[PATCH] NFS: Replace nfs_page insertion sort with a radix sort

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

fs/nfs/inode.c

@@ -135,7 +135,7 @@ nfs_write_inode(struct inode *inode, int sync)
 	int flags = sync ? FLUSH_WAIT : 0;
 	int ret;
 
-	ret = nfs_commit_inode(inode, 0, 0, flags);
+	ret = nfs_commit_inode(inode, flags);
 	if (ret < 0)
 		return ret;
 	return 0;

fs/nfs/pagelist.c

@@ -177,36 +177,6 @@ nfs_release_request(struct nfs_page *req)
 	nfs_page_free(req);
 }
 
-/**
- * nfs_list_add_request - Insert a request into a sorted list
- * @req: request
- * @head: head of list into which to insert the request.
- *
- * Note that the wb_list is sorted by page index in order to facilitate
- * coalescing of requests.
- * We use an insertion sort that is optimized for the case of appended
- * writes.
- */
-void
-nfs_list_add_request(struct nfs_page *req, struct list_head *head)
-{
-	struct list_head *pos;
-
-#ifdef NFS_PARANOIA
-	if (!list_empty(&req->wb_list)) {
-		printk(KERN_ERR "NFS: Add to list failed!\n");
-		BUG();
-	}
-#endif
-	list_for_each_prev(pos, head) {
-		struct nfs_page	*p = nfs_list_entry(pos);
-		if (p->wb_index < req->wb_index)
-			break;
-	}
-	list_add(&req->wb_list, pos);
-	req->wb_list_head = head;
-}
-
 static int nfs_wait_bit_interruptible(void *word)
 {
 	int ret = 0;
@@ -291,6 +261,62 @@ nfs_coalesce_requests(struct list_head *head, struct list_head *dst,
 	return npages;
 }
 
+#define NFS_SCAN_MAXENTRIES 16
+/**
+ * nfs_scan_lock_dirty - Scan the radix tree for dirty requests
+ * @nfsi: NFS inode
+ * @dst: Destination list
+ * @idx_start: lower bound of page->index to scan
+ * @npages: idx_start + npages sets the upper bound to scan.
+ *
+ * Moves elements from one of the inode request lists.
+ * If the number of requests is set to 0, the entire address_space
+ * starting at index idx_start, is scanned.
+ * The requests are *not* checked to ensure that they form a contiguous set.
+ * You must be holding the inode's req_lock when calling this function
+ */
+int
+nfs_scan_lock_dirty(struct nfs_inode *nfsi, struct list_head *dst,
+	      unsigned long idx_start, unsigned int npages)
+{
+	struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
+	struct nfs_page *req;
+	unsigned long idx_end;
+	int found, i;
+	int res;
+
+	res = 0;
+	if (npages == 0)
+		idx_end = ~0;
+	else
+		idx_end = idx_start + npages - 1;
+
+	for (;;) {
+		found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
+				(void **)&pgvec[0], idx_start, NFS_SCAN_MAXENTRIES,
+				NFS_PAGE_TAG_DIRTY);
+		if (found <= 0)
+			break;
+		for (i = 0; i < found; i++) {
+			req = pgvec[i];
+			if (req->wb_index > idx_end)
+				goto out;
+
+			idx_start = req->wb_index + 1;
+
+			if (nfs_set_page_writeback_locked(req)) {
+				radix_tree_tag_clear(&nfsi->nfs_page_tree,
+						req->wb_index, NFS_PAGE_TAG_DIRTY);
+				nfs_list_remove_request(req);
+				nfs_list_add_request(req, dst);
+				res++;
+			}
+		}
+	}
+out:
+	return res;
+}
+
 /**
  * nfs_scan_list - Scan a list for matching requests
  * @head: One of the NFS inode request lists

fs/nfs/write.c

@@ -352,7 +352,7 @@ int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
 		if (err < 0)
 			goto out;
 	}
-	err = nfs_commit_inode(inode, 0, 0, wb_priority(wbc));
+	err = nfs_commit_inode(inode, wb_priority(wbc));
 	if (err > 0) {
 		wbc->nr_to_write -= err;
 		err = 0;
@@ -446,6 +446,8 @@ nfs_mark_request_dirty(struct nfs_page *req)
 	struct nfs_inode *nfsi = NFS_I(inode);
 
 	spin_lock(&nfsi->req_lock);
+	radix_tree_tag_set(&nfsi->nfs_page_tree,
+			req->wb_index, NFS_PAGE_TAG_DIRTY);
 	nfs_list_add_request(req, &nfsi->dirty);
 	nfsi->ndirty++;
 	spin_unlock(&nfsi->req_lock);
@@ -537,12 +539,15 @@ static int
 nfs_scan_dirty(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
-	int	res;
+	int res = 0;
-	res = nfs_scan_list(&nfsi->dirty, dst, idx_start, npages);
+
-	nfsi->ndirty -= res;
+	if (nfsi->ndirty != 0) {
-	sub_page_state(nr_dirty,res);
+		res = nfs_scan_lock_dirty(nfsi, dst, idx_start, npages);
-	if ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty))
+		nfsi->ndirty -= res;
-		printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n");
+		sub_page_state(nr_dirty,res);
+		if ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty))
+			printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n");
+	}
 	return res;
 }
 
@@ -561,11 +566,14 @@ static int
 nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
-	int	res;
+	int res = 0;
-	res = nfs_scan_list(&nfsi->commit, dst, idx_start, npages);
+
-	nfsi->ncommit -= res;
+	if (nfsi->ncommit != 0) {
-	if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit))
+		res = nfs_scan_list(&nfsi->commit, dst, idx_start, npages);
-		printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n");
+		nfsi->ncommit -= res;
+		if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit))
+			printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n");
+	}
 	return res;
 }
 #endif
@@ -1209,36 +1217,24 @@ 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;
+	struct nfs_page		*first;
 	struct inode		*inode;
-	loff_t			start, end, len;
 
 	/* Set up the RPC argument and reply structs
 	 * NB: take care not to mess about with data->commit et al. */
 
 	list_splice_init(head, &data->pages);
 	first = nfs_list_entry(data->pages.next);
-	last = nfs_list_entry(data->pages.prev);
 	inode = first->wb_context->dentry->d_inode;
 
-	/*
-	 * Determine the offset range of requests in the COMMIT call.
-	 * We rely on the fact that data->pages is an ordered list...
-	 */
-	start = req_offset(first);
-	end = req_offset(last) + last->wb_bytes;
-	len = end - start;
-	/* If 'len' is not a 32-bit quantity, pass '0' in the COMMIT call */
-	if (end >= i_size_read(inode) || len < 0 || len > (~((u32)0) >> 1))
-		len = 0;
-
 	data->inode	  = inode;
 	data->cred	  = first->wb_context->cred;
 
 	data->args.fh     = NFS_FH(data->inode);
-	data->args.offset = start;
+	/* Note: we always request a commit of the entire inode */
-	data->args.count  = len;
+	data->args.offset = 0;
-	data->res.count   = len;
+	data->args.count  = 0;
+	data->res.count   = 0;
 	data->res.fattr   = &data->fattr;
 	data->res.verf    = &data->verf;
 	
@@ -1357,8 +1353,7 @@ static int nfs_flush_inode(struct inode *inode, unsigned long idx_start,
 }
 
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
-int nfs_commit_inode(struct inode *inode, unsigned long idx_start,
+int nfs_commit_inode(struct inode *inode, int how)
-		    unsigned int npages, int how)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	LIST_HEAD(head);
@@ -1366,15 +1361,13 @@ int nfs_commit_inode(struct inode *inode, unsigned long idx_start,
 				error = 0;
 
 	spin_lock(&nfsi->req_lock);
-	res = nfs_scan_commit(inode, &head, idx_start, npages);
+	res = nfs_scan_commit(inode, &head, 0, 0);
+	spin_unlock(&nfsi->req_lock);
 	if (res) {
-		res += nfs_scan_commit(inode, &head, 0, 0);
-		spin_unlock(&nfsi->req_lock);
 		error = nfs_commit_list(&head, how);
-	} else
+		if (error < 0)
-		spin_unlock(&nfsi->req_lock);
+			return error;
-	if (error < 0)
+	}
-		return error;
 	return res;
 }
 #endif
@@ -1396,7 +1389,7 @@ int nfs_sync_inode(struct inode *inode, unsigned long idx_start,
 			error = nfs_flush_inode(inode, idx_start, npages, how);
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 		if (error == 0)
-			error = nfs_commit_inode(inode, idx_start, npages, how);
+			error = nfs_commit_inode(inode, how);
 #endif
 	} while (error > 0);
 	return error;

include/linux/nfs_fs.h

@@ -395,10 +395,10 @@ extern void nfs_commit_done(struct rpc_task *);
  */
 extern int  nfs_sync_inode(struct inode *, unsigned long, unsigned int, int);
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
-extern int  nfs_commit_inode(struct inode *, unsigned long, unsigned int, int);
+extern int  nfs_commit_inode(struct inode *, int);
 #else
 static inline int
-nfs_commit_inode(struct inode *inode, unsigned long idx_start, unsigned int npages, int how)
+nfs_commit_inode(struct inode *inode, int how)
 {
 	return 0;
 }

include/linux/nfs_page.h

@@ -22,6 +22,7 @@
 /*
  * Valid flags for the radix tree
  */
+#define NFS_PAGE_TAG_DIRTY	0
 #define NFS_PAGE_TAG_WRITEBACK	1
 
 /*
@@ -31,6 +32,7 @@
 #define PG_NEED_COMMIT		1
 #define PG_NEED_RESCHED		2
 
+struct nfs_inode;
 struct nfs_page {
 	struct list_head	wb_list,	/* Defines state of page: */
 				*wb_list_head;	/*      read/write/commit */
@@ -59,8 +61,8 @@ extern	void nfs_clear_request(struct nfs_page *req);
 extern	void nfs_release_request(struct nfs_page *req);
 
 
-extern	void nfs_list_add_request(struct nfs_page *, struct list_head *);
+extern  int nfs_scan_lock_dirty(struct nfs_inode *nfsi, struct list_head *dst,
-
+				unsigned long idx_start, unsigned int npages);
 extern	int nfs_scan_list(struct list_head *, struct list_head *,
 			  unsigned long, unsigned int);
 extern	int nfs_coalesce_requests(struct list_head *, struct list_head *,
@@ -94,6 +96,18 @@ nfs_lock_request(struct nfs_page *req)
 	return 1;
 }
 
+/**
+ * nfs_list_add_request - Insert a request into a list
+ * @req: request
+ * @head: head of list into which to insert the request.
+ */
+static inline void
+nfs_list_add_request(struct nfs_page *req, struct list_head *head)
+{
+	list_add_tail(&req->wb_list, head);
+	req->wb_list_head = head;
+}
+
 
 /**
  * nfs_list_remove_request - Remove a request from its wb_list