct nfs_write_data *)task->tk_calldata;
 	nfs_writedata_free(wdata);
@@ -117,12 +119,6 @@ static __inline__ void nfs_commit_free(s
 	mempool_free(p, nfs_commit_mempool);
 }
 
-void nfs_commit_release(struct rpc_task *task)
-{
-	struct nfs_write_data	*wdata = (struct nfs_write_data *)task->tk_calldata;
-	nfs_commit_free(wdata);
-}
-
 /* Adjust the file length if we're writing beyond the end */
 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
 {
@@ -186,6 +182,7 @@ nfs_writepage_sync(struct file *file, st
 		.inode		= inode,
 		.args		= {
 			.fh		= NFS_FH(inode),
+			.lockowner	= current->files,
 			.pages		= &page,
 			.stable		= NFS_FILE_SYNC,
 			.pgbase		= offset,
@@ -299,8 +296,7 @@ nfs_writepage(struct page *page, struct 
 		goto out;
 do_it:
 	lock_kernel();
-	if (NFS_SERVER(inode)->wsize >= PAGE_CACHE_SIZE && !IS_SYNC(inode) &&
-			inode_referenced) {
+	if (!IS_SYNC(inode) && inode_referenced) {
 		err = nfs_writepage_async(NULL, inode, page, 0, offset);
 		if (err >= 0)
 			err = 0;
@@ -365,7 +361,7 @@ nfs_inode_add_request(struct inode *inod
 /*
  * Insert a write request into an inode
  */
-static inline void
+static void
 nfs_inode_remove_request(struct nfs_page *req)
 {
 	struct nfs_inode *nfsi;
@@ -416,7 +412,7 @@ nfs_find_request(struct inode *inode, un
 /*
  * Add a request to the inode's dirty list.
  */
-static inline void
+static void
 nfs_mark_request_dirty(struct nfs_page *req)
 {
 	struct inode *inode = req->wb_inode;
@@ -444,7 +440,7 @@ nfs_dirty_request(struct nfs_page *req)
 /*
  * Add a request to the inode's commit list.
  */
-static inline void
+static void
 nfs_mark_request_commit(struct nfs_page *req)
 {
 	struct inode *inode = req->wb_inode;
@@ -680,11 +676,7 @@ nfs_updatepage(struct file *file, struct
 		dentry->d_parent->d_name.name, dentry->d_name.name,
 		count, (long long)(page_offset(page) +offset));
 
-	/*
-	 * If wsize is smaller than page size, update and write
-	 * page synchronously.
-	 */
-	if (NFS_SERVER(inode)->wsize < PAGE_CACHE_SIZE || IS_SYNC(inode)) {
+	if (IS_SYNC(inode)) {
 		status = nfs_writepage_sync(file, inode, page, offset, count);
 		if (status > 0) {
 			if (offset == 0 && status == PAGE_CACHE_SIZE)
@@ -747,43 +739,146 @@ done:
 	return status;
 }
 
+static void nfs_writepage_release(struct nfs_page *req)
+{
+	end_page_writeback(req->wb_page);
+
+#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
+	if (!PageError(req->wb_page)) {
+		if (NFS_NEED_RESCHED(req)) {
+			nfs_mark_request_dirty(req);
+			goto out;
+		} else if (NFS_NEED_COMMIT(req)) {
+			nfs_mark_request_commit(req);
+			goto out;
+		}
+	}
+	nfs_inode_remove_request(req);
+
+out:
+	nfs_clear_commit(req);
+	nfs_clear_reschedule(req);
+#else
+	nfs_inode_remove_request(req);
+#endif
+	nfs_unlock_request(req);
+}
+
 /*
  * Set up the argument/result storage required for the RPC call.
  */
-static void
-nfs_write_rpcsetup(struct list_head *head, struct nfs_write_data *data, int how)
+static void nfs_write_rpcsetup(struct nfs_page *req,
+		struct nfs_write_data *data,
+		unsigned int count, unsigned int offset,
+		int how)
 {
 	struct rpc_task		*task = &data->task;
 	struct inode		*inode;
-	struct nfs_page		*req;
-	struct page		**pages;
-	unsigned int		count;
 
 	/* Set up the RPC argument and reply structs
 	 * NB: take care not to mess about with data->commit et al. */
 
-	pages = data->pagevec;
-	count = 0;
-	while (!list_empty(head)) {
-		req = nfs_list_entry(head->next);
-		nfs_list_remove_request(req);
-		nfs_list_add_request(req, &data->pages);
-		set_page_writeback(req->wb_page);
-		*pages++ = req->wb_page;
-		count += req->wb_bytes;
-	}
-	req = nfs_list_entry(data->pages.next);
+	data->req = req;
 	data->inode = inode = req->wb_inode;
 	data->cred = req->wb_cred;
 
-	NFS_PROTO(inode)->write_setup(data, count, how);
+	data->args.fh     = NFS_FH(inode);
+	data->args.offset = req_offset(req) + offset;
+	data->args.pgbase = req->wb_pgbase + offset;
+	data->args.count  = count;
+	data->args.pages  = data->pagevec;
+	data->res.fattr   = &data->fattr;
+	data->res.count   = count;
+	data->res.verf    = &data->verf;
+
+	NFS_PROTO(inode)->write_setup(data, how);
+
+	data->task.tk_calldata = data;
+	/* Release requests */
+	data->task.tk_release = nfs_writedata_release;
 
 	dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
 		task->tk_pid,
 		inode->i_sb->s_id,
 		(long long)NFS_FILEID(inode),
 		count,
-		(unsigned long long)req_offset(req));
+		data->args.offset);
+}
+
+static void nfs_execute_write(struct nfs_write_data *data)
+{
+	struct rpc_clnt *clnt = NFS_CLIENT(data->inode);
+	sigset_t oldset;
+
+	rpc_clnt_sigmask(clnt, &oldset);
+	lock_kernel();
+	rpc_execute(&data->task);
+	unlock_kernel();
+	rpc_clnt_sigunmask(clnt, &oldset);
+}
+
+/*
+ * Generate multiple small requests to write out a single
+ * contiguous dirty area on one page.
+ */
+static int nfs_flush_multi(struct list_head *head, struct inode *inode, int how)
+{
+	struct nfs_page *req = nfs_list_entry(head->next);
+	struct page *page = req->wb_page;
+	struct nfs_write_data *data;
+	unsigned int wsize = NFS_SERVER(inode)->wsize;
+	unsigned int nbytes, offset;
+	int requests = 0;
+	LIST_HEAD(list);
+
+	nfs_list_remove_request(req);
+
+	nbytes = req->wb_bytes;
+	for (;;) {
+		data = nfs_writedata_alloc();
+		if (!data)
+			goto out_bad;
+		list_add(&data->pages, &list);
+		requests++;
+		if (nbytes <= wsize)
+			break;
+		nbytes -= wsize;
+	}
+	atomic_set(&req->wb_complete, requests);
+
+	ClearPageError(page);
+	set_page_writeback(page);
+	offset = 0;
+	nbytes = req->wb_bytes;
+	do {
+		data = list_entry(list.next, struct nfs_write_data, pages);
+		list_del_init(&data->pages);
+
+		data->pagevec[0] = page;
+		data->complete = nfs_writeback_done_partial;
+
+		if (nbytes > wsize) {
+			nfs_write_rpcsetup(req, data, wsize, offset, how);
+			offset += wsize;
+			nbytes -= wsize;
+		} else {
+			nfs_write_rpcsetup(req, data, nbytes, offset, how);
+			nbytes = 0;
+		}
+		nfs_execute_write(data);
+	} while (nbytes != 0);
+
+	return 0;
+
+out_bad:
+	while (!list_empty(&list)) {
+		data = list_entry(list.next, struct nfs_write_data, pages);
+		list_del(&data->pages);
+		nfs_writedata_free(data);
+	}
+	nfs_mark_request_dirty(req);
+	nfs_unlock_request(req);
+	return -ENOMEM;
 }
 
 /*
@@ -794,25 +889,38 @@ nfs_write_rpcsetup(struct list_head *hea
  * This is the case if nfs_updatepage detects a conflicting request
  * that has been written but not committed.
  */
-static int
-nfs_flush_one(struct list_head *head, struct inode *inode, int how)
+static int nfs_flush_one(struct list_head *head, struct inode *inode, int how)
 {
-	struct rpc_clnt 	*clnt = NFS_CLIENT(inode);
+	struct nfs_page		*req;
+	struct page		**pages;
 	struct nfs_write_data	*data;
-	sigset_t		oldset;
+	unsigned int		count;
+
+	if (NFS_SERVER(inode)->wsize < PAGE_CACHE_SIZE)
+		return nfs_flush_multi(head, inode, how);
 
 	data = nfs_writedata_alloc();
 	if (!data)
 		goto out_bad;
 
+	pages = data->pagevec;
+	count = 0;
+	while (!list_empty(head)) {
+		req = nfs_list_entry(head->next);
+		nfs_list_remove_request(req);
+		nfs_list_add_request(req, &data->pages);
+		ClearPageError(req->wb_page);
+		set_page_writeback(req->wb_page);
+		*pages++ = req->wb_page;
+		count += req->wb_bytes;
+	}
+	req = nfs_list_entry(data->pages.next);
+
+	data->complete = nfs_writeback_done_full;
 	/* Set up the argument struct */
-	nfs_write_rpcsetup(head, data, how);
+	nfs_write_rpcsetup(req, data, count, 0, how);
 
-	rpc_clnt_sigmask(clnt, &oldset);
-	lock_kernel();
-	rpc_execute(&data->task);
-	unlock_kernel();
-	rpc_clnt_sigunmask(clnt, &oldset);
+	nfs_execute_write(data);
 	return 0;
  out_bad:
 	while (!list_empty(head)) {
@@ -851,18 +959,107 @@ nfs_flush_list(struct list_head *head, i
 	return error;
 }
 
+/*
+ * Handle a write reply that flushed part of a page.
+ */
+static void nfs_writeback_done_partial(struct nfs_write_data *data, int status)
+{
+	struct nfs_page		*req = data->req;
+	struct page		*page = req->wb_page;
+
+	dprintk("NFS: write (%s/%Ld %d@%Ld)",
+		req->wb_inode->i_sb->s_id,
+		(long long)NFS_FILEID(req->wb_inode),
+		req->wb_bytes,
+		(long long)req_offset(req));
+
+	if (status < 0) {
+		ClearPageUptodate(page);
+		SetPageError(page);
+		if (req->wb_file)
+			req->wb_file->f_error = status;
+		dprintk(", error = %d\n", status);
+	} else {
+#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
+		if (data->verf.committed < NFS_FILE_SYNC) {
+			if (!NFS_NEED_COMMIT(req)) {
+				nfs_defer_commit(req);
+				memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
+				dprintk(" defer commit\n");
+			} else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
+				nfs_defer_reschedule(req);
+				dprintk(" server reboot detected\n");
+			}
+		} else
+#endif
+			dprintk(" OK\n");
+	}
+
+	if (atomic_dec_and_test(&req->wb_complete))
+		nfs_writepage_release(req);
+}
+
+/*
+ * Handle a write reply that flushes a whole page.
+ *
+ * FIXME: There is an inherent race with invalidate_inode_pages and
+ *	  writebacks since the page->count is kept > 1 for as long
+ *	  as the page has a write request pending.
+ */
+static void nfs_writeback_done_full(struct nfs_write_data *data, int status)
+{
+	struct nfs_page		*req;
+	struct page		*page;
+
+	/* Update attributes as result of writeback. */
+	while (!list_empty(&data->pages)) {
+		req = nfs_list_entry(data->pages.next);
+		nfs_list_remove_request(req);
+		page = req->wb_page;
+
+		dprintk("NFS: write (%s/%Ld %d@%Ld)",
+			req->wb_inode->i_sb->s_id,
+			(long long)NFS_FILEID(req->wb_inode),
+			req->wb_bytes,
+			(long long)req_offset(req));
+
+		if (status < 0) {
+			ClearPageUptodate(page);
+			SetPageError(page);
+			if (req->wb_file)
+				req->wb_file->f_error = status;
+			end_page_writeback(page);
+			nfs_inode_remove_request(req);
+			dprintk(", error = %d\n", status);
+			goto next;
+		}
+		end_page_writeback(page);
+
+#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
+		if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) {
+			nfs_inode_remove_request(req);
+			dprintk(" OK\n");
+			goto next;
+		}
+		memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
+		nfs_mark_request_commit(req);
+		dprintk(" marked for commit\n");
+#else
+		nfs_inode_remove_request(req);
+#endif
+	next:
+		nfs_unlock_request(req);
+	}
+}
 
 /*
  * This function is called when the WRITE call is complete.
  */
-void
-nfs_writeback_done(struct rpc_task *task)
+void nfs_writeback_done(struct rpc_task *task)
 {
 	struct nfs_write_data	*data = (struct nfs_write_data *) task->tk_calldata;
 	struct nfs_writeargs	*argp = &data->args;
 	struct nfs_writeres	*resp = &data->res;
-	struct nfs_page		*req;
-	struct page		*page;
 
 	dprintk("NFS: %4d nfs_writeback_done (status %d)\n",
 		task->tk_pid, task->tk_status);
@@ -904,53 +1101,22 @@ nfs_writeback_done(struct rpc_task *task
 	/*
 	 * Process the nfs_page list
 	 */
-	while (!list_empty(&data->pages)) {
-		req = nfs_list_entry(data->pages.next);
-		nfs_list_remove_request(req);
-		page = req->wb_page;
-
-		dprintk("NFS: write (%s/%Ld %d@%Ld)",
-			req->wb_inode->i_sb->s_id,
-			(long long)NFS_FILEID(req->wb_inode),
-			req->wb_bytes,
-			(long long)req_offset(req));
+	data->complete(data, task->tk_status);
+}
 
-		if (task->tk_status < 0) {
-			ClearPageUptodate(page);
-			SetPageError(page);
-			if (req->wb_file)
-				req->wb_file->f_error = task->tk_status;
-			end_page_writeback(page);
-			nfs_inode_remove_request(req);
-			dprintk(", error = %d\n", task->tk_status);
-			goto next;
-		}
-		end_page_writeback(page);
 
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
-		if (argp->stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) {
-			nfs_inode_remove_request(req);
-			dprintk(" OK\n");
-			goto next;
-		}
-		memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
-		nfs_mark_request_commit(req);
-		dprintk(" marked for commit\n");
-#else
-		nfs_inode_remove_request(req);
-#endif
-	next:
-		nfs_unlock_request(req);
-	}
+static void nfs_commit_release(struct rpc_task *task)
+{
+	struct nfs_write_data	*wdata = (struct nfs_write_data *)task->tk_calldata;
+	nfs_commit_free(wdata);
 }
 
-
-#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 /*
  * Set up the argument/result storage required for the RPC call.
  */
-static void
-nfs_commit_rpcsetup(struct list_head *head, struct nfs_write_data *data, int how)
+static void nfs_commit_rpcsetup(struct list_head *head,
+		struct nfs_write_data *data, int how)
 {
 	struct rpc_task		*task = &data->task;
 	struct nfs_page		*first, *last;
@@ -979,7 +1145,18 @@ nfs_commit_rpcsetup(struct list_head *he
 	data->inode	  = inode;
 	data->cred	  = first->wb_cred;
 
-	NFS_PROTO(inode)->commit_setup(data, start, len, how);
+	data->args.fh     = NFS_FH(data->inode);
+	data->args.offset = start;
+	data->args.count  = len;
+	data->res.count   = len;
+	data->res.fattr   = &data->fattr;
+	data->res.verf    = &data->verf;
+
+	NFS_PROTO(inode)->commit_setup(data, how);
+
+	data->task.tk_calldata = data;
+	/* Release requests */
+	data->task.tk_release = nfs_commit_release;
 	
 	dprintk("NFS: %4d initiated commit call\n", task->tk_pid);
 }
@@ -990,10 +1167,8 @@ nfs_commit_rpcsetup(struct list_head *he
 int
 nfs_commit_list(struct list_head *head, int how)
 {
-	struct rpc_clnt		*clnt;
 	struct nfs_write_data	*data;
 	struct nfs_page         *req;
-	sigset_t		oldset;
 
 	data = nfs_commit_alloc();
 
@@ -1002,13 +1177,8 @@ nfs_commit_list(struct list_head *head, 
 
 	/* Set up the argument struct */
 	nfs_commit_rpcsetup(head, data, how);
-	clnt = NFS_CLIENT(data->inode);
 
-	rpc_clnt_sigmask(clnt, &oldset);
-	lock_kernel();
-	rpc_execute(&data->task);
-	unlock_kernel();
-	rpc_clnt_sigunmask(clnt, &oldset);
+	nfs_execute_write(data);
 	return 0;
  out_bad:
 	while (!list_empty(head)) {
diff -puN include/linux/nfs_fs.h~nfs-03-small_wsize include/linux/nfs_fs.h
--- 25/include/linux/nfs_fs.h~nfs-03-small_wsize	2004-04-03 15:02:43.867921600 -0800
+++ 25-akpm/include/linux/nfs_fs.h	2004-04-03 15:02:43.887918560 -0800
@@ -335,10 +335,8 @@ extern int  nfs_writepages(struct addres
 extern int  nfs_flush_incompatible(struct file *file, struct page *page);
 extern int  nfs_updatepage(struct file *, struct page *, unsigned int, unsigned int);
 extern void nfs_writeback_done(struct rpc_task *task);
-extern void nfs_writedata_release(struct rpc_task *task);
 
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
-extern void nfs_commit_release(struct rpc_task *task);
 extern void nfs_commit_done(struct rpc_task *);
 #endif
 
diff -puN include/linux/nfs_xdr.h~nfs-03-small_wsize include/linux/nfs_xdr.h
--- 25/include/linux/nfs_xdr.h~nfs-03-small_wsize	2004-04-03 15:02:43.868921448 -0800
+++ 25-akpm/include/linux/nfs_xdr.h	2004-04-03 15:02:43.888918408 -0800
@@ -253,7 +253,8 @@ struct nfs_readres {
 
 struct nfs_writeargs {
 	struct nfs_fh *		fh;
-	nfs4_stateid		stateid;
+	fl_owner_t		lockowner;
+	struct nfs4_state *	state;
 	__u64			offset;
 	__u32			count;
 	enum nfs3_stable_how	stable;
@@ -681,7 +682,6 @@ struct nfs_write_data {
 	struct rpc_task		task;
 	struct inode		*inode;
 	struct rpc_cred		*cred;
-	fl_owner_t		lockowner;
 	struct nfs_fattr	fattr;
 	struct nfs_writeverf	verf;
 	struct list_head	pages;		/* Coalesced requests we wish to flush */
@@ -742,8 +742,8 @@ struct nfs_rpc_ops {
 			     struct nfs_pathconf *);
 	u32 *	(*decode_dirent)(u32 *, struct nfs_entry *, int plus);
 	void	(*read_setup)   (struct nfs_read_data *);
-	void	(*write_setup)  (struct nfs_write_data *, unsigned int count, int how);
-	void	(*commit_setup) (struct nfs_write_data *, u64 start, u32 len, int how);
+	void	(*write_setup)  (struct nfs_write_data *, int how);
+	void	(*commit_setup) (struct nfs_write_data *, int how);
 	int	(*file_open)   (struct inode *, struct file *);
 	int	(*file_release) (struct inode *, struct file *);
 	void	(*request_init)(struct nfs_page *, struct file *);

_