Merge rsync://rsync.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6

9 files changed, 341 insertions, 338 deletions

diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index 3ddda6f7ecc..e7ffb4deb3e 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -690,7 +690,9 @@ int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
 			goto out_force;
 		/* This is an open(2) */
 		if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
-				!(server->flags & NFS_MOUNT_NOCTO))
+				!(server->flags & NFS_MOUNT_NOCTO) &&
+				(S_ISREG(inode->i_mode) ||
+				 S_ISDIR(inode->i_mode)))
 			goto out_force;
 	}
 	return nfs_revalidate_inode(server, inode);
diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index 4cdd1b499e3..76ca1cbc38f 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -67,25 +67,19 @@ struct nfs_direct_req {
 	struct kref		kref;		/* release manager */
 
 	/* I/O parameters */
-	struct list_head	list,		/* nfs_read/write_data structs */
-				rewrite_list;	/* saved nfs_write_data structs */
 	struct nfs_open_context	*ctx;		/* file open context info */
 	struct kiocb *		iocb;		/* controlling i/o request */
 	struct inode *		inode;		/* target file of i/o */
-	unsigned long		user_addr;	/* location of user's buffer */
-	size_t			user_count;	/* total bytes to move */
-	loff_t			pos;		/* starting offset in file */
-	struct page **		pages;		/* pages in our buffer */
-	unsigned int		npages;		/* count of pages */
 
 	/* completion state */
+	atomic_t		io_count;	/* i/os we're waiting for */
 	spinlock_t		lock;		/* protect completion state */
-	int			outstanding;	/* i/os we're waiting for */
 	ssize_t			count,		/* bytes actually processed */
 				error;		/* any reported error */
 	struct completion	completion;	/* wait for i/o completion */
 
 	/* commit state */
+	struct list_head	rewrite_list;	/* saved nfs_write_data structs */
 	struct nfs_write_data *	commit_data;	/* special write_data for commits */
 	int			flags;
 #define NFS_ODIRECT_DO_COMMIT		(1)	/* an unstable reply was received */
@@ -93,8 +87,18 @@ struct nfs_direct_req {
 	struct nfs_writeverf	verf;		/* unstable write verifier */
 };
 
-static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync);
 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
+static const struct rpc_call_ops nfs_write_direct_ops;
+
+static inline void get_dreq(struct nfs_direct_req *dreq)
+{
+	atomic_inc(&dreq->io_count);
+}
+
+static inline int put_dreq(struct nfs_direct_req *dreq)
+{
+	return atomic_dec_and_test(&dreq->io_count);
+}
 
 /**
  * nfs_direct_IO - NFS address space operation for direct I/O
@@ -118,50 +122,21 @@ ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_
 	return -EINVAL;
 }
 
-static void nfs_free_user_pages(struct page **pages, int npages, int do_dirty)
+static void nfs_direct_dirty_pages(struct page **pages, int npages)
 {
 	int i;
 	for (i = 0; i < npages; i++) {
 		struct page *page = pages[i];
-		if (do_dirty && !PageCompound(page))
+		if (!PageCompound(page))
 			set_page_dirty_lock(page);
-		page_cache_release(page);
 	}
-	kfree(pages);
 }
 
-static inline int nfs_get_user_pages(int rw, unsigned long user_addr, size_t size, struct page ***pages)
+static void nfs_direct_release_pages(struct page **pages, int npages)
 {
-	int result = -ENOMEM;
-	unsigned long page_count;
-	size_t array_size;
-
-	page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
-	page_count -= user_addr >> PAGE_SHIFT;
-
-	array_size = (page_count * sizeof(struct page *));
-	*pages = kmalloc(array_size, GFP_KERNEL);
-	if (*pages) {
-		down_read(&current->mm->mmap_sem);
-		result = get_user_pages(current, current->mm, user_addr,
-					page_count, (rw == READ), 0,
-					*pages, NULL);
-		up_read(&current->mm->mmap_sem);
-		if (result != page_count) {
-			/*
-			 * If we got fewer pages than expected from
-			 * get_user_pages(), the user buffer runs off the
-			 * end of a mapping; return EFAULT.
-			 */
-			if (result >= 0) {
-				nfs_free_user_pages(*pages, result, 0);
-				result = -EFAULT;
-			} else
-				kfree(*pages);
-			*pages = NULL;
-		}
-	}
-	return result;
+	int i;
+	for (i = 0; i < npages; i++)
+		page_cache_release(pages[i]);
 }
 
 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
@@ -173,13 +148,13 @@ static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
 		return NULL;
 
 	kref_init(&dreq->kref);
+	kref_get(&dreq->kref);
 	init_completion(&dreq->completion);
-	INIT_LIST_HEAD(&dreq->list);
 	INIT_LIST_HEAD(&dreq->rewrite_list);
 	dreq->iocb = NULL;
 	dreq->ctx = NULL;
 	spin_lock_init(&dreq->lock);
-	dreq->outstanding = 0;
+	atomic_set(&dreq->io_count, 0);
 	dreq->count = 0;
 	dreq->error = 0;
 	dreq->flags = 0;
@@ -220,18 +195,11 @@ out:
 }
 
 /*
- * We must hold a reference to all the pages in this direct read request
- * until the RPCs complete.  This could be long *after* we are woken up in
- * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
- *
- * In addition, synchronous I/O uses a stack-allocated iocb.  Thus we
- * can't trust the iocb is still valid here if this is a synchronous
- * request.  If the waiter is woken prematurely, the iocb is long gone.
+ * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
+ * the iocb is still valid here if this is a synchronous request.
  */
 static void nfs_direct_complete(struct nfs_direct_req *dreq)
 {
-	nfs_free_user_pages(dreq->pages, dreq->npages, 1);
-
 	if (dreq->iocb) {
 		long res = (long) dreq->error;
 		if (!res)
@@ -244,48 +212,10 @@ static void nfs_direct_complete(struct nfs_direct_req *dreq)
 }
 
 /*
- * Note we also set the number of requests we have in the dreq when we are
- * done.  This prevents races with I/O completion so we will always wait
- * until all requests have been dispatched and completed.
+ * We must hold a reference to all the pages in this direct read request
+ * until the RPCs complete.  This could be long *after* we are woken up in
+ * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
  */
-static struct nfs_direct_req *nfs_direct_read_alloc(size_t nbytes, size_t rsize)
-{
-	struct list_head *list;
-	struct nfs_direct_req *dreq;
-	unsigned int rpages = (rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-
-	dreq = nfs_direct_req_alloc();
-	if (!dreq)
-		return NULL;
-
-	list = &dreq->list;
-	for(;;) {
-		struct nfs_read_data *data = nfs_readdata_alloc(rpages);
-
-		if (unlikely(!data)) {
-			while (!list_empty(list)) {
-				data = list_entry(list->next,
-						  struct nfs_read_data, pages);
-				list_del(&data->pages);
-				nfs_readdata_free(data);
-			}
-			kref_put(&dreq->kref, nfs_direct_req_release);
-			return NULL;
-		}
-
-		INIT_LIST_HEAD(&data->pages);
-		list_add(&data->pages, list);
-
-		data->req = (struct nfs_page *) dreq;
-		dreq->outstanding++;
-		if (nbytes <= rsize)
-			break;
-		nbytes -= rsize;
-	}
-	kref_get(&dreq->kref);
-	return dreq;
-}
-
 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 {
 	struct nfs_read_data *data = calldata;
@@ -294,6 +224,9 @@ static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 	if (nfs_readpage_result(task, data) != 0)
 		return;
 
+	nfs_direct_dirty_pages(data->pagevec, data->npages);
+	nfs_direct_release_pages(data->pagevec, data->npages);
+
 	spin_lock(&dreq->lock);
 
 	if (likely(task->tk_status >= 0))
@@ -301,13 +234,10 @@ static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 	else
 		dreq->error = task->tk_status;
 
-	if (--dreq->outstanding) {
-		spin_unlock(&dreq->lock);
-		return;
-	}
-
 	spin_unlock(&dreq->lock);
-	nfs_direct_complete(dreq);
+
+	if (put_dreq(dreq))
+		nfs_direct_complete(dreq);
 }
 
 static const struct rpc_call_ops nfs_read_direct_ops = {
@@ -316,41 +246,56 @@ static const struct rpc_call_ops nfs_read_direct_ops = {
 };
 
 /*
- * For each nfs_read_data struct that was allocated on the list, dispatch
- * an NFS READ operation
+ * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
+ * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
+ * bail and stop sending more reads.  Read length accounting is
+ * handled automatically by nfs_direct_read_result().  Otherwise, if
+ * no requests have been sent, just return an error.
  */
-static void nfs_direct_read_schedule(struct nfs_direct_req *dreq)
+static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos)
 {
 	struct nfs_open_context *ctx = dreq->ctx;
 	struct inode *inode = ctx->dentry->d_inode;
-	struct list_head *list = &dreq->list;
-	struct page **pages = dreq->pages;
-	size_t count = dreq->user_count;
-	loff_t pos = dreq->pos;
 	size_t rsize = NFS_SERVER(inode)->rsize;
-	unsigned int curpage, pgbase;
+	unsigned int pgbase;
+	int result;
+	ssize_t started = 0;
+
+	get_dreq(dreq);
 
-	curpage = 0;
-	pgbase = dreq->user_addr & ~PAGE_MASK;
 	do {
 		struct nfs_read_data *data;
 		size_t bytes;
 
-		bytes = rsize;
-		if (count < rsize)
-			bytes = count;
+		pgbase = user_addr & ~PAGE_MASK;
+		bytes = min(rsize,count);
 
-		BUG_ON(list_empty(list));
-		data = list_entry(list->next, struct nfs_read_data, pages);
-		list_del_init(&data->pages);
+		result = -ENOMEM;
+		data = nfs_readdata_alloc(pgbase + bytes);
+		if (unlikely(!data))
+			break;
+
+		down_read(&current->mm->mmap_sem);
+		result = get_user_pages(current, current->mm, user_addr,
+					data->npages, 1, 0, data->pagevec, NULL);
+		up_read(&current->mm->mmap_sem);
+		if (unlikely(result < data->npages)) {
+			if (result > 0)
+				nfs_direct_release_pages(data->pagevec, result);
+			nfs_readdata_release(data);
+			break;
+		}
+
+		get_dreq(dreq);
 
+		data->req = (struct nfs_page *) dreq;
 		data->inode = inode;
 		data->cred = ctx->cred;
 		data->args.fh = NFS_FH(inode);
 		data->args.context = ctx;
 		data->args.offset = pos;
 		data->args.pgbase = pgbase;
-		data->args.pages = &pages[curpage];
+		data->args.pages = data->pagevec;
 		data->args.count = bytes;
 		data->res.fattr = &data->fattr;
 		data->res.eof = 0;
@@ -373,33 +318,37 @@ static void nfs_direct_read_schedule(struct nfs_direct_req *dreq)
 				bytes,
 				(unsigned long long)data->args.offset);
 
+		started += bytes;
+		user_addr += bytes;
 		pos += bytes;
+		/* FIXME: Remove this unnecessary math from final patch */
 		pgbase += bytes;
-		curpage += pgbase >> PAGE_SHIFT;
 		pgbase &= ~PAGE_MASK;
+		BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 
 		count -= bytes;
 	} while (count != 0);
-	BUG_ON(!list_empty(list));
+
+	if (put_dreq(dreq))
+		nfs_direct_complete(dreq);
+
+	if (started)
+		return 0;
+	return result < 0 ? (ssize_t) result : -EFAULT;
 }
 
-static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, unsigned int nr_pages)
+static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
 {
-	ssize_t result;
+	ssize_t result = 0;
 	sigset_t oldset;
 	struct inode *inode = iocb->ki_filp->f_mapping->host;
 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
 	struct nfs_direct_req *dreq;
 
-	dreq = nfs_direct_read_alloc(count, NFS_SERVER(inode)->rsize);
+	dreq = nfs_direct_req_alloc();
 	if (!dreq)
 		return -ENOMEM;
 
-	dreq->user_addr = user_addr;
-	dreq->user_count = count;
-	dreq->pos = pos;
-	dreq->pages = pages;
-	dreq->npages = nr_pages;
 	dreq->inode = inode;
 	dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
 	if (!is_sync_kiocb(iocb))
@@ -407,8 +356,9 @@ static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size
 
 	nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
 	rpc_clnt_sigmask(clnt, &oldset);
-	nfs_direct_read_schedule(dreq);
-	result = nfs_direct_wait(dreq);
+	result = nfs_direct_read_schedule(dreq, user_addr, count, pos);
+	if (!result)
+		result = nfs_direct_wait(dreq);
 	rpc_clnt_sigunmask(clnt, &oldset);
 
 	return result;
@@ -416,10 +366,10 @@ static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size
 
 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 {
-	list_splice_init(&dreq->rewrite_list, &dreq->list);
-	while (!list_empty(&dreq->list)) {
-		struct nfs_write_data *data = list_entry(dreq->list.next, struct nfs_write_data, pages);
+	while (!list_empty(&dreq->rewrite_list)) {
+		struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
 		list_del(&data->pages);
+		nfs_direct_release_pages(data->pagevec, data->npages);
 		nfs_writedata_release(data);
 	}
 }
@@ -427,14 +377,51 @@ static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 {
-	struct list_head *pos;
+	struct inode *inode = dreq->inode;
+	struct list_head *p;
+	struct nfs_write_data *data;
 
-	list_splice_init(&dreq->rewrite_list, &dreq->list);
-	list_for_each(pos, &dreq->list)
-		dreq->outstanding++;
 	dreq->count = 0;
+	get_dreq(dreq);
+
+	list_for_each(p, &dreq->rewrite_list) {
+		data = list_entry(p, struct nfs_write_data, pages);
+
+		get_dreq(dreq);
 
-	nfs_direct_write_schedule(dreq, FLUSH_STABLE);
+		/*
+		 * Reset data->res.
+		 */
+		nfs_fattr_init(&data->fattr);
+		data->res.count = data->args.count;
+		memset(&data->verf, 0, sizeof(data->verf));
+
+		/*
+		 * Reuse data->task; data->args should not have changed
+		 * since the original request was sent.
+		 */
+		rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
+				&nfs_write_direct_ops, data);
+		NFS_PROTO(inode)->write_setup(data, FLUSH_STABLE);
+
+		data->task.tk_priority = RPC_PRIORITY_NORMAL;
+		data->task.tk_cookie = (unsigned long) inode;
+
+		/*
+		 * We're called via an RPC callback, so BKL is already held.
+		 */
+		rpc_execute(&data->task);
+
+		dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
+				data->task.tk_pid,
+				inode->i_sb->s_id,
+				(long long)NFS_FILEID(inode),
+				data->args.count,
+				(unsigned long long)data->args.offset);
+	}
+
+	if (put_dreq(dreq))
+		nfs_direct_write_complete(dreq, inode);
 }
 
 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
@@ -471,8 +458,8 @@ static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
 	data->cred = dreq->ctx->cred;
 
 	data->args.fh = NFS_FH(data->inode);
-	data->args.offset = dreq->pos;
-	data->args.count = dreq->user_count;
+	data->args.offset = 0;
+	data->args.count = 0;
 	data->res.count = 0;
 	data->res.fattr = &data->fattr;
 	data->res.verf = &data->verf;
@@ -516,7 +503,7 @@ static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode
 
 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 {
-	dreq->commit_data = nfs_commit_alloc(0);
+	dreq->commit_data = nfs_commit_alloc();
 	if (dreq->commit_data != NULL)
 		dreq->commit_data->req = (struct nfs_page *) dreq;
 }
@@ -534,47 +521,6 @@ static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode
 }
 #endif
 
-static struct nfs_direct_req *nfs_direct_write_alloc(size_t nbytes, size_t wsize)
-{
-	struct list_head *list;
-	struct nfs_direct_req *dreq;
-	unsigned int wpages = (wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-
-	dreq = nfs_direct_req_alloc();
-	if (!dreq)
-		return NULL;
-
-	list = &dreq->list;
-	for(;;) {
-		struct nfs_write_data *data = nfs_writedata_alloc(wpages);
-
-		if (unlikely(!data)) {
-			while (!list_empty(list)) {
-				data = list_entry(list->next,
-						  struct nfs_write_data, pages);
-				list_del(&data->pages);
-				nfs_writedata_free(data);
-			}
-			kref_put(&dreq->kref, nfs_direct_req_release);
-			return NULL;
-		}
-
-		INIT_LIST_HEAD(&data->pages);
-		list_add(&data->pages, list);
-
-		data->req = (struct nfs_page *) dreq;
-		dreq->outstanding++;
-		if (nbytes <= wsize)
-			break;
-		nbytes -= wsize;
-	}
-
-	nfs_alloc_commit_data(dreq);
-
-	kref_get(&dreq->kref);
-	return dreq;
-}
-
 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 {
 	struct nfs_write_data *data = calldata;
@@ -604,8 +550,6 @@ static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 				}
 		}
 	}
-	/* In case we have to resend */
-	data->args.stable = NFS_FILE_SYNC;
 
 	spin_unlock(&dreq->lock);
 }
@@ -619,14 +563,8 @@ static void nfs_direct_write_release(void *calldata)
 	struct nfs_write_data *data = calldata;
 	struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 
-	spin_lock(&dreq->lock);
-	if (--dreq->outstanding) {
-		spin_unlock(&dreq->lock);
-		return;
-	}
-	spin_unlock(&dreq->lock);
-
-	nfs_direct_write_complete(dreq, data->inode);
+	if (put_dreq(dreq))
+		nfs_direct_write_complete(dreq, data->inode);
 }
 
 static const struct rpc_call_ops nfs_write_direct_ops = {
@@ -635,41 +573,58 @@ static const struct rpc_call_ops nfs_write_direct_ops = {
 };
 
 /*
- * For each nfs_write_data struct that was allocated on the list, dispatch
- * an NFS WRITE operation
+ * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
+ * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
+ * bail and stop sending more writes.  Write length accounting is
+ * handled automatically by nfs_direct_write_result().  Otherwise, if
+ * no requests have been sent, just return an error.
  */
-static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync)
+static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos, int sync)
 {
 	struct nfs_open_context *ctx = dreq->ctx;
 	struct inode *inode = ctx->dentry->d_inode;
-	struct list_head *list = &dreq->list;
-	struct page **pages = dreq->pages;
-	size_t count = dreq->user_count;
-	loff_t pos = dreq->pos;
 	size_t wsize = NFS_SERVER(inode)->wsize;
-	unsigned int curpage, pgbase;
+	unsigned int pgbase;
+	int result;
+	ssize_t started = 0;
+
+	get_dreq(dreq);
 
-	curpage = 0;
-	pgbase = dreq->user_addr & ~PAGE_MASK;
 	do {
 		struct nfs_write_data *data;
 		size_t bytes;
 
-		bytes = wsize;
-		if (count < wsize)
-			bytes = count;
+		pgbase = user_addr & ~PAGE_MASK;
+		bytes = min(wsize,count);
+
+		result = -ENOMEM;
+		data = nfs_writedata_alloc(pgbase + bytes);
+		if (unlikely(!data))
+			break;
+
+		down_read(&current->mm->mmap_sem);
+		result = get_user_pages(current, current->mm, user_addr,
+					data->npages, 0, 0, data->pagevec, NULL);
+		up_read(&current->mm->mmap_sem);
+		if (unlikely(result < data->npages)) {
+			if (result > 0)
+				nfs_direct_release_pages(data->pagevec, result);
+			nfs_writedata_release(data);
+			break;
+		}
+
+		get_dreq(dreq);
 
-		BUG_ON(list_empty(list));
-		data = list_entry(list->next, struct nfs_write_data, pages);
 		list_move_tail(&data->pages, &dreq->rewrite_list);
 
+		data->req = (struct nfs_page *) dreq;
 		data->inode = inode;
 		data->cred = ctx->cred;
 		data->args.fh = NFS_FH(inode);
 		data->args.context = ctx;
 		data->args.offset = pos;
 		data->args.pgbase = pgbase;
-		data->args.pages = &pages[curpage];
+		data->args.pages = data->pagevec;
 		data->args.count = bytes;
 		data->res.fattr = &data->fattr;
 		data->res.count = bytes;
@@ -693,19 +648,29 @@ static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync)
 				bytes,
 				(unsigned long long)data->args.offset);
 
+		started += bytes;
+		user_addr += bytes;
 		pos += bytes;
+
+		/* FIXME: Remove this useless math from the final patch */
 		pgbase += bytes;
-		curpage += pgbase >> PAGE_SHIFT;
 		pgbase &= ~PAGE_MASK;
+		BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 
 		count -= bytes;
 	} while (count != 0);
-	BUG_ON(!list_empty(list));
+
+	if (put_dreq(dreq))
+		nfs_direct_write_complete(dreq, inode);
+
+	if (started)
+		return 0;
+	return result < 0 ? (ssize_t) result : -EFAULT;
 }
 
-static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, int nr_pages)
+static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
 {
-	ssize_t result;
+	ssize_t result = 0;
 	sigset_t oldset;
 	struct inode *inode = iocb->ki_filp->f_mapping->host;
 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
@@ -713,17 +678,14 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
 	size_t wsize = NFS_SERVER(inode)->wsize;
 	int sync = 0;
 
-	dreq = nfs_direct_write_alloc(count, wsize);
+	dreq = nfs_direct_req_alloc();
 	if (!dreq)
 		return -ENOMEM;
+	nfs_alloc_commit_data(dreq);
+
 	if (dreq->commit_data == NULL || count < wsize)
 		sync = FLUSH_STABLE;
 
-	dreq->user_addr = user_addr;
-	dreq->user_count = count;
-	dreq->pos = pos;
-	dreq->pages = pages;
-	dreq->npages = nr_pages;
 	dreq->inode = inode;
 	dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
 	if (!is_sync_kiocb(iocb))
@@ -734,8 +696,9 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
 	nfs_begin_data_update(inode);
 
 	rpc_clnt_sigmask(clnt, &oldset);
-	nfs_direct_write_schedule(dreq, sync);
-	result = nfs_direct_wait(dreq);
+	result = nfs_direct_write_schedule(dreq, user_addr, count, pos, sync);
+	if (!result)
+		result = nfs_direct_wait(dreq);
 	rpc_clnt_sigunmask(clnt, &oldset);
 
 	return result;
@@ -765,8 +728,6 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
 ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
 {
 	ssize_t retval = -EINVAL;
-	int page_count;
-	struct page **pages;
 	struct file *file = iocb->ki_filp;
 	struct address_space *mapping = file->f_mapping;
 
@@ -788,14 +749,7 @@ ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count,
 	if (retval)
 		goto out;
 
-	retval = nfs_get_user_pages(READ, (unsigned long) buf,
-						count, &pages);
-	if (retval < 0)
-		goto out;
-	page_count = retval;
-
-	retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos,
-						pages, page_count);
+	retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos);
 	if (retval > 0)
 		iocb->ki_pos = pos + retval;
 
@@ -831,8 +785,6 @@ out:
 ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 {
 	ssize_t retval;
-	int page_count;
-	struct page **pages;
 	struct file *file = iocb->ki_filp;
 	struct address_space *mapping = file->f_mapping;
 
@@ -860,14 +812,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t
 	if (retval)
 		goto out;
 
-	retval = nfs_get_user_pages(WRITE, (unsigned long) buf,
-						count, &pages);
-	if (retval < 0)
-		goto out;
-	page_count = retval;
-
-	retval = nfs_direct_write(iocb, (unsigned long) buf, count,
-					pos, pages, page_count);
+	retval = nfs_direct_write(iocb, (unsigned long) buf, count, pos);
 
 	/*
 	 * XXX: nfs_end_data_update() already ensures this file's
diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index cc2b874ad5a..48e892880d5 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -312,7 +312,13 @@ static void nfs_invalidate_page(struct page *page, unsigned long offset)
 
 static int nfs_release_page(struct page *page, gfp_t gfp)
 {
-	return !nfs_wb_page(page->mapping->host, page);
+	if (gfp & __GFP_FS)
+		return !nfs_wb_page(page->mapping->host, page);
+	else
+		/*
+		 * Avoid deadlock on nfs_wait_on_request().
+		 */
+		return 0;
 }
 
 const struct address_space_operations nfs_file_aops = {
diff --git a/fs/nfs/idmap.c b/fs/nfs/idmap.c
index b81e7ed3c90..07a5dd57646 100644
--- a/fs/nfs/idmap.c
+++ b/fs/nfs/idmap.c
@@ -130,9 +130,7 @@ nfs_idmap_delete(struct nfs4_client *clp)
 
 	if (!idmap)
 		return;
-	dput(idmap->idmap_dentry);
-	idmap->idmap_dentry = NULL;
-	rpc_unlink(idmap->idmap_path);
+	rpc_unlink(idmap->idmap_dentry);
 	clp->cl_idmap = NULL;
 	kfree(idmap);
 }
diff --git a/fs/nfs/namespace.c b/fs/nfs/namespace.c
index 19b98ca468e..86b3169c8ca 100644
--- a/fs/nfs/namespace.c
+++ b/fs/nfs/namespace.c
@@ -51,7 +51,7 @@ char *nfs_path(const char *base, const struct dentry *dentry,
 		namelen = dentry->d_name.len;
 		buflen -= namelen + 1;
 		if (buflen < 0)
-			goto Elong;
+			goto Elong_unlock;
 		end -= namelen;
 		memcpy(end, dentry->d_name.name, namelen);
 		*--end = '/';
@@ -68,6 +68,8 @@ char *nfs_path(const char *base, const struct dentry *dentry,
 	end -= namelen;
 	memcpy(end, base, namelen);
 	return end;
+Elong_unlock:
+	spin_unlock(&dcache_lock);
 Elong:
 	return ERR_PTR(-ENAMETOOLONG);
 }
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index b4916b09219..153898e1331 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -2668,7 +2668,7 @@ out:
 	nfs4_set_cached_acl(inode, acl);
 }
 
-static inline ssize_t nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
+static ssize_t __nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
 {
 	struct page *pages[NFS4ACL_MAXPAGES];
 	struct nfs_getaclargs args = {
@@ -2721,6 +2721,19 @@ out_free:
 	return ret;
 }
 
+static ssize_t nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
+{
+	struct nfs4_exception exception = { };
+	ssize_t ret;
+	do {
+		ret = __nfs4_get_acl_uncached(inode, buf, buflen);
+		if (ret >= 0)
+			break;
+		ret = nfs4_handle_exception(NFS_SERVER(inode), ret, &exception);
+	} while (exception.retry);
+	return ret;
+}
+
 static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
 {
 	struct nfs_server *server = NFS_SERVER(inode);
@@ -2737,7 +2750,7 @@ static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
 	return nfs4_get_acl_uncached(inode, buf, buflen);
 }
 
-static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen)
+static int __nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen)
 {
 	struct nfs_server *server = NFS_SERVER(inode);
 	struct page *pages[NFS4ACL_MAXPAGES];
@@ -2763,6 +2776,18 @@ static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen
 	return ret;
 }
 
+static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen)
+{
+	struct nfs4_exception exception = { };
+	int err;
+	do {
+		err = nfs4_handle_exception(NFS_SERVER(inode),
+				__nfs4_proc_set_acl(inode, buf, buflen),
+				&exception);
+	} while (exception.retry);
+	return err;
+}
+
 static int
 nfs4_async_handle_error(struct rpc_task *task, const struct nfs_server *server)
 {
@@ -3144,9 +3169,6 @@ static int do_vfs_lock(struct file *file, struct file_lock *fl)
 		default:
 			BUG();
 	}
-	if (res < 0)
-		printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
-				__FUNCTION__);
 	return res;
 }
 
@@ -3258,8 +3280,6 @@ static struct rpc_task *nfs4_do_unlck(struct file_lock *fl,
 		return ERR_PTR(-ENOMEM);
 	}
 
-	/* Unlock _before_ we do the RPC call */
-	do_vfs_lock(fl->fl_file, fl);
 	return rpc_run_task(NFS_CLIENT(lsp->ls_state->inode), RPC_TASK_ASYNC, &nfs4_locku_ops, data);
 }
 
@@ -3270,30 +3290,28 @@ static int nfs4_proc_unlck(struct nfs4_state *state, int cmd, struct file_lock *
 	struct rpc_task *task;
 	int status = 0;
 
-	/* Is this a delegated lock? */
-	if (test_bit(NFS_DELEGATED_STATE, &state->flags))
-		goto out_unlock;
-	/* Is this open_owner holding any locks on the server? */
-	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
-		goto out_unlock;
-
 	status = nfs4_set_lock_state(state, request);
+	/* Unlock _before_ we do the RPC call */
+	request->fl_flags |= FL_EXISTS;
+	if (do_vfs_lock(request->fl_file, request) == -ENOENT)
+		goto out;
 	if (status != 0)
-		goto out_unlock;
+		goto out;
+	/* Is this a delegated lock? */
+	if (test_bit(NFS_DELEGATED_STATE, &state->flags))
+		goto out;
 	lsp = request->fl_u.nfs4_fl.owner;
-	status = -ENOMEM;
 	seqid = nfs_alloc_seqid(&lsp->ls_seqid);
+	status = -ENOMEM;
 	if (seqid == NULL)
-		goto out_unlock;
+		goto out;
 	task = nfs4_do_unlck(request, request->fl_file->private_data, lsp, seqid);
 	status = PTR_ERR(task);
 	if (IS_ERR(task))
-		goto out_unlock;
+		goto out;
 	status = nfs4_wait_for_completion_rpc_task(task);
 	rpc_release_task(task);
-	return status;
-out_unlock:
-	do_vfs_lock(request->fl_file, request);
+out:
 	return status;
 }
 
@@ -3461,10 +3479,10 @@ static int nfs4_lock_reclaim(struct nfs4_state *state, struct file_lock *request
 	struct nfs4_exception exception = { };
 	int err;
 
-	/* Cache the lock if possible... */
-	if (test_bit(NFS_DELEGATED_STATE, &state->flags))
-		return 0;
 	do {
+		/* Cache the lock if possible... */
+		if (test_bit(NFS_DELEGATED_STATE, &state->flags) != 0)
+			return 0;
 		err = _nfs4_do_setlk(state, F_SETLK, request, 1);
 		if (err != -NFS4ERR_DELAY)
 			break;
@@ -3483,6 +3501,8 @@ static int nfs4_lock_expired(struct nfs4_state *state, struct file_lock *request
 	if (err != 0)
 		return err;
 	do {
+		if (test_bit(NFS_DELEGATED_STATE, &state->flags) != 0)
+			return 0;
 		err = _nfs4_do_setlk(state, F_SETLK, request, 0);
 		if (err != -NFS4ERR_DELAY)
 			break;
@@ -3494,29 +3514,42 @@ static int nfs4_lock_expired(struct nfs4_state *state, struct file_lock *request
 static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock *request)
 {
 	struct nfs4_client *clp = state->owner->so_client;
+	unsigned char fl_flags = request->fl_flags;
 	int status;
 
 	/* Is this a delegated open? */
-	if (NFS_I(state->inode)->delegation_state != 0) {
-		/* Yes: cache locks! */
-		status = do_vfs_lock(request->fl_file, request);
-		/* ...but avoid races with delegation recall... */
-		if (status < 0 || test_bit(NFS_DELEGATED_STATE, &state->flags))
-			return status;
-	}
-	down_read(&clp->cl_sem);
 	status = nfs4_set_lock_state(state, request);
 	if (status != 0)
 		goto out;
+	request->fl_flags |= FL_ACCESS;
+	status = do_vfs_lock(request->fl_file, request);
+	if (status < 0)
+		goto out;
+	down_read(&clp->cl_sem);
+	if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
+		struct nfs_inode *nfsi = NFS_I(state->inode);
+		/* Yes: cache locks! */
+		down_read(&nfsi->rwsem);
+		/* ...but avoid races with delegation recall... */
+		if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
+			request->fl_flags = fl_flags & ~FL_SLEEP;
+			status = do_vfs_lock(request->fl_file, request);
+			up_read(&nfsi->rwsem);
+			goto out_unlock;
+		}
+		up_read(&nfsi->rwsem);
+	}
 	status = _nfs4_do_setlk(state, cmd, request, 0);
 	if (status != 0)
-		goto out;
+		goto out_unlock;
 	/* Note: we always want to sleep here! */
-	request->fl_flags |= FL_SLEEP;
+	request->fl_flags = fl_flags | FL_SLEEP;
 	if (do_vfs_lock(request->fl_file, request) < 0)
 		printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n", __FUNCTION__);
-out:
+out_unlock:
 	up_read(&clp->cl_sem);
+out:
+	request->fl_flags = fl_flags;
 	return status;
 }
 
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 1750d996f49..730ec8fb31c 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -3355,7 +3355,7 @@ static int decode_readdir(struct xdr_stream *xdr, struct rpc_rqst *req, struct n
 	struct kvec	*iov = rcvbuf->head;
 	unsigned int	nr, pglen = rcvbuf->page_len;
 	uint32_t	*end, *entry, *p, *kaddr;
-	uint32_t	len, attrlen;
+	uint32_t	len, attrlen, xlen;
 	int 		hdrlen, recvd, status;
 
 	status = decode_op_hdr(xdr, OP_READDIR);
@@ -3377,10 +3377,10 @@ static int decode_readdir(struct xdr_stream *xdr, struct rpc_rqst *req, struct n
 
 	BUG_ON(pglen + readdir->pgbase > PAGE_CACHE_SIZE);
 	kaddr = p = (uint32_t *) kmap_atomic(page, KM_USER0);
-	end = (uint32_t *) ((char *)p + pglen + readdir->pgbase);
+	end = p + ((pglen + readdir->pgbase) >> 2);
 	entry = p;
 	for (nr = 0; *p++; nr++) {
-		if (p + 3 > end)
+		if (end - p < 3)
 			goto short_pkt;
 		dprintk("cookie = %Lu, ", *((unsigned long long *)p));
 		p += 2;			/* cookie */
@@ -3389,18 +3389,19 @@ static int decode_readdir(struct xdr_stream *xdr, struct rpc_rqst *req, struct n
 			printk(KERN_WARNING "NFS: giant filename in readdir (len 0x%x)\n", len);
 			goto err_unmap;
 		}
-		dprintk("filename = %*s\n", len, (char *)p);
-		p += XDR_QUADLEN(len);
-		if (p + 1 > end)
+		xlen = XDR_QUADLEN(len);
+		if (end - p < xlen + 1)
 			goto short_pkt;
+		dprintk("filename = %*s\n", len, (char *)p);
+		p += xlen;
 		len = ntohl(*p++);	/* bitmap length */
-		p += len;
-		if (p + 1 > end)
+		if (end - p < len + 1)
 			goto short_pkt;
+		p += len;
 		attrlen = XDR_QUADLEN(ntohl(*p++));
-		p += attrlen;		/* attributes */
-		if (p + 2 > end)
+		if (end - p < attrlen + 2)
 			goto short_pkt;
+		p += attrlen;		/* attributes */
 		entry = p;
 	}
 	if (!nr && (entry[0] != 0 || entry[1] == 0))
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index 52bf634260a..7a9ee00e0c6 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -43,13 +43,15 @@ static mempool_t *nfs_rdata_mempool;
 
 #define MIN_POOL_READ	(32)
 
-struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
+struct nfs_read_data *nfs_readdata_alloc(size_t len)
 {
+	unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 	struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, SLAB_NOFS);
 
 	if (p) {
 		memset(p, 0, sizeof(*p));
 		INIT_LIST_HEAD(&p->pages);
+		p->npages = pagecount;
 		if (pagecount <= ARRAY_SIZE(p->page_array))
 			p->pagevec = p->page_array;
 		else {
@@ -63,7 +65,7 @@ struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
 	return p;
 }
 
-void nfs_readdata_free(struct nfs_read_data *p)
+static void nfs_readdata_free(struct nfs_read_data *p)
 {
 	if (p && (p->pagevec != &p->page_array[0]))
 		kfree(p->pagevec);
@@ -116,10 +118,17 @@ static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
 	pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
 	base &= ~PAGE_CACHE_MASK;
 	pglen = PAGE_CACHE_SIZE - base;
-	if (pglen < remainder)
+	for (;;) {
+		if (remainder <= pglen) {
+			memclear_highpage_flush(*pages, base, remainder);
+			break;
+		}
 		memclear_highpage_flush(*pages, base, pglen);
-	else
-		memclear_highpage_flush(*pages, base, remainder);
+		pages++;
+		remainder -= pglen;
+		pglen = PAGE_CACHE_SIZE;
+		base = 0;
+	}
 }
 
 /*
@@ -133,7 +142,7 @@ static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode,
 	int		result;
 	struct nfs_read_data *rdata;
 
-	rdata = nfs_readdata_alloc(1);
+	rdata = nfs_readdata_alloc(count);
 	if (!rdata)
 		return -ENOMEM;
 
@@ -329,25 +338,25 @@ static int nfs_pagein_multi(struct list_head *head, struct inode *inode)
 	struct nfs_page *req = nfs_list_entry(head->next);
 	struct page *page = req->wb_page;
 	struct nfs_read_data *data;
-	unsigned int rsize = NFS_SERVER(inode)->rsize;
-	unsigned int nbytes, offset;
+	size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
+	unsigned int offset;
 	int requests = 0;
 	LIST_HEAD(list);
 
 	nfs_list_remove_request(req);
 
 	nbytes = req->wb_bytes;
-	for(;;) {
-		data = nfs_readdata_alloc(1);
+	do {
+		size_t len = min(nbytes,rsize);
+
+		data = nfs_readdata_alloc(len);
 		if (!data)
 			goto out_bad;
 		INIT_LIST_HEAD(&data->pages);
 		list_add(&data->pages, &list);
 		requests++;
-		if (nbytes <= rsize)
-			break;
-		nbytes -= rsize;
-	}
+		nbytes -= len;
+	} while(nbytes != 0);
 	atomic_set(&req->wb_complete, requests);
 
 	ClearPageError(page);
@@ -395,7 +404,7 @@ static int nfs_pagein_one(struct list_head *head, struct inode *inode)
 	if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
 		return nfs_pagein_multi(head, inode);
 
-	data = nfs_readdata_alloc(NFS_SERVER(inode)->rpages);
+	data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize);
 	if (!data)
 		goto out_bad;
 
@@ -476,6 +485,8 @@ static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
 	unsigned int base = data->args.pgbase;
 	struct page **pages;
 
+	if (data->res.eof)
+		count = data->args.count;
 	if (unlikely(count == 0))
 		return;
 	pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
@@ -483,11 +494,7 @@ static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
 	count += base;
 	for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
 		SetPageUptodate(*pages);
-	/*
-	 * Was this an eof or a short read? If the latter, don't mark the page
-	 * as uptodate yet.
-	 */
-	if (count > 0 && (data->res.eof || data->args.count == data->res.count))
+	if (count != 0)
 		SetPageUptodate(*pages);
 }
 
@@ -502,6 +509,8 @@ static void nfs_readpage_set_pages_error(struct nfs_read_data *data)
 	count += base;
 	for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
 		SetPageError(*pages);
+	if (count != 0)
+		SetPageError(*pages);
 }
 
 /*
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index bca5734ca9f..8ab3cf10d79 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -90,22 +90,13 @@ static mempool_t *nfs_commit_mempool;
 
 static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion);
 
-struct nfs_write_data *nfs_commit_alloc(unsigned int pagecount)
+struct nfs_write_data *nfs_commit_alloc(void)
 {
 	struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS);
 
 	if (p) {
 		memset(p, 0, sizeof(*p));
 		INIT_LIST_HEAD(&p->pages);
-		if (pagecount <= ARRAY_SIZE(p->page_array))
-			p->pagevec = p->page_array;
-		else {
-			p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
-			if (!p->pagevec) {
-				mempool_free(p, nfs_commit_mempool);
-				p = NULL;
-			}
-		}
 	}
 	return p;
 }
@@ -117,13 +108,15 @@ void nfs_commit_free(struct nfs_write_data *p)
 	mempool_free(p, nfs_commit_mempool);
 }
 
-struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
+struct nfs_write_data *nfs_writedata_alloc(size_t len)
 {
+	unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 	struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS);
 
 	if (p) {
 		memset(p, 0, sizeof(*p));
 		INIT_LIST_HEAD(&p->pages);
+		p->npages = pagecount;
 		if (pagecount <= ARRAY_SIZE(p->page_array))
 			p->pagevec = p->page_array;
 		else {
@@ -137,7 +130,7 @@ struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
 	return p;
 }
 
-void nfs_writedata_free(struct nfs_write_data *p)
+static void nfs_writedata_free(struct nfs_write_data *p)
 {
 	if (p && (p->pagevec != &p->page_array[0]))
 		kfree(p->pagevec);
@@ -208,7 +201,7 @@ static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode,
 	int		result, written = 0;
 	struct nfs_write_data *wdata;
 
-	wdata = nfs_writedata_alloc(1);
+	wdata = nfs_writedata_alloc(wsize);
 	if (!wdata)
 		return -ENOMEM;
 
@@ -578,7 +571,7 @@ static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, un
 	return ret;
 }
 
-static void nfs_cancel_requests(struct list_head *head)
+static void nfs_cancel_dirty_list(struct list_head *head)
 {
 	struct nfs_page *req;
 	while(!list_empty(head)) {
@@ -589,6 +582,19 @@ static void nfs_cancel_requests(struct list_head *head)
 	}
 }
 
+static void nfs_cancel_commit_list(struct list_head *head)
+{
+	struct nfs_page *req;
+
+	while(!list_empty(head)) {
+		req = nfs_list_entry(head->next);
+		nfs_list_remove_request(req);
+		nfs_inode_remove_request(req);
+		nfs_clear_page_writeback(req);
+		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
+	}
+}
+
 /*
  * nfs_scan_dirty - Scan an inode for dirty requests
  * @inode: NFS inode to scan
@@ -986,24 +992,24 @@ static int nfs_flush_multi(struct inode *inode, struct list_head *head, 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;
+	size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
+	unsigned int offset;
 	int requests = 0;
 	LIST_HEAD(list);
 
 	nfs_list_remove_request(req);
 
 	nbytes = req->wb_bytes;
-	for (;;) {
-		data = nfs_writedata_alloc(1);
+	do {
+		size_t len = min(nbytes, wsize);
+
+		data = nfs_writedata_alloc(len);
 		if (!data)
 			goto out_bad;
 		list_add(&data->pages, &list);
 		requests++;
-		if (nbytes <= wsize)
-			break;
-		nbytes -= wsize;
-	}
+		nbytes -= len;
+	} while (nbytes != 0);
 	atomic_set(&req->wb_complete, requests);
 
 	ClearPageError(page);
@@ -1057,7 +1063,7 @@ static int nfs_flush_one(struct inode *inode, struct list_head *head, int how)
 	struct nfs_write_data	*data;
 	unsigned int		count;
 
-	data = nfs_writedata_alloc(NFS_SERVER(inode)->wpages);
+	data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize);
 	if (!data)
 		goto out_bad;
 
@@ -1365,7 +1371,7 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how)
 	struct nfs_write_data	*data;
 	struct nfs_page         *req;
 
-	data = nfs_commit_alloc(NFS_SERVER(inode)->wpages);
+	data = nfs_commit_alloc();
 
 	if (!data)
 		goto out_bad;
@@ -1381,6 +1387,7 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how)
 		nfs_list_remove_request(req);
 		nfs_mark_request_commit(req);
 		nfs_clear_page_writeback(req);
+		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
 	}
 	return -ENOMEM;
 }
@@ -1499,7 +1506,7 @@ int nfs_sync_inode_wait(struct inode *inode, unsigned long idx_start,
 		if (pages != 0) {
 			spin_unlock(&nfsi->req_lock);
 			if (how & FLUSH_INVALIDATE)
-				nfs_cancel_requests(&head);
+				nfs_cancel_dirty_list(&head);
 			else
 				ret = nfs_flush_list(inode, &head, pages, how);
 			spin_lock(&nfsi->req_lock);
@@ -1512,7 +1519,7 @@ int nfs_sync_inode_wait(struct inode *inode, unsigned long idx_start,
 			break;
 		if (how & FLUSH_INVALIDATE) {
 			spin_unlock(&nfsi->req_lock);
-			nfs_cancel_requests(&head);
+			nfs_cancel_commit_list(&head);
 			spin_lock(&nfsi->req_lock);
 			continue;
 		}